TUMOR MICROBIOME · 1/1/2020 · Cancer Epidemiol Biomarkers Prev Oct 1, 2019 28:10 1687–93; doi: 10.1158/1055-9965.EPI-19-0008 Chemoprevention of Colorectal Cancer by Anthocyanidins

Complimentary access to articles online:

aacrjournals.org/hot-topics

TUMOR MICROBIOMERecent Articles Published in the AACR Journals

Global Imaging CRO SolutionsPreclinical | Phase 0-IV | Imaging Core Lab | Image Analytics

Pathology Services | Radiochemistry | Data Management

Oncology

Neurology

Systemic Diseases

Rare Diseases

+1 617-904-2100

www.invicro.com

[email protected]

THE WAY TO ANSWERS:Operational Excellence & Scientific Expertise

Cross-Journal Collection:Tumor Microbiome

Table of Contents

Alterations to the Esophageal Microbiome Associated with Progression from Barrett's Esophagus toEsophageal AdenocarcinomaErik J. Snider, Griselda Compres, Daniel E. Freedberg, Hossein Khiabanian, Yael R. Nobel, Stephania Stump, Anne-Catrin Uhlemann,Charles J. Lightdale, and Julian A. AbramsCancer Epidemiol Biomarkers Prev Oct 1, 2019 28:10 1687–93; doi: 10.1158/1055-9965.EPI-19-0008

Chemoprevention of Colorectal Cancer by Anthocyanidins and Mitigation of Metabolic Shifts Inducedby Dysbiosis of the Gut MicrobiomeAshley M. Mudd, Tao Gu, Radha Munagala, Jeyaprakash Jeyabalan, Nejat K. Egilmez, and Ramesh C. GuptaCancer Prev Res Jan 1, 2020 13:1 41–52; doi: 10.1158/1940-6207.CAPR-19-0362

Gastrointestinal Tract Dysbiosis Enhances Distal Tumor Progression through Suppression of LeukocyteTraffickingSamir V. Jenkins, Michael S. Robeson II, Robert J. Griffin, Charles M. Quick, Eric R. Siegel, Martin J. Cannon, Kieng B. Vang, andRuud P.M. DingsCancer Res Dec 1, 2019 79:23 5999–6009; doi: 10.1158/0008-5472.CAN-18-4108

The Microbiome in Lung Cancer Tissue and Recurrence-Free SurvivalBrandilyn A. Peters, Richard B. Hayes, Chandra Goparaju, Christopher Reid, Harvey I. Pass, and Jiyoung AhnCancer Epidemiol Biomarkers Prev Apr 1, 2019 28:4 731–740; doi: 10.1158/1055-9965.EPI-18-0966

Microbiota- and Radiotherapy-Induced Gastrointestinal Side-Effects (MARS) Study: A Large Pilot Studyof the Microbiome in Acute and Late-Radiation EnteropathyMiguel Reis Ferreira, H. Jervoise N. Andreyev, Kabir Mohammed, Lesley Truelove, Sharon M. Gowan, Jia Li, Sarah L. Gulliford,Julian R. Marchesi, and David P. DearnaleyClin Cancer Res Nov 1, 2019 25:21 6487–500; doi: 10.1158/1078-0432.CCR-19-0960

Editors of the AACR journals reviewed recently published content toidentify hot topics across the entire portfolio. This publication focuseson the tumormicrobiome and highlights articles based on a number of keymetrics, such as usage and citations. We hope that you enjoy this com-plimentary cross-journal collection.

Tumor Microbiome

Neoadjuvant Chemotherapy Shifts Breast Tumor Microbiota Populations to Regulate Drug Responsivenessand the Development of MetastasisAkiko Chiba, Alaa Bawaneh, Christine Velazquez, Kenysha Y.J. Clear, Adam S. Wilson, Marissa Howard-McNatt, Edward A. Levine,Nicole Levi-Polyachenko, Shaina A. Yates-Alston, Stephen P. Diggle, David R. Soto-Pantoja, and Katherine L. CookMol Cancer Res Jan 1, 2020 18:1 130–9; doi: 10.1158/1541-7786.MCR-19-0451

Oral Microbiome Profiling in Smokers with and without Head and Neck Cancer Reveals Variations BetweenHealth and DiseaseAshok Kumar Sharma, William T. DeBusk, Irina Stepanov, Andres Gomez, and Samir S. KhariwalaCancer Prev Res May 1, 2020 13:5 463–74; doi: 10.1158/1940-6207.CAPR-19-0459

To read a full-text article within this pdf, please click on its title above. While viewing the full-text article, you may access it onlineby clicking its title on the article’s title page.

Tumor Microbiome

Research Article

Alterations to the Esophageal MicrobiomeAssociated with Progression from Barrett'sEsophagus to Esophageal AdenocarcinomaErik J. Snider1, Griselda Compres2, Daniel E. Freedberg2, Hossein Khiabanian3,Yael R. Nobel2, Stephania Stump2,4, Anne-Catrin Uhlemann2,4, Charles J. Lightdale2, andJulian A. Abrams2

Abstract

Background: The incidence of esophageal adenocarcinomahas risen dramatically over the past half century, and theunderlying reasons are incompletely understood. Broad shiftsto the upper gastrointestinal microbiome may be partlyresponsible. The goal of this study was to describe alterationsin the esophageal microbiome that occur with progressionfrom Barrett's esophagus to esophageal adenocarcinoma.

Methods: A case–control study was performed of patientswith and without Barrett's esophagus who were scheduled toundergo upper endoscopy. Demographic, clinical, and dietaryintake data were collected, and esophageal brushings werecollected during the endoscopy. 16S rRNA gene sequencingwas performed to characterize the microbiome.

Results: A total of 45 patients were enrolled and includedin the analyses [16 controls; 14 Barrett's esophagus withoutdysplasia (NDBE); 6 low-grade dysplasia (LGD); 5 high-grade dysplasia (HGD); and 4 esophageal adenocarcino-ma]. There was no difference in alpha diversity between

non–Barrett's esophagus and Barrett's esophagus, but therewas evidence of decreased diversity in patients with esoph-ageal adenocarcinoma as assessed by Simpson index. Therewas an apparent shift in composition at the transition fromLGD to HGD, and patients with HGD and esophagealadenocarcinoma had decreased Firmicutes and increasedProteobacteria. In addition, patients with HGD or esoph-ageal adenocarcinoma had increased Enterobacteriaceaeand Akkermansia muciniphila and reduced Veillonella. In thestudy population, patients taking proton pump inhibitorshad increased Streptococcus and decreased Gram-negativebacteria overall.

Conclusions: Shifts in the Barrett's esophagus–associatedmicrobiomewere observed in patients withHGD and esoph-ageal adenocarcinoma, with increases in certain potentiallypathogenic bacteria.

Impact: The microbiome may play a role in esophagealcarcinogenesis.

IntroductionThe incidence of esophageal adenocarcinoma has increased

10-fold since the late 1960s (1), andBarrett's esophagus incidencelikely began to rise as early as the 1950s. Known modifiable riskfactors for esophageal adenocarcinomadonot adequately explainthese incidence trends. Gastroesophageal reflux disease (GERD)prevalence began to rise in the 1970s (2, 3), andmodeling studiessuggest that only a minority of esophageal adenocarcinoma casesare attributable to GERD (4). The obesity epidemic did not begin

until 1980, and obesity may only account for a small fraction ofthe rise in esophageal adenocarcinoma (5).

Helicobacter pylori infection is associated with a 30% to 40%reduced risk of Barrett's esophagus and esophageal adenocarci-noma (6), and H. pylori prevalence has plummeted since themid-20th century (7). When present, H. pylori dominates thegastric microbiome, and its absence results in major shifts togastricmicrobiome composition (8, 9). Thus, dramatic changes inthe upper GI microbiome in western populations likely occurredat the same time that Barrett's esophagus and subsequentlyesophageal adenocarcinoma began to rise in incidence. Any roleof the microbiome in the development of esophageal adenocar-cinoma is likely complex and multifactorial, and may representa cofactor in the development of Barrett's esophagus, the pro-gression fromBarrett's esophagus to esophageal adenocarcinoma,or both.

There is ample evidence that elements of the microbiome candirectly contribute to the development of colon cancer (10).However, the role of the microbiome in the progression ofBarrett's esophagus to esophageal adenocarcinoma has not beenwell described. In health, the esophageal microbiome is broadlysimilar in composition to the oral microbiome, with a highrelative abundance of the phylum Firmicutes (11). Previouslypublished data suggest that the esophageal microbiome inpatients with reflux esophagitis or Barrett's esophagus is heavily

1Department of Medicine, University of Washington School of Medicine, Seattle,Washington. 2Department of Medicine, Columbia University Irving MedicalCenter, New York, New York. 3Rutgers Cancer Institute of New Jersey, RutgersUniversity, New Brunswick, New Jersey. 4Microbiome Core Facility, ColumbiaUniversity Irving Medical Center, New York, New York.

Note: Supplementary data for this article are available at Cancer Epidemiology,Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).

Corresponding Author: Julian A. Abrams, Columbia University Medical Center,630 W. 168th Street, P&S 3-401, New York, NY 10032. Phone: 212-342-0476;Fax: 212-342-5759; E-mail: [email protected]

Cancer Epidemiol Biomarkers Prev 2019;28:1687–93

doi: 10.1158/1055-9965.EPI-19-0008

�2019 American Association for Cancer Research.

CancerEpidemiology,Biomarkers& Prevention

www.aacrjournals.org 1687

http://cebp.aacrjournals.org/cgi/content/full/1055-9965.EPI-19-0008?ijkey=sIo05eUcgXyZw&keytype=ref&siteid=aacrjnls

populated with Gram-negative bacteria, which may contribute toa chronic inflammatory, proneoplastic state (12, 13). More recentanalyses of esophageal adenocarcinoma surgical resections haveshown that the tumor-associated microbiome demonstratesdecreased microbial richness and diversity compared with non-dysplastic Barrett's esophagus and normal squamous tissue (14).

In order to understand the potential role of the microbiome inesophageal carcinogenesis, knowledge of microbiome alterationsthat occur along the neoplastic pathway from Barrett's esophagusto esophageal adenocarcinoma is needed. The current studyaimed to elucidate shifts in the esophagealmicrobiome that occurin the setting of progression from Barrett's esophagus to associ-ated dysplasia and adenocarcinoma.

Materials and MethodsStudy population

This was a case–control study of patients �18 years old,enrolling subjects without or with a diagnosis of Barrett's esoph-aguswhowere scheduled to undergo upper endoscopy for clinicalindications. Analysis of the salivary microbiome in these patientshas been previously reported (15). Subjects were prospectivelyenrolled over 18 months at a single academic medical center(Columbia University Medical Center, New York, NY). Barrett'sesophagus subjects had histologically confirmed Barrett's esoph-agus measuring �2 cm, had never received endoscopic therapy,and were taking at least once daily proton pump inhibitors (PPI)for the priormonth. Barrett's esophagus subjects were categorizedbased on worst prior or current confirmed pathology: no dyspla-sia (NDBE), low-grade dysplasia (LGD), high-grade dysplasia(HGD), or esophageal adenocarcinoma. Controls were patientswith no prior history of Barrett's esophagus and were includedif taking at least once daily PPI or no acid suppression (PPIs orH2-receptor antagonists) for the priormonth. Other details of theexclusion criteria have been described previously (15).

Demographics, clinical data, and anthropometric measureswere collected. History of reflux symptoms was assessed usingquestionnaire (16), and dietary fat and fiber intake over thepreceding 4 weeks was analyzed using a food frequency ques-tionnaire (17, 18). All participants provided written-informedconsent. The Institutional Review Board of Columbia Universityapproved the study on February 25, 2015.

Sample collectionDetails of the sample collections have been described previ-

ously (15, 19). The microbiome was sampled by brushingthe squamous esophagus as well as Barrett's esophagus tissue(Barrett's esophagus patients) or gastric cardia, within 1 cm of thesquamo-columnar junction (controls). Sampling of any nodules,masses, or other focal lesions was avoided, in case grossly alteredtopography affected bacterial colonization. Biopsies were alsotaken from themid–Barrett's esophagus segment or gastric cardiafor subsequent gene expression analyses.

Microbiome characterizationAfter DNA extraction from esophageal brushings, the V4 hyper-

variable ribosomal RNA region was amplified using primers 515Fand 806R (20). Sequencing of the 16S rRNA gene V4 region wasperformed, and sequence data were uploaded to the NCBISequence Read Archive (BioProjectID PRJNA517734). Greengeneswas used as reference database (21). Clustering of taxonomic units

was made at 97% sequence similarity using USEARCH. The func-tions classify.seqs and classify.otu (bothwith default settings) fromthemothur project (22)were used tomake taxonomic assignmentsto operational taxonomic units (OTU). FastTree version 2.1.7 wasused to generate a phylogenetic tree of the contigs (23). Usingmothur and the phylogenetic tree, weighted and unweightedUniFrac distances as well as diversity indices were calculated (24).

Semiquantitative PCR (SsoAdvanced Universal SYBR GreenSupermix, Bio-Rad) was also performed from esophageal brush-ing DNA for Enterobacteriaceae to further assess key findings from16S rRNA gene sequencing analyses using previously publishedprimer pairs (25). DDCt values were calculated, using as a referencethe Ct value for Eubacteria for the corresponding sample. qPCRfor Eubacteria represents the entire bacterial DNA in the sample;thus, the DDCt values were analogous to relative abundance datafrom 16S rRNA gene sequencing.

Statistical analysesContinuous variables were analyzed using t tests and rank sum

tests, and categorical variables were analyzed using the Fisherexact tests. ANOVA or Kruskal–Wallis tests were used to comparecontinuous variables across multiple categories. The main anal-yses for this study were of brushings from Barrett's mucosa(Barrett's esophagus patients) or gastric cardia (controls). With-in-individual correlations were assessed between paired swabsfrom esophageal squamous lining and from paired swabs fromBarrett's esophagus or cardia by calculating Spearman rank cor-relation coefficients at the genus level for all genera with non-zeroread counts in both of the paired swabs. There were high correla-tions between paired swabs from the same site within the sameindividual (esophageal squamous, mean rho 0.85, SD 0.15;Barrett's esophagus or cardia, mean rho 0.86, SD 0.12). For thepurpose of these analyses, the mean relative abundance for eachtaxon from paired swabs was calculated from each sampling site.Of note, there was also high within-individual correlationbetween esophageal squamous and Barrett's esophagus or cardiabrushings (mean rho 0.82, SD 0.13).

Alpha diversity was assessed by observed OTUs and Shannonand Simpson indices. Pair-wise weighted and unweightedUniFrac beta diversity was calculated using functions implemen-ted inQIIME.Nonparametric permutationalMANOVA, as imple-mented in the FATHOM Toolbox for MATLAB, was used tocompare beta-diversity measures between Barrett's esophagusversus controls and betweenNDBE/LGD versusHGD/esophagealadenocarcinoma groups. Principal coordinate analyses for thesetests were also performed using functions implemented in theFATHOM Toolbox for MATLAB. Differentially abundant taxabetween groups were identified using linear discriminant analysiseffect size (LEfSe; https://huttenhower.sph.harvard.edu/galaxy/).Functional composition of the esophageal microbiome wasassessed using predicted metabolic pathways derived by phylo-genetic investigation of communities by reconstruction of unob-served states (PICRUSt) analysis (26). Analyses were performedfocused on the relative abundance of Gram-negative bacteria;Gram-negative genera and species were identified using a refer-ence list assembled by our group (Supplementary Table S1), andthe relative abundances of these taxa were summed for eachsample. Additional analyses were performed on relative abun-dance of Streptococcus, themost abundant genus in the esophagus;alterations in the relative abundance of this genus have beenassociated with a variety of esophageal conditions (13, 27, 28).

Snider et al.

Cancer Epidemiol Biomarkers Prev; 28(10) October 2019 Cancer Epidemiology, Biomarkers & Prevention1688

https://huttenhower.sph.harvard.edu/galaxy/

Upon visual observation of relative abundance of phylaacross levels of Barrett's esophagus and associated neoplasia, itappeared that therewere shifts in relative abundance of Firmicutesand Proteobacteria, the two most abundant phyla in the esoph-ageal samples, with the transition from LGD to HGD (Supple-mentary Fig. S1). Thus, additional analyses were performed withBarrett's esophagus subjects categorized as NDBE/LGD or HGD/esophageal adenocarcinoma. Multivariable linear regressionanalyses were performed to assess for covariates independentlyassociated with relative abundance of differentially abundantphyla and other select taxa. Full models were created includingall covariates with a univariate P value < 0.10. Variables with thehighest P value and >0.15 were then sequentially removed togenerate afinal reducedmodel. Statistical significancewasdefinedas P < 0.05. Analyses were performed using Stata 14.1 (StataCorp)and MATLAB (The MathWorks, Inc.).

ResultsA total of 45 subjectswere enrolled andhadbrushings collected

for analysis. The characteristics of the subjects are shownin Table 1. There were 16 non–Barrett's esophagus subjects and29 subjects with Barrett's esophagus (14 without dysplasia, 6LGD, 5 HGD, and 4 intramucosal esophageal adenocarcinoma).

Microbiome analysesThere were no significant differences in alpha diversity com-

paring Barrett's esophaguswith non–Barrett's esophagus patients,both in terms of richness and evenness (Supplementary Fig. S2).There was decreased diversity assessed by Simpson index, but notby Shannon index or observed OTUs, across levels of Barrett'sesophagus–associated neoplasia (NDBE, LGD, HGD, and esoph-ageal adenocarcinoma; Supplementary Fig. S3). In post hocpairwise comparisons, the Simpson index in esophageal adeno-carcinoma was significantly reduced compared with NDBE(P ¼ 0.006), LGD (P ¼ 0.01), and HGD (P ¼ 0.01). None ofthe other pairwise comparisons were significant. On beta-diversity analyses, there was no evidence of significant clusteringcomparing Barrett's esophagus versus controls (SupplementaryFig. S4).

The most abundant phyla in the samples from Barrett's esoph-agus and gastric cardia were Firmicutes (46.2%), Proteobacteria(22.9%), Bacteroidetes (19.6%), Actinobacteria (5.6%), andFusobacteria (5.1%). Barrett's esophagus subjects had significant-ly reduced relative abundance of Bacteroidetes compared withcontrols (16.3% vs. 25.5%, P ¼ 0.04), although there was noassociation after adjusting for patient characteristics (Supplemen-tary Table S2).

Therewere nooverall differences in relative abundance of phylaacross levels of Barrett's esophagus–associated neoplasia. How-ever, upon visual inspection of the results, it appeared that therewas a shift in composition with regard to Firmicutes and Proteo-bacteria, the two predominant phyla, with the transition fromLGD toHGD (Supplementary Fig. S1). Thus, subsequent analyseswere performed with Barrett's esophagus subjects categorized as(NDBE or LGD) and (HGD or esophageal adenocarcinoma).Compared with NDBE/LGD, subjects with HGD/esophageal ade-nocarcinoma had decreased relative abundance of Firmicutes(38.3% vs. 55.0%, P ¼ 0.04) and increased relative abundanceof Proteobacteria (32.1% vs. 17.7%, P ¼ 0.04; Fig. 1). In multi-variable analyses, HGD/esophageal adenocarcinoma remained

independently associated both with increased Firmicutes (P ¼0.03) and decreased Proteobacteria (P ¼ 0.01; SupplementaryTable S2). On beta-diversity analyses, there was no evidence ofsignificant clustering comparing HGD/esophageal adenocarcino-ma versus NDBE/LGD (Supplementary Fig. S4).

Taxonomic differencesAs compared with controls, subjects with Barrett's esophagus

had increased relative abundance of Sphingomonas and an unclas-sified species of Campylobacter. Non–Barrett's esophagus subjectshad increased relative abundance of various taxa includingPrevotella pallens, Porphyromonas endodontalis, and Aggregatibactersegnis (Supplementary Table S3). Based on the observations thatthere was a shift with transition from LGD to HGD at the phylumlevel, additional differences in relative abundance of taxa wereassessed by LEfSe with subjects again categorized as NDBE/LGDand HGD/esophageal adenocarcinoma (Fig. 2A). Patients withNDBE/LGD had significantly increased Veillonella. Several taxawere increased in theHGD/esophageal adenocarcinoma subjects,notably in Enterobacteriaceae and Verrucomicrobiaceae, specificallyAkkermansia muciniphila (Fig. 2B).

Table 1. Characteristics of patients who underwent upper endoscopy and hadmicrobiome analyses, comparing those without with those with Barrett'sesophagus

Non–Barrett'sesophagus(n ¼ 16)

Barrett'sesophagus(n ¼ 29) P

Age, mean (SD) 60.1 (14.9) 63.6 (11.7) 0.39Sex, male 9 (56%) 25 (86%) 0.04WHR, mean (SD) 0.95 (0.08) 0.97 (0.05) 0.37GERD 10 (63%) 27 (93%) 0.02Ever smoker 7 (44%) 19 (66%) 0.21PPI use 6 (38%) 29 (100%) <0.001Aspirin use 3 (19%) 11 (38%) 0.31Dietary fibera, grams per day;mean (SD) 15.2 (5.3) 16.5 (4.5) 0.42Dietary fata, % daily calories; mean (SD) 33.6 (2.3) 34.3 (3.2) 0.46

Abbreviation: WHR, waist-to-hip ratio.aDietary data missing in 1 subject.

Figure 1.

Relative abundance of the major phyla comparing subjects with no dysplasiaor LGD with those with HGD or esophageal adenocarcinoma (EAC).Compared with NDBE/LGD subjects, those with HGD or EAC had decreasedFirmicutes (P¼ 0.04) and increased Proteobacteria (P¼ 0.04).

Microbiome Alterations in Barrett's Progression

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(10) October 2019 1689

As members of Enterobacteriaceae can promote gut inflamma-tion and neoplasia, the data on this family were examinedin greater detail. Compared with NDBE/LGD, patients withHGD/esophageal adenocarcinoma were more likely to be smo-kers (P ¼ 0.03) and had higher dietary fat intake (P ¼ 0.05).After adjusting for these two factors, HGD/esophagealadenocarcinoma remained significantly associated with therelative abundance of Enterobacteriaceae (P ¼ 0.02; Supplemen-tary Table S2). Two subjects had very high relative abundance ofEnterobacteriaceae: one of these had HGD and a relative abun-dance of 38.3%, and one had intramucosal esophageal adeno-carcinoma and a relative abundance of 30.4%. These findingswere replicated in the esophageal squamous brushings, wherethese two subjects again had the highest relative abundance ofEnterobacteriacea in the study population. For each of these sub-jects, a single distinct OTU drove the high relative abundance. Onfurther evaluation of theseOTUs usingNCBI BLAST, onematchedpredominantly to species in the genera Klebsiella and Enterobacter,and the other matched to species in genera including Escherichiaand Shigella.

Esophageal and cardia biopsies were then analyzed by qPCRto assess whether they harbored differences compared withbrushings in relative abundance of Enterobacteriaceae. There was

no significant difference by qPCR comparing patients withNDBE/LGD and HGD/esophageal adenocarcinoma (medianDDCt 12.5 vs. 12.8, respectively; P ¼ 0.57).

Gram-negative bacteriaIn brushings, the mean relative abundance of Gram-negative

bacteria in all of the subjects was 54.7% (SD 23.0). There was nosignificant difference in the relative abundance of Gram-negativebacteria comparing non–Barrett's esophagus controls withBarrett's esophagus subjects (61.6% vs. 50.9%, P ¼ 0.14). Therewere also no significant alterations in the relative abundance ofGram-negative bacteria across levels of Barrett's esophagus–associated neoplasia (ANOVA, P ¼ 0.66). In the entire studypopulation (Barrett's esophagus and non–Barrett's esophagus),PPI users had decreased relative abundance of Gram-negativebacteria compared with PPI nonusers (51.1% vs. 67.3%;P ¼ 0.05; Fig. 3A).

StreptococcusThe mean relative abundance of Streptococcus in the study

population was 32.6% (SD 20.9%). There was no significantdifference in the relative abundance of Streptococcus comparingBarrett's esophagus patients with non–Barrett's esophagus

Figure 2.

A, Cladogram from LEfSe analysesof differentially abundant taxacomparing Barrett's esophaguspatients without dysplasia (NDBE)or LGD versus HGD or esophagealadenocarcinoma (EAC). B, Subjectswith HGD or EAC had reducedrelative abundance of Veillonella(left) and had increasedproportion of samples withpresence of the other differentiallyabundant taxa (right), which wererelatively rare. (Presence definedas having any reads, except forEnterobacteriaceae, whichwas defined as relativeabundance > 0.1%.)

Snider et al.


controls (35.7% vs. 26.9%, P ¼ 0.18) and no significant overallalteration in the relative abundance of Streptococcus across levels ofBarrett's esophagus–related neoplasia (ANOVA P ¼ 0.51). Withregard to PPI use, all subjects (Barrett's esophagus and non–Barrett's esophagus) on PPIs had greater relative abundance ofStreptococcus compared with controls not on PPIs (36.2% vs.19.9%, P ¼ 0.03; Fig. 3B).

Functional profilingPICRUSt analyses were performed to assess for functional

alterations to the esophageal microbiome. Several gene pathwayswere significantly altered comparing patientswith Barrett's esoph-agus with non–Barrett's esophagus controls (SupplementaryFig. S5A). Controls had increased RNA degradation and vitaminB6 metabolism, whereas Barrett's esophagus patients hadincreased glycerolipid metabolism. Compared with patients withNDBE or LGD, those with HGD or esophageal adenocarcinomaexhibited increased glycerophospholipid metabolism anddecreased other glycan degradation (Supplementary Fig. S5B).

DiscussionIn the current study, we assessed the Barrett's esophagusmicro-

biome with progression to dysplasia and adenocarcinoma. Weobserved a shift in composition with progression, notably at the

transition from LGD to HGD. This was manifested by significantclustering in beta-diversity analyses, as well as alterations to thetwo predominant phyla, with reductions in Firmicutes andincreases in Proteobacteria.

There are little previous data describing esophageal micro-biome changes that occur in the development of esophagealadenocarcinoma. Elliott and colleagues reported microbiomealterations comparing esophageal squamous samples fromnon–Barrett's esophagus controls, Barrett's samples from patientswithout dysplasia, and tumor tissue from patients with esoph-ageal adenocarcinoma (14). The authors noted that esophagealadenocarcinoma tumors had decreased alpha diversity comparedwith Barrett's esophagus, and in the present study, there was someevidence of a decline in diversity with progression. However,many of the specific taxonomic alterations were distinct. Thismaybe explained in part by the fact that the esophageal adenocarci-noma tumor-microbiome was analyzed in this prior study (14),whereas in the current study, sampling was performed only ofnormal appearing Barrett's mucosa, avoiding any nodules orlesions, in patients with esophageal adenocarcinoma. Also in thecurrent study, there were high within-individual correlationsbetween squamous and Barrett's esophagus or cardia brushings,but the across-group alterations were less marked in squamous ascompared with Barrett's esophagus or cardia (data not shown).Finally, the esophageal adenocarcinoma subjects in the current

Figure 3.

Compared with controls not takingPPIs, patients taking PPIs had (A)reduced relative abundance of Gram-negative bacteria (P¼ 0.05) and (B)increased relative abundance ofStreptococcus (P¼ 0.03).



study all had very early lesions (T1a), and thus microbiomealterations in these patients would not have been caused by stasisdue to tumor obstruction.

The increased relative abundance of Enterobacteriaceae inesophageal brushings from patients with HGD and esophagealadenocarcinoma has potential biological significance. Certainspecies within Enterobacteriaceae harbor the pks genomic islandand can produce colibactin, a genotoxin that induces DNAdamage (29). Colibactin-producing Escherichia coli promotetumor growth in xenograftmousemodels (30),modify the tumormicroenvironment (31), and have been found in high abundancein colonic biofilms in patients with familial adenomatous poly-posis (32). Members of the family Enterobacteriaceae have alsobeen implicated in gut inflammation in inflammatory boweldisease (33–35). Thus, it is plausible that increased levels ofEnterobacteriaceae in Barrett's esophagus may promote progres-sion to esophageal adenocarcinoma, either directly via colibactinor other bacterial products or indirectly by triggering an immuneresponse and local inflammation.

Interestingly, the Enterobacteriaceae findings from 16S analysesof esophageal brushings were not replicated by qPCR of esoph-ageal biopsies. However, the two subjects with high relativeabundance of Enterobacteriaceae had similar findings in the squa-mous esophagus, in linewith priorwork demonstrating that thereis little within-individual variability in the microbiome in thesquamous and Barrett's lining in patients with Barrett's esopha-gus (36). Further, our group previously showed that patients withHGDor esophageal adenocarcinoma have increased Enterobacter-iaceae in saliva, and that there is strong within-individual corre-lation between the salivary and esophageal microbiome (15).Thus, possible explanations for the discrepant findings are thatesophageal brushings are superior to biopsies for microbiomeassessment, as previously reported by Gall and colleagues (36),and that Enterobacteriaceae may reside predominantly within theesophageal biofilm rather than within the mucosa (37).

The increased relative abundance of A. muciniphila in subjectswith HGD or esophageal adenocarcinoma was also notable. Inthe colon,A.muciniphilahas been associatedwithmanybeneficialeffects related to obesity andmetabolic syndrome (38). However,depending on the context, this species also can degrade mucinsand thin the mucus layer (39), potentially leading to increasedinteraction between pathobionts and the underlying epithelium.In this fashion, the presence of A. muciniphila could conceivablylead to increased Barrett's tissue inflammation and promoteprogression to esophageal adenocarcinoma.

Yang and colleagues previously described a microbiome asso-ciated with reflux esophagitis and Barrett's esophagus that wascharacterized bydecreased relative abundance of Streptococcus andincreased relative abundance of Gram-negative bacteria (13). Inthe current study, there were no differences in relative abundanceof Streptococcus or in overall Gram-negative bacteria comparingnondysplastic Barrett's esophagus with controls (data not shown)or with progression from Barrett's esophagus to esophagealadenocarcinoma. However, controls not taking PPIs hadincreased Gram-negative bacteria and decreased Streptococcuscompared with subjects on PPIs, and our group has previouslydemonstrated that PPIs cause significant increases in Streptococcusin the distal gut (40). If Gram-negative bacteria in the esophaguspromote chronic inflammation and increase the risk of Barrett'sesophagus and esophageal adenocarcinoma (12), then PPIs mayprovide a chemoprotective effect by reducing overall levels of

Gram-negative bacteria. However, the PPI results from the currentstudy should be interpreted with caution, as the PPI users were amix of Barrett's esophagus and non–Barrett's esophagus patients.

The current study has several strengths. There were patientsfromall stages of Barrett's esophagus–associated neoplasia,whichpermitted the ascertainment of microbiome shifts prior to thedevelopment of esophageal adenocarcinoma. During the endos-copy, only flat Barrett's esophagus tissue was sampled, avoidinglesions so as tominimize confounding by the presence of bacteriathat may have been mere colonizers due to an altered tumormacro- and microenvironment. Care was taken with regard toexclusion criteria to minimize the effects of certain factors on themicrobiome such as antibiotics and immunosuppressants.Detailed clinical information and dietary intake data wererecorded and assessed in the analyses.

There were also certain limitations. The sample size was rela-tively small, and the studymayhave been underpowered to detectadditional important microbiome alterations associated withneoplastic progression in Barrett's esophagus. The current studydescribes associations with various stages of Barrett's esophagusneoplasia but no information on causative effects on progression.However, the findings provide key hypothesis-generating data forfollow-up functional studies.

In conclusion, there were pronounced shifts in themicrobiomein Barrett's esophagus associated with progression to esophagealadenocarcinoma, particularly at the transition fromLGD toHGD.Notably, patients with HGD and esophageal adenocarcinomahad increased relative abundance of Enterobacteriaceae, andmem-bers of this family have been implicated in gut inflammation andcarcinogenesis. Further studies are indicated to identify specificbacteria that may promote the development of esophageal ade-nocarcinoma, and also whether therapies targeting the micro-biome can be developed to modify the risk of esophagealadenocarcinoma.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: E.J. Snider, Y.R. Nobel, J.A. AbramsDevelopment of methodology: E.J. Snider, D.E. Freedberg, H. Khiabanian,S. Stump, J.A. AbramsAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): E.J. Snider, D.E. Freedberg, S. Stump,C.J. Lightdale, J.A. AbramsAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): E.J. Snider, H. Khiabanian, A.-C. Uhlemann,C.J. Lightdale, J.A. AbramsWriting, review, and/or revision of themanuscript: E.J. Snider, D.E. Freedberg,H. Khiabanian, Y.R. Nobel, A.-C. Uhlemann, C.J. Lightdale, J.A. AbramsAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): E.J. Snider, S. StumpStudy supervision: J.A. AbramsOther (research coordinator): G. Compres

AcknowledgmentsThe authors were supported in part by a Columbia Physician's and Surgeon's

Dean's Research Fellowship (E.J. Snider), a Career Development Award fromNIDDK (K23 DK111847; D.E. Freedberg), a U54 award from NCI (U54CA163004; J.A. Abrams), a R01 from NIAID (AI116939; A.-C. Uhlemann),and the Price Family Foundation (J.A. Abrams).

The authors would like to acknowledge the New York Genome Center forperforming the 16S rRNA gene sequencing on the samples collected as part ofthis study. The authors would also like to acknowledge Nora C. Toussaint (ETH

Snider et al.


Zurich, NEXUS Personalized Health Technologies; New York Genome Center)for assistance with processing and analyses of the sequencing data.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby marked

advertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received January 2, 2019; revised April 17, 2019; accepted July 10, 2019;published first August 29, 2019.

References1. Abrams JA, Sharaiha RZ, Gonsalves L, Lightdale CJ, Neugut AI. Dating

the rise of esophageal adenocarcinoma: analysis of Connecticut TumorRegistry data, 1940-2007. Cancer Epidemiol Biomarkers Prev 2011;20:183–6.

2. el-SeragHB, Sonnenberg A.Opposing time trends of peptic ulcer and refluxdisease. Gut 1998;43:327–33.

3. Friedenberg FK, Hanlon A, Vanar V, Nehemia D, Mekapati J, Nelson DB,et al. Trends in gastroesophageal reflux disease asmeasuredby theNationalAmbulatory Medical Care Survey. Dig Dis Sci 2010;55:1911–7.

4. HazeltonWD, Curtius K, Inadomi JM, Vaughan TL,Meza R, Rubenstein JH,et al. The role of gastroesophageal reflux and other factors during progres-sion to esophageal adenocarcinoma. Cancer Epidemiol Biomarkers Prev2015;24:1012–23.

5. Kong CY, Nattinger KJ, Hayeck TJ, Omer ZB, Wang YC, Spechler SJ, et al.The impact of obesity on the rise in esophageal adenocarcinoma incidence:estimates from a disease simulation model. Cancer Epidemiol BiomarkersPrev 2011;20:2450–6.

6. Islami F, Kamangar F. Helicobacter pylori and esophageal cancer risk: ameta-analysis. Cancer Prev Res 2008;1:329–38.

7. Banatvala N, Mayo K, Megraud F, Jennings R, Deeks JJ, Feldman RA. Thecohort effect and Helicobacter pylori. J Infect Dis 1993;168:219–21.

8. Bik EM, Eckburg PB, Gill SR, Nelson KE, Purdom EA, Francois F, et al.Molecular analysis of the bacterial microbiota in the human stomach.Proc Natl Acad Sci U S A 2006;103:732–7.

9. Abreu MT, Peek RM Jr. Gastrointestinal malignancy and the microbiome.Gastroenterology 2014;146:1534–46.e3.

10. Brennan CA, Garrett WS. Gut microbiota, inflammation, and colorectalcancer. Annu Rev Microbiol 2016;70:395–411.

11. Pei Z, Bini EJ, Yang L, Zhou M, Francois F, Blaser MJ. Bacterial biota in thehuman distal esophagus. Proc Natl Acad Sci U S A 2004;101:4250–5.

12. Yang L, Francois F, Pei Z. Molecular pathways: pathogenesis and clinicalimplications of microbiome alteration in esophagitis and Barrett esoph-agus. Clin Cancer Res 2012;18:2138–44.

13. Yang L, Lu X, Nossa CW, Francois F, Peek RM, Pei Z. Inflammation andintestinalmetaplasia of the distal esophagus are associatedwith alterationsin the microbiome. Gastroenterology 2009;137:588–97.

14. Elliott DRF, Walker AW, O'Donovan M, Parkhill J, Fitzgerald RC. A non-endoscopic device to sample the oesophageal microbiota: a case-controlstudy. Lancet Gastroenterol Hepatol 2017;2:32–42.

15. Snider EJ, Compres G, Freedberg DE, GiddinsMJ, KhiabanianH, LightdaleCJ, et al. Barrett's esophagus is associated with a distinct oral microbiome.Clin Transl Gastroenterol 2018;9:135.

16. Locke GR, Talley NJ, Weaver AL, Zinsmeister AR. A new questionnaire forgastroesophageal reflux disease. Mayo Clin Proc 1994;69:539–47.

17. Thompson FE, Midthune D, Subar AF, Kahle LL, Schatzkin A, Kipnis V.Performance of a short tool to assess dietary intakes of fruits and vege-tables, percentage energy from fat and fibre. Public Health Nutr 2004;7:1097–105.

18. Thompson FE, Midthune D, Subar AF, McNeel T, Berrigan D, Kipnis V.Dietary intake estimates in the National Health Interview Survey, 2000:methodology, results, and interpretation. J Am Diet Assoc 2005;105:352–63; quiz 487.

19. Nobel YR, Snider EJ, Compres G, Freedberg DE, Khiabanian H, LightdaleCJ, et al. Increasing dietary fiber intake is associated with a distinctesophageal microbiome. Clin Transl Gastroenterol 2018;9:199.

20. Nossa CW, Oberdorf WE, Yang L, Aas JA, Paster BJ, Desantis TZ, et al.Design of 16S rRNA gene primers for 454 pyrosequencing of the humanforegut microbiome. World J Gastroenterol 2010;16:4135–44.

21. Edgar RC. Search and clustering orders of magnitude faster than BLAST.Bioinformatics 2010;26:2460–1.

22. Schloss PD,Westcott SL, Ryabin T, Hall JR, HartmannM,Hollister EB, et al.Introducing mothur: open-source, platform-independent, community-supported software for describing and comparingmicrobial communities.Appl Environ Microbiol 2009;75:7537–41.

23. Price MN, Dehal PS, Arkin AP. FastTree 2–approximately maximum-likelihood trees for large alignments. PLoS One 2010;5:e9490.

24. LozuponeC, Knight R. UniFrac: a newphylogeneticmethod for comparingmicrobial communities. Appl Environ Microbiol 2005;71:8228–35.

25. Fan D, Coughlin LA, Neubauer MM, Kim J, Kim MS, Zhan X, et al.Activation of HIF-1alpha and LL-37 by commensal bacteria inhibitsCandida albicans colonization. Nat Med 2015;21:808–14.

26. Langille MG, Zaneveld J, Caporaso JG, McDonald D, Knights D, Reyes JA,et al. Predictive functional profiling of microbial communities using 16SrRNA marker gene sequences. Nat Biotechnol 2013;31:814–21.

27. Harris JK, Fang R, Wagner BD, Choe HN, Kelly CJ, Schroeder S, et al.Esophageal microbiome in eosinophilic esophagitis. PLoS One 2015;10:e0128346.

28. Benitez AJ, HoffmannC,Muir AB,Dods KK, Spergel JM, Bushman FD, et al.Inflammation-associatedmicrobiota in pediatric eosinophilic esophagitis.Microbiome 2015;3:23.

29. Bossuet-Greif N, Vignard J, Taieb F, Mirey G, Dubois D, Petit C, et al. Thecolibactin genotoxin generates DNA interstrand cross-links in infectedcells. mBio 2018;9. pii: e02393-17.

30. Cougnoux A, Dalmasso G, Martinez R, Buc E, Delmas J, Gibold L, et al.Bacterial genotoxin colibactin promotes colon tumour growth by inducinga senescence-associated secretory phenotype. Gut 2014;63:1932–42.

31. Dalmasso G, Cougnoux A, Delmas J, Darfeuille-Michaud A, Bonnet R. Thebacterial genotoxin colibactin promotes colon tumor growth bymodifyingthe tumor microenvironment. Gut Microbes 2014;5:675–80.

32. DejeaCM, Fathi P,Craig JM, Boleij A, Taddese R,GeisAL, et al. Patientswithfamilial adenomatous polyposis harbor colonic biofilms containingtumorigenic bacteria. Science 2018;359:592–7.

33. Wright EK, KammMA, Teo SM, Inouye M,Wagner J, Kirkwood CD. Recentadvances in characterizing the gastrointestinal microbiome in Crohn'sdisease: a systematic review. Inflamm Bowel Dis 2015;21:1219–28.

34. Lewis JD, Chen EZ, Baldassano RN, Otley AR, Griffiths AM, Lee D, et al.Inflammation, antibiotics, and diet as environmental stressors of the gutmicrobiome in pediatric Crohn's disease. Cell Host Microbe 2015;18:489–500.

35. Garrett WS, Gallini CA, Yatsunenko T, MichaudM, DuBois A, DelaneyML,et al. Enterobacteriaceae act in concert with the gut microbiota to inducespontaneous and maternally transmitted colitis. Cell Host Microbe 2010;8:292–300.

36. Gall A, Fero J,McCoyC,Claywell BC, SanchezCA, Blount PL, et al. Bacterialcomposition of the human upper gastrointestinal tract microbiome isdynamic and associated with genomic instability in a Barrett's esophaguscohort. PLoS One 2015;10:e0129055.

37. Blackett KL, Siddhi SS, Cleary S, Steed H, Miller MH, Macfarlane S, et al.Oesophageal bacterial biofilm changes in gastro-oesophageal reflux dis-ease, Barrett's and oesophageal carcinoma: association or causality?Aliment Pharmacol Ther 2013;37:1084–92.

38. Derrien M, Belzer C, de Vos WM. Akkermansia muciniphila and its role inregulating host functions. Microb Pathog 2017;106:171–81.

39. Shono Y, Docampo MD, Peled JU, Perobelli SM, Velardi E, Tsai JJ, et al.Increased GVHD-related mortality with broad-spectrum antibiotic useafter allogeneic hematopoietic stem cell transplantation in humanpatientsand mice. Sci Transl Med 2016;8:339ra71.

40. FreedbergDE, ToussaintNC, Chen SP, Ratner AJ,Whittier S,Wang TC, et al.Proton pump inhibitors alter specific taxa in the human gastrointestinalmicrobiome: a crossover trial. Gastroenterology 2015;149:883–5e9.



CANCER PREVENTION RESEARCH | RESEARCH ARTICLE

Chemoprevention of Colorectal Cancer byAnthocyanidins and Mitigation of Metabolic ShiftsInduced by Dysbiosis of the Gut MicrobiomeAshley M. Mudd1, Tao Gu2, Radha Munagala3,4, Jeyaprakash Jeyabalan3, Nejat K. Egilmez2, andRamesh C. Gupta1,3

ABSTRACT◥

Diets rich in fat, smoking, as well as exposure to envi-ronmental pollutants and dysbiosis of gut microbiota,increase the risk of developing colorectal cancer. Muchprogress has been made in combating colorectal cancer.However, options for chemoprevention from environmen-tal insult and dysbiosis of gut microbiota remain elusive.We investigated the influence of berry-derived anthocya-nidins (Anthos), with and without encapsulating them inbovine milk–derived exosomes (ExoAnthos), on the che-moprevention of bacteria-driven colon tumor develop-ment. Anthos and ExoAnthos treatment of colon cancercells showed dose-dependent decreases in cell viability.Calculated selectivity index (SI) values for Anthos andExoAnthos suggest that both treatments selectively tar-geted cancer over normal colon cells. In addition,ExoAnthos treatment yielded higher SI values than Anthos.Anthos and ExoAnthos treatment of ApcMin/þ mice

inoculated with enterotoxigenic Bacteriodes fragilis (ETBF)bacteria led to significant decreases in colon tumor num-bers over mice receiving vehicle treatments. Western blotanalysis of normal colon, colon tumor, and liver tissuelysates showed that mice inoculated with ETBF featuredincreased expression of phase I enzymes in normal colontissue and decreased expression of phase II enzymes in livertissue. Treatment with the Anthos and ExoAnthos revertedthe modulation of phase I and phase II enzymes, respec-tively; no significant changes in phase II enzyme expressionoccurred in colon tumor tissue. Treatment of HCT-116cells with the ubiquitous carcinogen, benzo[a]pyrene(B[a]P) led to similarmodulation of phase I and II enzymes,which was partially mitigated by treatment with Anthos.These results provide a promising outlook on the impact ofberry Anthos for prevention and treatment of bacteria- andB[a]P-driven colorectal cancer.

IntroductionAlthoughmuch progress has beenmade in the diagnosis and

treatment of cancer over the last 30 years, colorectal cancerremains a looming threat on the horizon. For instance, althoughthe incidence of colorectal cancer has been trending downwardsince the mid to late 1980s for individuals 55 years or older, arecent study has found a rather disconcerting uptick in thecolorectal cancer incidence for individuals below 55 yearsold (1). According to the Centers for Disease Control, the thirdmost common form of cancer in the United States is colorectal

cancer. Furthermore, colorectal cancer is the third leading formattributed to cancer-related deaths each year. In fact, accordingto the American Cancer Society, it is estimated that 140,250individualswill be diagnosed and50,630will die fromcolorectalcancer in theUnited States alone in 2018. Epidemiologic studiessuggest that diets rich in fat, smoking, and increased alcoholconsumption, as well as exposure to environmental pollutantsand dysbiosis of gut microbiota, increase the risk of developingcolorectal cancer (2, 3). Much progress has been made incombating the disease due to advancements made in the earlydetection of colorectal cancer. However, options for chemo-prevention from environmental insult and an understanding ofhow such treatments could alter the dialogue between one'smicrobiome and environmental toxins remain largely elusive.Environmental factors such as air pollution, cigarette smoke,

and dietary contaminants have been mechanistically linked toan increased risk of colorectal cancer (4–6).One particular classof environmental pollutants that is especially pervasive ispolycyclic aromatic hydrocarbons (PAH). One of the mostubiquitous members of this family is benzo[a]pyrene (B[a]P),which is found in cigarette smoke, as a contaminant in manyfoods, car exhaust fumes, wood-burning, and coal tar. Toinitiate the carcinogenic process, B[a]P undergoes bioactiva-tion by enzymes such as the cytochrome P450 (CYP) 1A1 and/or 1B1 and microsomal epoxide hydrolase (mEH), resulting in

1Department of Pharmacology and Toxicology, University of Louisville, Louis-ville, Kentucky. 2Department of Microbiology and Immunology, University ofLouisville, Louisville, Kentucky. 3JamesGrahamBrownCancer Center, Universityof Louisville, Louisville, Kentucky. 4Department of Medicine, University ofLouisville, Louisville, Kentucky.

Current address for R. Munagala: 3P Biotechnologies, Inc. 580 S. Preston Street,320 Delia Baxter II, Louisville, KY 40202.

Corresponding Author: Ramesh C. Gupta, University of Louisville, 580 S.Preston, St # 304E, Baxter II Research Building, Louisville, KY 40202.Phone: 502-852-6980; Fax: 502-852-3842; E-mail: [email protected]

Cancer Prev Res 2020;13:41–52

doi: 10.1158/1940-6207.CAPR-19-0362


AACRJournals.org | 41

http://cancerpreventionresearch.aacrjournals.org/cgi/content/full/1940-6207.CAPR-19-0362?ijkey=JMQaYyelVpED2&keytype=ref&siteid=aacrjnls

the formation of the ultimate carcinogen benzo[a]pyrene-7,8-diol-9,10-epoxide (BPDE). Various enzymes including mem-bers of the glutathione S-transferase (GST), uridine 50-dipho-spho-glucuronosyltransferase (UGT), and sulfotransferase(SULT) families are involved in detoxification of the inter-mediates along this bioactivation pathway (Fig. 1). The meta-bolites of B[a]P are classified as group 1 carcinogens by theInternational Agency for Research on Cancer. BPDE inter-calates DNA and ultimately covalently binds with guaninebases. This acts to distort the structure of DNA, disrupting thecopying of DNA, which in turn causes mutations. BPDE hasalso been found to target p53 thereby altering the tumorsuppression of cells, which may ultimately lead to cancer (7).Interestingly, research that measures all of the exposures of anindividual in a lifetime and how those exposures relate tohealth, referred to as “exposome,” has found that exposure toB[a]P leads to the enhanced susceptibility of macrophagemembranes to bacterial infection and ultimately may lead toimmunosuppression (8). Although one of the first carcinogensto be studied, B[a]P, remains a continued threat due to itswidespread presence in the environment. Given the continuedrelevance of B[a]P, how this toxicant and its metabolitesinteract with endogenous factors such as the gut microbiomeis of high importance in addressing the management of severaldiseases including colon cancer.In addition to traditional carcinogens such as B[a]P, recent

research has begun to uncover the importance of the gutmicrobiome in the development of colorectal cancer. Researchhas shown that imbalances in intestinalmicrobiota lead to bothan increase in inflammatory conditions, as well as an increasedproduction of carcinogenic metabolites, which may ultimatelylead to neoplasia. Several bacteria have been associated withincreased risk of developing colorectal cancer including, S.gallolyticus, H. pylori, virulent forms of Escherichia coli (E. coli),Fusobacterium nucleatum (F. nucleatum), Salmonella enterica(S. enterica), and enterotoxigenic Bacteriodes fragilis (ETBF)(9). ETBF in particular is a highly relevant model for devel-opment of colorectal cancer due to its contribution to bothfamilial and sporadic forms of cancer (10, 11). ETBF existsasymptomatically in 12.4% of individuals overall and in 27% ofindividuals with diarrhea symptoms (11). Furthermore, pres-ence of ETBF in the gut is a well-known cause of diarrhealdisease globally that is accompanied by colitis in both humansand animals. The pathogenicity associated with ETBF is due tothe secretion of a 20 kDa zinc-dependent metalloproteasetoxin, B. fragilis toxin (BFT), which binds to colonic epithelialcells and leads to the cleavage of the tumor suppressor protein,E-cadherin, and the secretion of IL-8 (12). Overall, this processleads to the stimulation of proliferation and migration ofhuman colon cancer cells (13). It should be noted that BFThas also been shown to induce proinflammatory cytokinesecretion by further activating the NF-kB pathway (13).Interestingly, a bidirectional dialogue has been found to exist

between the gut microbiome and environmental chemicals,with bacteria metabolizing the pollutants contributing to host

toxicity and the contaminants altering the composition of gutmicrobiota (3). This dynamic interaction between the hostmicrobiome and environmental carcinogens is becoming evermore prevalent and relevant in themodern era. Understandingthe impact of gut bacteria such as ETBF on the expression ofphase I/II enzymes and identifying a chemopreventive methodto combat this omnipresent insult is of great importance.Several plant bioactives have been an invaluable source of

medicines for humans. The family of plant pigments, known asthe anthocyanins, have been identified with a variety of healthbenefits including chemopreventive and therapeutic effects dueto their roles as anti-inflammatory, antioxidant agents andmodulators of CYP enzymes, CYP1A1 and CYP1B1 (14, 15).Found in dark-colored vegetables, fruits, grains, and flowers,anthocyanins provide the characteristic red, purple, and bluehues. Anthocyanins are, in part, converted to anthocyanidins(Anthos), the aglycone moieties, and, in fact, have higherantiproliferative and anti-inflammatory activities than theanthocyanins (16) presumably due to higher cell uptake.The berry Anthos presents a potential chemopreventive

option for individuals to avoid developing colorectal cancer.Berries were shown to reduce the oral dysplasia and carcinomain situ by approximately 50% in animals previously treatedwitha mixture of the cigarette smoke carcinogens, B[a]P andNNK (17). Previous work from our laboratory against breastand lung cancer has shown that Anthos possess both chemo-preventive and therapeutic effects due to their roles as anti-inflammatory, antioxidant agents and modulators of CYP1A1and CYP1B1 (18).We demonstrated that intervention with theanthocyanidin, delphinidin favorably modulated the underly-ing mechanisms of potent PAHs (19). Furthermore, althoughwork has been conducted to research the impact of anthocya-nins on colorectal cancer (20, 21), no data have been reported tostudy the impact of Anthos or ETBF bacteria on alterations inphase I and II enzyme expression.With this in mind, the aim of the series of studies presented

in this article was to assess how treatment with a nativemixtureof anthocyanidins derived from bilberry (i.e., Anthos) and anexosomal formulation of Anthos (ExoAnthos) influence pro-liferation and modulation of expression of key phase I and IIenzymes both in vitro and in a bacterially induced in vivomodelof colorectal cancer. Furthermore, the influence of gut bacterialdysbiosis induced by ETBF on phase I/II enzyme expressionwas also investigated.

Materials and MethodsChemicalsB[a]P was handled carefully with all safety procedures as it is

a highly carcinogenic and hazardous chemical. B[a]P (B-1760)was purchased from Sigma-Aldrich.

Isolation of bilberry-derived AnthosThe native bilberry Anthos mixture (>85%), composed of

delphinidin, cyanidin, petunidin, malvidin, and peonidin, was

Mudd et al.

Cancer Prev Res; 13(1) January 2020 CANCER PREVENTION RESEARCH42

generously provided by 3P Biotechnologies, Inc. The nativeAnthos was further enriched using C18 Sep-Pak cartridges(Waters) and eluted in acidified (0.1% HCl) ethanol. Theenriched extract was then dried using a Savant SC210ASpeed-Vac (Thermo Fisher Scientific) and stored at �20�C.Purity, batch to batch consistency, and reproducibility of theAnthos was determined by HPLC-PDA-UV. The enrichedbilberry extract available commercially is highly standard-ized and it provided similar ratios of the individual antho-cyanidins isolated from different batches of the bilberryextract. Briefly, 15 mL samples were analyzed using a Shi-madzu Premier C18 reverse-phase column (250 � 4.6 mm i.d., 5 mm). Mobile phase A was composed of water:formicacid:acetonitrile (87:10:3) and mobile phase B was composedof water:formic acid:acetonitrile (40:10:50). The flow ratewas 0.6 mL/minute and the gradient condition was 0–5minutes 5% B; 5–15 minutes 15% B; 15–20 minutes 25%B; 20–30 minutes 35% B; 30–40 minutes 45% B; 40–45minutes 100% B; 45–50 minutes 5% B. Detection of Anthoswas at 520 nm by PDA-UV and total Anthos concentrationwas calculated using a standard curve. The reference com-

pounds were purchased from ChromaDex and CaymanChemical Company.

Isolation of milk-derived exosomesExosomes from cow colostrum, isolated using differential

centrifugation as described by Munagala and colleagues (22),were generously provided by 3P Biotechnologies.

Protein determinationProtein estimation for exosomes was assessed using a

bicinchoninic acid (BCA) assay (Thermo Fisher Scientific).To determine protein concentration, diluted exosomal pre-parations were compared, in triplicate, to a serially diluted BSAstandard curve.

Preparation of ExoAnthosAnthos were loaded onto exosomes by mixing Anthos

(dissolved in 1:1 mixture of ethanol and water) with theexosomes suspension in a 1:5 (Anthos:Exosomal protein,w/w) ratio at room temperature (22�C). Unbound Anthos andany coagulated exosomes were removed by low-speed

O

OHOH

OH

O

OH

CYP450s (1A1, 1B1)

Epoxide hydrolase

CYP450s (1A1, 1B1, 3A4)

GST

GSH

OH

OH

OH

SG

OH

OH

OH

UDPGUGT

UDPGAExcretion Excretion

B[a]P triol-dG

BPDE

GSTGSH

OH

OH

SG

ExcretionOH

OH

AKR 1A1/AKR 1C1-1C4

OR1OR2

SULTs

R1, R2= H or SO2H

Excretion

Catechol

B[a]P

(-)B[a]P 7,8-dihydrodiol

(+)B[a]P 7,8-oxide

NADP+

O2-

H2O2

Oxidative DNA damage OO B[a]P 7,8-dione

Adduct formation:GSHRNADNA

O-Sulfated Catechol

NADPHAKRs

Activation oncogenes (KRAS)Alteration antioncogenes (TP53)

GSH conjugates

GSH conjugates glucuronides

OH

OH

NHN

NH

O

N

N

OHO

OH OH

Figure 1.

Metabolism of B[a]P to carcinogenic BPDE versus detoxification pathways.

Anthos Prevent Metabolic Shifts from Microbiome and B[a]P

AACRJournals.org Cancer Prev Res; 13(1) January 2020 43

centrifugation (10,000 � g for 10 minutes). The exosomalformulation of Anthos was collected by ultracentrifugation(135,000 � g for 1.5 hours). The pellet was then suspended inPBS and passed through a 0.22 mm syringe filter and storedat �80�C. The percent loading was determined using solventextraction as described previously (22, 23).

Analysis of Anthos loadingTo determine the load of Anthos in the ExoAnthos formu-

lation, the protein and Anthos concentrations were measured.Briefly, a 50 mL aliquot of ExoAnthos formulation was mixedwith 950 mL of acidified ethanol (0.1% HCl) and incubated at4 �C for 30–60 minutes. The precipitated proteins were sep-arated by centrifugation (10,000 � g for 10 minutes). TheAnthos contained in the supernatant was then analyzed using aSpectraMax M2 spectrometer (Molecular Devices). Anthoswere detected at 520 nm and total Anthos concentration wascalculated using a standard curve. Anthos concentrations wereconfirmed via HPLC-PDA against reference compounds. Thepelleted exosomal proteins were determined by the BCAmethod described above. The percent Anthos load was calcu-lated by dividing the amount of Anthos by exosomal proteins�100 (22). Individual anthocyanidins present in the Anthosmixture were loaded onto exosomes equally as confirmed usingHPLC-DAD.

Cells, culture conditions, and treatmentsThe APC wild-type HCT 116 (ATCC CCL-247) and APC

mutant HT-29 (ATCC HTB-38D) colon cancer cell lines andCCD-18Co (ATCC CRL-1459) normal colon cells wereacquired from ATCC. HCT-116 and HT-29 cells were main-tained inMcCoy's 5Amedium (Gibco) supplementedwith 10%FBS, 100 U/mL penicillin, and 100 mg/mL streptomycin in ahumidified atmosphere containing 5% CO2 at 37�C. Basedupon dosages derived from cell viability assays, cells werepretreated with Anthos alone (25, 50, 100, and 200 mmol/L)for 24 hours and treated with a mixture of Anthos (25, 50, 100,and 200 mmol/L) and B[a]P (20 mmol/L) for 24 hours.

Measurement of cell viabilityThe cytotoxicity of bilberry Anthos in colon cancer cell lines

was assessed by enzymatic reduction of the tetrazolium dyeMTT. Briefly, 3.0� 103 cells/well were grown in 96-well tissueculture plates and were then exposed to varying concentrationsof the Anthos, ExoAnthos, or vehicle control 24 hours afterseeding. After 72 hours treatment, cells were incubated with5mg/mLMTT reagent for 2 hours. Resulting formazan crystalswere subsequently solubilized in DMSO and measured spec-trophotometrically at 570 nm (Bio-Rad). IC50 values were thendetermined using CalcuSyn Software version 2.1 (Biosoft).

Western blot analysisFor Western blot analysis, 40 mg of protein from in vivo and

in vitro tissue lysates was resolved using SDS-PAGE andelectro-transferred to polyvinylidene difluoride membranes bysemidry transfer (Bio-Rad Trans-blot SD). Blots were blocked

with 4% dry powdermilk or BSA for 1 hour and then incubatedwith primary antibodies b-actin, UGT1A6, SULT1, GSTM1,GSTM2, CYP1A1, CYP1B1, PXR, Nrf2, AhR, AhRR, andARNT1 which were all acquired from Santa Cruz Biotechnol-ogy at 4�C overnight and secondary antibodies (Santa CruzBiotechnology) conjugated to peroxidase for 1 hour at roomtemperature. Blots were then developed with an ECL detectionsystem. Densitometric analysis was then performed usingImageJ 1.x software (24).

In vivo colorectal cancer studiesAnimal experiments were performed in agreement with an

approved protocol by the Institutional Animal Care and UseCommittee at the University of Louisville (Louisville, KY).Breeding colonies were established in collaboration with Dr.NejatK. Egilmez's laboratory (25) at theUniversity of Louisvilleusing C57BL/6J Min/þ (ApcMin/þ) mice that were originallyprocured from Jackson Laboratories. Mice were genotyped forthe APCmutation using PCR according to the protocol estab-lished by Jackson Laboratories. Mice were fed a standard chowdiet and received water ad libitum and were maintained on astandard light/dark cycle for the duration of the study. At 5–6 weeks of age, animals were administered antibiotics[clindamycin (0.1 g/L) and streptomycin (5 g/L)]. Four dayslater the animals were administered ETBF to promote tumor-igenesis in the colon and 1 week following ETBF inoculation,animals began their respective treatment regimen.Male (n¼ 2)and female (n¼ 3) ApcMin/þmice were orally administered (bygavage) an average of 8.6 mg/kg/day Anthos or ExoAnthos orvehicle control 3 days a week for 4 weeks. The Anthos dosageselected is roughly equivalent to 400 g of fresh blueberries perday for a 70 kg individual, based upon the level of anthocya-nidins commonly found in this fruit (26). Animals were culledin the fed state at 12 weeks, colon tumors were counted, andtissues were harvested.

Data analysisStatistical analysis was performed using Graph Pad Prism

statistical software version 4.03 and R Studio software version1.0.153 Lattice package (27, 28). One-way ANOVA was usedfor assessing the significance of mean differences across thevarious treatments for animal tumor and Western blot data.IC50 values were determined using CalcuSyn Software version2.1 (Biosoft). Heatmaps were constructed using R Studiosoftware version 1.0.153 gplot package (28, 29).

ResultsAntiproliferative effects of Anthos and ExoAnthos oncolon cancer cellsPrevious work from our laboratory has shown that exosomal

formulation yielded enhanced therapeutic potency and efficacyfor therapeutics such as paclitaxel, celastrol, curcumin, andAnthos against lung cancer because of increased stability andbioavailability of these compounds (22, 30–32). Prior to car-rying out our in vivo work, we first compared the relative

Mudd et al.


activity of ExoAnthos andAnthos treatment on proliferation ofHCT-116 and HT-29 colon cancer cell lines and CCD-18Conormal colon cells (Fig. 2). Results from these studies showed aclear increase in the antiproliferative properties of Anthosagainst colon cancer cells, with 4- to 16-fold decreases in theIC50 values of ExoAnthos as compared with the free Anthos(Fig. 2). One can posit that the improved antiproliferativeeffects of the ExoAnthos formulation overAnthos alone ismostlikely due to the increased cell uptake and stability in media ofthe ExoAnthos over Anthos. Part of the higher efficacy ofExoAnthos may be attributed to the intrinsic effect of theexosomes alone (22, 23).To determine whether Anthos and ExoAnthos were selective

toward colon cancer over normal colon cells in vitro, wedetermined the selectivity index (SI) values for both HCT116 andHT-29 colon cancer cells compared with normal colonCCD-18Co cells. The results (Table 1) showed that not onlywere Anthos and ExoAnthos selective for colon cancer cellsover normal colon cells but that ExoAnthos enhanced thisselectivity, with the greatest increase yielded in HT-29 cells,which increased from an SI value of 9 for Anthos to 51 for

ExoAnthos. Overall, these results confirm that Anthos andExoAnthos did not show any significant toxicity for the normalCCD-18Co colon cells and that the cytotoxicity is specific forcolon cancer cells.

Impact of Anthos treatment on tumor number in vivoAfter the promising results attained in vitro, we next sought

to test our ExoAnthos formulation, which makes use of anexosomal nano deliverymethod to determinewhether a greatertherapeutic effect could be achieved. Results from the com-parison of Anthos and ExoAnthos treatment showed thatExoAnthos lead to a similar reduction in colon tumor burdenas Anthos alone, when given at the same dose (P¼ 0.30).Whencompared with control, a significant reduction in colon tumorswas noted in the ExoAnthos-treated animals versus exosomevehicle control (P¼ 0.019) and Anthos-treated animals versusvehicle control (P ¼ 0.0025), when given at 8.6 mg/kg/day(Fig. 3). No significant difference was found between the tumornumbers in control versus exosome-alone–treated animals(P ¼ 0.728).

Impact of ETBF bacteria on phase I/II enzyme expressionin colon/liverApcMin/þ mice treated with the bacteria showed significant

increases in the expression of phase I enzymes, CYP1A1 andCYP1B1, in normal colon tissue (Fig. 4A). In Addition, asignificant decrease in the expression of the phase II enzyme,GSTM1, in normal colon tissue was noted. To assess howETBFbacteria influences the expression of phase I and II enzymes, weassessed the expression of AhR, AhRR, ARNT1, Nrf2, VDR,and PXR in ApcMin/þmice, with and without ETBF treatment.Results from this survey (Fig. 4A) showed that mice treatedwith the bacteria had significant decreases in the expression of

020406080

100

0 50 100 150 200

% C

ell s

urvi

val

Anthos (µmol/L)

HCT-116

020406080

100

0 50 100 150 200

% C

ell s

urvi

val

Anthos (µmol/L)

HT-29

020406080

100120

0 50 100 150 200

% C

ell s

urvi

val

Anthos (µmol/L)

CCD-18Co

A B

C

Anthos ExoAnthosFigure 2.

Antiproliferative activity of Anthos and ExoAnthosagainst colon normal cells and cancer cells in vitro. Colonnormal cells, CCD-18Co (C), and colon cancer cells, HCT-116 (A) and HT-29 (B) were treated with various con-centrations of bilberry-derived Anthos or ExoAnthos for72 hours and the effect on cell growth inhibition wasassessed using an MTT assay. Data represent average �SEM (n ¼ 4).

Table 1. SI values for Anthos and ExoAnthos treatments.

Cellline

IC50

Anthos(mmol/L)

SIAnthos

IC50

ExoAnthos(mmol/L)

SIExoAnthos

Fold differ-ence in IC50

values

CCD-18Co

1,050 — 407 — —

HCT116 75 14 20 20 4HT-29 124 9 8 51 16

Note: The SI values were calculated by dividing the IC50 value of normal CCD-18Co colon cells by the IC50 value of HCT116 or HT-29 colon cancer cells.



the AhR repressor, AhRR. No significant changes in theexpression of PXR, Nrf2, and VDR expression were noted incolon tissue samples taken from the mice that received bacteriacompared with untreated mice. Liver samples taken frombacteria-treated mice also featured decreased expression ofphase II enzymes UGT1A6 and GSTM1 and increased expres-sion of AhR when compared with ApcMin/þ mice that did notreceive bacteria treatment (Fig. 4B).

Impact of Anthos treatment on phase I/II enzymeexpression in colon and liver tissuesThere have been no prior literature reports to assess

whether Anthos modulate phase I and II enzyme expressionin the colon or liver tissue in animals treated with ETBFbacteria. Results from our survey of the impact of Anthos andExoAnthos on key phase I and II enzymes involved in themetabolism of the environmental carcinogen B[a]P, as well asother carcinogens, demonstrated significant modulation ofthe phase I enzymes CYP1A1 and CYP1B1 and phase IIenzymes GSTM1 and SULT1 by Anthos and ExoAnthos innormal colon tissue (Fig. 4A). Our survey of nuclear tran-scription factors and associated proteins including AhR,AhRR, ARNT1, Nrf2, VDR, and PXR found that Anthos led

to increases in the expression of the AhRR when comparedwith ETBF-alone–treated animals. Interestingly, ExoAnthostreatment did not affect the expression of AhRR; however, itled to decreased expression of the aryl hydrocarbon receptornuclear translocator, ARNT1. No significant changes werenoted in the expression of PXR, Nrf2, or VDR in colonsamples fromAnthos- or ExoAnthos-treated animals. Resultsfrom the enzyme expression analysis of liver tissues takenfrom the same animals noted similar decreases in the expres-sion of CYP1A1 in Anthos-treated mice and increases in thephase II enzymes GSTM1 and UGT1A6 (Fig. 4B). No sig-nificant changes in expression of mEH or SULT1 were notedin liver tissue. A significant decrease in AhR was noted inExoAnthos-treated mice and a significant increase in AhRRexpression was noted with the Anthos treatment as comparedwith ETBF ApcMin/þ control mice.

Impact of Anthos treatment on phase I/II enzymeexpression in tumor tissueAlterations of phase I and II enzymes have been implicated in

the development of chemo-resistance in cancer (33, 34).Because modulation of phase I and II enzymes was demon-strated in normal colon and liver tissues, we sought todetermine whether additional chemo-resistance could arisefor individuals undergoing chemotherapy if they were takingAnthos or whether Anthos could help alleviate offsite tox-icity associated with chemotherapeutic drugs if phase IIenzymes would be favorably modulated in normal versustumor tissue. To do this, we assessed the expression of selectphase II enzymes in the colon tumor tissue taken fromAnthos- and ETBF-alone–treated mice and showed thatAnthos treatment did not result in any significant changesin the expression of UGT1A6 (P > 0.5), GSTM1 (P > 0.1), orSULT1 (P > 0.1; Fig. 5).

Impact of Anthos treatment on alterations induced byB[a]P treatment in vitroAlterations in the expression of phase I enzymes had been

previously reported by our laboratory in an ACI rat model forbreast cancer (18). Given this background, we next sought todetermine whether Anthos and ExoAnthos treatment wouldalter this shift in phase I/II enzyme expression induced byB[a]P treatment. Results showed that cells treated overnightwith B[a]P (20 mmol/L) featured increased expression ofCYP1A1 accompanied by decreased expression of GSTM1(Fig. 6). Furthermore, treatment with Anthos led to decreasesin the expression of CYP1A1, while increasing the phase IIenzymes GSTM1 and SULT1. Anthos treatment, at the highestdose tested, led to approximately 3-fold reductions in theexpression of a key transcription factor AhR, which is involvedin the expression of CYP1A1, in HCT-116 colon cancer cells.Furthermore, Anthos treatment simultaneously led to a nearly2-fold increase in the expression of AhRR. Anthos treatmentalso led to 2- to 10-fold increase in the expression of a keynuclear receptor, PXR.

5-6 week 4 days 1 week 4 week

Mice bornAntibiotic ETBF

Treatment:-ExoAnthos-Anthos -Vehicle control

-Euthanasia at 12 week-Quantitation of colon adenomas

A

B

Control/-E

TBF

Control/+

ETBF

Anthos

Exoalo

ne

ExoAnthos

0

10

20

30

Treatment group

Tum

ornu

mbe

r

Figure 3.

Antitumor activities of Anthos and ExoAnthos against colon tumors. A, Studyoverview, ApcMin/þmice inoculatedwith ETBFwere treated via oral gavagewithAnthos or ExoAnthos at 8.6mg/kg/dayor vehicle controls.B,Data represent thedistribution of animal colon tumor counts, with the average noted. ControlApcMin/þ mice without ETBF bacteria versus ApcMin/þ mice with ETBF bacteria(P ¼ 0.0003), Anthos versus control (P ¼ 0.0025), control versus exosomes-alone–treated animals (P ¼ 0.728), ExoAnthos-treated animals versus exo-somes vehicle control (P ¼ 0.019), and Anthos versus ExoAnthos (P ¼ 0.30).�� , P < 0.0001; �� , P < 0.001; �� , P < 0.01; � , P < 0.05.

Mudd et al.


DiscussionGiven the recent uptick in cases of colorectal cancer diag-

nosed in younger individuals (1), colorectal cancer appears tobe making a comeback and thus warrants research into iden-tifying potential chemopreventive methods to combat thisdisease. Although much research has been conducted on therole of carcinogens such as B[a]P on the development ofcolorectal cancer, little has been achieved in the successfulprevention of this disease in populations who are exposed to

such environmental carcinogens on a daily basis. It should benoted that many studies have been conducted on a variety ofplant-derived chemopreventive agents but few agents havesuccessfully been translated. For instance, anthocyanins havebeen attributed to a variety of health-promoting benefitsincluding chemopreventive and therapeutic effects due to theirroles as anti-inflammatory and antioxidant agents (14, 15).However, no studies investigating the impact of Anthos on thebalance of phase I/II enzymes in the colon have been reported.Furthermore, the impact of bacteria such as ETBF on this

Figure 4.

In vivo changes in phase I and II enzymeexpression and related xenobiotic-sensing nuclear receptors in normalcolon and liver tissue following treat-ment with ETBF bacteria alone, bilber-ry-derived Anthos or ExoAnthos.Changes in the expression of key phaseI and II enzymes along with key nuclearreceptors including CYP1A1, CYP1B1,GSTM1, SULT1, AhR,AhRR,ARNT1,Nrf2,PXR, and VDR in normal colon (A) andliver (B) tissue, taken from ETBF-inoculated ApcMin/þ mice after treat-ment with bilberry-derived Anthos,ExoAnthos, or vehicle control asassessed using Western blot analysiscomparedwithwith ApcMin/þmice thatreceived no bacteria; b-actin served asloading control. Representative heat-maps depict fold changes in expressionbetween each group and the corre-sponding statistical significance(�� , P < 0.001; �� , P < 0.01; � , P < 0.05).



balance in the colon and liver has also yet to be elucidated.Given the lack of understanding in these areas, we sought touncover the impact of ETBF bacteria, Anthos, and B[a]P onthis crucial enzymatic balance.Several recent studies have confirmed the link between

dysbiosis of the gut microbiome and colorectal cancer.Several bacteria, in particular, have been associated withincreased risk of developing colorectal cancer including S.gallolyticus, H. pylori, and virulent forms of E. coli, F.nucleatum, S. enterica, and ETBF (9). Although muchresearch has been conducted on the role of bacteria oninflammation and carcinogenesis (10, 35, 36), no reportshave been published regarding the role of ETBF on theexpression of phase I and II enzymes. Results from ourassessment of the impact that ETBF has on the expressionof phase I and II enzymes in ApcMin/þ mice showed thattreatment with the bacteria led to significant increases in theexpression of phase I enzymes CYP1A1 and CYP1B1 innormal colon tissue. Furthermore, significant increases inthe expression of AhR and significant decreases in theexpression of phase II enzymes GSTM1 and UGT1A6 werenoted in liver tissue samples taken from ETBF mice whencompared with mice that did not receive the bacteria. Theresults gathered in this study provide an additional linkbetween how “bad” bacteria such as ETBF can ultimatelycontribute to the development of cancer beyond the initiallyelucidated inflammatory and gut barrier breakdownpathways (13, 35).After elucidating how ETBF bacteria increases the expres-

sion of phase I enzymes while decreasing the expression of thephase II enzymes in normal colon and liver tissue, we nextdetermined whether Anthos or ExoAnthos treatment couldalter this enzyme imbalance. Previous work from our labora-tory has shown that Anthos treatment led to decreased expres-sion and activity of CYP1A1 and CYP1A2 in an estrogen-driven ACI rat model for breast cancer (18). Results from this

series of studies showed that Anthos treatment significantlydecreased the expression of phase I enzymes CYP1A1 andCYP1B1, while increasing the expression of the phase IIenzymes GSTM1 and SULT1. Our survey of AhR, ARNT1,and AhRR expression suggest that the modulation of phase Iand II enzymes could be attributable to the altered expression ofAhRR and ARNT1 induced by Anthos treatment. It should benoted that AhRR is a key protein in the AhR signaling cascadethat acts as a repressor of AhR-dependent gene expression.Structural work shows that AhRR acts by competitively repres-singAhR binding toARNTand target DNA (37). Furthermore,AhRR levels have been shown to decline in a variety of diseasestates ranging from rheumatoid arthritis (38) to lung can-cer (39). Interestingly, DNAmethylation atAhRRhas also beenshown to be a marker for smoking and was correlated withfuture smoking morbidity and mortality (40). As noted above,ARNT1 is also a key component of the AhR signaling cascadeand functions by binding to the ligand-bound form of AhR andaiding in the movement of the AhR complex to the nucleus.ARNT has also been shown to be upregulated under hypoxicconditions by a HIF-1a–dependent mechanism in Hep3Bcells (41).Phase II enzymes such as UGT1A6, SULT1, and GSTM1

play an important role in the breakdown of chemotherapeuticdrugs such as irinotecan and cisplatin (42, 43). Therefore,increased expression of phase II enzymes in target tumor tissuewould not be desirable. Importantly results from this studyshowed that no significant changes in the expression of phase IIenzymes UGT1A6, SULT1, or GSTM1 occurred in colontumor tissue taken from animals treated with Anthos. There-fore, potential negative effects due to increased breakdown ofchemotherapeutic drugs that may result from increased phaseII enzyme expression, as evolved inmany tumors, should not bea cause for concern with Anthos treatment. With this in mind,the selective increase in phase II enzyme expression in normaltissue over tumor tissue may actually be an advantage in

Figure 5.

Impact of in vivo Anthos treatment onphase II enzyme expression in colontumor tissue: changes in the expres-sion of key phase II enzymes includingGSTM1 (P > 0.1), SULT1 (P > 0.1), andUGT1A6 (P > 0.5) taken from ETBF-inoculated ApcMin/þ mice after treat-ment with bilberry-derived Anthos orvehicle control as assessed usingWestern blot analysis compared withb-actin loading control.

Mudd et al.


decreasing off-target toxicity to healthy tissue for drugs such asirinotecan (42). However, additional studies would be neededto confirm this hypothesis.Exposure to environmental pollutants is now considered to

be one of the reasons behind the increasing rates of individualswith disorders ranging from obesity and type 2 diabetes tocancer (4, 6). Up to 90% of colorectal cancer cases are ofsporadic origin and it is estimated that diet contributes to 80%of known cases of colorectal cancer. The role of chemicals thatcontaminate food and ultimately contribute to the develop-

ment of colorectal cancer has been of great interest (6). ThePAH, B[a]P, is of special relevance due to its presence in avariety of common sources of exposure ranging from charcoal-cooked food to cigarette smoke, as well as several environ-mental sources and importantly, its epidemiologic correlationwith increased risk of colorectal cancer (44, 45). Similar to thealterations in phase I and II metabolism found to exist in vivofollowing ETBF inoculation, we found that cells treated withB[a]P (20 mmol/L) featured increased expression of the phase Ienzyme CYP1A1 with decreased expression of the phase IIenzyme GSTM1, as well as the AhR repressor, AhRR. Theseresults suggest that dysbiosis of the gut microbiome andexposure to the environmental carcinogen B[a]P both lead todysfunction of the balance between phase I and II enzymes incolon tissue. Furthermore, treatment with Anthos effectivelyshifted this balance in expression levels of AhRR, AhR, PXR,CYP1A1, and SULT1 to greater favor a state of detoxification inB[a]P-treated cells. It should be noted that the beneficial effectsattributed to the consumption of Anthos in this studymay varyfrom person to person depending upon the presence of geneticpolymorphisms of these enzymes. However, because theexpression of several enzymes was shown to be modulated bythese compounds, individualsmay still benefit overall byway ofalternative enzymes.Although our initial hypothesis was that ExoAnthos would

enhance therapeutic efficacy over Anthos, the findings pre-sented in this article showed no significant difference in tumornumbers between the Anthos- and ExoAnthos-treated ani-mals.Ultimately, bothAnthos andExoAnthos yielded similarlysignificant decreases in tumor numbers compared with vehicletreatments, respectively. This lack of enhanced therapeuticefficacy could perhaps be due to higher absorption ofExoAnthos prior to reaching the gut and enhanced deliveryto distant sites (31). Further studies would be needed to furtherelucidate this notion.Overarching results from this series of studies stress the

importance of integrating the gut microbiome into the study ofcarcinogen metabolism and carcinogenesis. With the everomnipresent threat and build-up of carcinogens within theindustrialized environment and the resulting inevitable dailyexposure of the colon to these compounds, a more integratedapproach to the prevention of cancer is needed. Future studiesassessing potential synergy thatmay arise when dysbiosis of thegut microbiome is combined with exposure to environmentalcarcinogens such as B[a]P could lead to a better understandingand a potential explanation for the current upward tick incolorectal cancer cases.


Authors’ ContributionsConception and design: A.M. Mudd, J. Jeyabalan, N.K. Egilmez,R.C. GuptaDevelopment of methodology: A.M. Mudd, T. Gu, J. Jeyabalan,R.C. Gupta

CYP1A1

AhRR

AhR

β-Ac�n

0 0 25 50 100 200

20 μmol/L B[a]P

Anthos [μmol/L]

1.0 0.9 0.8 0.8 0.6 0.3

1.0 0.9 0.9 0.7 0.3

1.0 0.5 0.3 0.6 0.7 0.9

1.0 0.5 1.2 1.4 1.7

1.0 2.9 1.9 0.4 0.8 0.3

1.0 0.7 0.1 0.2 0.1

1.0 1.1 1.1 1.6 1.9 2.5

1.0 0.9 1.3 1.6 2.2

1.0 0.4 0.3 0.4 0.6 0.5

1.0 0.8 1.0 1.5 1.3

SULT1

GSTM1

a:

b:

a:

b:

a:

b:

a:

b:

a:

b:

Figure 6.

Changes in expression of key phase I and II enzymes and related nuclearreceptors following treatment with B[a]P versus B[a]P and Anthos. Coloncancer cell line HCT-116 was pretreated with various concentrations of Anthosfor 24 hours followed by cotreatment with B[a]P and Anthos for 24 hours andthe effect on the expression of AhR, AhRR, CYP1A1, SULT1, and GSTM1 wasassessed using Western blot analysis and compared with b-actin loadingcontrol. Densitometry values listed are the ratio of each dose to the vehiclecontrol (A) and B[a]P control (B), corrected for b-actin loading control. Treat-ment and controls were run on the same gels.



Acquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): A.M. Mudd, T. Gu, N.K. Egilmez, R.C. GuptaAnalysis and interpretation of data (e.g., statistical analysis,biostatistics, computational analysis): A.M. Mudd, T. Gu,R. Munagala, N.K. EgilmezWriting, review, and/or revision of the manuscript: A.M. Mudd,R. Munagala, N.K. Egilmez, R.C. GuptaAdministrative, technical, or material support (i.e., reporting ororganizing data, constructing databases): A.M. MuddStudy supervision: R. Munagala, R.C. Gupta

AcknowledgmentsThis work was supported by Agnes Brown Duggan Endowment,

Helmsley Trust Fund, U.S. NIH (grant numbers AI092133 andCA100656), and the National Institute of Environmental Sciences

(grant number USPHS T32-ES011564). A.M. Mudd was supported bythe IPIBS fellowship and the NIEHS training grant. The authors thankDrs. Farrukh Aqil and Manicka Vadhanam for useful discussionduring the course of the work. We also thank Dr. Wendy Spencerof 3P Biotechnologies, Inc. for generously providing highly enrichedberry Anthos, as well as colostrum-derived exosomes used in ourstudies.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received July 26, 2019; revised October 17, 2019; acceptedNovember 26,2019; published first December 3, 2019.

References1. Siegel RL, Fedewa SA, Anderson WF, Miller KD, Ma J, Rosenberg PS,

et al. Colorectal cancer incidence patterns in the United States, 1974-2013. J Natl Cancer Inst 2017;109:djw322.

2. Shivappa N, Zucchetto A, Montella M, Serraino D, Steck SE, LaVecchia C, et al. Inflammatory potential of diet and risk of colorectalcancer: a case-control study from Italy. Br J Nutr 2015;114:152–8.

3. Claus SP, Guillou H, Ellero-Simatos S. The gut microbiota: a majorplayer in the toxicity of environmental pollutants? 2017;3:17001.

4. Beamish LA, Osornio-Vargas AR, Wine E. Air pollution: an environ-mental factor contributing to intestinal disease. J Crohns Colitis 2011;5:279–86.

5. Wei P-L, Lin S-Y, Chang Y-J. Cigarette smoking and colorectal cancer:from epidemiology to bench. J Exp Clin Med 2011;3:257–61.

6. Diggs DL, Huderson AC, Harris KL, Myers JN, Banks LD, RekhadeviPV, et al. Polycyclic aromatic hydrocarbons and digestive tractcancers: a perspective. J Environ Sci Health C Environ CarcinogEcotoxicol Rev 2011;29:324–57.

7. Pfeifer GP, Denissenko MF, Olivier M, Tretyakova N, Hecht SS,Hainaut P. Tobacco smoke carcinogens, DNA damage and p53mutations in smoking-associated cancers. Oncogene 2002;21:7435–51.

8. Clark RS, Pellom ST, Booker B, Ramesh A, Zhang T, Shanker A, et al.Validation of research trajectory 1 of an exposome framework:exposure to benzo(a)pyrene confers enhanced susceptibility to bac-terial infection. Environ Res 2016;146:173–84.

9. Sun J, Kato I. Gut microbiota, inflammation and colorectal cancer.Genes Dis 2016;3:130–43.

10. Dejea CM, Fathi P, Craig JM, Boleij A, Taddese R, Geis AL, et al.Patients with familial adenomatous polyposis harbor colonic biofilmscontaining tumorigenic bacteria. Science 2018;359:592–7.

11. Zhang G, Svenungsson B, Karnell A, Weintraub A. Prevalence ofenterotoxigenic Bacteroides fragilis in adult patients with diarrhea andhealthy controls. Clin Infect Dis 1999;29:590–4.

12. Rhee K-J,Wu S,WuX,HusoDL, Karim B, FrancoAA, et al. Inductionof persistent colitis by a human commensal, enterotoxigenic Bacter-oides fragilis, in wild-type C57BL/6mice. Infect Immun2009;77:1708–18.

13. Wu S, Powell J, Mathioudakis N, Kane S, Fernandez E, Sears CL.Bacteroides fragilis enterotoxin induces intestinal epithelial cell secre-tion of interleukin-8 through mitogen-activated protein kinases anda tyrosine kinase-regulated nuclear factor-kappaB pathway. InfectImmun 2004;72:5832–9.

14. Li L, Wang L, Wu Z, Yao L, Wu Y, Huang L, et al. Anthocyanin-richfractions from red raspberries attenuate inflammation in both

RAW264.7 macrophages and a mouse model of colitis. Sci Rep2014;4:6234.

15. He J, Giusti MM. Anthocyanins: natural colorants with health-promoting properties. Annu Rev Food Sci Technol 2010;1:163–87.

16. Aqil F, Jeyabalan J, Kausar H, Munagala R, Singh IP, Gupta R. Lungcancer inhibitory activity of dietary berries and berry polyphenolics.J Berry Res 2016;6:105–14.

17. Casto BC, Kresty LA, Kraly CL, Pearl DK, Knobloch TJ, Schut HA,et al. Chemoprevention of oral cancer by black raspberries.Anticancer Res 2002;22:4005–15.

18. AiyerHS, Gupta RC. Berries and ellagic acid prevent estrogen-inducedmammary tumorigenesis by modulating enzymes of estrogen metab-olism. Cancer Prev Res 2010;3:727–37.

19. Russell GK, Gupta RC, Vadhanam MV. Effect of phytochemicalintervention on dibenzo[a,l]pyrene-induced DNA adduct formation.Mutat Res 2015;774:25–32.

20. Thomasset S, BerryDP,CaiH,WestK,MarczyloTH,MarsdenD, et al.Pilot study of oral anthocyanins for colorectal cancer chemopreven-tion. Cancer Prev Res 2009;2:625–33.

21. Lala G, Malik M, Zhao C, He J, Kwon Y, Giusti MM, et al. Antho-cyanin-rich extracts inhibit multiple biomarkers of colon cancer inrats. Nutr Cancer 2006;54:84–93.

22. Munagala R, Aqil F, Jeyabalan J, Gupta RC. Bovine milk-derivedexosomes for drug delivery. Cancer Lett 2016;371:48–61.

23. Munagala R, Aqil F, Jeyabalan J, Agrawal AK, Mudd AM, KyakulagaAH, et al. Exosomal formulation of anthocyanidins against multiplecancer types. Cancer Lett 2017;393:94–102.

24. Schneider CA, Rasband WS, Eliceiri KW. NIH Image to ImageJ:25 years of image analysis. Nat Meth 2012;9:671–5.

25. Gu T, De Jesus M, Gallagher HC, Burris TP, Egilmez NK. Oral IL-10suppresses colon carcinogenesis via elimination of pathogenicCD4þ

T-cells and induction of antitumor CD8þ T-cell activity. Oncoimmu-nology 2017;6:e1319027.

26. Corona G, Tang F, Vauzour D, Rodriguez-Mateos A, Spencer J.Assessment of the anthocyanidin content of common fruits anddevelopment of a test diet rich in a range of anthocyanins. J BerryRes 2011;1:45–52.

27. Warnes GR, Bolker B, Bonebakker L, Gentleman R, Liaw WHA,Lumley T, et al. gplots: Various R programming tools for plottingdata. Rpackage version 2.6.0. Available from: https://CRAN.R-project.org/package=gplots.

28. Team R. RStudio: Integrated Development Environment forR. Boston, MA: RStudio, Inc. 2016. Available from: http://www.rstudio.com/.

Mudd et al.


https://CRAN.R-project.org/package=gplots



http://www.rstudio.com/



29. Sarkar D. Lattice: multivariate data visualization with R. New York,NY: Springer-Verlag New York; 2008.

30. Aqil F, Kausar H, Agrawal AK, Jeyabalan J, Kyakulaga AH, MunagalaR, et al. Exosomal formulation enhances therapeutic response ofcelastrol against lung cancer. Exp Mol Pathol 2016;101:12–21.

31. Aqil F, Jeyabalan J, AgrawalAK,KyakulagaAH,Munagala R, Parker L,et al. Exosomal delivery of berry anthocyanidins for the managementof ovarian cancer. Food Funct 2017;8:4100–7.

32. Agrawal AK, Aqil F, Jeyabalan J, Spencer WA, Beck J, Gachuki BW,et al. Milk-derived exosomes for oral delivery of paclitaxel. Nanome-dicine 2017;13:1627–36.

33. Housman G, Byler S, Heerboth S, Lapinska K, Longacre M, Snyder N,et al. Drug resistance in cancer: an overview. Cancers 2014;6:1769–92.

34. Mansoori B,MohammadiA,Davudian S, Shirjang S, Baradaran B. Thedifferent mechanisms of cancer drug resistance: a brief review.Adv Pharm Bullet 2017;7:339–48.

35. Chung L, Thiele Orberg E, Geis AL, Chan JL, Fu K, DeStefano ShieldsCE, et al. Bacteroides fragilis toxin coordinates a pro-carcinogenicinflammatory cascade via targeting of colonic epithelial cells. Cell HostMicrobe 2018;23:203–14.

36. Wu S, Rhee KJ, Albesiano E, Rabizadeh S, Wu X, Yen HR, et al. Ahuman colonic commensal promotes colon tumorigenesis viaactivation of T helper type 17 T cell responses. Nat Med 2009;15:1016–22.

37. Sakurai S, Shimizu T, Ohto U. The crystal structure of the AhRR-ARNT heterodimer reveals the structural basis of the repression ofAhR-mediated transcription. J Biol Chem 2017;292:17609–16.

38. Cheng L, Qian L, Wang GS, Li XM, Li XP. Genetic association ofaromatic hydrocarbon receptor and its repressor gene polymorphismswith risk of rheumatoid arthritis in Han Chinese populations. Med-icine 2017;96:e6392.

39. Zhang Y, Elgizouli M, Schottker B, Holleczek B, Nieters A, Brenner H.Smoking-associated DNA methylation markers predict lung cancerincidence. Clin Epigenet 2016;8:127.

40. Bojesen SE, Timpson N, Relton C, Davey Smith G, Nordestgaard BG.AHRR (cg05575921) hypomethylation marks smoking behaviour,morbidity and mortality. Thorax 2017;72:646–53.

41. Mandl M, Lieberum MK, Depping R. A HIF-1alpha-driven feed-forward loop augments HIF signalling in Hep3B cells by upregulationof ARNT. Cell Death Dis 2016;7:e2284.

42. Chen S, Yueh MF, Bigo C, Barbier O, Wang K, Karin M, et al.Intestinal glucuronidation protects against chemotherapy-inducedtoxicity by irinotecan (CPT-11). Proc Natl Acad Sci U S A 2013;110:19143–8.

43. Rose MC, Kostyanovskaya E, Huang RS. Pharmacogenomics ofcisplatin sensitivity in non-small cell lung cancer. GenomicsProteomics Bioinformatics 2014;12:198–209.

44. Kazerouni N, Sinha R, Hsu C-H, Greenberg A, Rothman N.Analysis of 200 food items for benzo[a]pyrene and estimation ofits intake in an epidemiologic study. Food Chem Toxicol 2001;39:423–36.

45. Sinha R, Kulldorff M, Gunter MJ, Strickland P, RothmanN. Dietary benzo[a]pyrene intake and risk of colorectal adenoma.Cancer Epidemiol Biomarkers Prev 2005;14:2030–4.



Tumor Biology and Immunology

Gastrointestinal Tract Dysbiosis Enhances DistalTumor Progression through Suppression ofLeukocyte TraffickingSamir V. Jenkins1, Michael S. Robeson II2, Robert J. Griffin1, Charles M. Quick3,Eric R. Siegel4, Martin J. Cannon5, Kieng B. Vang6, and Ruud P.M. Dings1

Abstract

The overall use of antibiotics has increased significantly inrecent years. Besides fighting infections, antibiotics also alterthe gut microbiota. Commensal bacteria in the gastrointesti-nal tract are crucial to maintain immune homeostasis, andmicrobial imbalance or dysbiosis affects disease susceptibilityand progression. We hypothesized that antibiotic-induceddysbiosis of the gut microbiota would suppress cytokineprofiles in the host, thereby leading to changes in the tumormicroenvironment. The induced dysbiosis was characterizedby alterations in bacterial abundance, composition, and diver-sity in our animal models. On the host side, antibiotic-induced dysbiosis caused elongated small intestines and ceca,and B16-F10melanoma and Lewis lung carcinomaprogressedmore quickly than in control mice. Mechanistic studiesrevealed that this progression was mediated by suppressedTNFa levels, both locally and systemically, resulting inreduced expression of tumor endothelial adhesionmolecules,

particularly intercellular adhesionmolecule-1 (ICAM-1) and asubsequent decrease in the number of activated and effectorCD8þT cells in the tumor.However, suppression of ICAM-1orits binding site, the alpha subunit of lymphocyte function-associated antigen-1, was not seen in the spleen or thymusduring dysbiosis. TNFa supplementation in dysbiotic micewas able to increase ICAM-1 expression and leukocyte traf-ficking into the tumor. Overall, these results demonstrate theimportance of commensal bacteria in supporting anticancerimmune surveillance, define an important role of tumorendothelial cells within this process, and suggest adverseconsequences of antibiotics on cancer control.

Significance: Antibiotic-induced dysbiosis enhances distaltumor progression by altering host cytokine levels, resulting insuppression of tumor endothelial adhesion molecules andactivated and effector CD8þ T cells in the tumor.

IntroductionBacteria colonize many parts of the body, and the cross-talk

between the microbiota and the host is crucial to maintainingimmune homeostasis. A growing body of literature supports theidea that microbial imbalance affects disease susceptibility andprogression. For instance, intestinal dysbiosis has been associatedwith a growing list of diseases of inflammatory, autoimmune,allergic, metabolic, and psychologic/neurologic nature (1–3).

A common and significant influence on the microbiota in thegastrointestinal (GI) tract is the use of antibiotics. Oral anti-biotics severely alter the bacteria in the GI tract by destroyingbeneficial bacteria as well as potentially pathogenic ones, pro-ducing a state of microbial imbalance called dysbiosis. Theoverall use of antibiotics has increased by more than 30% inrecent years (4), and moreover many patients with cancer areprescribed antibiotics during treatment, as infection is a frequentcomplication. A cohort study including over 3 million indivi-duals showed that there is a positive correlation between anti-biotic use and cancer risk (5). Specifically, individuals whoreceived 2–5 prescriptions over 2 years had an increased relativerisk (RR) of 1.27 for getting cancer [with a 95% confidenceinterval (CI) of 1.26–1.29], as compared with individuals whowere prescribed one or fewer antibiotic treatments in that period.The cancer risk was even greater, RR (95% CI) of 1.37 times(1.34–1.40) for individuals with more than six prescriptions (5).In addition, a growing body of evidence in murine studiesindicates that dysbiosis of the GI tract affects local colon carci-nogenesis due to the initial chronic inflammation and subse-quent immune suppression that dysbiosis produces (6–8). How-ever, whether commensal bacteria also play a role in stromalimmune surveillance of distal tumors remains unclear.

The bidirectional communication between tumor endothelialcells and the immune system is increasingly appreciated, yet howleukocyte trafficking is impacted by dysbiosis is unknown. At thecenter of this interaction are adhesion molecules on endothelialcells, including intercellular adhesion molecule 1 (ICAM-1),

1Department ofRadiationOncology, University ofArkansas forMedical Sciences,Little Rock, Arkansas. 2Department of Biomedical Informatics, University ofArkansas for Medical Sciences, Little Rock, Arkansas. 3Department of Pathology,University of Arkansas for Medical Sciences, Little Rock, Arkansas. 4Departmentof Biostatistics, University of Arkansas for Medical Sciences, Little Rock,Arkansas. 5Department of Microbiology and Immunology, University ofArkansas for Medical Sciences, Little Rock, Arkansas. 6Center for IntegrativeNanotechnology Sciences, University of Arkansas, Little Rock, Arkansas.

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

K.B. Vang and R.P.M. Dings contributed equally to this article.

Corresponding Author: Ruud P.M. Dings, University of Arkansas for MedicalSciences, 4301 W. Markham Street, Mail Slot #771, Little Rock, AR 72205.Phone: 501-526-7876; Fax: 501-526-5934; E-mail: [email protected]

Cancer Res 2019;79:5999–6009

doi: 10.1158/0008-5472.CAN-18-4108


CancerResearch


http://cancerres.aacrjournals.org/cgi/content/full/0008-5472.CAN-18-4108?ijkey=zctOtHVHfbH1E&keytype=ref&siteid=aacrjnls

vascular cell adhesion molecule (VCAM-1), and selectins (9).These molecules promote rolling, adherence, and transmigrationof leukocytes into tumor tissue, making tumors more vulnerableto host immunity. This communication is mediated by cytokines,the levels of which are controlled by cellular cross-talk betweenhost cells and commensal bacteria (10). We therefore examinedthe consequences of antibiotic-induced dysbiosis on stromalimmune surveillance in distal tumors, as a mechanism by whichGI tract microbiota influence immune surveillance. Mechanisticstudies revealed that this progressionwasmediated by suppressedTNFa levels, both locally and systemically, resulting in suppres-sion of tumor endothelial adhesion molecules, and a subsequentdecrease in the number of tumor-infiltrating activated CD8þ

T cells.

Materials and MethodsCell lines and cell culture

B16-F10 (murine melanoma; #CRL-6475) and Lewis lungcarcinoma (LLC; murine lung carcinoma; #CRL-1642) cell lineswere purchased from ATCC and cultured according to the man-ufacturer's instructions. All cell lines were cultured and main-tained as described previously (11).

Mice and tumor mouse modelsAll mice (C57BL/6J, #0664; Foxn1�/�, #2019; ICAM-1�/�,

#2867) were purchased from Jackson Laboratory and allowed toacclimatize to local conditions for at least 1 week. Animals wereprovided water and standard chow ad libitum and were main-tained on a 12-hour light/dark cycle. For tumor cell inoculation, a100-mL solution of 2 � 105 B16-F10 or 1 � 106 LLC cells wasinjected subcutaneously in the right rear leg of each mouse, asdescribed previously (11). Mice (sex- and age-matched litter-mates) were randomly inoculated. Tumor volume was deter-mined by measuring the diameters of tumors with calipersand calculated by the equation for volume of a spheroid: (a2 �b � p)/6, where a is the short axis and b is the long axis ofthe tumor. Dysbiosis was induced 2 weeks prior to the tumorinoculations by administering a cocktail of antibiotics, that is,ampicillin (250 mg/L), vancomycin (125 mg/L), neomycin(250 mg/L), and metronidazole (250 mg/L) in their drinkingwater, available ad libitum during the experiment (12). Micereceiving TNFa were randomized on day 7 after tumorinoculation, and murine TNFa treatment was initiated (everythree days with a total of 4 doses; 120 mg/kg in sterile PBS;intraperitoneally). After euthanization, organs were promptlyharvested, measured, and processed. Experiments were approvedby the University of Arkansas for Medical Sciences InstitutionalAnimal Care and Use Committee (protocol #3610 and #3836).

Bacterial diversity analysisStool samples less than 6 hours old were collected from

individual mice and stored at �80�C (13). DNA extraction wasperformed using ZymoBiomics DNA Miniprep Kit (#D4300;Zymo Research) according to the manufacturer's instructions.Briefly, samples were suspended in lysis buffer and heated to60�C for 20 minutes prior to 20 minutes of horizontal vortexingwith beads to homogenize the samples. Samples were centrifugedand the supernatant collected. From this, several processing stepswere performed to remove residual protein and the final DNAsample was eluted in 100 mL of nuclease-free H2O. The concen-

tration and purity was determined by the A260/A280 value (Cyta-tion 5; BioTek).

16S rRNA gene sequencingThe extracted sample DNA was sent to the ZymoBIOMICS

Targeted Sequencing Service for Microbiome Analysis (ZymoResearch) and sequenced using the Quick-16S Primer set V3-V4 (Zymo Research) via the IlluminaMiSeq v3 reagent kit using a10% PhiX spike-in.

Summary of the sequencing service. PCR products were quantifiedwith qPCR fluorescence readings and pooled together based onequalmolarity. The pooled library was cleaned using the Select-a-Size DNA Clean & Concentrator (Zymo Research), then quanti-fied with TapeStation (Agilent Technologies, Thermo FisherScientific).

Data processing and analysisDemultiplexed FASTQ files were received from ZymoBIOMICS

(GenBankBioProject accessionno. PRJNA561567) andprocessedwithQIIME2 version 2019.1 and 2019.4 (14). Primer sequence iscontained within the first 16 bp of the forward read and the first24 bp of the reverse read. Because of difficulties with retaininghigh qualitymerged reads, we opted to only use the forward readsherein. Forward reads were denoised and converted to AmpliconSequence Variants (ASV) via DADA2 (15) through the q2-dada2QIIME 2 plugin (all plugins are noted by q2-�). DADA2 wasinitiated by trimming the first 16 bp (to remove the proprietaryZymoBIOMICS primer sequence), using the "pooled" option forchimera detection and removal, and truncating the reads at263 bp.

Taxonomy assignment was achieved by mapping against theQIIME formatted SILVA (v132) reference database (availablefrom https://www.arb-silva.de). To increase the robustness oftaxonomy assignment, the corresponding V3-V4 amplicon regionwas extracted from the clustered (99% similarity) SILVA referencealignment. This region corresponds to alignment positions 5,045through 17,652 of the 99_alignment.fna file, provided throughthe QIIME-formatted SILVA reference database. Alignment gapswere subsequently removed and the resulting sequences used asinput to the q2-feature-classifier plugin, for both classifier trainingthrough fit-classifier-naive-bayes and taxonomy assignment viaclassify-sklearn to return 7-level consensus taxonomy. Any ampli-con sequence variants (ASV) that were classified as "Eukaryota,""Chloroplast," "Mitochondria," and "Unclassified" were exclud-ed. Only ASVs with at least a Phylum-level taxonomy assignmentwere retained.

The remaining ASVs were evaluated via q2-quality-control toassess possible ASV quality exclusion criteria based on visualinspection of the BLAST output of hits against the SILVA referencesequences via the evaluate-seqs method. After which, the pluginmethod exclude-seqswas used to removeASVswith less than95%sequence identity and 97% query coverage to any referencesequences contained within the SILVA reference set. Data wasexported to R for analysis and plotting via phyloseq and ggplotand differential abundance analyses was performed via lineardiscriminant analysis effect size (LEfSe; ref. 18).

Flow cytometryTumors were mechanically dissociated with shears until pieces

were approximately 1 mm3. This was followed by enzymatic

Jenkins et al.

Cancer Res; 79(23) December 1, 2019 Cancer Research6000

https://www.arb-silva.de

dissociation (1 mg/mL Collagenase Invitrogen #17101-015; 2.5U/mL Dispase Invitrogen 17105-041 and 20 mg/mL DNaseISigma #D-4527) for 30 minutes with continuous agitation bya MACSmix tube rotator in a 37�C incubator. Subsequently, thetissue suspensions were put on ice and 5 mL cold FACS buffer(sterile 2% FBSþ 2 mmol/L EDTA in PBS) was added. Single-cellsuspensions were made by filtering through a 70-mm cell strainer(BD Falcon #352350) to remove undigested cell clumps followedby a second filtration step via a 40-mm cell strainer (BD Falcon#352340). After lysing the red blood cells (ACS lysis buffer Gibco;#A1049201), the cell suspension was washed and collected bycentrifugation at 300 � g for 5 minutes at 4�C (11), followed bysurface staining as described in ref. 19. For intracellular staining,cell suspensions were stimulated with cell activation cocktail (BDBiosciences; #423304) and 250 ng/mL anti-CD3e for 5 hours in a37�C tissue culture incubator. Subsequently, the suspensionswere fixed and permeabilized using the Perm/Wash kit, accordingto the manufacturer's instructions (BD Biosciences).

Anti-mouse antibodies targeting CD4 (#0041), CD8a (#0081),CD27 (#563365)CD31 (#0311), CD54 (#0541), CD44 (#0441),CD45 (#0451), CD69 (#0691), CD62E (#553751), CD62P(#0626), CD106 (#1061), CD146 (#1469), CD197(#562675), TNFa (#7321), IFNg (#5945), and isotype controlantibodies were purchased from Thermo Fisher Scientific(eBioscience), BioLegend, Invitrogen, and BD Biosciences andused for FACS analysis.

Samples were acquired bymultiparameter flow cytometry on aLSR IIflowcytometer (BDBiosciences) andanalyzedusing Flowjosoftware (Tree Star, Inc.). Singlets were gated on by doubletexclusion and dead cells were excluded from the analysis usingFixable Viability Dye (FVD; #0865 eBioscience). Fluorescenceminus one (FMO) controls were utilized to determine gatingstrategy. Absolute numbers of cells are expressed per million cellsto compare tumors of different sizes.

HistologyThe small intestines and colons were stained with hematoxylin

and eosin after being formalin-fixed, paraffin-embedded and cutin tissue 5-mm sections. Images of the sections and staining wereacquired on Olympus IX71 microscope at �200 magnificationand digitally analyzed and differentially quantified by morpho-metric analysis, as described previously (20). The size of thelumen was estimated by the formula to calculate the area of anellipse: a � b � p.

ELISATo obtain chemokine and growth factor levels ELISA kits for

TNFa (#MTA00B), INFg (#MIF00), and VEGF (#MMV00) werepurchased from Biotechne and used according to the manufac-turer's instructions.

Statistical analysisData are reported as mean � SEM unless otherwise stated and

were analyzedby anunpaired two-tailed t test.P values<0.05wereconsidered statistically significant.

ResultsTumorigenesis and antibiotics change the microbiome

The microbial community composition among the sub-groups not treated with antibiotics displayed a prevalence of

Bacteroidetes and Firmicutes over other groups of bacteria,whereas Proteobacteria were noticeably more abundant withinthe dysbiotic subgroups (Fig. 1). The antibiotics caused asubstantial reduction in bacterial quantity and diversity, where-as the order Enterobacteriales was the most dominant in thedysbiotic mice bearing B16-F10 tumors, Betaproteobacterialesprevailed in the tumor-free or LLC-bearing mice (Fig. 1A). Onthe family level, this manifested in Enterobacteriaceae in thedysbiotic mice bearing B16-F10 tumors, and Burkholderiaceae inthe tumor-free or LLC-bearing mice (Fig. 1B and C). Lineardiscriminant analysis (LDA) revealed that there were severaldifferentially enriched groups, as visualized through taxonomichierarchy, of microbiota between the tumor types and treat-ment type (Fig. 1D and E).

Both weighted and unweighted UniFrac distance matriceswere produced by rarefying the data to 10,500 reads persample. PERMANOVA of weighted UniFrac revealed no differ-ences in beta-diversity between the subgroups not treated withantibiotics. However, there were significant differencesbetween mice with LLC to both B16-F10–bearing mice (q �0.042) and tumor-free mice (q � 0.042) when investigated viaunweighted UniFrac. During dysbiotic conditions, there weresignificant differences observed by weighted Unifrac betweenB16-F10 and LLC-bearing mice (q � 0.023) and tumor-freemice (q � 0.045), while no significant differences wereobserved by unweighted UniFrac. Through Kruskal–Wallisanalysis of Faith's phylodiversity (PD), we observed significantdifferences in alpha diversity between the orthobiotic controlsand dysbiotic groups (P � 0.0001). However, no significantdifferences of PD were observed within each of the orthobioticor dysbiotic subgroups.

Dysbiosis alters the host's GI tract and enhances tumorprogression at distal sites

Antibiotic-induced dysbiosis caused the small intestine andcecum to enlarge and elongate, yet the colon was not affected(Fig. 2). The length of the small intestines increased on average by15% in theB16-F10model from35.9�0.95 cm to41.7�1.11 cm(P < 0.01) during dysbiosis, and on average by 20% in the LLCmodel from 33.6� 0.54 cm to 40.7� 0.99 cm (P < 0.001) duringdysbiosis (Fig. 2A).

In termsof relative enlargement, the ceca increased themost, onaverage by approximately 400% during dysbiosis. Namely, thececa increased inweight in the B16-F10model from 0.47� 0.02 gto 1.96 � 0.07 g (P < 0.001) during dysbiosis, and on averageby 615% in the LLC model from 0.38 � 0.02 g to 2.34 � 0.11 g(P < 0.001) during dysbiosis (Fig. 2B). Images of representativececa are shown in Supplementary Fig. S1.

In contrast, the colons' lengths were not significantly affectedby dysbiosis. Namely, the colons were comparable in length inthe B16-F10 model in the controls from 6.48 � 0.18 cm versus6.55 � 0.17 cm during dysbiosis, and in the LLC model from6.70 � 0.16 cm versus 6.77 � 0.23 cm during dysbiosis(Fig. 2C).

The change in size of the small intestine was also reflected on amicroscopic scale as assessed by histology. The lumen of the smallintestines increased from 2.84 � 0.14 mm2 in the controls to3.98�0.47mm2 (P<0.05) during dysbiosis (Fig. 2D). The lumenof the colons, in contrast, changed only negligibly from 2.58 �0.23 mm2 in the controls to 2.51 � 0.53 mm2 during dysbiosis(Fig. 2E).

Stromal Modulation by GI Tract Dysbiosis

www.aacrjournals.org Cancer Res; 79(23) December 1, 2019 6001

Tumor growth curves and tumor mass determinationsindicated that both melanoma and lung carcinoma pro-gressed roughly twice as fast, on average, in mice withantibiotic-induced dysbiosis as compared with control mice

(Fig. 2F and G). Namely, the tumor masses were increasedby almost 250% in the B16-F10 model from 0.13 � 0.05 gto 0.32 � 0.07 g (P < 0.05) during dysbiosis, and byapproximately 165% on average in the LLC model from

Figure 1.

Microbial community composition in tumor-free or mice bearing B16-F10 melanoma or LLC with or without dysbiosis. Microbial composition in healthy controls(orthobiotic, Ortho) and antibiotic exposed (dysbiotic, Dys) mice with and without B16-F10 melanoma or LLC at the order (A) and family (B) levels. C, Log2abundance heatmap of microbial families. The relative abundance for the microbial families is indicated by hue. D, Taxonomic cladograms obtained from LEfSeanalysis of 16S rRNA sequences (blue, no tumor; red, B16-F10; green, LLC). Yellow circles represent nonsignificant differences in abundance between the groups.E, LEfSe analysis revealing differentially enriched microbial groups across tumor types (blue, no tumor; red, B16-F10; green, LLC) within each of the orthobioticand dysbiotic groups. For A–C, independent samples (n¼ 4 each) were merged (i.e., ASV read counts summed) into their respective groups prior tovisualization. The letters (a)–(i) in D refer to the microbial groups listed in E.

Jenkins et al.


0.67 � 0.13 g to 1.09 � 0.2 g during dysbiosis (P < 0.05) onthe day of sacrifice.

Dysbiosis did not affect body weights as mice bearing eitherB16-F10or LLC tumorsmaintained theirweights (SupplementaryFig. S2A and S2B).

Dysbiosis suppresses tumor vascular adhesion moleculesAs tumors progressedmore rapidly under dysbiotic conditions,

we investigated the changes in vascular adhesion molecules,which play a crucial role in immune surveillance and leukocyteextravasation. Indeed, dysbiosis caused ICAM-1 (CD54), VCAM-1(CD106), and MCAM (CD146) suppression on tumor endothe-lial cells (TEC; CD45� CD31þ) in B16-F10 tumors (Fig. 3).Namely, ICAM-1 expression reduced by 82% on average from2.73%� 0.95% to 0.49%� 0.12%, VCAMby 79% from0.14%�0.04% to 0.03% � 0.01%, and MCAM by 60% from 0.15% �0.03% to 0.06% � 0.01% during dysbiosis on TEC (all P values<0.05; Fig. 3A–D). In addition, selectins also showed changeduring dysbiosis, namely E-selectin (CD62E) levels changed by40% from 0.05% � 0.01% to 0.03% � 0.01% and P-selectin(CD62P) by 15% from0.53%� 0.10% to 0.45%� 0.06%duringdysbiosis on TECs (Fig. 3E and F). However, these relative lowexpression levels and mild changes in selectins failed to attain

statistical significance. Besides percentages, these same trendswere seen for the number of TECs expressing these adhesionmolecules in B16-F10 tumors (Fig. 3B-F), and for the number andpercentage of ICAM-1–positive TECs in LLC tumors (Supplemen-tary Fig. S3A–S3C).

This dysbiosis-induced suppression of ICAM-1 was not seen onestablished vasculature of tissues, such as in the spleen or in thethymus (Supplementary Fig. S4A–S4I; Fig. 5A–I). In addition,integrin alpha L (CD11a), the alpha subunit of lymphocyte func-tion–associated antigen-1 (LFA-1) and binding site of ICAM-1, wasnot suppressed on T cells in the spleen or thymus during dysbiosis(Supplementary Figs. S4D–S4F and S5D–S5F; SupplementaryTable S1). In concordance, dysbiosis did not affect the grossmorphology or mass of the spleens in B16-F10–bearing mice(0.09 � 0.01 g vs. 0.08 � 0.01 g during dysbiosis), or LLC-bearing mice (0.12 � 0.02 g vs. 0.10 � 0.01 g during dysbiosis)as measured on the day of sacrifice (Supplementary Fig. S6A andS6B).

Dysbiosis decreases the number of activated and effector T cellsin tumors

As tumor endothelial adhesion molecules are crucial for cyto-toxic leukocyte trafficking and extravasation into the tumor, we

Ortho Dys0

30

35

40

45

50

Smal

l int

estin

e (c

m) **

Ortho Dys0

30

35

40

45

50 ***

Ortho Dys0.0

1.0

2.0

3.0

Cec

um (g

)

***

Ortho Dys0.0

1.0

2.0

3.0***

0 2 4 6 8 10 12 140

500

1,000

1,500

Time (days)

Tum

or v

olum

e (m

m3 )

Ortho

DysLLC

Ortho Dys0.0

0.4

0.8

Tum

or m

ass

(g)

*

1210864200

250

500

750

Time (days)

Tum

or v

olum

e (m

m3 )

Ortho

DysB16-F10

Ortho Dys0

6.0

8.0

Col

on (c

m)

ns

GF

__ __

LLCSm

all i

ntes

tine

__Ortho

A

D

Col

on

Dys

B

E

Dys __

B16-F10 LLCB16-F10C

Ortho

LLCB16-F10

Ortho Dys0

6.0

8.0

ns

Ortho Dys0.0

0.5

1.0

1.5

Tum

or m

ass

(g)

*

Ortho Dys0.0

2.0

4.0

6.0

Lum

en s

urfa

ce (m

m2 ) *

Ortho Dys0.0

2.0

4.0

6.0

Lum

en s

urfa

ce (m

m2 ) ns

Figure 2.

Dysbiosis alters the gastrointestinal tract and enhances melanoma and lung carcinoma progression. The small intestines (A), ceca (B), and colons (C) of micebearing B16-F10 melanoma or LLC cells with or without antibiotic-induced dysbiosis. Representative hematoxylin and eosin images of small intestines (D) andcolons (E) in B16-F10melanoma–bearing mice. Growth curves and masses of B16-F10melanoma (F) and LLC (G). Tumor growth curves are shown in meanvolumes� SEMwith nonlinear regression-fit lines and tumor weights are shown as mean mass� SEM (n¼ 10–18 animals per group pooled from two orthree individual experiments). Dysbiosis was induced by exposing mice to antibiotics (ampicillin, neomycin, metronidazole, each at 250mg/L; vancomycin at125 mg/L). Scale bar, 200 mm. � , P < 0.05, �� , P < 0.01, �� , P < 0.05, two-sided t test; ns, not significant.*, orthobiotic (Ortho);&, dysbiotic (Dys) mice.



subsequently investigated the tumor infiltrate (Fig. 4). Dysbioticconditions significantly decreased the average number of cyto-toxic CD3þ CD8þ T cells that infiltrate tumors from 2.18 � 0.53(�104 cells) to 0.54 � 0.12 (�104 cells) during dysbiosis (P <0.05). The percentage of CD8þ T cells changed by fewer than 8percentage points from 91.3% � 0.7% to 83.6% � 5.1% duringdysbiosis (P < 0.23). Similarly, CD3þCD4þ T cells did not changein percentage (0.37% � 0.2% vs 1.4% � 0.5% during dysbiosis),nor in total amount, which was two orders of magnitude lower tobegin with [0.7 � 0.3 (�102 cells) vs. 0.6 � 0.01 (�102 cells)during dysbiosis] as compared with the cytotoxic CD3þ CD8þ Tcells (Fig. 4A and B).

Next, we assessed whether the T cells were active within thetumor microenvironment by evaluating activation markers hya-luronate receptor CD44, early activation antigen CD69, TNFreceptor CD27 and effector function marker chemokine receptortype 7 (CCR7; CD197; Fig. 4C–F). Dysbiosis reduced the totalnumber of CD3þ CD8þ CD44þ T cells by 67% [from 1.0 � 0.09(�104 cells) to 0.33 � 0.07 (�104 cells) during dysbiosis; P <0.001], CD3þ CD8þ CD69þ T cells by 79% [from 1.10 � 0.38(�104 cells) to 0.23 � 0.06 (�104 cells) during dysbiosis; P ¼0.04], CD3þ CD8þ CD27þ T cells by 83% [from 0.52 � 0.20(�104 cells) to 0.09 � 0.02 (�104 cells) during dysbiosis; P <0.05], andCD3þCD8þCCCR7þ T cells by 77% [from1.95�0.50(�104 cells) to 0.45 � 0.11 (�104 cells) during dysbiosis; P ¼0.01]. Dysbiosis only significantly reduced the proportion ofCD3þ CD8þ CCR7þ T cells by 6% (from 88.6% � 1.6 to 83.3� 0.8% during dysbiosis; Fig. 4D).

The overall decreased number of activated and effector CD8þ Tcells in the tumorwas not a result of a reduced number of T cells inthe spleen or the thymus (Supplementary Figs. S4A–S4C andS5A–S5C; Supplementary Table S1). Dysbiosis did not change

overall cell viability in either tumor model (SupplementaryFig. S7A and S7B).

TNFa is suppressed under dysbiotic conditionsNext, we measured effector molecules involved in ICAM-1

regulation, TNFa and IFNg (Fig. 5). Dysbiosis significantlyreduced the total amount of CD3þ CD8þ CD44þ T cellsexpressing TNFa by 78% [from 0.46 � 0.08 (�104 cells) to0.10 � 0.03 (�104 cells) during dysbiosis; P < 0.002], andreduced total amount of CD3þ CD8þ CD44þ T cells expressingIFNg by 71% [from 0.35 � 0.08 (�104 cells) to 0.10 � 0.02(�104 cells) during dysbiosis; P < 0.01]. Dysbiosis only sig-nificantly reduced the proportion of CD3þ CD8þ CD44þ

TNFa-expressing T cells (from 45.9% � 5.6 to 29.6 � 3.3%during dysbiosis; P < 0.03; Fig. 5A and B).

In addition, serum levels of TNFa also decreased from2.2 � 0.5 pg/mL to 1.3 � 0.4 pg/mL, resulting in decreasedTNFa levels per tumor tissue from 0.8 � 0.2 pg/mm3 to0.04 � 0.02 pg/mm3 during dysbiosis (Fig. 5C). However,serum concentrations of IFNg and VEGF did not changesignificantly during dysbiosis (IFNg , from 24.5 � 10.1 pg/mLto 14.7 � 7.6 pg/mL and VEGF, from 156.5 � 18.6 pg/mL to155.8 � 22.6 pg/mL; Supplementary Fig. S8A and S8B).

Dysbiosis-induced tumor progression is ICAM-1 mediatedTo confirm the importance of ICAM-1 in tumors during dys-

biosis, we next grewmelanoma inwild-type and ICAM-1�/�mice(Fig. 6). Tumors progressed up to twice as fast, on average, inICAM-1�/� mice, as compared with wild-type mice. However,dysbiosis did not enhance this tumor progression, suggestingICAM-1 involvement in dysbiotic-induced tumor progression(Fig. 6A).

Ortho Dys0.0

5.0

10.0

CD

54+

TEC

(%)

*

Ortho Dys0.0

0.1

0.2

0.3

CD

146+

TEC

(%)

*

Ortho Dys0.0

1.0

2.0

3.0

4.0

CD

146+

TEC

(×10

4 )

*

Ortho Dys0.00

0.05

0.10

CD

62E+

TEC

(%)

ns

Ortho Dys0.0

0.5

1.0

CD

62E+

TEC

(×10

4 )

ns

Ortho Dys0.0

0.5

1.0

CD

62P+

TEC

(%)

ns

Ortho Dys0.0

2.0

4.0

6.0

8.0

CD

62P+

TEC

(×10

4 )

ns

Ortho Dys0.0

20.0

40.0

CD

54+

TEC

(×10

4 )

*

Ortho Dys0.0

0.1

0.2

0.3

CD

106+

TEC

(%)

*

Ortho Dys0.0

1.0

2.0

3.0

CD

106+

TEC

(×10

4 )

*

CD54

+ TE

C

A Ortho Dys

FSC

B ICAM-1 VCAM-1

MCAM E-selec�n P-selec�n

C

D E F

9.9% 0.12%

Figure 3.

Tumor vascular adhesionmolecules are suppressed under dysbiotic conditions.A, Representative analysis plots for ICAM-1 (CD54) on TECs (CD45� CD31þ).Quantification in percentage and absolute number of ICAM-1 (CD54; B), VCAM-1 (CD106; C), MCAM (CD146; D), E-selectin (CD62E; E), and P-selectin (CD62P)expressing TECs (F). Data are the mean� SEM (n¼ 5–11 B16-F10 tumors per group, pooled from 2–3 individual FACS experiments). � , P < 0.05, two-sided t test;ns, not significant.*, orthobiotic (Ortho);&, dysbiotic (Dys) mice.

Jenkins et al.


TNFa supplementation overcomes dysbiosis-induced tumorprogression and enhances T-cell trafficking into the tumor

To further validate the importance of T-cell trafficking andextravasation into tumors during dysbiosis, we injected B16-F10 cells simultaneously into immunocompetent (C57BL/6J)and T-cell–deficient (Foxn1�/�) mice. We found that the growthrate of tumors was an order of magnitude lower in immunocom-petent mice than in the T-cell–deficient mice, and that dysbiosisonly disrupted tumor growth in immunocompetent mice, indi-cating that dysbiosis-induced tumor progression is dependent onT-cell trafficking (Fig. 6B).

Finally, based on the reduced levels of TNFa under dysbioticconditions, we hypothesized that the suppressed ICAM-1 expres-sion and subsequent tumor-infiltrating T-cells, could be restoredby TNFa administration. Indeed, TNFa supplementation (120mg/kg in sterile PBS; every three days with a total of 4 doses wasinitiated onday7 after B16-F10 inoculation) resulted in analmost6-fold increase on average in ICAM-1 expression on tumor endo-

thelial cells (P<0.02), and a4-fold increase on average inCD8þT-cell infiltrate in the tumor (P < 0.03), and a decrease of more than60% in tumor growth (Fig. 6C–E).

DiscussionThe immune system can recognize and eliminate developing

tumors, butmany cancers ultimately escape immune surveillance.Studies have shown that a disruption of the microbiota can bluntantitumor immune response and the efficacy of cancer therapiesby modulating circulating inflammatory chemo- and cytokinesand thereby the host inflammatory response (21, 22). The micro-biota of theGI tract is commonly perceived as diverse, robust, andlong-term stable to environmental disturbances (23). However,the individual bacterial communities within are dynamic incomposition and susceptible to various external factors like anti-biotics (24). Although the altered GI tract microbiome can berecovered and recolonized to a certain degree, loss of specific

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D8+

CD

44+

(×10

4 ) ***

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D8+

CD

69+

(×10

4 ) *

Ortho Dys0

20

40

60

80

100

CD

3+ C

D8+

CD

27+

(%) ns

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D8+

CD

27+

(×10

4 ) *

Ortho Dys0

20

40

60

80

100

CD

3+ C

D8+

CD

69+

(%) ns

Ortho Dys0

70

80

90

100C

D3+

CD

8+ (%

)ns

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D8+

(×10

4 )

*

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D4+

(%)

ns

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D4+

(×10

2 )

ns

Ortho Dys0

20

40

60

80

100

CD

3+ C

D8+

CD

44+

(%) ns

Ortho Dys0

70

80

90

100

CD

3+ C

D8+

CC

R7+

(%) *

Ortho Dys0.0

1.0

2.0

3.0

CD

3+ C

D8+

CC

R7+

(×10

4 ) *

C

E

D

F

A B

Figure 4.

Dysbiosis decreases the abundance of activated and effector T cells in tumors. B16-F10–infiltrated CD3þ CD8þ T cells (A) and CD3þ CD4þ T cells (B) with orwithout dysbiotic conditions. Percentage and amount of CD3þ CD8þ T cells expressing CD44 (C), CD69 (D), CD27 (E), and CCR7 (F). Data presented asmeans� SEM and is representative of two independent experiments n¼ 4–5 per group. � , P < 0.05; �� , P < 0.001, two-sided t test; ns, not significant.*, orthobiotic (Ortho);&, dysbiotic (Dys) mice.



strains and overall diversity follows antibiotic selection pres-sure (25). Consequently, even after the completion of treatment,the use of certain antibiotics may have direct and long-lastingdeleterious effects on the host by altering the composition andfunctions of the microbiota (26). With the sharp rise of multi-regimen, high-dose antibiotics being prescribedworldwide, this isa particularly sobering possibility and our results and those ofothers suggest that much additional investigation is needed.

Here, we induced dysbiosis in mice with clinically relevantdoses of antibiotics (27). Ampicillin and neomycin are bothbroad-spectrum antibiotics, while vancomycin is indicated forthe treatment of Gram-positive bacteria. When given orally, theseantibiotics are very poorly absorbed, or not at all for neomycin,without any systemic effects (28, 29). Metronidazole is absorbedin theGI tract, but has no systemic effect as it only functions whenit is reduced by anaerobic bacteria (30). In previous studies,dysbiosis was induced with high doses of antibiotics (4 times as

high) and/or acutely (1–3 days), resulting in changes in spleno-cytes and significant weight loss (> 30%), which are possibleconfounding variables (2, 31). The doses applied herein did notinduce these alterations.While both approaches address clinicallyrelevant aspects and considerations, one examines the acuteeffects of high-dose antibiotics, and the other assesses the con-tinuing consequences of long-term or multiple courses of anti-biotics on the microbiome and the host.

Cancer treatment is affected by changes of the GI tract micro-biome as well, as microbiota can hinder the efficacy of conven-tional cancer therapy by modulating the host inflammatoryresponse. For instance, Zitvogel and colleagues have shown thatcolon cancer only minimally responds to the immune-modulating anticancer drug cyclophosphamide during dysbio-sis (21, 22). It was suggested that this effect is mediated by adecrease in leukocyte-derived circulating inflammatory cytokines.More recent clinical anticancer strategies are affected by themicrobiome as well. Namely, a number of recent studies in miceand humans have shown the importance of GI tract microbiotaand immunotherapy efficacy by immune checkpoint inhibi-tors (32). For example, CTL-associated protein 4 (CTLA-4;CD152) and programmed cell death protein-1 (PD-1) blockageonly reduced tumor growth in mice harboring Bacteroides andBifidobacterium species, respectively (33, 34). Chaput and collea-gues showed that patients with metastatic melanoma, treatedwith CTLA-4 inhibition, had longer progression-free and overallsurvival when they were Faecalibacterium rich at the start of thetreatment (35). Similarly, Routy and colleagues found thatmicro-biota disruption with antibiotics in patients with cancer imme-diately prior to checkpoint inhibitor treatment led to shorterprogression-free andoverall survival (36).Wargo's andGajewski'sgroup reported that a more robust microbiota is associated withenhanced anti-PD-1 efficacy in patients with melanoma (37, 38).Interestingly, however, each group identifieddifferent "favorable"bacteria, suggesting that either certain mechanisms are sharedamong bacterial species or that the overall bacterial abundance,composition, and diversity is of importance rather than a specificbacterial strain. Along those lines, Honda and colleagues recentlydefined a commensal consortiumof 11 strains (7Bacteroidales and4non-Bacteroidales species) derived fromhumanvolunteers beingable to induce effector IFNgþ CD8þ T cells in germ-free mice.Moreover, these 11 strains were also able to enhance immunecheckpoint inhibitors' efficacy in murine tumor models, whereasa different 10-strain mix not associated with IFNgþ CD8þ T cellsfailed to do so (39).

We noted that dysbiosis caused by this specific regimen ofantibiotics resulted in a decline in bacterial amount and diversity,particularly inducing a change from Bacteroidetes and Firmicutesphyla dominance toward strains within the Proteobacteria phy-lum. This is in agreement with the above-mentioned human andmurine studies. Moreover, we found that dysbiosis also causedenlargement and elongation of the small intestine and cecumonamacro- and microscopic scale. The loss of anaerobic fusiform-shaped bacteria, which are abundant in the cecum and imbeddedin the mucus layer of its epithelium, are likely at the center of thisprocess as functionally they maintain the integrity of the watertransportmechanism (40). Losing these bacteria due to antibiotictreatment will disrupt this homeostasis and cause water and fluidretention in the cecum causing enlargement and elongation,which is also seen in germ-free mice. Restoring the cecum to itsnormal anatomic and physiologic state can be attained by

Ortho Dys0.0

2.0

4.0

TNFa

(pg/

mL

seru

m)

*

Ortho Dys0.0

0.5

1.0

1.5

2.0

TNFa

(pg/

mm

3 tu

mor

) *

Ortho Dys0

20

40

60

CD

3+ C

D8+

CD

44+

TNFa

+ (%

)*

Ortho Dys0.0

0.2

0.4

0.6

CD

3+ C

D8+

CD

44+

TNFa

(×10

4 )

**

Ortho Dys0

20

40

60

CD

3+ C

D8+

CD

44+

IFN

g+ (%

)

ns

Ortho Dys0.0

0.2

0.4

0.6

CD

3+ C

D8+

CD

44+

IFN

g+ (×

104 )

*

A

B

C

Figure 5.

TNFa is suppressed under dysbiotic conditions. Percentage and amount ofTNFa-producing (A) or IFNg-producing (B) tumor-infiltrating CD3þ CD8þ

CD44þ T cells and systemic TNFa serum levels (C) under orthobiotic anddysbiotic conditions. Data presented as means� SEM (n¼ 4–9 C57BL/6mice bearing B16-F10 tumors per group pooled from one to three individualexperiments). � , P < 0.05; �� , P < 0.01; two-sided t test; ns, not significant.*, orthobiotic (Ortho);&, dysbiotic (Dys) mice.

Jenkins et al.


reintroducing intestinal flora from unperturbed mice, but only toa certain degree as abnormalities of the cecum persist (40).

Several types of immunosuppression and tolerance,obstructing adequate T-cell effector function, have beendescribed within the microenvironment during orthobioticconditions. This includes coinhibitory effects of cytokines onT-cell activation, proliferation, and survival (41, 42), interfer-ence of activated T-cell migration (43), or direct suppression ofthe effector function by iNOS (inducible nitric oxide synthase),TGFb, and Tregs (44). Whether or not dysbiotic conditionsexacerbate any or all of these immune suppression and escapemechanisms has yet to be elucidated. Our data supports thenotion that dysbiosis negatively impacts T-cell trafficking, acti-vation, and effector function. Moreover, a low frequency ofCD8þ CCR7þ effector T cells is a significant risk factor fordisease recurrence in the clinic (45).

Here, we found that dysbiosis reduces the levels of TNFa andIFNg in the tumor microenvironment, and TNFa was suppressedsystemically in the serum as well. TNFa is a pleiotropic regulatorof ICAM-1 (46), an essential adhesion molecule for leukocytetrafficking and extravasation into tumors (9). Whereas selectinsare predominantly involved in the initial tethering, rolling, andarrest of leukocytes on endothelial cells, it is particularly ICAM-1that facilitates transcellular diapedesis into the tissue (47, 48).While dysbiosis enhanced tumor growth in ICAM-1þ/þ andimmunocompetent mice, it did not affect tumor growth inICAM-1�/� or T-cell–deficient mice, highlighting the influence

and impact of dysbiosis on T-cell trafficking and extravasation. Asexpected from these observations, supplementing TNFa duringdysbiosis caused elevated ICAM-1 expression on the tumor vas-culature, which increased the amount of effector T cells in thetumor and inhibited tumor growth inhibition.

We focused on melanoma and lung carcinoma as exemplaryimmunogenic solid tumor models. Because dysbiosis enhancedprogression in both models, our findings may have a broaderimpact for a variety of non-GI tumors, as it pertains to tumorvasculature modulation and thus could be pertinent for solidtumors in general. An important next step will be defining thecommensal bacterial species, or mixture, and the metabolomeresponsible for maintaining a more competent immunogenictumor microenvironment (39, 49, 50).

Overall, ourwork expands on the increasing appreciation of theimportance of commensal bacteria in the GI tract in maintaininghost immune homeostasis via conditioning of the tumor stroma.It indicates an additional deleterious effect of certain classes ofantibiotics on the host's ability to elicit an effective antitumorimmune response.


DisclaimerThe content is solely the responsibility of the authors and does not neces-

sarily represent the official views of the NIH, NSF or UAMS/WPRCI.

1

10

100

Tum

or g

row

th ra

te

(mg/

d)

*

Ortho OrthoDys Dys

ns

T-cell impaired

Immuno- competent

60 1614121080

500

1,000

Time (days)

DysDys + TNFα

0.0

2.0

4.0

6.0

CD

3 C

D8

(×10

4 )

Dys Dys

*

0.0

2.0

4.0

6.0

CD

54 (×

104 )

Dys Dys+ TNFa + TNFa

*

0 4 6 8 10 120

200

400

600

Time (days)

Tum

or v

olum

e (m

m3 )

Tum

or v

olum

e (m

m3 )

wt

ICAM-1-/-

ICAM-1-/- Dys

A B

C D E

Figure 6.

ICAM-1 and T cells are needed for dysbiotic-induced tumor progression and TNFa supplementation rescues T-cell trafficking under dysbiotic conditions.A,Tumor growth curves of B16-F10 melanoma in ICAM-1�/�mice with and without dysbiotic conditions. B, Tumor growth rates of B16-F10 melanoma inimmunocompetent (C57BL/6J) and T-cell–impaired Foxn1�/�mice with and without dysbiotic conditions. C, Tumor growth curves of B16-F10 melanoma duringdysbiotic conditions with or without TNFa supplementation (every three dayswith a total of 4 doses; 120 mg/kg). Quantification of ICAM-1 (CD54) expressingCD45� CD31þ TECs (D) and tumor-infiltrated CD3þ CD8þ T cells (E) in B16-F10 tumors during dysbiotic conditions with or without TNFa. Data presented asmeans� SEM (n¼ 4 or 5 B16-F10 tumors per group). �, P < 0.05, two-sided t test; ns, not significant. A,r, ICAM-1�/�;~, dysbiotic (Dys) ICAM-1�/�;*, wild-type(wt) mice; C, dysbiotic (Dys) mice (&) and dysbiotic mice (&) supplemented with TNFa (Dysþ TNFa).



Authors' ContributionsConception and design: R.J. Griffin, M.J. Cannon, K.B. Vang, R.P.M. DingsDevelopment of methodology: R.J. Griffin, K.B. Vang, R.P.M. DingsAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): S.V. Jenkins, K.B. Vang, R.P.M. DingsAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): S.V. Jenkins, M.S. Robeson, C.M. Quick, E.R. Siegel,K.B. Vang, R.P.M. DingsWriting, review, and/or revision of the manuscript: S.V. Jenkins,M.S. Robeson, R.J. Griffin, C.M. Quick, E.R. Siegel, M.J. Cannon, K.B. Vang,R.P.M. DingsAdministrative, technical, or material support (i.e., reporting ororganizing data, constructing databases): M.S. Robeson, K.B. Vang,R.P.M. DingsStudy supervision: M.S. Robeson, K.B. Vang, R.P.M. Dings

AcknowledgmentsThe study was supported by P20GM103625: the Center for Microbial

Pathogenesis andHost Inflammatory Responses grant through theNIHNation-al Institute of General Medical Sciences Centers of Biomedical Research Excel-lence (to R.P.M.Dings). Thisworkwas also supported inpart by a grant from theArkansas Biosciences Institute and the Winthrop P. Rockefeller Cancer Institute(to R.P.M. Dings). K.B. Vang received grant NSF OIA-1457888 through theCenter for Advanced Surface Engineering.

The costs of publicationof this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received January 2, 2019; revised August 21, 2019; acceptedOctober 1, 2019;published first October 7, 2019.

References1. Erturk-Hasdemir D, Kasper DL. Resident commensals shaping immunity.

Curr Opin Immunol 2013;25:450–5.2. Cheng M, Qian L, Shen G, Bian G, Xu T, Xu W, et al. Microbiota modulate

tumoral immune surveillance in lung through a gdT17 immune cell-dependent mechanism. Cancer Res 2014;74:4030–41.

3. Willing BP, Russell SL, Finlay BB. Shifting the balance: antibioticeffects on host-microbiota mutualism. Nat Rev Microbiol 2011;9:233–43.

4. Centers for Disease Control and Prevention. Antibiotic use in the UnitedStates, 2017: progress and opportunities. Atlanta, GA: Centers for DiseaseControl and Prevention; 2017.

5. Kilkkinen A, RissanenH, Klaukka T, Pukkala E, HeliovaaraM, Huovinen P,et al. Antibiotic use predicts an increased risk of cancer. Int J Cancer 2008;123:2152–5.

6. Gagliani N, Hu B, Huber S, Elinav E, Flavell RA. The fire within: microbesinflame tumors. Cell 2014;157:776–83.

7. Grivennikov SI, Wang K, Mucida D, Stewart CA, Schnabl B, Jauch D, et al.Adenoma-linked barrier defects andmicrobial products drive IL-23/IL-17-mediated tumour growth. Nature 2012;491:254–8.

8. Wu S, Rhee KJ, Albesiano E, Rabizadeh S, Wu X, Yen HR, et al. A humancolonic commensal promotes colon tumorigenesis via activation of Thelper type 17 T cell responses. Nat Med 2009;15:1016–22.

9. Springer TA. Traffic signals for lymphocyte recirculation and leukocyteemigration: the multistep paradigm. Cell 1994;76:301–14.

10. Naik S, Bouladoux N, Wilhelm C, Molloy MJ, Salcedo R, Kastenmuller W,et al. Compartmentalized control of skin immunity by resident commen-sals. Science 2012;337:1115–9.

11. Dings RP, Vang KB, Castermans K, Popescu F, Zhang Y, Oude EgbrinkMG,et al. Enhancement of T-cell-mediated antitumor response: angiostaticadjuvant to immunotherapy against cancer. Clin Cancer Res 2011;17:3134–45.

12. Ichinohe T, Pang IK, Kumamoto Y, Peaper DR, Ho JH, Murray TS,et al. Microbiota regulates immune defense against respiratory tractinfluenza A virus infection. Proc Natl Acad Sci U S A 2011;108:5354–9.

13. Jenkins SV, Vang KB, Gies A, Griffin RJ, Jun SR, Nookaew I, et al. Samplestorage conditions induce post-collection biases in microbiome profiles.BMC Microbiol 2018;18:227.

14. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA,et al. Reproducible, interactive, scalable and extensible microbiome datascience using QIIME 2. Nat Biotechnol 2019;37:852–7.

15. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP.DADA2: High-resolution sample inference from Illumina amplicon data.Nat Methods 2016;13:581–3.

16. Nguyen LT, Schmidt HA, von Haeseler A, Minh BQ. IQ-TREE: a fast andeffective stochastic algorithm for estimating maximum-likelihood phy-logenies. Mol Biol Evol 2015;32:268–74.

17. Kalyaanamoorthy S, Minh BQ, Wong TKF, von Haeseler A, Jermiin LS.ModelFinder: fast model selection for accurate phylogenetic estimates.Nat Methods 2017;14:587–9.

18. R Core Team. R: A Language and Environment for Statistical Computing.Vienna, Austria: R Foundation for Statistical Computing; 2014.

19. BurchillMA, Yang J, Vang KB,Moon JJ, ChuHH, LioCW, et al. Linked T cellreceptor and cytokine signaling govern the development of the regulatory Tcell repertoire. Immunity 2008;28:112–21.

20. Dings RP, Chen X, Hellebrekers DM, van Eijk LI, Zhang Y, Hoye TR, et al.Design of nonpeptidic topomimetics of antiangiogenic proteins withantitumor activities. J Natl Cancer Inst 2006;98:932–6.

21. Iida N, Dzutsev A, Stewart CA, Smith L, BouladouxN,Weingarten RA, et al.Commensal bacteria control cancer response to therapy bymodulating thetumor microenvironment. Science 2013;342:967–70.

22. Viaud S, Saccheri F, Mignot G, Yamazaki T, Daillere R, Hannani D, et al.The intestinal microbiota modulates the anticancer immune effects ofcyclophosphamide. Science 2013;342:971–6.

23. Faith JJ, Guruge JL, CharbonneauM, Subramanian S, Seedorf H, GoodmanAL, et al. The long-term stability of the human gut microbiota. Science2013;341:1237439.

24. Vazquez-Baeza Y, Callewaert C, Debelius J, Hyde E, Marotz C, Morton JT,et al. Impacts of the human gut microbiome on therapeutics. Annu RevPharmacol Toxicol 2018;58:253–70.

25. Antonopoulos DA,Huse SM,MorrisonHG, Schmidt TM, SoginML, YoungVB. Reproducible community dynamics of the gastrointestinal microbiotafollowing antibiotic perturbation. Infect Immun 2009;77:2367–75.

26. Becattini S, Taur Y, Pamer EG. Antibiotic-induced changes in the intestinalmicrobiota and disease. Trends Mol Med 2016;22:458–78.

27. Uptodate.com. Drug information - Dosing; 2018.28. Ferrier L, Berard F, Debrauwer L, Chabo C, Langella P, Bueno L, et al.

Impairment of the intestinal barrier by ethanol involves enteric microfloraand mast cell activation in rodents. Am J Pathol 2006;168:1148–54.

29. Moellering RC Jr. Pharmacokinetics of vancomycin. J Antimicrob Che-mother 1984;14:43–52.

30. Lofmark S, Edlund C, Nord CE.Metronidazole is still the drug of choice fortreatment of anaerobic infections. Clin Infect Dis 2010;50:S16–23.

31. Buffie CG, Bucci V, Stein RR, McKenney PT, Ling L, Gobourne A, et al.Precisionmicrobiome reconstitution restores bile acidmediated resistanceto Clostridium difficile. Nature 2015;517:205–8.

32. Kroemer G, Zitvogel L.Cancer immunotherapy in 2017: The breakthroughof the microbiota. Nat Rev Immunol 2018;18:87–8.

33. Vetizou M, Pitt JM, Daillere R, Lepage P, Waldschmitt N, Flament C, et al.Anticancer immunotherapy by CTLA-4 blockade relies on the gut micro-biota. Science 2015;350:1079–84.

34. Sivan A, Corrales L, Hubert N, Williams JB, Aquino-Michaels K, Earley ZM,et al. Commensal Bifidobacterium promotes antitumor immunity andfacilitates anti-PD-L1 efficacy. Science 2015;350:1084–9.

35. Chaput N, Lepage P, Coutzac C, Soularue E, Le Roux K, Monot C, et al.Baseline gut microbiota predicts clinical response and colitis in metastaticmelanoma patients treated with ipilimumab. Ann Oncol 2017;28:1368–79.

36. Routy B, Le Chatelier E, Derosa L, Duong CPM, Alou MT, Daillere R, et al.Gut microbiome influences efficacy of PD-1-based immunotherapyagainst epithelial tumors. Science 2018;359:91–7.

37. Gopalakrishnan V, Spencer CN, Nezi L, Reuben A, AndrewsMC, KarpinetsTV, et al. Gut microbiomemodulates response to anti-PD-1 immunother-apy in melanoma patients. Science 2018;359:97–103.

Jenkins et al.


Uptodate.com

38. Matson V, Fessler J, Bao R, Chongsuwat T, Zha Y, Alegre ML, et al. Thecommensal microbiome is associated with anti-PD-1 efficacy inmetastaticmelanoma patients. Science 2018;359:104–8.

39. Tanoue T, Morita S, Plichta DR, Skelly AN, Suda W, Sugiura Y, et al. Adefined commensal consortium elicits CD8 T cells and anti-cancer immu-nity. Nature 2019;565:600–5.

40. Savage DC, Dubos R. Alterations in the mouse cecum and its floraproduced by antibacterial drugs. J Exp Med 1968;128:97–110.

41. Dunn GP, Bruce AT, Ikeda H, Old LJ, Schreiber RD. Cancer immunoe-diting: from immunosurveillance to tumor escape. Nat Immunol 2002;3:991–8.

42. Khong HT, Restifo NP. Natural selection of tumor variants in the gener-ation of "tumor escape" phenotypes. Nat Immunol 2002;3:999–1005.

43. Lizee G, CantuMA, Hwu P. Less yin, more yang: confronting the barriers tocancer immunotherapy. Clin Cancer Res 2007;13:5250–5.

44. Sakaguchi S. Naturally arising Foxp3-expressing CD25þCD4þ regulatoryT cells in immunological tolerance to self and non-self. Nat Immunol2005;6:345–52.

45. Czystowska M, Gooding W, Szczepanski MJ, Lopez-Abaitero A, Ferris RL,Johnson JT, et al. The immune signature of CD8(þ)CCR7(þ) T cells in theperipheral circulation associates with disease recurrence in patients withHNSCC. Clin Cancer Res 2013;19:889–99.

46. Kalliolias GD, Ivashkiv LB. TNF biology, pathogenic mechanisms andemerging therapeutic strategies. Nat Rev Rheumatol 2016;12:49–62.

47. Schnoor M, Alcaide P, Voisin MB, van Buul JD. Crossing the vascular wall:common and uniquemechanisms exploited by different leukocyte subsetsduring extravasation. Mediat Inflamm 2015;2015:946509.

48. Celli L, Ryckewaert JJ, Delachanal E, Duperray A. Evidence of a functionalrole for interaction between ICAM-1 and nonmuscle alpha-actinins inleukocyte diapedesis. J Immunol 2006;177:4113–21.

49. Weir TL, Manter DK, Sheflin AM, Barnett BA, Heuberger AL, Ryan EP. Stoolmicrobiome and metabolome differences between colorectal cancerpatients and healthy adults. PLoS One 2013;8:e70803.

50. Luu M, Weigand K, Wedi F, Breidenbend C, Leister H, Pautz S, et al.Regulation of the effector function of CD8(þ) T cells by gut microbiota-derived metabolite butyrate. Sci Rep 2018;8:14430.



Research Article

The Microbiome in Lung Cancer Tissue andRecurrence-Free SurvivalBrandilyn A. Peters1, Richard B. Hayes1,2, Chandra Goparaju3, Christopher Reid3,Harvey I. Pass2,3, and Jiyoung Ahn1,2

Abstract

Background:Humanmicrobiota have many functions thatcould contribute to cancer initiation and/or progression atlocal sites, yet the relation of the lung microbiota to lungcancer prognosis has not been studied.

Methods: In a pilot study, 16S rRNA gene sequencing wasperformed on paired lung tumor and remote normal samplesfrom the same lobe/segment in 19 patients with non–smallcell lung cancer (NSCLC). We explored associations of tumoror normal tissue microbiome diversity and composition withrecurrence-free (RFS) and disease-free survival (DFS), andcompared microbiome diversity and composition betweenpaired tumor and normal samples.

Results: Higher richness and diversity in normal tissuewere associated with reduced RFS (richness P ¼ 0.08,Shannon index P ¼ 0.03) and DFS (richness P ¼ 0.03,Shannon index P ¼ 0.02), as was normal tissue overallmicrobiome composition (Bray–Curtis P ¼ 0.09 for RFS

and P ¼ 0.02 for DFS). In normal tissue, greater abundanceof family Koribacteraceae was associated with increased RFSand DFS, whereas greater abundance of families Bacteroi-daceae, Lachnospiraceae, and Ruminococcaceae were asso-ciated with reduced RFS or DFS (P < 0.05). Tumor tissuediversity and overall composition were not associated withRFS or DFS. Tumor tissue had lower richness and diversity(P � 0.0001) than paired normal tissue, though overallmicrobiome composition did not differ between the pairedsamples.

Conclusions: We demonstrate, for the first time, a poten-tial relationship between the normal lung microbiota andlung cancer prognosis, which requires confirmation in alarger study.

Impact: Definition of bacterial biomarkers of prognosismay lead to improved survival outcomes for patients withlung cancer.

IntroductionLung cancer is the most common cancer, excluding nonmela-

noma skin cancer, and the most common cause of cancer-relateddeath in theworld, with approximately 1.8million diagnoses and1.6 million deaths per year (1). Although incidence rates for lungcancer have been declining in the United States due to reductionsin smoking, challenges in early detection have left lung cancer asthe leading cause ofU.S. cancer-related deaths (5-year survival rate18%, on average, in the United States; ref. 2). Non–small cell lungcancer (NSCLC), the most common form of lung cancer, istypically treated at the early stages with surgical resection, withor without chemotherapy or chemoradiotherapy (3). These early-stage cancers have better 5-year survival rates (50%–90%), how-ever a substantial proportion of patients still die of diseaserecurrence (4). Improvements in early detection with low-doseCT (5) will inevitably increase the identification of early-stage

lung cancers and offer more opportunities for curative resection,making it extremely timely to investigate factors contributing tolong-term disease-free survival (DFS) following resection. Betteridentification of patients with early-stage at high risk of recurrencecan improve survival by indicating which patients may benefitfrom increased surveillance and/or adjuvant therapy.

The healthy human lung is host to a unique and dynamicbacterial community, determined by bidirectional movement ofnonsterile air and mucus in and out of the airways (6). In lungdisease, regional changes in the lung environment create permis-sive niches for bacterial growth, resulting in significant differencesin community composition between healthy and diseasedlungs (7). Studies have explored the oral or airway microbiomein lung cancer cases compared with controls (8–12), noting lowermicrobial diversity and altered abundance of specific bacterialgroups in cases. However, few studies have characterized themicrobiome in lung tumor tissue (13, 14), and no studies haveexplored the relationship between the microbiome of resectedlung tissue and lung cancer prognosis. Bacteria have many func-tions that could contribute to cancer initiation and/or progressionat local sites, including genotoxic pathways, bacterial metabolitesignaling, and induction of host inflammatory pathways (15).Investigation of potential bacterial involvement in lung cancerprognosis may lead to new biomarkers and therapies to improvesurvival outcomes for patients with lung cancer.

We conducted a pilot study of paired tumor and remote normallung tissue samples from19patientswithNSCLCatNYULangoneHealth (New York, NY), 17 of them with prospective follow up.Using 16S rRNA gene sequencing, we exploredwhether the tumor

1Department of Population Health, NYU School of Medicine, New York, NewYork. 2NYU Perlmutter Cancer Center, New York, New York. 3Department ofCardiothoracic Surgery, NYU School of Medicine, New York, New York.

Note: Supplementary data for this article are available at Cancer Epidemiology,Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).

Corresponding Author: Jiyoung Ahn, New York University School of Medicine,650 First Ave, New York, NY, 10016. Phone: 212-263-3390; Fax: 301-496-6829;E-mail: [email protected]

doi: 10.1158/1055-9965.EPI-18-0966


CancerEpidemiology,Biomarkers& Prevention


http://cebp.aacrjournals.org/cgi/content/full/1055-9965.EPI-18-0966?ijkey=zdw9AIfVM.YSc&keytype=ref&siteid=aacrjnls

or normal lung microbiome was associated with recurrence-freesurvival (RFS) and DFS, and compared the lung microbiome ofpaired tumor and normal samples.

Materials and MethodsPatients and sample collection

Samples were selected from the NYU Thoracic Surgery Archives(NTSA). Established in 2006, the NTSA has prospectively collect-ed serum, plasma, buffy coat, peripheral blood mononuclearcells, along with lung cancer and matching normal lung speci-mens under the Institutional Review Board–approved 8896 pro-tocol. Patients identified on preoperative workup as having apulmonary nodule suspicious for lung cancer were consented forcollection of blood and snap frozen tissues (tumor and remotelung from the same lobe/segment) in the operating room at thetime of their resection. Lung and matching tumor are sterilely cutat the operating room table, transferred to prelabeled nunc vialsand immediately snap frozen in liquid nitrogen within 10 min-utes of resection. Samples are deidentified for storage at �80�Cuntil use. Because these specimens remain sterile and are imme-diately frozen, they are ideal for microbiome analysis, as imme-diate freezing does not impact microbiome composition (16).Less is known regarding long-term (i.e., years) storage at �80�C,which may impact certain aspects of the microbiome (17, 18);however, the length of storage time in our samples was notassociated with overall microbiome diversity and composition(a- and b-diversity).

Clinical and pathologic demographics are recorded in anencrypted Research Electronic Data Capture spreadsheet. Patientsare seen at 3-month intervals for 2 years, and then at 6-monthintervals for 1 year, and then annually, with CT scans performedfor surveillance to document any systemic and loco-regionalrecurrences, or the development of a second primary tumor. The19 patients' samples included in this study were originally chosenas pilot samples to test whether sufficient material was availablefor DNA extraction from tumor and matching normal lung. Thesamples were also chosen to represent patients with differentstages of NSCLC and patients with recurrence, to explore the lungmicrobiome in relation to these factors.

DefinitionsEndpoints were defined according to the consensus agreement

in Punt and colleagues (19). DFS includes recurrences (loco-regional and systemic), new primaries (same or other cancer),and death from any cause as events. RFS includes recurrences(loco-regional and systemic) and death from any cause as events,ignoringnewprimaries as events. For both endpoints, person timeis defined as time from surgery to event or loss to follow-up(censored).

Microbiome assayLung tissue samples underwent 16S rRNA gene sequencing at

the Environmental Sample Preparation and Sequencing Facility atArgonne National Laboratory. DNA extraction and amplificationsteps occurred in twobatches (batch 1: 10 samples andbatch 2: 28samples; tumor–normal pairs from same patient kept together),but all samples were sequenced in the same batch. DNA wasextracted from tissue using the Mo Bio PowerSoil DNA IsolationKit, following the manufacturer's protocol. This protocoluses mechanical bead beating and chemical methods to achieve

sample homogenization and cell lysis, ensuring that sample fea-tures do not interfere with DNA extraction. The V4 region of the16S rRNA gene was PCR amplified with the 515F/806R primerpair, which included sequencer adapter sequences used inthe Illumina flowcell and sample-specific barcodes (20, 21). Each25-mL PCR reaction contained 9.5 mL of Mo Bio PCR Water(Certified DNA-Free), 12.5 mL of QuantaBio's AccuStart II PCRToughMix (2� concentration, 1� final), 1 mL Golay barcode-tagged Forward Primer (5 mmol/L concentration, 200 pmol/Lfinal), 1 mL Reverse Primer (5 mmol/L concentration, 200 pmol/L final), and1 mL of templateDNA. The conditions for PCRwere asfollows: 94�C for 3 minutes to denature the DNA, with 35 cyclesat 94�C for 45 seconds, 50�C for 60 seconds, and 72�C for90 seconds; with a final extension of 10 minutes at 72�C. PCRproducts were quantified using PicoGreen (Invitrogen) and a platereader (Infinite 200 PRO, Tecan). Each batch included two extrac-tion blanks and 10 amplification blanks, all of which did notamplify. In addition, amplification levels for samples were in thesame range for both batches. Sample PCR products werethen pooled in equimolar amounts, purified using AMPureXP Beads (Beckman Coulter), and then quantified using afluorometer (Qubit, Invitrogen). Molarity was then diluted to2 nmol/L, denatured, and then diluted to a final concentrationof 6.75 pmol/L with a 10% PhiX spike for sequencing on theIlluminaMiSeq. Ampliconswere sequencedona151bp�12bp�151 bp MiSeq run (21).

Sequence read processingSequence reads were processed using QIIME 2 (22). Briefly,

sequence reads were demultiplexed and paired-end reads werejoined, followed by quality filtering as described in Bokulich andcolleagues (23). Next, the Deblur workflow was applied, whichuses sequence error profiles to obtain putative error-freesequences, referred to as "sub" operational taxonomic units(s-OTU; ref. 24). s-OTUs were assigned taxonomy using a na€�veBayes classifier pretrained on the Greengenes (25) 13_8 99%OTUs, where the sequences have been trimmed to only include250 bases from the 16S V4 region, bound by the 515F/806Rprimer pair. A phylogenetic tree was constructed via sequencealignmentwithMAFFT (26),filtering the alignment, and applyingFastTree (27) to generate the tree. One tumor sample withoutdetectable s-OTUs was dropped, leaving 37 samples (19 normal,18 tumor) from 19 patients for final analysis. The number ofsequence reads per sample prior to the Deblur workflow wassimilar in tumor compared with the normal tissue samples(Wilcoxon signed-rank P¼ 0.61), and marginally higher in batch1 compared with the batch 2 (Wilcoxon rank-sum P ¼ 0.08;Supplementary Fig. S1). Because of the amplification of humanmitochondrial DNA in these tissue samples, the majority ofsequence reads belonged to the human mitochondria and weredropped when not matching to the bacterial 16S database duringDeblur. Thenumber of sequence reads per sample after theDeblurworkflow was marginally lower in tumor compared with normaltissue samples (Wilcoxon signed-rank P ¼ 0.09), and higher inbatch 1 compared with batch 2 (Wilcoxon rank-sum P ¼ 0.001;Supplementary Table S1; Supplementary Fig. S1).

a-diversitya-diversity (within-samplemicrobiome diversity) was assessed

using richness (number of s-OTUs) and the Shannon diversityindex, calculated in 100 iterations for rarefied s-OTU tables

Peters et al.

Cancer Epidemiol Biomarkers Prev; 28(4) April 2019 Cancer Epidemiology, Biomarkers & Prevention732

[(63 sequence reads per sample (lowest sequencing depth amongsamples)] using the QIIME 2 diversity plugin. Rarefaction curvessuggested that this depth reflected the general ranking of com-munity richness and diversity of the samples (SupplementaryFig. S2). We used Cox proportional hazard models to determinewhether a-diversity was associated with RFS and DFS. We exam-ined whether a-diversity differed between paired tumor andnormal samples using the Wilcoxon signed-rank test.

b-diversityb-diversity (between-sample microbiome diversity) was

assessed using unweighted and weighted UniFrac distances (28),the Bray–Curtis dissimilarity, and the Jaccard index. Principalcoordinate analysis (29) was used for visualization. Thecommunity-level test of association between the microbiota andsurvival times (MiRKAT-S; ref. 30) and the optimal microbiome-based survival analysis test (OMiSA; ref. 31) were used to test theassociation of overall bacterial composition with RFS and DFS.We also assigned samples to clusters by applying Ward Hierar-chical Agglomerative Clustering method (32) to the distancematrices, and then tested whether these clusters were relatedRFS and DFS using log-rank tests. Permutational multivariateANOVA (33) was used to examine statistically whether overallbacterial composition differed between paired tumor and normalsamples, using patient ID as strata. We also compared between-pair distances in overall bacterial composition for tumor andnormal tissue sample pairs with distances for all possible pairingsof tumor and normal samples from different subjects (i.e., truepairs vs. not-true pairs) using the Wilcoxon rank-sum test, todetermine whether true sample pairs are more similar to eachother than random pairings. These analyses were performed withandwithout rarefying s-OTU tables to an evendepth (63 sequencereads per sample), as b-diversity can be sensitive to sequencingdepth (34).

Differential abundanceRelative abundance of s-OTUs (total sum scaling) was calcu-

lated, and s-OTUs were agglomerated to phylum, class, order,family, genus, and species levels. We filtered taxa to include inanalysis only those present in 25% of the samples. We usedCox proportional hazard models to assess whether taxa centeredlog ratio (clr)-transformed (35, 36) abundance or carriage was

associated with RFS and DFS. We used the Wilcoxon signed-ranktest and McNemar test to assess differences in taxon relativeabundance and carriage, respectively, between paired tumorand normal samples. P values were adjusted for the falsediscovery rate.

Sensitivity analysesWe checked whether results for overall a-diversity and

b-diversity were consistent when restricting to adenocarcinomacases only, restricting to patients in the larger extraction batch(batch 2), excluding stage III and IV cases, excluding currentsmokers, and excluding samples with low sequencing depths(�124 reads/sample). We did not perform analyses within otherhistology groups or the smaller extraction batch due to smallsample size (n ¼ 4 squamous cell carcinoma, n ¼ 1 sarcomatoidcarcinoma, n ¼ 5 in batch 1).

ResultsPatient characteristics

Demographic and clinical characteristics of the 19 patients arepresented in Table 1. The average patient age was 71.6 years old,and 37% were male, 100% were white, and 95% formerly orcurrently smoked. The majority of patients had lung adenocarci-nomas (74%), whereas a minority had other histologic types(squamous cell carcinoma 21%; sarcomatoid carcinoma 5%).Two patients with no follow-up due to postoperative deathwere excluded from survival analysis; of the remaining 17patients, 3 had new primaries and 9 had recurrences (loco-regional or systemic) during follow-up (follow-up times rangedfrom 1–12 years).

Normal lung tissue microbiome diversity and composition isassociated with RFS and DFS

Patients with recurrence or a newprimary during follow-up hadgreater bacterial richness (P ¼ 0.01) and diversity (P ¼ 0.06), intheir normal lung tissue, than disease-free patients, at the evenlyrarefied depth of 63 sequences per sample (Fig. 1; SupplementaryTable S2). Consistently, higher richness and diversity in normaltissue were significantly associated with reduced RFS and DFSin Cox proportional hazard models (RFS: P ¼ 0.08 for richness,P¼ 0.03 for Shannon index; DFS: P¼ 0.03 for richness, P¼ 0.02

Table 1. Characteristics of 19 patients with lung cancer

All Disease-free Recurrence New primaryCharacteristic (n ¼ 19) (n ¼ 5) (n ¼ 9) (n ¼ 3)

Age, mean � SD 71.6 � 6.7 73.6 � 6.3 73.6 � 6.3 64.3 � 6.3Male, n (%) 7 (36.8) 3 (60.0) 2 (22.2) 0 (0)White, n (%) 19 (100.0) 5 (100.0) 9 (100.0) 3 (100.0)Smoking status, n (%)Never 1 (5.3) 1 (20.0) 0 (0) 0 (0)Former 16 (84.2) 4 (80.0) 8 (88.9) 2 (66.7)Current 2 (10.5) 0 (0) 1 (11.1) 1 (33.3)

Histology, n (%)Adenocarcinoma 14 (73.7) 4 (80.0) 9 (100.0) 1 (33.3)Squamous cell carcinoma 4 (21.1) 0 (0) 0 (0) 2 (66.7)Sarcomatoid carcinoma 1 (5.3) 1 (20.0) 0 (0) 0 (0)

Stage, n (%)I 10 (52.6) 3 (60.0) 4 (44.4) 2 (66.7)II 5 (26.3) 2 (40.0) 2 (22.2) 1 (33.3)III 2 (10.5) 0 (0) 1 (11.1) 0 (0)IV 2 (10.5) 0 (0) 2 (22.2) 0 (0)

Lung Microbiome and Lung Cancer Recurrence

www.aacrjournals.org Cancer Epidemiol Biomarkers Prev; 28(4) April 2019 733

for Shannon index; Supplementary Table S2). Results remainedlargely consistent in the sensitivity analyses (SupplementaryTable S3).

Overall microbiome composition in normal lung tissue wasassociated with RFS and DFS according to several distancemeasures with the MiRKAT-S test (RFS P � 0.09 and DFSP � 0.04 for unweighted and weighted UniFrac distances,Bray–Curtis dissimilarity, and Jaccard index; SupplementaryTable S4), though not with the OMiSA test (RFS P ¼ 0.20,DFS P ¼ 0.12). Results were similar in the sensitivity analyses(Supplementary Table S3). Results were also similar whenrarefying to an even depth for the UniFrac distances, butsomewhat attenuated for the Bray–Curtis dissimilarity andJaccard index (Supplementary Table S4). Principal coordinateanalysis of the Bray–Curtis dissimilarity from the normal tissuerevealed clustering of patients by recurrence status (Fig. 2A andB); results were similar for the unweighted and weightedUniFrac distances and the Jaccard index (Supplementary Fig.S3). We grouped patients into four discrete clusters based onthe Bray–Curtis dissimilarity in normal tissue (Fig. 2C), andobserved that these clusters were significantly related to RFSand DFS as well (RFS P ¼ 0.03, DFS P ¼ 0.015; Fig. 2D and E).

We observed several taxa in normal tissue for which relativeabundance and/or carriage were associated with both RFSand DFS in Cox proportional hazard models at P < 0.05(Supplementary Table S5; Fig. 3); these taxa were not significantafter FDR adjustment. Greater abundance of family Koribacter-aceae in normal tissuewas associatedwith increased RFS andDFS,whereas greater abundance of family Lachnospiraceae, and generaFaecalibacterium and Ruminococcus (from Ruminococcaceae fam-ily), and Roseburia and Ruminococcus (from Lachnospiraceae fam-ily) were associated with reduced RFS and DFS. Taxa associatedonly with RFS (P < 0.05) included family S24-7 (increased RFS),and family Bacteroidaceae and genus Bacteroides (reduced RFS).Taxa associated only with DFS (P < 0.05) included family Sphin-gomonadaceae and genus Sphingomonas (increased DFS), andfamily Ruminococcaceae (reduced DFS). A heatmap of these12 taxa in normal tissue clustered patients somewhat by recur-rence status (Fig. 3).

Lung tumor tissue microbiome is not associated with survivalTumor tissue richness and diversity were not associated with

recurrence status or with RFS and DFS (Supplementary Fig. S4;Supplementary Table S2), and thiswas consistent in the sensitivity

Figure 1.

a-diversity in normal lung tissue and survival. A, Distribution of number of OTUs at an even depth of 63 sequence reads per sample in normal lung tissue byrecurrence status of patients (P values are from Kruskal–Wallis tests). B and C, RFS and DFS curves for patients grouped in tertiles of number of OTUs at an evendepth of 63 sequence reads per sample in normal lung tissue (P values are from log-rank tests for trend). D, Distribution of the Shannon index at an even depth of63 sequence reads per sample in normal lung tissue by recurrence status of patients (P values are from Kruskal–Wallis tests). E and F, RFS and DFS survivalcurves for patients grouped in tertiles of the Shannon index at an even depth of 63 sequence reads per sample in normal lung tissue (P values are from log-ranktests for trend).

Peters et al.


Figure 2.

b-diversity in normal lung tissue and survival. Principal coordinate analysis of the Bray–Curtis dissimilarity, with samples annotated according to recurrencestatus, histology, and person days: nonrarefied (A), rarefied to an even depth of 63 sequence reads per sample (B). C, Unsupervised clustering (ward.D2method) of the Bray–Curtis dissimilarity grouped patients into four clusters. These clusters were significantly related to RFS (D; log-rank P¼ 0.031) and DFS(E; log-rank P¼ 0.015).



analyses (Supplementary Table S3). In addition, tumor overallmicrobiome composition was not associated with RFS or DFS(Supplementary Table S4; Supplementary Fig. S5), and this wasconsistent when rarefying to an even depth and in the sensitivityanalyses (Supplementary Table S3; Supplementary Table S4). Intumor tissue, only families Koribacteraceae and Lachnospiraceaewere associated with reduced RFS and DFS (P < 0.05; Supple-mentary Table S5).

Lung tumor tissue microbiome is less diverse than, butcompositionally similar to, paired normal tissue microbiome

Tumor tissue samples had significantly lower bacterial richness(observed OTUs; P ¼ 0.0001) and diversity (Shannon index;P < 0.0001) than paired normal tissue samples at the evenlyrarefied depth of 63 sequences per sample (Fig. 4; SupplementaryTable S6). Significance remained at higher sequencing depthsdespite dropped samples with lower depths (15 tumor/normalpairs at 124 sequence reads per sample: P ¼ 0.02 for number ofOTUs, P < 0.0001 for Shannon index). Results were consistentwhen restricting to adenocarcinoma histology, restricting topatients in batch 2, excluding stage III and IV cases, or excludingcurrent smokers (P < 0.05).

Overall microbiome composition did not differ significantlybetween paired lung tumor and normal samples according tounweighted and weighted UniFrac distance, Bray–Curtis dissim-ilarity, or the Jaccard index (Supplementary Fig. S6; Supplemen-tary Table S6). Results were consistent when rarefying to an evendepth and when restricting to adenocarcinoma histology, restrict-ing to patients in batch 2, excluding stage III and IV cancers,excluding current smokers, or excluding samples with lowsequencing depth. Moreover, paired tumor and normal sampleswere significantly more alike than random pairings of tumor andnormal samples from different patients, according to theunweighted and weighted UniFrac distance, Bray–Curtis dissim-ilarity, and Jaccard index (all P� 0.02). Lung tumor samples had

higher abundance of family Veillonellaceae, lower abundanceof genus Cloacibacterium, and lower carriage of family Erysipe-lotrichaceae, than paired normal samples (P < 0.05; Fig. 5;Supplementary Table S7); these taxa were not significant afterFDR adjustment.

DiscussionIn this pilot study of the lung microbiome and lung cancer

prognosis, we showed, for the first time, that increased diversityand altered composition of the normal lung tissue was associatedwith reduced DFS and RFS. This important novel observationsuggests that the microbiome of normal lung tissue may be usedas a biomarker of lung cancer prognosis, which could guideclinical practice to improve survival outcomes for patients withlung cancer. We also observed a clear reduction in bacterialrichness and diversity in lung tumor samples compared withpaired normal tissue samples, indicating dysbiosis of the lungtumor microbiome.

Few studies have reported on the microbiome in lung cancer,andeven fewer characterized themicrobiome in actual lung tumortissue.Wehave reported that lower airwaybrushesof patientswithlung cancer (n¼ 39) were enriched in Veillonella and Streptococcuscompared with patients with benign lung disease (n ¼ 36) andhealthy controls (n¼ 10; ref. 9). A study of lung cancer attributedto household coal burning in China found that sputum samplesof lung cancer cases (n¼ 8) had lower diversity and enrichment ofGranulicatella, Abiotrophia, and Streptococcus compared withhealthy controls (n ¼ 8; ref. 10). Similarly, another study fromChina reported decreased diversity and increased Streptococcusabundance in bronchial brush specimens from cancerous lungsites compared to paired noncancerous lung sites (n ¼ 24) andhealthy controls (n¼ 18; ref. 8). A third report from China foundfamily Veillonellaceae and genera Veillonella, Capnocytophaga, andSelenomonas were more abundant in saliva of patients with lung

Figure 3.

Taxa in normal lung tissue associated with RFS or DFS.Heatmap shows relative abundance of families andgenera (F, family; G, genus) with P� 0.05 from Coxproportional hazard models of clr-transformedabundance or carriage (Supplementary Table S4).Heatmap was generated with average linkageclustering and the Manhattan distance method;samples are annotated with recurrence status andBray–Curtis cluster (from Fig. 2).

Peters et al.


cancer (n¼ 20) compared with controls (n¼ 10; ref. 12). A studyin Korea observed that Veillonella and Megasphaera were moreabundant in bronchoalveolar lavage fluid frompatients with lungcancer (n ¼ 20) compared with patients with benign lung mass-like lesions (n ¼ 8; ref. 11). A study of Italian patients with lungcancer found lower bacterial diversity in lung tumor tissue sam-ples (n¼ 31) comparedwith nonmalignant lung tissue (n¼ 165),and no differences in overall composition (b-diversity) betweenthe tumor and nonmalignant samples (13). Finally, a recent studyof lung tissue samples from patients with lung cancer (tumor andadjacent normal) and hospital controls observed increased bac-terial diversity in tumor and adjacent normal tissue from patientswith lung cancer compared with the controls (14).

From this previous literature, it is apparent that the airway andlungmicrobiome is perturbed in patients with lung cancer, whichmay have implications for prognosis. We observed that greaterbacterial diversity and greater abundance of families Bacteroi-daceae, Lachnospiraceae, and Ruminococcaceae, and generaBacteroides, Faecalibacterium, Roseburia, (Ruminococcus), and Rumi-nococcus in normal lung tissue were associated with reducedsurvival, whereas greater abundance of Koribacteraceae and

Sphingomonadaceae were associated with increased survival.Interestingly, the majority of our findings were similar for theRFS and DFS outcomes; this may suggest that the normal lungmicrobiome is related to both recurrences and new primarycancers. Members of Lachnospiraceae and Ruminococcaceae,particularly Roseburia and Faecalibacterium, are known to produceantiinflammatory short-chain fatty acids (e.g., butyrate; ref. 37),making the association of these bacteria with reduced survivalunexpected. Bacteroides abundance in the gut has been associatedwith impaired antitumor immune responses in patients withmelanoma (38), and may play a similar cancer-promoting rolein the lungs. Though our conclusions are limited by small samplesize, these valuable preliminary results suggest that bacteria inresected normal lung tissuemay serve as biomarkers of recurrencerisk in early-stageNSCLC.Moreover, if these identifiedmicrobiotaare determined to be causally related to cancer recurrence in futureinvestigations, they may serve as novel targets for therapeuticintervention (7) to improve RFS in patients with lung cancer.

The results of our analysis comparing paired tumor and normalsamples are similar to the previous literature in that we observedsignificant reductions in bacterial diversity and enrichment of

Figure 4.

a-diversity in relation to lung tissue type (tumor vs. normal). A, Number of OTUs for tumor/normal pairs by patient ID at an even depth of 63 sequencereads per sample. B, Distribution of number of OTUs at 63 sequence reads per sample for normal and tumor samples (P values are fromWilcoxon signed-ranktest). C, Shannon index for tumor/normal pairs by patient ID at an even depth of 63 sequence reads per sample. D, Distribution of the Shannon index at 63sequence reads per sample for normal and tumor samples (P values are fromWilcoxon signed-rank test).



Veillonellaceae in lung tumor compared with normal lung tissue,which has been observed bymany (8–13), but not all (14) studiescomparing lung cancer cases to controls. It is not clear from ourobservational study whether the identified bacterial differencesare causally related to lung carcinogenesis, or are merely reflectiveof disease processes in the lung. However, there are severalmechanisms by which the lung microbiota could contribute tolung carcinogenesis, including genotoxic pathways, bacterialmetabolite effects, and induction of host inflammatory path-ways (15). For example, intranasal administration of lipopoly-saccharide (a membrane component of Gram-negative bacteria)in a mouse model of lung cancer significantly enhanced pulmo-nary inflammation and lung tumorigenesis (39). We previouslyshowed in a human study that airway Veillonella and Streptococcuswere associated with upregulation of ERK and PI3K signalingpathways in the airway, pathways regulating cell proliferation,survival, and differentiation, which are upregulated in patients

with lung cancer (9). Interestingly, we have previously reportedthat these two genera are enriched in the mouths of currentsmokers compared with never smokers (40), suggesting a furthermechanismbywhich smoking causes lung cancer. Taken together,there is an accumulating support for specific bacteria as biomar-kers of lung cancer presence; further study of the causal role ofthese bacteria in lung carcinogenesis may provide therapeutictargets for lung cancer prevention.

In summary,we showed in a small pilot study that diversity andcomposition of the normal lung tissue microbiome may beassociated with RFS and DFS, and observed differential micro-biome signatures between lung tumor andnormal tissue thatwereconsistent with previous research. The strengths of our studyinclude the availability of fresh-frozen tumor and normal lungtissue for paired analysis, and prospective long-term follow-up forsurvival analysis. However, our study conclusions were limited bysmall sample size and lack of a replication dataset, and therefore

Figure 5.

Taxa associated with lung tissue type (tumor vs. normal). Heatmap shows relative abundance of families (A), genera (B), and species (C) in paired normal (N)and tumor (T) samples (only taxa present in >25% of samples are shown). Normal and tumor samples are sorted left to right by patient ID. Taxa with � indicateP < 0.05 fromWilcoxon signed-rank test for pair difference in relative abundance or McNemar test for pair difference in carriage (Supplementary Table S6).

Peters et al.


findings will require confirmation in a larger study. In addition,though the 16S rRNA gene sequencing assay provides a snapshotof what bacteria are present in the normal and lung tumorsamples, localization of bacteria in these tissue samples (e.g.,using fluorescence in situ hybridization; ref. 41) could provideadditional insight into bacterial mechanisms of action in lungcancer. Continued study of the role of the lung microbiome inlung cancer may yield several promising future applications,including biomarkers of lung cancer risk, recurrence, and prog-nosis, and therapeutic targets for lung cancer primary and tertiaryprevention.


DisclaimerThe funders had no involvement in the study design, the collection, analysis,

and interpretation of data, the writing of this report, and the decision to submitfor publication.

Authors' ContributionsConception and design: B.A. Peters, H.I. Pass, J. AhnDevelopment of methodology: J. Ahn

Acquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): C. Reid, H.I. Pass, J. AhnAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): B.A. Peters, R.B. Hayes, H.I. Pass, J. AhnWriting, review, and/or revision of the manuscript: B.A. Peters, R.B. Hayes,H.I. Pass, J. AhnAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): C. Goparaju, C. Reid, H.I. PassStudy supervision: H.I. Pass, J. Ahn

AcknowledgmentsLung tissue samples underwent 16S rRNA gene sequencing at the Environ-

mental Sample Preparation and Sequencing Facility at Argonne NationalLaboratory. This work was supported by the NIH NCI Early Detection ResearchNetwork grant 5U01CA111295-07 (toH.I. Pass), andNCI grant R01CA164964(to J. Ahn).


Received August 30, 2018; revised November 5, 2018; accepted January 28,2019; published first February 7, 2019.

References1. Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo M, et al.

Cancer incidence and mortality worldwide: sources, methods and majorpatterns in GLOBOCAN 2012. Int J Cancer 2015;136:E359–86.

2. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin2017;67:7–30.

3. American Cancer Society. Cancer Facts & Figures 2017. Atlanta, GA:American Cancer Society; 2017.

4. Detterbeck FC, Chansky K, Groome P, Bolejack V, Crowley J, Shemanski L,et al. The IASLC lung cancer staging project: methodology and validationused in the development of proposals for revision of the stage classificationof NSCLC in the forthcoming (eighth) edition of the TNM classification oflung cancer. J Thorac Oncol 2016;11:1433–46.

5. Aberle DR, Adams AM, Berg CD, BlackWC, Clapp JD, FagerstromRM, et al.Reduced lung-cancer mortality with low-dose computed tomographicscreening. N Engl J Med 2011;365:395–409.

6. Dickson RP, Huffnagle GB. The lung microbiome: new principles forrespiratory bacteriology in health and disease. PLoS Pathog 2015;11:e1004923.

7. Dickson RP, Erb-Downward JR, Huffnagle GB. The role of the bacterialmicrobiome in lung disease. Expert Rev Respir Med 2013;7:245–57.

8. Liu HX, Tao LL, Zhang J, Zhu YG, Zheng Y, Liu D, et al. Difference of lowerairway microbiome in bilateral protected specimen brush between lungcancer patients with unilateral lobar masses and control subjects. Int JCancer 2018;142:769–78.

9. Tsay JJ, Wu BG, Badri MH, Clemente JC, Shen N, Meyn P, et al. Airwaymicrobiota is associated with up-regulation of the PI3K pathway in lungcancer. Am J Respir Crit Care Med 2018;198:1188–98.

10. HosgoodHD 3rd, Sapkota AR, RothmanN, Rohan T, HuW, Xu J, et al. Thepotential role of lung microbiota in lung cancer attributed to householdcoal burning exposures. Environ Mol Mutagen 2014;55:643–51.

11. Lee SH, Sung JY, Yong D, Chun J, Kim SY, Song JH, et al. Character-ization of microbiome in bronchoalveolar lavage fluid of patients withlung cancer comparing with benign mass like lesions. Lung Cancer2016;102:89–95.

12. Yan X, Yang M, Liu J, Gao R, Hu J, Li J, et al. Discovery and validation ofpotential bacterial biomarkers for lung cancer. Am J Cancer Res 2015;5:3111–22.

13. Yu G, Gail MH, Consonni D, Carugno M, Humphrys M, Pesatori AC, et al.Characterizing human lung tissue microbiota and its relationship toepidemiological and clinical features. Genome Biol 2016;17:163.

14. Greathouse KL, White JR, Vargas AJ, Bliskovsky VV, Beck JA, vonMuhlinenN, et al. Interaction between the microbiome and TP53 in human lungcancer. Genome Biol 2018;19:123.

15. Mao Q, Jiang F, Yin R, Wang J, Xia W, Dong G, et al. Interplay between thelung microbiome and lung cancer. Cancer Lett 2018;415:40–8.

16. Fouhy F, Deane J, ReaMC,O'Sullivan �O, Ross RP, O'CallaghanG, et al. Theeffects of freezing on faecal microbiota as determined using MiSeqsequencing and culture-based investigations. PLoS One 2015;10:e0119355.

17. Shaw AG, Sim K, Powell E, Cornwell E, Cramer T, McClure ZE, et al.Latitude in sample handling and storage for infant faecal microbiotastudies: the elephant in the room? Microbiome 2016;4:40.

18. Kia E, Wagner Mackenzie B, Middleton D, Lau A, Waite DW, Lewis G, et al.Integrity of the human faecal microbiota following long-term samplestorage. PLoS One 2016;11:e0163666.

19. PuntCJ, BuyseM,KohneCH,Hohenberger P, Labianca R, SchmollHJ, et al.Endpoints in adjuvant treatment trials: a systematic review of the literaturein colon cancer and proposed definitions for future trials. J Natl Cancer Inst2007;99:998–1003.

20. Caporaso JG,LauberCL,WaltersWA,Berg-LyonsD,LozuponeCA,TurnbaughPJ, et al. Global patterns of 16S rRNA diversity at a depth of millions ofsequences per sample. Proc Natl Acad Sci U S A 2011;108:4516–22.

21. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Huntley J, Fierer N,et al. Ultra-high-throughput microbial community analysis on the Illu-mina HiSeq and MiSeq platforms. ISME J 2012;6:1621–4.

22. Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, CostelloEK, et al. QIIME allows analysis of high-throughput community sequenc-ing data. Nat Methods 2010;7:335–6.

23. Bokulich NA, Subramanian S, Faith JJ, Gevers D, Gordon JI, Knight R, et al.Quality-filtering vastly improves diversity estimates from Illumina ampli-con sequencing. Nat Methods 2013;10:57–9.

24. Amir A,McDonaldD, Navas-Molina JA, Kopylova E,Morton JT, Zech Xu Z,et al. Deblur rapidly resolves single-nucleotide community sequencepatterns. mSystems 2017;2:pii: e00191-16.

25. DeSantis TZ, Hugenholtz P, Larsen N, Rojas M, Brodie EL, Keller K, et al.Greengenes, a chimera-checked 16S rRNA gene database and workbenchcompatible with ARB. Appl Environ Microbiol 2006;72:5069–72.

26. Katoh K, Misawa K, Kuma K, Miyata T. MAFFT: a novel method forrapid multiple sequence alignment based on fast Fourier transform.Nucleic Acids Res 2002;30:3059–66.



27. Price MN, Dehal PS, Arkin AP. FastTree: computing large minimumevolution trees with profiles instead of a distance matrix. Mol Biol Evol2009;26:1641–50.

28. Lozupone C, Lladser ME, Knights D, Stombaugh J, Knight R. UniFrac: aneffective distance metric for microbial community comparison. ISME J2011;5:169–72.

29. Gower JC. Somedistance properties of latent root and vectormethods usedin multivariate analysis. Biometrika 1966;53:325–38.

30. Plantinga A, Zhan X, Zhao N, Chen J, Jenq RR, Wu MC. MiRKAT-S: acommunity-level test of association between the microbiota and survivaltimes. Microbiome 2017;5:17.

31. Koh H, Livanos AE, Blaser MJ, Li H. A highly adaptive microbiome-based association test for survival traits. BMC Genomics 2018;19:210.

32. Murtagh F, Legendre P. Ward's hierarchical agglomerative clustering meth-od: which algorithms implement Ward's criterion? J Classification 2014;31:274–95.

33. Anderson MJ. A new method for non-parametric multivariate analysis ofvariance. Austral Ecol 2001;26:32–46.

34. Weiss S, Xu ZZ, Peddada S, Amir A, Bittinger K, Gonzalez A, et al.Normalization and microbial differential abundance strategies dependupon data characteristics. Microbiome 2017;5:27.

35. Kurtz ZD, Muller CL, Miraldi ER, Littman DR, Blaser MJ, Bonneau RA.Sparse and compositionally robust inference of microbial ecological net-works. PLoS Comput Biol 2015;11:e1004226.

36. Fernandes AD, Reid JN, Macklaim JM, McMurrough TA, Edgell DR, GloorGB. Unifying the analysis of high-throughput sequencing datasets: char-acterizing RNA-seq, 16S rRNA gene sequencing and selective growthexperiments by compositional data analysis. Microbiome 2014;2:15.

37. Louis P, Flint HJ. Diversity,metabolism andmicrobial ecology of butyrate-producing bacteria from the human large intestine. FEMS Microbiol Lett2009;294:1–8.

38. Gopalakrishnan V, Spencer CN, Nezi L, Reuben A, Andrews MC, KarpinetsTV, et al. Gut microbiome modulates response to anti-PD-1 immunother-apy in melanoma patients. Science 2018;359:97–103.

39. Melkamu T, Qian X, Upadhyaya P, O'Sullivan MG, Kassie F. Lipopolysac-charide enhances mouse lung tumorigenesis: a model for inflammation-driven lung cancer. Vet Pathol 2013;50:895–902.

40. Wu J, Peters BA, Dominianni C, Zhang Y, Pei Z, Yang L, et al. Cigarettesmoking and the oral microbiome in a large study of American adults.ISME J 2016;10:2435–46.

41. Dejea CM,Wick EC,Hechenbleikner EM,White JR,MarkWelch JL, RossettiBJ, et al. Microbiota organization is a distinct feature of proximal colorectalcancers. Proc Natl Acad Sci U S A 2014;111:18321–6.


Peters et al.

Translational Cancer Mechanisms and Therapy

Microbiota- and Radiotherapy-InducedGastrointestinal Side-Effects (MARS) Study: ALarge Pilot Study of the Microbiome in Acute andLate-Radiation EnteropathyMiguel Reis Ferreira1,2,3,4, H. Jervoise N. Andreyev2, Kabir Mohammed2, Lesley Truelove1,2,Sharon M. Gowan1, Jia Li5, Sarah L. Gulliford1,6, Julian R. Marchesi5,7, andDavid P. Dearnaley1,2

Abstract

Purpose: Radiotherapy is important in managing pelviccancers. However, radiation enteropathy may occur and canbe dose limiting. The gut microbiota may contribute to thepathogenesis of radiation enteropathy. We hypothesized thatthe microbiome differs between patients with and withoutradiation enteropathy.

Experimental Design: Three cohorts of patients (n ¼ 134)were recruited. The early cohort (n ¼ 32) was followedsequentially up to 12months post-radiotherapy to assess earlyradiation enteropathy. Linear mixed models were used toassess microbiota dynamics. The late cohort (n ¼ 87) wasassessed cross-sectionally to assess late radiation enteropathy.The colonoscopy cohort compared the intestinal mucosamicroenvironment in patients with radiation enteropathy(cases, n ¼ 9) with healthy controls (controls, n ¼ 6). Fecalsamples were obtained from all cohorts. In the colonoscopycohort, intestinal mucosa samples were taken. Metataxo-nomics (16S rRNA gene) and imputed metataxonomics(Piphillin) were used to characterize the microbiome. Clini-cian- and patient-reported outcomes were used for clinicalcharacterization.

Results: In the acute cohort, we observed a trend for higherpreradiotherapy diversity in patientswith no self-reported symp-toms (P¼ 0.09). Dynamically, diversity decreased less over timein patients with rising radiation enteropathy (P ¼ 0.05). Aconsistent association between low bacterial diversity and lateradiation enteropathywas also observed, albeit nonsignificantly.Higher counts ofClostridium IV, Roseburia, and Phascolarctobacter-ium significantly associatedwith radiation enteropathy.Homeo-static intestinal mucosa cytokines related to microbiota regula-tion and intestinal wall maintenance were significantly reducedin radiation enteropathy [IL7 (P ¼ 0.05), IL12/IL23p40 (P ¼0.03), IL15 (P ¼ 0.05), and IL16 (P ¼ 0.009)]. IL15 inverselycorrelated with counts of Roseburia and Propionibacterium.

Conclusions: The microbiota presents opportunities to pre-dict, prevent, or treat radiation enteropathy.We report the largestclinical study to date into associations of the microbiota withacute and late radiation enteropathy. An altered microbiotaassociateswith early and late radiation enteropathy, with clinicalimplications for risk assessment, prevention, and treatment ofradiation-induced side-effects.

See related commentary by Lam et al., p. 6280

IntroductionPelvic radiotherapy is an important curative treatment

option for patients with pelvic cancers. However, acute(�90 days of starting radiotherapy) and chronic (thereafter)gastrointestinal side-effects, collectively summarized by theterm "radiation enteropathy", may develop. Indeed, risk ofgastrointestinal toxicity limits the radiation dose that can bedelivered (1). Radiation enteropathy can be defined as a pro-gressive, ischemic, and profibrotic process occurring afterabdominal or pelvic irradiation, driven by pathophysiologicprocesses which are incompletely defined (1, 2). Mechanismsinvolving the microbiota may contribute to the spectrum ofradiation enteropathy (1). However, published research con-centrates on acute radiation enteropathy, whereas it is lateradiation enteropathy that is usually dose limiting, and oftenuses animal models, which have limitations (3–6).

To better understand the role of the microbiota in radiationenteropathy, we prospectively collected fecal samples from threecomplementary cohorts of patients, collectively assessing thewhole spectrum of radiation enteropathy. We hypothesized that

1The Institute of Cancer Research, London, United Kingdom. 2The Royal MarsdenNHS Foundation Trust, London, United Kingdom. 3Guys and St Thomas NHSFoundation Trust, London, United Kingdom. 4King's College London, London,United Kingdom. 5Imperial College, London, United Kingdom. 6University Col-lege London Hospitals NHS Foundation Trust, London, United Kingdom. 7CardiffUniversity, Cardiff, United Kingdom.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

J.R. Marchesi and D.P. Dearnaley are the cosenior authors of this article.

Current address for H.J.N. Andreyev: Dept Gastroenterology, United Lincoln-shire Hospitals Trust and the School of Medicine, University of Nottingham.

Corresponding Authors: Miguel Reis Ferreira, Institute of Cancer Research, 15Cotswold Road, Sutton, London SM2 5NG, United Kingdom. Phone: 745-329-3233; Fax: 44 (0)2077249369; E-mail: [email protected]; and JulianR. Marchesi, Division of Integrative Systems Medicine and Digestive Disease,Imperial College London, LondonW2 1NY, United Kingdom. Phone: 4402-0331-26197; E-mail: [email protected]

Clin Cancer Res 2019;25:6487–500

doi: 10.1158/1078-0432.CCR-19-0960


ClinicalCancerResearch


http://clincancerres.aacrjournals.org/cgi/content/full/1078-0432.CCR-19-0960?ijkey=CEP5GCbe.4jEc&keytype=ref&siteid=aacrjnls

the microbiome differs between patients with and without radi-ation enteropathy after pelvic radiotherapy.

Materials and MethodsThe MARS study

The MARS study was an observational, noninterventionalstudy. Three cohorts were recruited in parallel (SupplementaryFig. S1). All patients attending relevant clinics were invited toparticipate during a 2-year period (see Section 1C in Supplemen-tary Data for sample size justification).

The first (termed "early cohort") assessed the development ofearly radiation enteropathy in a group of patients recruited beforeundergoing high-dose intensity-modulated radiotherapy to theprostate and pelvic lymph nodes (PLN-IMRT) and followedlongitudinally up to a year thereafter. Patients undergoingPLN-IMRT were chosen because they are at increased risk ofradiation enteropathy when compared with prostate-only radio-therapy (7). Clinical assessment and sampling was performed atbaseline (preradiotherapy), at 2/3 weeks, 4/5 weeks, 12 weeks,6 months, and 12 months post-radiotherapy initiation (Supple-mentary Fig. S2).

The second (termed "late cohort") explored late radiationenteropathy and included patients with �2 years of follow-upafter PLN-IMRT who were evaluated cross-sectionally. Patientsin this cohort were recruited from the population of a previ-ously reported dose-escalation trial of PLN-IMRT (7). Theirradiotherapy followed an identical protocol to the longitudinalcohort.

The third (termed "colonoscopy cohort") assessed the intesti-nal mucosa immune environment in radiation enteropathy andits relationships with the microbiome. It included patients with

�1 year of follow-up after radiotherapy for prostate cancer andattending a specialist clinical service for managing radiation-induced gastrointestinal symptoms who were undergoing colo-noscopy for symptom investigation (termed "cases"), as well asnonirradiated control subjects ("controls"), undergoing colonos-copy for colon cancer screening and confirmed free of gastroin-testinal diseases. We sampled anterior rectum (cases/controls)and distal sigmoid (cases only). The anterior rectum is thegastrointestinal location receiving maximal irradiation in radio-therapy for prostate cancer, while the distal sigmoid is lessirradiated and was thus used as a self-control in cases.

All subjects provided written informed consent prior to entryinto the study. The study was approved by the Committee forClinical Research at the Royal Marsden (no.: 4010) and by theLondon-Bromley Research Ethics Committee (no.: 13/LO/1527),and registered by the NHS Health Research Authority (ID:130287). All study procedures were conducted in accordance tothe Declaration of Helsinki.

AssessmentsClinician-reported outcomes (CRO) included items of the

Radiation Therapy Oncology Group (RTOG) and Late Effects ofNormal Tissues (LENT-SOM) scales with an impact on qualityof life [bowel problem/distress measured with the University ofCalifornia, Los Angeles Prostate Cancer Index (UCLA-PCI)scale; refs. 8–10]. The criteria used were RTOG diarrhea andproctitis, and LENT-SOM sphincter control (subjective); tenes-mus (subjective), bleeding (objective), pain (objective), andbleeding (management). Two summary figures (RTOG maxi-mum and LENT/SOMmaximum) were created from maximumtoxicity scores. Both scales are graded 1–5, with increasingscores representing worse symptoms. Patient-reported out-comes (PRO) were analyzed with the bowel subset of a gas-trointestinal symptom score validated for radiation enteropa-thy and graded 1–7 for 10 items (Supplementary Table S1),with scores ranging from 10 (very symptomatic) to 70 (nosymptoms; ref. 11).

In the late cohort, peak cumulative late toxicity scores (from6 month after radiotherapy onwards) were available as per theIMRT for Prostate Cancer study protocol. CRO included RTOGdiarrhea and RTOG proctitis. PRO included UCLA-PCI bowelproblem and distress. For convenience, we have termed preva-lence data at the time of sampling "actual toxicity", and peakcumulative data "historic toxicity".

Patient comorbidity and diet, were also assessed (Supplemen-tary Materials and Methods; Supplementary Table S2). Intestinalmucosa histology (colonoscopy cohort), was evaluated with asemiquantitative histopathology score (Supplementary Table S3;ref. 12).

Definition of symptom groupsPatients in the early cohort were divided in three groups, which

were (i) no symptoms (no symptoms at either 4/5 weeks or6 months); (ii) nonpersistent symptoms (symptoms at either4/5 weeks or 6 months); and (iii) persistent symptoms (symp-toms at 4/5 weeks and 6 months). To not lose data, the CRO-based symptom classification was substituted for 13 patients(41%)where PROdataweremissing at either of these timepoints,which were chosen as representative of maximal acute enterop-athy (4/5 weeks) and early late enteropathy (6 months; ref. 7).This strategy enables identification of patients experiencing

Translational Relevance

Clinical evidence shows that gutmicrobiota changes duringradiotherapy and suggests associations with radiation enter-opathy, but this evidence is limited. Clinical studies ofteninclude patients receiving concurrent cytotoxic systemic ther-apies. Experiments in animal models indicate that gut micro-biota is necessary for radiation enteropathy to occur and thatan irradiated microbiota promotes enteropathy. However,animal models have different radioresistance and microbiotacompared with humans, and usually receive high-dose single-fraction radiation, limiting clinical translation. Moreover, allevidence focuses on acute radiation enteropathy and does notaddress dose-limiting late radiation enteropathy. We reportthe largest clinical study to date into associations of themicrobiota with acute and late radiation enteropathy. It is theonly study where patients received homogeneous treatmentand where no patients received cytotoxic systemic therapies.Our novel methodology allowed assessment of acute and lateradiation enteropathy. We demonstrate that some bacteriaproducing short-chain fatty acids are associated with radia-tion-induced side-effects and that this relates to an alteredintestinal microenvironment. We demonstrate that an alteredmicrobiota associates with early and late radiation enteropa-thy, with clinical implications for risk assessment, prevention,and treatment of radiation-induced side-effects.

Reis Ferreira et al.

Clin Cancer Res; 25(21) November 1, 2019 Clinical Cancer Research6488

nonhealing acute toxicity, which may be related to a consequen-tial reaction and determines a higher risk of long-term radiationenteropathy (13).

In the late cohort, CROgroupswere definedby symptomgrade.PRO-based groupings were based on the data, by dividingpatients in quartiles defining increasing symptoms. For conve-nience, these categorieswere identified as no,mild,moderate, andsevere symptoms (Supplementary Table S4).

In the colonoscopy cohort, cases were compared with controls.

Sampling procedures and processingSampling of stool. Sampling of stool was performed according topublished guidance (14). Details are given in SupplementaryMaterials and Methods.

Sampling of intestinal mucosa (colonoscopy cohort only). Threebiopsies were taken per site (Supplementary Fig. S3) for meta-taxonomics, cytokine analysis, and pathology assessment. Incases and controls, samples were obtained from the anteriorrectum, which is the part of the gastrointestinal tract whichreceives the greatest radiation dose during radiotherapy for pros-tate cancer (15). In cases only, another three biopsies wereobtained from a macroscopically unaffected region as close toaffected areas as possible, whichwas in all the distal sigmoid, to beused as a self-control. Further details are given in SupplementaryMaterials and Methods.

Data acquisition. DNA extraction procedure, data acquisition, andprocessing Genomic DNA was extracted from fecal (250 mg) andgut biopsy (whole biopsy) samples using the Qiagen Stool Kit(Qiagen) according to the manufacturer's instructions with anadditional bead beating step for homogenization of sample andlysis of bacterial cells. Library preparation and Illumina (MiSeq)sequencing of the V1–2 regions of the 16S rRNA gene wereperformed at RTL Genomics. Details are given in SupplementaryMaterials and Methods.

Cytokine detection Total protein was extracted from mucosal sam-ples and cytokine detection was carried out with theMSD V-PLEXHuman Cytokine 30-Plex kit according to the manufacturer'sinstructions. The manufacturer states that all cytokine isoformsare detected. Details are given in Supplementary Materials andMethods.

Statistical considerationsBioinformatic processing of 16S rRNA gene data. Sequences gen-erated from Illumina (MiSeq) sequencing were analyzed withMOTHUR (version 1.36.0) for identification of operational tax-onomic units (OTU), taxonomic assignment, community com-parison, and data cleaning by adapting its standard operationalprocedure (16). Details are given in SupplementaryMaterials andMethods.

Inferred metagenomes were obtained by using the Piphillinweb tool by Second Genome, using the Kyoto Encyclopedia ofGenes and Genomes (KEGG) May 2017 database, and a 97%identity cutoff.

Significance testing. The significance of taxonomic differences wasassessed with one-way ANOVA (�3 group comparisons), orWhite nonparametric two-sided t test (two-group comparisons).The Benjamini–Hochberg method was used for FDR correction.

However, a pragmatic approach was taken, with uncorrected Pvalues taken into account given the exploratory context of thiswork (17). Uncorrected P values are termed "p�", while P valuesafter correction are termed "p". Statistically significant resultsexplained by large peaks in <10% of a group were considerednonbiologically relevant.

The Kruskal–Wallis H test was used to assess differencesobserved when comparing a-diversity indices, diet, and histologyscores (colonoscopy cohort).

Longitudinal dynamics in the early cohort were evaluatedwith linear mixed models. Linear mixed models use fixed andrandom effects in the same analysis. Unlike univariate ormultivariate linear regression, one can assess individual vari-ation by subject per timepoint by analyzing the longitudinalchange of a variable of interest over time by symptomgroup (18). Also, mixed mods allow for missing observations,as other data endpoints can be still be used as long as themissing data meets the missing-at-random definition (18). Thisanalysis was performed in R using the "nlme" package and thefollowing formulation:

Yti ¼ b0 þ b1 timepointtið Þ þ b2 symptom classifierið Þþ b3 symptom classifierið Þ timepointtið Þ þ b0i

þ b1i timepointtið Þ þ "ti

Where b0 is the population estimate of the intercept for thecontrol group (no symptoms), b1 is the population estimate ofthe linear slope of the control group, b2 and b3 capture theestimates of the mean difference in intercept and slope betweensymptom groups, b0i and b1i are random effects that allow theintercepts and slopes to vary across individuals, and "ti is a time-specific residual that expresses the difference between and indi-vidual's fitter linear trajectory and the observed data. Thus, b3

represents the "symptom group by timepoint" interaction (19).To assess significance, t tests using Satterthwaite method wereimplemented.

Variable transformations were used according to the data andare discussed with results. The Akaike Information Criterion wasused to assess whether models with transformed variablesimproved goodness of fit. Full results (including all effect esti-mates and significance) are provided in Supplementary Materialsand Methods.

Multivariate analysis was performed with robust linear modelsin R using the "MASS" and "sfsmisc" packages with the followingformulation:

Y ¼ b0 þ b1x1 þ b2x2

Where b0 is the mean intercept, and b1 and b2 are the coefficientsfor variables x1 and x2, respectively. Significance of coefficientswas assessed with a robust F test (Wald test) using the f.robftest()function.

Comparison of cytokine levels and correlationswithmicrobiome.Thesignificance of differences between cytokine levels was assessedwith the Kruskal–Wallis H test. A significance of P < 0.1 (uncor-rected formultiple comparisons) was defined for post hoc (Mann–Whitney) testing.We report results of post hoc tests. Correlations ofcytokines with the microbiome were explored in bacterial generawhere OTU counts were >0 in �20% of subjects with Spearmanrank correlation coefficient.

Microbiota- and Radiotherapy-Induced GI Side-Effects (MARS)

www.aacrjournals.org Clin Cancer Res; 25(21) November 1, 2019 6489

Data availabilityAll data generated or analyzed during this study are included in

the published article and its Supplementary Information Files.

ResultsDemographics and symptoms

A total of 134men were enrolled betweenMarch 18, 2014 andFebruary 01, 2016 (Table 1): 32 in the early cohort, 87 in the latecohort, and 15 in the colonoscopy cohort (9 cases/6 controls). All

patients in the early and late cohorts underwent prostate andpelvic radiotherapy following a previously published proto-col (7). In the colonoscopy cohort, 6 cases had undergoneradiotherapy to the prostate and seminal vesicles, 1 had under-gone radiotherapy to the prostate, seminal vesicles, and pelviclymph nodes, and 2 had undergone postprostatectomy radio-therapy to the prostate bed and pelvic lymph nodes. Controlsubjects had not been treated with any radiotherapy.

In the early cohort, patients with nonpersistent symptomsmostly experienced symptoms at 4/5 weeks (84%/PRO, 92%/

Table 1. Demographics

Colonoscopy cohortItem Early cohort Late cohort Cases Controls

Median age at date of enrollment in years (IQR) 66 (63–72) 74 (68–79) 75 (71–76) 68 (57–69)Median time in years between radiotherapy commencement and sampling NA 6.05 (4.57–7.28) 4.2 (1.9–10.4) NARadiotherapy detailsPatients treatedwith conventionally fractionated radiotherapya: 70–74Gyto prostate and seminal vesicles (35–37 fractions) or 64 Gy to prostatebed (32 fractions); 50–60 Gy to pelvic lymph nodes (35–37 fractions)–n (%)

31 (97%) 48 (55%) 3 (33%) NA

Patients treated with hypofractionated radiotherapya: 60 Gy to prostateand seminal vesicles or 55 Gy to prostate bed (20 fractions); 47 Gy topelvic lymph nodes)–n (%)

1 (3%) 39 (45%) 0 (0%) NA

Patients treatedwith conventionally fractionated radiotherapy toprostateand seminal vesicles only: 70–74 Gy in 35–37 fractions

NA NA 6 (67%) NA

Prostate cancer detailsMedian presenting PSA (IQR) in ng/mL 26.2 (13.4–47) 18.1 (11.05–34.50) 7.05 (5.43–13.40) NAMedian PSA at time of sampling (IQR) in ng/mL NA NA 8.4 (5.7–14.6) NAGleason 6–n (%) 1 (3%) 3 (3%) 2 (22%) NAGleason 7–n (%) 12 (37%) 33 (38%) 6 (67%) NAGleason 8–n (%) 3 (9%) 14 (16%) 0 (0%) NAGleason 9–n (%) 16 (50%) 37 (43%) 1 (1%) NAN0–n (%) 16 (50%) 62 (71%) 7 (78%) NAN1–n (%) 16 (50%) 24 (28%) 2 (22%) NANX–n (%) 0 (0%) 1 (1%) 0 (0%) NAT1–n (%) 1 (3%) 1 (1%) 0 (0%) NAT2–n (%) 7 (22%) 18 (21%) 2 (22%) NAT3–n (%) 24 (75%) 65 (75%) 7 (78%) NAT4–n (%) 0 (0%) 2 (2%) 0 (0%) NATX–n (%) 0 (0%) 1 (1%) 0 (0%) NASubjects on short-course antiandrogen and long-term LHRH analogues 22 (69%) NA 1 (1%) NASubjects on bicalutamide monotherapy 1 (3%) NA 0 (0%) NASubjects on maximum androgen blockade 9 (28%) NA 0 (0%) NASubjects with recurrent tumors at time of sampling–n (%)b NA 11 (13%) 1 (1%) NASubjects on ADT at time of sampling–n (%)b 32 (100%)c 10 (11%) 1 (1%) NASubjects with recovered testosterone levels (�6 nmol/L)–n (%)b NAc 47 (54%) 5 (56%) NA

Other comorbiditiesb

Subjects with history of abdominal or pelvic surgery–n (%) 19 (59%) 40 (46%) 6 (67%) 3 (50%)Median body mass index (IQR) 27 (25–32) 26.5 (24.7–29.8) 26 (25–27) 24 (24–25)Subjects with dyslipidemia and on statins–n (%) 10 (31%) 45 (52%) 4 (44%) 2 (33%)Subjects with history of diabetes–n (%) 7 (22%) 15 (17%) 0 (0%) 0 (0%)Subjects with history of hypertension and on medical treatment–n (%) 13 (41%) 49 (56%) 7 (78%) 1 (16%)Subjects with history of IBS–n (%) 0 (0%) 3 (3.4%) 2 (22%) 2 (33%)Subjects with history of diverticular disease–n (%) 1 (3%) 10 (11%) 2 (22%) 0 (0%)Nonsmokers/ex-smokers/smokers–n (%) 19 (59%)/11 (34%)/

2 (6%)37 (42%)/38 (44%)/

12 (14%)4 (44%)/4 (44%)/

1 (1%)1 (17%)/5 (83%)/

0 (0%)

NOTE: The reader is reminded that cohortswere not directly compared, but independently assessed to investigate themicrobiota of patientswith early and late side-effects.Abbreviations: IQR, inter quartile range; NA ¼ not applicable.aConventional and hypofractionated radiotherapy schedules used to treat patientswere shown toproduce comparable rates of tumor recurrence, aswell as early andlate toxicities in a phase II trial (see ref. 7).bA detailed comparison of comorbidities between toxicity groups in each cohort is reported in the main text and in Supplementary Tables S6–S8 in SupplementaryMaterials and Methods.cAll subjects in the early cohort were under neo-adjuvant ADT from the time of recruitment, as per the protocol for treating high-risk prostate cancer (including ADTstarting before radiotherapy and extending for 2–3 years in total) and their testosterone levels were therefore undetectable. Some patients in the late cohort(�2 years after RT) were under long-term ADT for the same reason. ADT was not found to significantly impact the microbiome in this study (see SupplementaryData).



CRO). Classification was concordant (i.e., patients classified inthe same group with both PRO and CRO) in 21 patients (66%;Supplementary Table S5).

Symptoms and diet are described detail in SupplementaryData. We did not detect biologically relevant dietary differencesbetween groups.

As per our study design, cohorts were not compared with oneanother, but used to assess different aspects of radiation enter-opathy. Therefore, we analyzedwhether comorbidities [includingbody mass index, smoking status/history, metabolic diseases,androgen deprivation therapy (ADT), and other comedications]were different between symptom groups to be analyzed in eachcohort. Overall, comorbidities were well balanced betweengroups (Supplementary Tables S6–S8). No significant differenceswere found in the early cohort. In the late cohort, actual CRO-stratified groups showed significantly higher proportions of irri-table bowel syndrome (P ¼ 0.0004) in patients with risingsymptoms. In the colonoscopy cohort, proportions of controlsunder hypertensive medication were higher than in cases (P ¼0.02). The overall lowproportions of patients with irritable bowelsyndrome (IBS) in all cohorts may be attributed to eligibilitycriteria for pelvic radiotherapy, which is relatively contraindicatedfor patients with gastrointestinal conditions.

Comparison of stool and mucosal microbiome in thecolonoscopy cohort

We did not find significant differences when comparing stooland intestinal mucosa microbiomes in the colonoscopy cohort,although there was a trend for higher a-diversity, measured withthe Chao index, in stools compared with intestinal mucosalmicrobiome in cases [median (IQR): 77 (59.3–117.1) vs. 54.1(51.3–65.2)]. However, no significant differences in b-diversity orin individual phyla or genera were found. Results are described indetail in Supplementary Data.

Low bacterial diversity associates with radiation enteropathyLow bacterial diversity has been consistently associated with

acute radiation enteropathy (6, 20–23). Relationships with lateenteropathy have never been explored. We therefore assessedbacterial diversity between irradiated patients with and withoutgastrointestinal side-effects.

In the early cohort, we first explored associations betweenbaseline diversity and radiation enteropathy in the acute cohort.We found a trend for higher diversity at baseline in patients withno self-reported symptoms [P ¼ 0.09; median (IQR) Chaorichness for no radiation enteropathy: 89.1 (78.9–114.0); non-persistent radiation enteropathy: 55.2 (42.6–72.5); and persistentradiation enteropathy: 68.6 (41.8–75.1)]. This observation wasrecapitulated with CRO, albeit nonsignificantly [P ¼ 0.61; 76.3(65–86.1); nonpersistent radiation enteropathy: 55.2 (48.0–84.5); and persistent radiation enteropathy: 65.3 (39.1–75.1)].We next examined dynamics of a-diversity over time with linearmixed models. The variable of interest was Chao abundance ofbacterial species, with predictors of dynamics specified as time-point and symptom group (Supplementary Fig. S1A and S1B;Supplementary Table S9).Whennot stratifiedby symptomgroup,diversity appeared to decrease in the whole cohort over time,albeit nonsignificantly (effect of timepoint: �0.02; P ¼ 0.35;Supplementary Fig. S5). With PRO stratification, diversity gener-ally decreased over time (P¼ 0.03). A positive effect of timepointby symptom group indicates differential dynamics of diversity

over time (P ¼ 0.05; Fig. 1A). This pattern was similar when anidenticalmodel based onCROwas used, although it did not reachstatistical significance. We next examined differences in bacterialdiversity between patients with and without late enteropathy inthe late cohort (Fig. 1C–J). No significant differences were foundwith PRO or CRO in the late (Fig. 1C–J; Supplementary Fig S4) orcolonoscopy (Fig. 1K–L) cohorts. However, a nonsignificantpattern of higher diversity in symptomatic patients was observedin both cohorts.

Patients with radiation enteropathy have higher counts ofRoseburia, Clostridium IV, and Faecalibacterium

Enrichment in specific microbial taxa has been described inpatients with primary inflammatory bowel disease (IBD; ref. 24).Similarly, associations between specific bacterial taxa and acuteradiation enteropathy have been reported (6, 21, 23, 25). Wetherefore investigated whether specific bacterial taxa were asso-ciatedwith radiation enteropathy.We first compared proportionsof phyla and genera between patients with and without radiationenteropathy at each timepoint in the early cohort. No microbialfeatures showed statistically significant relationships. However,because of the limited power of this cohort for detecting differ-ences based on direct comparisons per timepoint, we definedbiologically plausible relationships as progressive changes inproportions of microbial features (i.e., either increasing ordecreasing) with rising symptoms, irrespective of statistical sig-nificance. Results are summarized in Supplementary Table S10.We used linear mixed models to evaluate longitudinal dynamicsof specific microbial taxa taking into account the results above.Bacterial taxa with biologically plausible relationships whereuncorrected P values (p�) < 0.05 were retained. They were Clos-tridium IV, Roseburia, and Phascolarctobacterium, which are shortchain fatty acid (SCFA) producers. Sutterella dynamics were alsoanalyzed in light of a biologically plausible relationship and apublished evidence suggesting that amicrobiome enriched in thistaxon associates with acute radiation proctitis in an animalmodel (23). Results are summarized in Fig. 2 and SupplementaryTable S11. Clostridium IV proportions increased significantlywith PRO (effect ¼ 0.4; P ¼ 0.007), with a trend toward aprogressively more negative slope of proportions over time withincreasing symptoms group (estimate ¼ �0.04; P ¼ 0.11). Thisbehavior was reflected with CRO. A trend was also observed forincreased Roseburia counts in direct proportion with patient-reported symptoms (effect¼ 0.37; P¼ 0.08), whichwere reflectedwith CRO. Plotting the models shows a comparatively steepdecrease in Roseburia proportions in patients with persistentsymptoms. A trend of higher proportions of Phascolarctobacteriumindirect proportion toCROwasobserved (effect¼0.26;P¼0.09)and reflected with PRO. Proportions of Sutterella appeared toincrease with symptoms with minimal change over time, albeitnonsignificantly.

We next examined microbial taxa in the late cohort. It is notedthat this analysis was completely independent of the early cohort,so all microbial taxa (and not only SCFA producers) were includ-ed, with ensuing FDR correction. No significant differences werefound at either phylum or genus levels when stratifying patientsaccording to either actual or historical PROs. However, whenstratifying patients according to CROs, Roseburia significantlyassociated with toxicity (Fig. 3; Supplementary Table S12). Pro-portions ofRoseburia rosewithmaximumactual (p� <0.000001; p< 0.00001) and historical (p� ¼ 0.001; p ¼ 0.06) CRO symptom



A B

30405060708090

100

Grade 0/1 Grade 2+

Chao

30405060708090

100

Grade 0/1 Grade 2+

Chao

30405060708090

100

Grade 0/1 Grade 2+

Chao

C D E

30405060708090

100

Grade 0/1 Grade 2+

Chao

30405060708090

100

Grade 0/1 Grade 2+

Chao

30

50

70

90

110

Grade 0/1 Grade 2+Ch

ao

F G H

30405060708090

100110

Noproblem

Very smallproblem

Smallproblem

Moderateproblem

Bigproblem

Chao

0102030405060708090

100110

Nosymptoms

Mildsymptoms

Moderatesymptoms

Severesymptoms

Chao

I J

30

40

50

60

70

80

90

100

Cases Controls

Chao

30

50

70

90

110

130

150

170

Cases rectum Cases sigmoid Controls rectum

Chao

K L

Early

coh

ort

Late

coh

ort

Colo

nosc

opy

coho

rt

Figure 1.

Bacterial diversity in the early (A and B), late (C–J), and colonoscopy (K and L) cohorts of the MARS study. Dynamics of Chao diversity over time in PRO (A) andCRO (B) stratified groups, where the effect of timepoint (P¼ 0.03) and timepoint by symptom group (P¼ 0.05) were significant in PRO-stratified groups.Groups: 0, no symptoms; 1, nonpersistent symptoms; and 2, persistent symptoms. Timepoints: 1, baseline; 2, 2/3 weeks; 3, 4/5 weeks; 4, 12 weeks; 5, 6 months;and 6, 12 months after radiotherapy initiation. A log transformation was used because of a positive skew of the data, which was confirmed to provide superiorgoodness of fit when compared with square-root transformations. Chao diversity in the late cohort in groups stratified by CRO actual/historical diarrhea (C andF), proctitis (D and G), and maximum toxicity (E and H); and by PRO actual (I) and late (J) toxicity. P > 0.05 in all comparisons. The reader is reminded that scalesfor PRO stratification differed between actual and historical toxicity (see Materials and Methods). Chao diversity in the colonoscopy cohort with stool (K) andintestinal mucosa (L) samples (P > 0.05 in both comparisons).



Figure 2.

Dynamics of proportions of Clostridium IV (A and B), Roseburia (C and D), Phascolarctobacterium (E and F), and Sutterella (G and H) over time in PRO (left) andCRO (right) stratified groups. The effect of PRO symptom group was significant for Clostridium IV (P¼ 0.007). There was a trend for significance for the effect ofPRO and CRO symptom group for Roseburia (P¼ 0.08) and Phascolarctobacterium, respectively. Groups: 0, no symptoms; 1, nonpersistent symptoms; and 2,persistent symptoms. Timepoints: 1, baseline; 2, 2/3 weeks; 3, 4/5 weeks; 4, 12 weeks; 5, 6 months; and 6, 12 months after radiotherapy initiation. A square-roottransformation was used because of a positive skew of the data, which was confirmed to provide superior goodness of fit when compared with a logtransformation in all models.



grade.No relevant differences at either phylumor genus levelweredetected for proctopathy. Roseburia significantly rose with bothactual (P < 0.000001) and historical (P < 0.00001) clinician-reported diarrhea grade. To test whether significance was due tovery high peaks in patients with grade 3 diarrhea, all patients withgrade 3 toxicity were removed and differences retested includingonly patients with grade 0–2 diarrhea. Results with actual (P ¼0.056) and historical (P ¼ 0.04) diarrhea remained significant.Proportions of Roseburia also rose with historical PRO-stratifiedsymptoms, albeit nonsignificantly (Supplementary Table S12). Ashigher proportions of IBS were found in patients with CRO-stratified actual symptoms, we used robust linear regression toadjust for these parameters in two independent multivariatemodels. The model predicted actual CRO grade with IBS (P ¼0.002) andRoseburia (P¼ 0.02) as significant variables. To furtherassess whether a relationship between Roseburia and IBS waspresent, we also examined correlation between the two variables,which was not present (Spearman Rho ¼ 0.09; P ¼ 0.43).Moreover, no significant differences in genus-level taxa werefound between patients with and without IBS in the late cohort,including Roseburia (p� ¼ 0.60; p > 0.1) and Clostridium IV (p� ¼0.59, p > 0.1). Because ADT has been associated with a modifiedmicrobiota in a previous report, we also examined in the micro-biota of the late cohort stratified by active ADT or testosteronerecovery status (26). No significant differences were found in

a-diversity or in taxa (see Supplementary Data). We note that wedid not carry out such analyses in the early cohort due to allpatients being on active ADT since before baseline sampling andconsequently having undetectable testosterone levels.

In the colonoscopy cohort, no significant differences werefound when comparing cases and controls. However, the size ofthis cohort limited statistical power (see Supplementary Data).

We then hypothesized that metagenomic abundance of micro-bial SCFA metabolism pathways differed between patients withand without symptoms of radiation enteropathy where signifi-cant associations were detected. We combined community com-position with annotated genomes from the KEGG catalog andselected pathways related to microbial SCFA metabolism foranalysis (27). We again used linear mixed models to evaluatedynamics in the early cohort (Supplementary Fig. S6; Supple-mentary Table S13). Abundances of SCFA-related microbial met-abolic pathways increased consistently with symptoms, mostnoticeably with PRO, although this effect only trended for sig-nificance for propionate metabolism (P ¼ 0.07). Propionate andother SCFA fuel colonocytes and upregulate colonic regulatory Tlymphocytes, thereby promoting gut homeostasis (28). Its ben-efits to gut health have been reviewed elsewhere (29). Theconsistently negative effect of timepoint by symptom groupsuggests that microbial SCFA pathways may decrease more overtimewith rising symptoms. In the late cohort, only the abundance

Figure 3.

Proportions of Roseburia in actual (A–C) and historical (D–F) CRO-stratified groups of the late cohort by CRO grade. A and D,Maximum toxicity. B and E,Diarrhea. C and F, Proctitis. Higher grades reflect more serious symptoms. �, P¼ 0.06; �� , P� 0.01; �� , P� 0.001. All P values shown are corrected for FDR. Thex-axis shows CRO grade.



of fatty acid metabolism pathways decreased consistently withrising CRO diarrhea grade (P < 0.0001; Supplementary Fig. S7;Supplementary Table S14).

Patients with radiation enteropathy have depletion of rectalmucosa cytokines regulating gut microbiota and homeostasis,correlating with higher counts of Roseburia andPropionibacterium

Cytokines are small molecules involved in cell signaling andhave immunomodulatory, paracrine, and autocrine functionswith pathophysiologic implications. However, gastrointestinalmucosal cytokine changes have never been studied in late radi-ation enteropathy. We therefore investigated differences in theconcentrations of 29 cytokines, divided in three panels, betweencases and controls in the colonoscopy cohort. A distinct generalpattern of highest concentration in controls and lowest concen-trations in the anterior rectum of cases was observed, except forproinflammatory cytokines, where no differences were found.When analyzing differences between sample types by cytokine,IL7 (P ¼ 0.05), IL12/IL23p40 (P ¼ 0.03), IL15 (P ¼ 0.05), andIL16 (P¼ 0.009) were significantly higher in control than in caserectal biopsies, while eotaxin (P ¼ 0.03) followed an inversepattern (Fig. 4A–C). We did not find significant differences inpathology (including fibrosis) or in proinflammatory cytokinesbetween cases and controls, which argues against the hypothesisof difficulty in tissue permeation in cases or subclinical inflam-mation in controls (Supplementary Table S15). Interestingly,cytokines observed to be lower in cases have intestinal homeo-static properties by regulating the microbiota and the intestinalbarrier (Supplementary Table S16).

We then examined correlations between themicrobiome of theanterior rectal mucosa and cytokine concentrations. Rectal Rose-buria and Propionibacterium, which are SCFA producers, and Strep-tococcus, an acetate producer, were inversely correlated with IL15(decreased in patients with radiation enteropathy) in our dataset(rho ¼ �0.54 and �0.52; P ¼ 0.04 and 0.05, respectively; Fig. 5;ref. 30). Flavonifractor, a butyrate-producing genus, correlatedpositively with eotaxin (increased in patients with radiationenteropathy; ref. 31). These observations suggests an associationbetween mucosal SCFA producers and radiation enteropathy inthe anterior rectal mucosa, which is the gastrointestinal locationreceiving the highest levels of radiation in prostate radiotherapy.We observed similar correlations, albeit not so evidently, withsigmoid and stool microbiota.

DiscussionIn this study, we have shown that a modified microbiota is

associated with radiation enteropathy and that key homeostaticintestinal mucosa cytokines related to microbiota regulation andintestinal wall maintenance are also significantly reduced inpatients with radiation enteropathy. Our study confirms previousobservations in small cohorts of patients where acute radiationinjurywasassociatedwithanalteredmicrobiota (6,20,21,23,25).However, our data are not compromised by the delivery ofconcurrent cytotoxic systemic treatments, which made the find-ings from these smaller studies difficult to interpret. In addition,we have shown for the first time that a modified microbiota isassociated with late radiation enteropathy.

We previously reviewed the potential of the microbiota in theprediction and treatment of radiation enteropathy (1). Although

the importance of the gastrointestinal microbiota in radiation-induced intestinal toxicity is highlighted by the recognition ofcauses of enteropathy such as small intestinal bacterial over-growth, the studies evaluating the microbiome in patients withradiation-induced gastrointestinal symptoms are limited by lowpatient numbers. Also, all previous authors focused only on acuteradiation enteropathy. Although bacteriotherapy has been stud-ied by administering probiotics or prebiotics, interventions weremarred by insufficient knowledge of the microbiota in radiationenteropathy, which is reflected in modest and often conflictingresults. Also, although preclinical studies provide useful infor-mation, they do not reflect the clinical reality of radiation enter-opathy in the modern era of precision radiotherapy. We thusintended to provide a comprehensive characterization of themicrobiota in radiation enteropathy, which provides a founda-tion for further studies in this field.

We acknowledge the limitations of our study. Radiationenteropathy has multiple causes, which are likely to havedifferential contributions from the microbiota (2). As yet, noobjective markers of radiation enteropathy have been definedand there is no option but to rely on abnormal symptoms.However, symptom scales have limitations for detecting radi-ation enteropathy, hence our approach of using both CRO andPRO for better characterization of patients. Also, although ourpatients had comorbidities, as expected in the aged populationof patients with prostate cancer, they were globally well dis-tributed between symptom groups. We nevertheless adjustedfor their effect where significant differences in comorbiditiescould have an impact, and our results were robust to theseanalyses. The relatively younger age of patients in the earlycompared with the late cohort reflects that patients undergoingtreatment are younger than patients on long-term follow-up.However, this is unlikely to significantly affect the microbiota,given its overall stability with time (1). Moreover, we did notdirectly compare these cohorts, which were used to analyzedifferent phases of radiation enteropathy as per our studydesign (32). We also acknowledge that, although we detectedconsistent results across all cohorts, high intersubject variabil-ity of the microbiota is known to affect cohort studies and isthe main conundrum bedeviling all microbiota research inhumans (33). Unfortunately, studies in animals are limited byadministration of extreme (often lethal) radiation doses andvery different radioresistance and microbiota when comparedwith humans. Although findings may appear more clear-cut,such models poorly represent clinical radiotherapy (5, 23).Furthermore, we acknowledge the limitation of not measuringdiet longitudinally in the acute cohort, which was due toethical concerns of study procedure–related patient exhaus-tion. We did not find, however, biologically relevant dietarydifferences between any of the cohorts. Also, although radia-tion-induced gastrointestinal side-effects remain the maindose-limiting factor in modern prostate cancer radiotherapy,their severity has been much reduced by successful improve-ments in treatment delivery. Limitations of metataxonomicsare also acknowledged, such as PCR bias and artificial over-representation of some species carrying multiple copies of 16SrRNA genes (34).

We observed associations of microbiota endpoints with acute(mostly with PRO) and late (mostly with CRO) toxicity. Usingboth types of instruments is known to provide a full representa-tion of toxicity and is the reason why radiotherapy trialists now



0.0

50.0

100.0

150.0

200.0

250.0

Eotaxin* MIP-1β Eotaxin-3 TARC IP-10 MIP-1α MCP-1 MDC MCP-4

Conc

entr

a�on

A

0

2

4

6

8

10

12

14

Conc

entr

a�on

B

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

Conc

entr

a�on

C

Controls - Rectum Cases - Distal Sigmoid Cases - Rectum

IL8

IFNγ IL1β IL2 IL4 IL6 IL8 IL10 IL12p70 IL13 TNFα

Figure 4.

Mean absolute cytokine concentrations by sample group. Blue, controls (rectum); green, cases (sigmoid); and red, cases (rectum).A, Chemokine panel. B,Cytokine panel. C, Proinflammatory panel. For scaling purposes, all concentrations are pg/mL except TARC, IL7, IL12/IL23p40, and IL17a (� 10 pg/mL;A); IL16(� 10 ng/mL); and IL8, VEGFa, IFNg , IL1b, and IL2 (� 0.1 ng/mL; B); and IL8 and IL13 (� 0.01 pg/mL; C; � , P� 0.05).



report both separately (35). We hypothesize that this discrepancyis due to three factors: (i) differences in perception of side-effectsfrom thepoint of viewof patients and clinicians; (ii) limitations ofboth PROandCRO instruments; and (iii) the overall low grade oftoxicity produced by modern radiotherapy. In the acute setting,where patients are na€�ve to radiotherapy, their perception ofsymptoms may be higher and therefore PROs may be moresensitive. In the late setting, both successful ongoing treatmentof toxicity and increased patient tolerance to side-effects maymake CROs more sensitive. For example, a patient successfullyusing loperamide for diarrhea may not report symptoms, butclinicians would classify such a patient as having diarrhea. Fur-

thermore, we acknowledge limitations in using PRO instruments.PROs in the acute cohort were analyzed as difference to baseline,andwill thus reflect better each patient's longitudinal evolution interms of symptoms (36). However, patients were assigned togroups of increasing late patient–reported toxicity (late andcohort) based on dividing them in four quartiles, as there are isno "normal threshold" in our validated PRO score. Given thesmall range in PRO scores in the late cohort (described inSupplementary Materials and Methods), patients with differenttoxicity phenotypes may have been grouped together thereforemaking results more difficult to interpret. Despite these limita-tions and in the absence of a reliable biomarker of radiation

Figure 5.

Correlation matrices of microbiome of stools (A), rectal mucosa (B), and sigmoid mucosa (C) and concentrations of cytokines. The size of circles representssignificance and the color code represents Spearman correlation coefficient (rho). Only significant results (P� 0.05) are shown. Stools and rectal mucosamicrobiomeswere correlated with rectal cytokine levels, whereas sigmoid microbiome was correlated with sigmoid cytokine levels. Class is defined as eithercases (coded 0) or controls (coded 1) and therefore a positive correlation denotes higher concentration/proportion in controls and vice versa.



enteropathy, our approach provides the most comprehensiveclinical characterization of radiation enteropathy ever carried outin a study in this field.

Decreased bacterial diversity was consistently associated withradiation enteropathy in all three cohorts, and we conclude thatthis observation is not random, although results in two of ourcohorts were nonsignificant. A less diverse microbiota associateswith other forms of colitis, including IBD, IBS, and infectivecolitis, as well as with diseases such as obesity and autoimmunediseases (37, 38). Associations between acute radiation enterop-athy and reduced diversity have also been reported by otherauthors (6, 20–23). In animal models, a less diverse irradiatedmicrobiota (which is enriched in Sutterella among other bacteria)is sufficient for the induction of higher susceptibility to intestinalinflammation, suggesting that reduced bacterial diversity maycause patients to be at risk of enteropathy in the short and longterms (23). It is noteworthy that Sutterella was higher in patientswith radiation enteropathy in the early cohort, albeit nonsignif-icantly (23). Our results suggest that strategies for increasingbacterial diversity in patients at risk could be trialed to seewhetherthey modify the course of radiation enteropathy.

We found significant associations between some organismsproducing SCFA and radiation enteropathy in all cohorts, againsuggesting that this association is nonrandom. Imbalances in themicrobiota, often termed dysbiosis, associate with many gastro-intestinal diseases, including IBD, IBS, and viral colitis. Generally,such imbalances are characterized by an increase in bacteria,which are recognized to be pathogens, such as Escherichia coli, orthe Shigella and Klebsiella genera (1). However, SCFA producerspromote intestinal homeostasis and their depletion has beenassociated with IBD, so increased proportions in patients withsymptoms are surprising (39). Mechanistic exploration is beyondthe scope of our study, but some hypotheses can be suggested.These bacteria are part of intestinal mucosa–associated commu-nities and it is possible that, in patients at risk of symptoms,increased competition by potentially pathogenic bacteria leads toincreased shedding in the stools. This shedding would be con-sistent with differential dynamics observed between groups. Analternative hypothesis would be that chronic, subclinical prera-diotherapy intestinal dysfunction may lead to a dependence onmicrobiota-derived nutrients for epithelial health (2). Radiother-apy led to decreased SCFA production capacity, associating withsymptom onset. The high counts of Roseburia associating withCRO-stratified but not PRO-stratified late symptoms support thathigher proportions of these bacteria relate to decreased symptomperception by patients in the presence of clinician-perceiveddisease. This hypothesis is consistent with the limited clinicaleffectiveness of oral or topical butyrate when treating radiationenteropathy (40). Althoughwe acknowledge the low comparativeproportions of these bacteria when compared with other SCFAproducers such as Faecalibacterium, the trend of patients withradiation enteropathy having higher, but dynamically decreasing,SCFA production capacity (early cohort) and significantlydecreased levels of homeostatic rectal mucosa cytokines involvedinmucosal barrier maintenance andmicrobiota regulation (colo-noscopy cohort) would support this assumption. These hypoth-eses need to be further explored.

Our study provides evidence of structural and functional shiftsin the microbiota in patients with radiation enteropathy. How-ever, whether these changes are a cause or consequence of intes-tinal symptoms is a matter of considerable debate even in well

researched fields of noninfectious colitis such as IBD (41). Also,unlike IBD, radiation enteropathy is characterized by noninflam-matory mechanisms, which is well illustrated by evidence of arecent placebo-controlled randomized trial where sulfasalazine,an anti-inflammatory drug often used to treat IBD, actually had adetrimental effect in terms of diarrhea for patients undergoingradiation enteropathy (42). We also did not find evidence ofincreased inflammatory cytokines in patients with late radiationenteropathy. Other authors have provided complementarymech-anistic evidence, which suggests a causative role for the micro-biota in radiation enteropathy (23). We provide a framework forfurther downstream studies assessing a causative role for themicrobiota, which could provide further scope for microbialinterventions such as fecal transplantation, which has recentlybeen suggested as a successful treatment of immunotherapy-induced colitis (43). Other bacteriotherapy interventions, suchas the administration of probiotics (live organisms that, whenconsumed in an adequate amount, confer a health effect on thehost) or prebiotics (nondigestible foods that promote the growthor activity of specific microorganisms, promoting a health effect),have also been trialed in patients undergoing pelvic radiothera-py (1). The mixed results observed may stem from the fact thatmany of these therapies modulate bacteria, which do not have animpact in radiation enteropathy. However, Garcia-Peris and col-leagues showed in a randomized trial that the delivery of a fibermixture containing inulin, which promotes the growth of SCFAproducers such as Roseburia, improves diarrhea in patients under-going pelvic radiotherapy, supporting our observations (44, 45).

We conclude that radiotherapy may upset the balance ofmicrobiota which supports intestinal health, by decreasing theinfluence of key microorganisms, probably more susceptible toradiation effects. The microbiota may be used to predict,prevent, or treat clinical radiation enteropathy and our studyprovides an evidence base for developing preclinical and clin-ical studies.

Disclosure of Potential Conflicts of InterestD.P. Dearnaley is listed as a co-inventor on a patent application regarding

a plastic rectal obturator to improve accuracy and reduce the side effects ofprostate radiotherapy that will be owned by The Institute of Cancer Researchand Sussex Development; is a consultant/advisory board member forTakeda, Amgen, Astellas, Sandoz, Janssen, EMA, MsC Lecture Fees, EMUCPresentation Prize, Norway Oncology Society Meeting Honorarium, andQueens University Belfast; and reports receiving other remuneration fromthe Institute of Cancer Research. No potential conflicts of interest weredisclosed by the other authors.

DisclaimerThis article is independent research funded by the National Institute for

Health Research (NIHR) Biomedical Research Centre, and the views expressedin this article are those of the authors and not necessarily those of the NHS,NIHR, or the Department of Health.

Authors' ContributionsConception anddesign:M.Reis Ferreira,H.J.N. Andreyev, J. Li, J.Marchesi, D.P.DearnaleyDevelopment of methodology:M. Reis Ferreira, H.J.N. Andreyev, S.M. Gowan,J. Marchesi, D.P. DearnaleyAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): M. Reis Ferreira, H.J.N. Andreyev, L. Truelove, S.M.Gowan, J. Marchesi, D.P. DearnaleyAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis):M. Reis Ferreira, H.J.N. Andreyev, K. Mohammed, S.M. Gowan, J. Li, S.L. Gulliford, J. Marchesi, D.P. Dearnaley



Writing, review, and/or revision of the manuscript: M. Reis Ferreira, H.J.N.Andreyev, K. Mohammed, L. Truelove, S.M. Gowan, J. Li, S.L. Gulliford, J.Marchesi, D.P. DearnaleyAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): M. Reis Ferreira, K. Mohammed, L. TrueloveStudy supervision: H.J.N. Andreyev, D.P. DearnaleyOther (sample collection, day-to-day management, design of figures andtables): M. Reis Ferreira

AcknowledgmentsWe thank the patients and the trials unit staff at the Bob Champion Unit and

RMH Trial Unit who contributed to the coordination of the study. We acknowl-edge support of Cancer Research UK awarded to D. Dearnaley (C8262/A7253,C1491/A9895, C1491/A15955, and SP2312/021), funding from the NIHRCancer Research Network through the NIHR BRC at the Royal Marsden NHSFoundation Trust and The Institute of Cancer Research, London awarded to D.Dearnaley andM. Reis Ferreira (A53/CCR4010), the Department of Health, theNational Institute for Health Research (NIHR) Cancer Research Network, and

NHS funding to the NIHR Biomedical Research Centre (BRC) at the RoyalMarsden NHS Foundation Trust and The Institute of Cancer Research, London.The Division of Integrative Systems Medicine and Digestive Disease at ImperialCollege London (to J. Marchesi) received financial support from the NIHRImperial BRC based at Imperial College Healthcare NHS Trust and ImperialCollege London. M. Reis Ferreira acknowledges support from the CalousteGulbenkian Foundation, the Fundac~ao para a Ciencia e a Tecnologia, and theChampalimaud Foundation (SFRH/BDINTD/51547/2011). J. Li acknowledgesMedical Research Council and European Research Commission starting grantsfor salary support.


Received March 21, 2019; revised June 18, 2019; accepted July 22, 2019;published first July 25, 2019.

References1. Ferreira MR, Muls A, Dearnaley DP, Andreyev HJN. Microbiota and

radiation-induced bowel toxicity: lessons from inflammatory bowel dis-ease for the radiation oncologist. Lancet Oncol 2014;15:e139–47.

2. Hauer-Jensen M, Denham JW, Andreyev HJN. Radiation enteropathy—pathogenesis, treatment and prevention. Nat Rev Gastroenterol Hepatol2014;11:470–9.

3. Andreyev J.Gastrointestinal complications of pelvic radiotherapy: are theyof any importance? Gut 2005;54:1051–4.

4. Conlon AM, Bird RA.The impact of diet and lifestyle on gutmicrobiota andhuman health. Nutrients 2015;7:17–44.

5. Crawford PA,Gordon JI.Microbial regulation of intestinal radiosensitivity.Proc Natl Acad Sci U S A 2005;102:13254–9.

6. Manichanh C, Varela E, Martinez C, Antolin M, Llopis M, Dor�e J, et al. Thegut microbiota predispose to the pathophysiology of acute postradiother-apy diarrhea. Am J Gastroenterol 2008;103:1754–61.

7. Reis Ferreira M, Khan A, Thomas K, Truelove L, McNair H, Gao A, et al.Phase 1/2dose-escalation studyof the useof intensitymodulated radiationtherapy to treat the prostate and pelvic nodes in patients with prostatecancer. Int J Radiat Oncol 2017;99:1234–42.

8. Cox JD, Stetz J, Pajak TF. Toxicity criteria of the radiation therapy oncologygroup (RTOG) and the European organization for research and treatmentof cancer (EORTC). Int J Radiat Oncol 1995;31:1341–6.

9. Litwin MS, Hays RD, Fink A, Ganz PA, Leake B, Brook RH. The UCLAprostate cancer index: development, reliability, and validity of a health-related quality of life measure. Med Care 1998;36:1002–12.

10. LENT SOMA scales for all anatomic sites. Int J Radiat Oncol 1995;31:1049–91.

11. Olopade FA, Norman A, Blake P, Dearnaley DP, Harrington KJ, Khoo V,et al. Amodified inflammatory bowel disease questionnaire and the vaizeyincontinence questionnaire are simple ways to identify patients withsignificant gastrointestinal symptoms after pelvic radiotherapy. Br J Cancer2005;92:1663–70.

12. Langberg CW, Sauer T, Reitan JB, Hauer-Jensen M. Tolerance of rat smallintestine to localized single dose and fractionated irradiation. Acta Oncol1992;31:781–7.

13. Pinkawa M, Holy R, Piroth MD, Fischedick K, Schaar S, Sz�ekely-Orb�an D,et al. Consequential late effects after radiotherapy for prostate cancer - aprospective longitudinal quality of life study. Radiat Oncol 2010;5:27.

14. Eppinga H, Fuhler GM, Peppelenbosch MP, Hecht GA. Gut microbiotadevelopments with emphasis on inflammatory bowel disease: report fromthe gutmicrobiota for health world summit 2016. Gastroenterology 2016;151:e1–4.

15. Peterson JL, Buskirk SJ, HeckmanMG,DiehlNN, Bernard JR, Tzou KS, et al.Image-guided intensity-modulated radiotherapy for prostate cancer: Doseconstraints for the anterior rectal wall to minimize rectal toxicity.Med Dosim 2014;39:12–7.

16. Schloss PD,Westcott SL, Ryabin T, Hall JR, HartmannM,Hollister EB, et al.Introducing mothur: open-source, platform-independent, community-

supported software for describing and comparingmicrobial communities.Appl Environ Microbiol 2009;75:7537–41.

17. Rothman KJ. No adjustments are needed for multiple comparisons.Epidemiology 1990;1:43–6.

18. Seltman HJ. Chapter 15 mixed models. Exp Des Anal 2014;357–78.19. Hesser H. Modeling individual differences in randomized experiments

using growth models: Recommendations for design, statistical analysisand reporting of results of internet interventions. Internet Interv 2015;2:110–20.

20. Cui M, Xiao H, Li Y, Zhou L, Zhao S, Luo D, et al. Faecal microbiotatransplantation protects against radiationinduced toxicity. EMBO MolMed 2017;9:448–61.

21. Wang A, Ling Z, Yang Z, Kiela PR, Wang T, Wang C, et al. Gut microbialdysbiosis may predict diarrhea and fatigue in patients undergoing pelviccancer radiotherapy: a pilot study. PLoS One 2015;10:e0126312.

22. Zhu XX, Yang XJ, Chao YL, Zheng HM, Sheng HF, Liu HY, et al. Thepotential effect of oral microbiota in the prediction of mucositis duringradiotherapy for nasopharyngeal carcinoma. EBioMedicine 2017;18:23–31.

23. Gerassy-Vainberg S, Blatt A, Danin-Poleg Y, Gershovich K, Sabo E,Nevelsky A, et al. Radiation induces proinflammatory dysbiosis: trans-mission of inflammatory susceptibility by host cytokine induction. Gut2018;67:97.

24. Kostic AD, Xavier RJ, Gevers D. The microbiome in inflammatory boweldisease: current status and the future ahead. Gastroenterology 2014;146:1489–99.

25. Nam YD, Kim HJ, Seo JG, Kang SW, Bae JW. Impact of pelvic radiotherapyon gut microbiota of gynecological cancer patients revealed by massivepyrosequencing. PLoS One 2013;8:e82659.

26. Sfanos KS, Markowski MC, Peiffer LB, Ernst SE, White JR, Pienta KJ, et al.Compositional differences in gastrointestinal microbiota in prostate can-cer patients treated with androgen axis-targeted therapies. Prostate CancerProstatic Dis 2018;21:539–48.

27. Louis P, Flint HJ. Formation of propionate and butyrate by the humancolonic microbiota. Environ Microbiol 2017;19:29–41.

28. Smith PM, Howitt MR, Panikov N, Michaud M, Gallini CA, Bohlooly-Y M,et al. The microbial metabolites, short-chain fatty acids, regulate colonicTreg cell homeostasis. Science 2013;341:569–73.

29. Gill PA, van ZelmMC,Muir JG, Gibson PR. Review article: short chain fattyacids as potential therapeutic agents in human gastrointestinal and inflam-matory disorders. Aliment Pharmacol Ther 2018;48:15–34.

30. Koh A, De Vadder F, Kovatcheva-Datchary P, B€ackhed F. From dietary fiberto host physiology: short-chain fatty acids as key bacterialmetabolites. Cell2016;165:1332–45.

31. Kl€aring K, Hanske L, Bui N, Charrier C, Blaut M, Haller D, et al. Intesti-nimonas butyriciproducens gen. nov., sp. nov., a butyrate-producingbacterium from the mouse intestine. Int J Syst Evol Microbiol 2013;63:4606–12.



32. O'Toole PW, Jeffery IB. Gut microbiota and aging. Science 2015;350:1214–5.

33. Waldor MK, Tyson G, Borenstein E, OchmanH,Moeller A, Finlay BB, et al.Where next for microbiome research? PLOS Biol 2015;13:e1002050.

34. Carlos N, Tang Y-W, Pei Z. Pearls and pitfalls of genomics-based micro-biome analysis. Emerg Microbes Infect 2012;1:1–3.

35. Gilbert A, Ziegler L, Martland M, Davidson S, Efficace F, Sebag-MontefioreD, et al. Systematic review of radiation therapy toxicity reporting inrandomized controlled trials of rectal cancer: a comparison of patient-reported outcomes and clinician toxicity reporting. Int J Radiat Oncol2015;92:555–67.

36. Andreyev HJN, Benton BE, Lalji A, Norton C, Mohammed K, Gage H, et al.Algorithm-basedmanagement of patients with gastrointestinal symptomsin patients after pelvic radiation treatment (ORBIT): a randomised con-trolled trial. Lancet 2013;382:2084–92.

37. Mosca A, Leclerc M, Hugot JP. Gut microbiota diversity and humandiseases: should we reintroduce key predators in our ecosystem?Front Microbiol 2016;7:455.

38. Casta~no-Rodríguez N, Underwood AP, Merif J, Riordan SM, RawlinsonWD, Mitchell HM, et al. Gut microbiome analysis identifies potentialetiological factors in acute gastroenteritis. Infect Immun 2018;86:e00060–18.

39. Machiels K, Joossens M, Sabino J, De Preter V, Arijs I, Eeckhaut V, et al. Adecrease of the butyrate-producing species Roseburia hominis and Faecali-

bacterium prausnitziidefines dysbiosis in patientswithulcerative colitis. Gut2014;63:1275–83.

40. Teo MTW, Sebag-Montefiore D, Donnellan CF. Prevention and manage-ment of radiation-induced late gastrointestinal toxicity. Clin Oncol 2015;27:656–67.

41. Ni J, Wu GD, Albenberg L, Tomov VT. Gut microbiota and IBD: causationor correlation? Nat Rev Gastroenterol Hepatol 2017;14:573.

42. Miller RC, Petereit DG, Sloan JA, Liu H, Martenson JA, Bearden JD III, et al.N08C9 (Alliance): A phase 3 randomized study of sulfasalazine versusplacebo in the prevention of acute diarrhea in patients receiving pelvicradiation therapy. Int J Radiat Oncol Biol Phys 2016;95:1168–74.

43. Wang Y, Wiesnoski DH, Helmink BA, Gopalakrishnan V, Choi K, DuPontHL, et al. Fecal microbiota transplantation for refractory immune check-point inhibitor-associated colitis. Nat Med 2018;24:1804–8.

44. Garcia-Peris P, Velasco C,HernandezM, LozanoMA, Paron L, de la CuerdaC, et al. Effect of inulin and fructo-oligosaccharide on the prevention ofacute radiation enteritis in patients with gynecological cancer and impacton quality-of-life: a randomized, double-blind, placebo-controlled trial.Eur J Clin Nutr 2015;70:170.

45. Falony G, Verschaeren A, De Bruycker F, De Preter V, Verbeke K, Leroy F,et al. In vitro kinetics of prebiotic inulin-type fructan fermentation bybutyrate-producing colon bacteria: implementation of online gas chro-matography for quantitative analysis of carbon dioxide and hydrogen gasproduction. Appl Environ Microbiol 2009;75:5884.



MOLECULAR CANCER RESEARCH | NEW HORIZONS IN CANCER BIOLOGY

Neoadjuvant Chemotherapy Shifts Breast TumorMicrobiota Populations to Regulate DrugResponsiveness and the Development of MetastasisAkiko Chiba1,2, Alaa Bawaneh3, Christine Velazquez1,2, Kenysha Y.J. Clear1, Adam S. Wilson1,Marissa Howard-McNatt1,2, Edward A. Levine1,2, Nicole Levi-Polyachenko4, Shaina A. Yates-Alston4,Stephen P. Diggle5, David R. Soto-Pantoja1,2,6, and Katherine L. Cook1,2,6

ABSTRACT◥

Breast tumors have their own specific microbiota, distinct fromnormal mammary gland tissue. Patients with breast cancer thatpresent with locally advanced disease often undergo neoadjuvantchemotherapy to reduce tumor size prior to surgery to allow breastconservation or limit axillary lymphnode dissection. The purpose ofour study was to evaluate whether neoadjuvant chemotherapymodulates the tumor microbiome and the potential impact ofmicrobes on breast cancer signaling. Using snap-frozen asepticallycollected breast tumor tissue from women who underwent neoad-juvant chemotherapy (n ¼ 15) or women with no prior therapy attime of surgery (n ¼ 18), we performed 16S rRNA-sequencing toidentify tumoral bacterial populations. We also stained breasttumor microarrays to confirm presence of identified microbiota.Using bacteria-conditioned media, we determined the effect ofbacterial metabolites on breast cancer cell proliferation and doxo-rubicin therapy responsiveness. We show chemotherapy adminis-

tration significantly increased breast tumor Pseudomonas spp.Primary breast tumors from patients who developed distant metas-tases displayed increased tumoral abundance of Brevundimonasand Staphylococcus. We confirmed presence of Pseudomonas inbreast tumor tissue by IHC staining. Treatment of breast cancer cellswith Pseudomonas aeruginosa conditioned media differentiallyeffected proliferation in a dose-dependent manner and modulateddoxorubicin-mediated cell death. Our results indicate chemother-apy shifts the breast tumor microbiome and specific microbescorrelate with tumor recurrence. Further studies with a largerpatient cohort may provide greater insights into the role of micro-biota in therapeutic outcome and develop novel bacterial biomar-kers that could predict distant metastases.

Implications:Breast tumormicrobiota aremodified by therapy andaffects molecular signaling.

IntroductionPrevious studies have identified the presence of breast-specific

microbiome (1–5). Mammary gland samples obtained from Canadianor Irish women undergoing lumpectomies, mastectomies, or breastreduction surgeries displayed bacterial taxa differences. Canadianwomen's breast tissue had a high proportional abundance of Bacillus,Acinetobacter, Enterobacteriaceae, Pseudomonas, Staphylococcus, Pro-pionibacterium, and Prevotella.When compared with Canadian breastsamples themammary tissue from Irishwomen displayed high Listeriawelshimeri, over a 3-fold higher abundance of Enterobacteriaceae,2-fold higher Staphylococcus, and a 2-fold higher abundance of

Propionibacterium.Mammary gland samples from Irish andCanadianwomen displayed similar (5%–6%) abundance of Pseudomonas genuslevel taxa (1).

Another study investigating the microbiome of normal mam-mary gland tissue compared with breast tumor adjacent mammarygland tissue, indicated increased relative abundance of Staphylo-coccus in mammary gland tissue of patients with cancer. This studydid not control for patient body weight (2). Another study whichreported patient body mass index (BMI), breast tissue was obtainedadjacent to either benign or malignant breast tumors and mammarygland tissue was sequenced to identify changes in microbiota.Mammary gland tissue from patients with malignant disease haddecreased levels of Lactobacillus, suggesting breast tumorigenesismay modify mammary gland microbiota (6). Recently our groupdemonstrated the plasticity of the mammary gland microbiomeusing a dietary approach in a nonhuman primate model. Weshowed that monkeys consuming a Mediterranean diet had dis-tinctively different breast microbiota populations than monkeysconsuming a Western diet that was modulated independently of thegut microbiome and systemic effects (5).

Breast tumors also display a distinct microbiota population whencompared with mammary tissue. Noncancerous tumor adjacenttissue (n ¼ 72) and breast tumor (n ¼ 668) The Cancer GenomeAtlas data were mined and nonoverlapping sequences were mappedto bacterial genomes (7). Bacterial 16S ribosomal genes wereidentified from these nonoverlapping sequences and prevalentbacterial populations quantified (exclusion criteria included: malebreast cancers, neoadjuvant chemotherapy treated, metastatic sam-ples, and history of prior breast cancer). Tumor tissue was enrichedwith bacteria from Proteobacteria phylum when compared with

1Department of Surgery, Wake Forest University School of Medicine, Winston-Salem, North Carolina. 2Comprehensive Cancer Center, Wake Forest UniversitySchool of Medicine, Winston-Salem, North Carolina. 3Department of IntegrativePhysiology and Pharmacology, Wake Forest University School of Medicine,Winston-Salem, North Carolina. 4Department of Plastic and ReconstructiveSurgery, Wake Forest University School of Medicine, Winston-Salem, NorthCarolina. 5School of Biological Sciences, Georgia Institute of Technology,Atlanta, Georgia. 6Department of Cancer Biology, Wake Forest UniversitySchool of Medicine, Winston-Salem, North Carolina.

Note: Supplementary data for this article are available at Molecular CancerResearch Online (http://mcr.aacrjournals.org/).

Corresponding Author: Katherine L. Cook, Wake Forest University School ofMedicine, 575 N. Patterson Ave, Suite 340, BioTech Place, Winston Salem, NC27101. Phone: 336-716-2234; Fax: 336-716-1456; E-mail: [email protected]

Mol Cancer Res 2020;18:130–9

doi: 10.1158/1541-7786.MCR-19-0451



http://mcr.aacrjournals.org/cgi/content/full/1541-7786.MCR-19-0451?ijkey=wMMCpnp2scB12&keytype=ref&siteid=aacrjnls

noncancerous adjacent tissue. This study also did not investigate theimpact of patient body weight on tumor microbial populations.Another study that isolated DNA and performed 16S sequencingfrom formalin-fixed, paraffin-embedded breast tumors (ERþ n ¼50, HER2þ n ¼ 34, and triple negative n ¼ 40) and normal tissuefrom breast reduction surgeries (n ¼ 20) indicate elevated Proteo-bacteria in tumor samples and specific changes in bacteria popula-tions by subtype (8). This study did not have access to deidentifieddata and could not correlate breast tumor bacterial populations withBMI or relapse-free survival.

Neoadjuvant chemotherapy, usually a combination of anthracy-cline [adriamycin or doxorubicin (DOX)], alkylating agents (cyclo-phosphamide), and taxanes (taxotere), are often used in patientswith locally advanced breast cancer to reduce tumor size and limitlymph node involvement before surgery to enable breast conser-vative surgery and minimize surgery in the axilla (9, 10). Whetherchemotherapy can modulate breast tumor microbiota is unknown.Moreover, whether the tumor bacterial populations impact thera-peutic responsiveness or the development of distant metastases isundetermined.

To determine whether chemotherapy modulates breast tumormicrobiota, we obtained snap-frozen aseptically collected tumortissue from patients with breast cancer that underwent neoadjuvantchemotherapy or tumors from women with no prior therapyat time of surgery. Our sample set is clinically annotated enablingcorrelation of primary tumor microbiota populations with tumorrecurrence and obesity. Our data indicate that neoadjuvant che-motherapy increased tumoral Pseudomonas abundance anddecreased Streptococcus populations. The development of distantmetastases correlated with increased primary tumor abundance ofBrevundimonas and Staphylococcus. Moreover, using a secondarybreast tumor microarray sample set, we validated Pseudomonaspresence by IHC. We also used Pseudomonas aeruginosa condi-tioned media to demonstrate differential molecular signaling acti-vation in breast cancer cells depending on the concentrationof bacterial-secreted metabolites to impact cell growth, apoptosis,and chemotherapy responsiveness. Taken together these data sug-gest that distinct intratumoral microbiota populations are present,and that the composition of bacterial populations can be modifiedby therapy to impact breast cancer cell signaling and drugresponsiveness.

Material and MethodsStudy approval and breast tumor tissue procurement

This study was approved by our Institutional Review Board(IRB00045734) in accordance with HHS regulations for the protectionof human research subjects. Subjects were retrospectively identified asthose in the Sentinel Lymph Node Mapping and Surgical Outcomes(IRB00008131) database whowere female and diagnosedwith invasiveductal carcinoma. In order for subjects to be included, they must haveprovided written consent for the Advanced Tumor/Tissue Bank(BG04-104) and have tumor tissue for research in the Tumor Bank.Patient demographics, preoperative variables, surgical details, andclinical outcomes were also collected. Patient characteristics aredescribed in Table 1. Breast tumor specimens provided to us by thepathologist in the OR Path Lab within 45 to 60 minutes from surgeryare snap-frozen. A designation of tumor or normal is made by visualgross inspection by the pathologist before tumor samples are submit-ted to the Tumor Bank.

DNA extraction, PCR, 16s rRNA-sequencing, and dataprocessing

Breast tumor specimens (n ¼ 33) were placed into a MoBioPowerMag Soil DNA Isolation Bead Plate. DNA was extracted fol-lowing the manufacturer's protocol. Bacterial 16S sequencing, dataanalysis, and interpretation were performed by Microbiome Insights.Bacterial 16S rRNA genes were PCR-amplified with dual-barcodedprimers targeting the V4 region, following the methods of ref. 11.Amplicons were sequenced with an Illumina MiSeq using the 250-bpPaired-End Kit (v.2). Sequences were taxonomically classified usingGreengenes (v.13_8) reference database, and clustered into 97%-similarity operational taxonomic units (OTU) with the mothur(v.1.39.5) software package (12), following the recommended proce-dure (https://www.mothur.org/wiki/MiSeq_SOP). The potential forcontamination was addressed by cosequencing DNA amplified fromspecimens and from four each of template-free controls and extractionkit reagents processed the same way as the specimens. Two positivecontrols, consisting of cloned SUP05 DNA, were also included (num-ber of copies¼ 2� 106). Samples with less than 100 OTU counts wereexcluded from analysis.

To estimate beta diversity across samples, we excluded OTUsoccurring in fewer than 10% of the samples with a count of less thanthree and computed Bray–Curtis indices. We visualized beta diver-sity, emphasizing differences across samples, using nonmetricmultidimensional ordination. Variation in community structurewas assessed with permutational multivariate analyses of variancewith treatment group as the main fixed factor. All analyses wereconducted using R software. Proportional abundance of selectedmicrobiota were graphed and multiple unpaired t tests performed(correcting for multiple comparisons using Holm–Sidak method)was used and a value of P � 0.05 was considered statisticallysignificant.

Table 1. Patient with breast cancer characteristics.

Pretreatmentgroup, n ¼ 18

Neoadjuvantchemotherapygroup, n ¼ 15

Tumorrecurrencegroup, n ¼ 9

Age (years) 65.3 � 8.9 58.9 � 10.1 64.3 � 7.9BMI 32.7 � 9.3 30.4 � 7.5 33.7 � 7.7Ethnicity

Caucasian 83% 80% 100%African American 17% 20% 0%

Tumor classificationDuctal 89% 80% 88%Lobular 11% 7% 12%Other 0% 13% 0%Estrogenreceptor-apositive (ERþ)

39% 21.5% 38%

ER�,progesteronereceptorþ (PRþ)

0% 7% 0%

TNBC (ER�,PR�, HER2�)

39% 43% 25%

ERþ/HER2þ 11% 7% 25%HER2þ 11% 21.5% 12%

Note: Numbers for patient age and BMI represent mean � SD. There are nosignificant differences between the treatment groups.

Neoadjuvant Chemotherapy and Breast Tumor Microbiota

AACRJournals.org Mol Cancer Res; 18(1) January 2020 131

https://www.mothur.org/wiki/MiSeq_SOP

IHCA tissuemicroarray (TMA) consisting of normally mammary gland

tissue (n¼ 10), primary breast tumor tissue (n¼ 50), and lymph nodemetastases (n¼ 40) was purchased from US Biomax, Inc. (catalog no.BR1008a). TMA was stained against P. aeruginosa antibody (Abcam;catalog no. ab68538) in a 1:100 dilution using the Dako Envision PlusIHC Staining Kit.

P. aeruginosa conditioned mediaP. aeruginosa (ATCC 27853) was cultured overnight in tryptic soy

broth at 37�C in an orbital shaker.Media were spun down at 1,000 rpmfor 10 minutes to pellet the P. aeruginosa, which was discarded. Thesupernatant was filtered using a Corning disposable vacuum filtrationsystem with a 0.22 mm pore size any bacterial particulates. Media wasstored at 4�C and used within 24 to 48 hours of harvest.

Cell line acquisition and growth conditionsMDA-MB-231, 4T1, and the ZR-75-1 breast cancer cell lines were

previously purchased fromATCC. The 67NR cell line were generouslyprovided byDr. AndreaMastro at Pennsylvania State University (StateCollege, PA).MCF7 clone 2 cells were obtained fromDr. Robert Clarkeat Georgetown University (Washington DC). MDA-MB-231 humantriple-negative breast cancer (TNBC) cells, 4T1 murine TNBC cells,ZR-75-1 human ERþ breast cancer cells, and 67NR murine ERþbreast cancer cells were grown in RMPI media with 10% FBS at 37�C(basal growth conditions). MCF-7 human ERþ breast cancer cellswere grown in DMEM media with 10% FBS at 37�C (basal growthconditions). Cells were passaged for approximately 3 months beforenew cell stock was obtained from liquid nitrogen storage. 4T1 and theMDA-MB-231 cell lines were last authenticated by IDEXX BioAna-lytics using short tandem repeat analysis, in November 2018. The 4T1was confirmed to be of mouse origin and no mammalian interspeciescontamination was detected. The MDA-MB-231 was confirmed to beof human origin and no interspecies contamination was detected. TheMDA-MB-231 cells 100% identity match consistent with the cell lineof origin. Mycoplasma contamination was measured in cell culturemedia supernatant by PCR. Primer sequences are as follows:CGCCTGAGTAGTACGTTCGC;CGCCTGAGTAGTACGTACGC;TGCCTGAGTAGTACATTCGC;TGCCTGGGTAGTACATTCGC;CGCCTGGGTAGTACATTCGC;CGCCTGAGTAGTATGCTCGC;GCGGTGTGTACAAGACCCGA;GCGGTGTGTACAAAACCCGA;GCGGTGTGTACAAACCCCGA.

In vitro cell index measurementBreast cancer cells (5 � 105) were plated in an ACEA xCELLi-

gence system for 24 hours before treated with 5%, 10%, or 20% P.aeruginosa conditioned media (P-CM) for an additional 48 hours.In another set of experiments, breast cancer cells were plated in anACEA xCELLigence system for 24 hours before treated with 10% or20% P-CM � 1 mg/mL doxorubicin for an additional 48 hours. Cellindex was measured by electrical impedance every 6 hours.

In vitro P. aeruginosa metabolite replacement cell indexmeasurement

MDA-MB-231 cells (5 � 105) in basal growth conditions cellswere plated in an ACEA xCELLigence system for 24 hours beforetreated with 10 mg/mL pyocyanin (catalog no. R9532; Sigma-Aldrich),10 mg/mL lectin (catalog no. L9895; Sigma-Aldrich), 10 ng/mL exo-toxin A (catalog no. P0184; Sigma-Aldrich), 10 mg/mL phospholipaseD (catalog no. P0065; Sigma-Aldrich), or 1 mg/mL lipopolysaccharide(LPS; catalog no. L9143; Sigma-Aldrich)� 1mg/mLdoxorubicin for an

additional 24 hours. Cell index was measured by electrical impedanceevery 6 hours.

Western blot hybridizationMDA-MB-231, 4T1, MCF7, and ZR-75-1 cells were plated for

24 hours before treated with 5%, 10%, or 20% P-CM � 1 mg/mLdoxorubicin for an additional 24 hours. Treated cells were harvested inRIPA buffer, protein was size fractionated by gel electrophoresis, andtransferred to a nitrocellulose membrane. Membranes were blocked inblotto for 30 minutes then incubated overnight at 4�C with primaryantibodies [pAkt Ser473, Akt (total), cleaved caspase-7, and b-actinfrom Cell Signaling Technology, dilution 1:1,000]. The next day,membranes were washed and incubation with polyclonal horseradishperoxidase–conjugated secondary antibodies. Immunoreactive pro-ducts were visualized by chemiluminescence (SuperSignal FemtoWest; Pierce Biotechnology) and quantified by densitometry usingthe Bio-Rad digital densitometry software. Western blots are shown infigures as cropped images.

ResultsNeoadjuvant chemotherapy shifts tumor microbiota

Bacterial load measured by qPCR targeting the V4 region of 16Sgene indicate similar levels of total bacteria in untreated and neoad-juvant chemotherapy treated patients (Fig. 1A), demonstrating thatchemotherapy administration does not reduce total bacterial presencein the tumor tissue. However, analysis of microbiota diversity (Shan-non index) indicates that neoadjuvant chemotherapy administrationsignificantly reduces bacterial diversity within the tumor (Fig. 1B).OTU abundances were aggregated on the basis of genus-level orga-nization and grouped by treatment [untreated (n ¼ 18) versusneoadjuvant chemotherapy (n¼ 15)]. Relative abundance of bacterialgenera in different tumor samples is visualized by bar plots. Each barrepresents a single tumor and each colored box a bacterial taxon.The height of a color box represents the relative abundance of thatorganism within the sample (Fig. 1C). Analysis of phylum propor-tional abundance indicates that tumors from neoadjuvant chemother-apy treated patients do not display any significant regulationof bacteria at phylum-level classification (Fig. 1D). However, analysisof genus-level differences indicate significant increase in tumorPseudomonas abundance (Fig. 1E) and decreased tumoral Prevo-tella abundance (Fig. 1F) from patients with breast cancer treatedwith neoadjuvant chemotherapy.

Specific primary tumor microbiota correlate with distantmetastases

No significant differences in bacterial load (Fig. 2A), Shannondiversity (Fig. 2B), or phylum-level proportional abundance(Fig. 2D) were observed in patients who developed metastases. WeaggregatedOTUabundances into genus-level grouping and segregatedtumor samples based upon whether patients developed metastaticdisease (Fig. 2C). At the genus level (Fig. 2E), primary breast tumorproportional abundance Brevunimonas and Staphylococcus wereincreased in patients who developed metastatic disease. There wereno observed changes in the proportional abundances of Pseudomonasor Prevotella genus taxa (microbes regulated by neoadjuvant chemo-therapy) by tumor recurrence.

Patient obesity modifies tumoral microbiome populationsWe aggregated OTU abundances into genus-level grouping and

segregated tumor samples based upon patient BMI (Supplementary

Chiba et al.

Mol Cancer Res; 18(1) January 2020 MOLECULAR CANCER RESEARCH132

Fig. S1A). Looking at broad phylum level differences, obesity signif-icantly decreased bacteriodetes phylum-level abundance in breasttumor tissue (Supplementary Fig. S1B). There was also a trend forincreased tumoralfirmicutes abundance (P¼ 0.1), partially correlatingwith observed obesity effects in the gutmicrobiome.At the family level,obesity significantly decreased tumor Comamonadaceae (Supplemen-tary Fig. S1C) and Ruminococcacea (Supplementary Fig. S1D) pro-portional abundance. At the genus level, obesity increased tumoralEnterobacteriaceae_unclassified abundance (Supplementary Fig. S1E).

P. aeruginosa laden breast tumor epithelial cells are observedin primary breast tumors and lymph node metastases

Paraffin-embedded primary breast tumors, breast tumor lymphnodemetastases, and tumor adjacent breast tissuewas stained against a

P. aeruginosa specific antibody. See Fig. 3A for primary breast tumorsrepresentative images demonstrating gradient of P. aeruginosa stain-ing positivity associated with each IHC score. Representative primarybreast tumor, lymph node metastases, and normal adjacent mammarygland tissue stained for P. aeruginosa is shown in Fig. 3B. IHCquantification is shown in Fig. 3C. Approximately, 56% of the primarytumors stained positive for P. aeruginosa compared with 20% of thenormal surrounding mammary gland tissue. Lymph node metastasesalso had higher levels of P. aeruginosa with 48% of the cases stainingpositive. There was also differential staining intensity within positivelystained tumor sections. Approximately 26% of tissue sections fromprimary breast tumors had a 2 or higher IHC score, suggesting agradient of tumoral bacterial infection that may result in variedmolecular signaling impacting cancer cell survival or apoptosis.

Figure 1.

Neoadjuvant chemotherapy modulates breast tumor microbiota. Tumor 16S sequencing results were grouped by whether patient was treated withneoadjuvant chemotherapy before definitive surgery (n ¼ 15) or whether patient underwent surgery first (n ¼ 18). A, Total bacterial load was quantifiedby 16S RT-PCR and graphed as 16S copies per mL of DNA. Neoadjuvant chemotherapy did not change total levels of bacteria within a tumor. B, Shannondiversity. Neoadjuvant chemotherapy significantly reduced the bacterial diversity within the tumor. C, Relative abundance of bacterial genera in differenttumor samples is visualized by bar plots. Each bar represents a single tumor and each colored box a bacterial taxon. The height of a color box represents therelative abundance of that organism within the sample. ‘‘Other’’ represents lower abundance taxa. Approximately 25% of total tumor microbiota comprises of105 genus level taxa with less than 1% proportional abundance. D, Primary breast tumors from patients that received neoadjuvant chemotherapy display nobroad spectrum differences in bacterial phylum level proportional abundances. At the genus level, neoadjuvant chemotherapy–treated tumors have elevatedproportional abundance of Pseudomonas (E) and decreased abundance of Prevotella (F) taxa when compared with untreated breast tumors. � , P < 0.05 anderror bars show the min to max distribution.



Bacteria-produced bioactive compounds may modify breastcancer proliferation and chemotherapy responsiveness

To determine whether P. aeruginosa secreted metabolites modulatebreast cancer cell signaling, we used varying doses of P-CM to treatMDA-MB-231, 4T1, MCF7, ZR-75-1, and 67NR cells. We demon-strated that low doses of P-CM (5–10%) stimulate breast cancer cellgrowth within 24 hours of administration in MDA-MB-231, MCF7,andZR-75-1 cell lines, whereas higher doses of P-CM(20%) reduce cellviability in the MDA-MB-231, 4T1, MCF7, and 67NR cell lines(Fig. 4A). High-dose P-CM (20%) stimulated cell growth in theZR-75-1 cell line only. We also combined chemotherapy with P-CM to determine whether bacterial metabolites modify drug respon-siveness. P-CM enhanced doxorubicin-mediated cell death in the

MDA-MB-231, 4T1, and theMCF7 cell lines. Combining doxorubicinwith 10% P-CM had no negative effect in the 67NR cell line. However,P-CM reduced doxorubicin-killing efficacy in the ZR-75-1 cell line,possibly suggesting a cell-line–specific phenotype. However, for themajority of the cell line tested, the data indicate that although low-doseP-CM may stimulate growth, other secreted factors potentiate che-motherapy cancer cell killing (Fig. 4B). To further identifywhich of thepotential P. aeruginosa metabolites may be having these effects, wetreatedMDA-MB-231 cells with LPS, pyocyanin, lectin, phospholipaseD, and exotoxin A in the presence or absence of doxorubicin (Fig. 4C).LPS stimulated MDA-MB-231 cell proliferation whereas pyocyanin(a P. aeruginosa-derived metabolite) potentiated cancer cell death.Lectin while having no significant overall effect alone on MDA-MB-

Figure 2.

Primary breast tumor microbiota may predict tumor recurrence. Tumor 16S sequencing results were grouped by whether patient developed metastases within5 years of primary tumor resection surgery (n¼ 9) or whether patient did not developmetastases (n¼ 24).A, Total bacterial loadwas not different between primarytumors that displayed recurrence. B, Tumor recurrence had no effect on Shannon diversity within the primary tumor. C, Relative abundance of bacterial genera indifferent tumor samples is visualized by bar plots. Each bar represents a single tumor and each colored box a bacterial taxon. The height of a color box represents therelative abundance of that organism within the sample. ‘‘Other’’ represents lower abundance taxa. D, Primary breast tumors from patients who later on developedmetastases did not display differences in bacterial phylum level proportional abundances. E, At the genus level, primary breast tumors from patients who later ondevelop metastases have elevated proportional abundance of Brevundimonas and Staphylococcus taxa. � , P < 0.05 and error bars show the min to max distribution.

Chiba et al.


231 proliferation, when combined with doxorubicin potentiated che-motherapy-mediated breast cancer cell killing.

Potential molecular mechanisms driving the dose-dependentduality of P-CM

Treatment of MDA-MB-231 and MCF7 breast cancer cells with P-CM resulted in a dose-dependent stimulation phosphorylated Akt(Ser473), whereas all doses of P-CM stimulated pAkt in the ZR-75-1breast cancer cells (Fig. 5A). We did not observe pAkt activation in4T1 or 67NR cell lines, both of which were not growth stimulated bylow-dose P-CM (Fig. 4A). When combining P-CM with doxorubicin,protein analysis indicates that P-CM enhances chemotherapy-mediated killing through modulation of apoptosis. Treatment with10% P-CM doxorubicin significantly increased doxorubicin-inducedcleaved caspase-7 protein levels in the MDA-MB-231 and MCF7breast cancer cell lines when compared with doxorubicin treatmentalone (Fig. 5B). P-CM treatment did not negatively affect doxorubicinapoptosis induction in the 4T1 cell line. However, treatment with 10%P-CM reduced doxorubicin-mediated cleaved caspase-7 in the ZR-75-1breast cancer cell line.

DiscussionThe human body contains slightly more bacterial cells than human

cells (1.3 bacterial cells to each human cell in the body), according tonew estimates refuting the previously estimated 10:1 ratio establishedin the 1970s (13). Although the majority of the bacteria biomass iscontained within the intestinal track, microbes in lower abundancehave been identified in other organs located outside the gastrointes-tinal tract, including the mammary gland (1). Our group recentlyidentified that diet can modulate the mammary gland microbiotapopulation, suggesting the plasticity of the breast microbiome (5).Moreover, recent studies demonstrate that breast tumors displaydistinct bacterial populations when compared with the surroundingmammary gland tissue microbiota (2–4, 7). Although other reportshave highlighted the presence of bacteria in breast tumor tissues, thesestudies have not shown the impact of therapy on the tumor micro-biome. Moreover, these initial breast tumor microbiome studies, forthe most part, fail to indicate the possible functional relevance of thesemicrobes in the tumor microenvironment. In the study by Urbaniakand colleagues (2), authors demonstrated that the Escherichia coli

Figure 3.

Elevated P. aeruginosa abundance in primary breasttumors and lymph node metastases when com-pared with normal surrounding mammary glandtissue. A, Representative images of primary breasttumorswith differentialP. aeruginosa IHC scoring.B,Representative images of primary breast tumorsand lymph node metastases, and normal tumoradjacent mammary gland tissue stained with ananti-P. aeruginosa antibody. C, Table of the quan-tified P. aeruginosa tissue staining. Percentages andtotal number of cases shown (n ¼ 10–50).



found in the surrounding mammary gland tissue can stimulate DNAbreakage which could induce breast tumorigenesis indicating a pos-sible role of the microbiota in breast tumor formation. Our studyaimed to determine impact of breast tumor microbiota on therapeuticresponsiveness and breast cancer outcomes. Our 16S sequencing dataof breast tumors from untreated or neoadjuvant chemotherapy treatedpatients indicates that chemotherapy increases the tumor proportionalabundance of Pseudomonas. Normal mammary gland tissue display alow proportional abundance [approximately 5%; for reference seeUrbaniak and colleagues (1)] of Pseudomonas, whereas breast tumortissue contains elevated (approximately 20%) Pseudomonas. Neoad-juvant chemotherapy further increases the proportional abundance ofPseudomonas to 85%, suggesting chemotherapy induces preferentialgrowth or survival of these bacteria.

Although Pseudomonas genus contains over 140 species, most ofthese are found in water or soil (14). Pseudomonas genus contains onlya few animal pathogenic strains: P. aeruginosa (wound infections,cystic fibrosis, and hospital-based infections in humans), P. oryziha-bitans (sepsis in humans), and P. plecoglossicida (hemorrhagic ascites

in fish). Although we currently do not know the strain of Pseudomonasfound in the 16S sequencing results from the neoadjuvant chemo-therapy treated breast tumors, we anticipate that it is most likelyP. aeruginosa. The TMA staining data using a P. aeruginosa-specificantibody indicates that this species is found in breast tumor samples.Future studies will focus on the isolation and characterization ofbreast tumor–specific Pseudomonas populations. P. aeruginosa isimplicated in hospital infections (15). Although P. aeruginosa infec-tion is well studied in ventilator-based infections, chronic wounds, andcystic fibrosis, the impact in tumorigenesis is unknown. Molecularevidence supplied by the P. aeruginosa wound healing field suggeststhat P. aeruginosa biofilm and secreted factors may play importantroles in various signaling pathways including cell death, proliferation,chemotherapy responsiveness, and/or development of metastases.The signaling mechanism stimulated by P. aeruginosa secretedfactor may be dependent on concentration driven by tumoralP. aeruginosa abundance. P. aeruginosa secretes numerous proteases,lipases, lipopolysaccharides, quinolones, pyocyanin, Cif, or hemoly-sin-coregulated proteins (16–20). It also forms multicellular biofilms

Figure 4.

Secreted P. aeruginosa metabolites effects breast cancer cell survival and doxorubicin (DOX) responsiveness. A, Differential effects of P. aeruginosa conditionedmedia (P-CM) on MDA-MB-231 human triple negative breast cancer cell, 4T1 murine triple negative breast cancer cell, MCF7 human ERþ breast cancer cell, ZR-75–1human ERþ breast cancer cell, and 67NR murine ERþ breast cancer cell viability (n ¼ 4–6; � , P < 0.05). B, Combining P-CM and doxorubicin enhances anticancerchemotherapy effects onMDA-MB-231 human triple negative breast cancer cells, 4T1 murine triple negative breast cancer cells, and MCF7 human ERþ breast cancercells (n¼ 4). C,Administration of individual P. aeruginosa secretedmetabolites pyocyanin, lectin (PA-I), LPS, exotoxin A, and phospholipase D onMDA-MB-231 cellsviability and doxorubicin responsiveness (n ¼ 6; � , P < 0.05).

Chiba et al.


which are resistant to biotic and abiotic stressors. Literature showedthat P. aeruginosa can grow in the presence of chemotherapy andthat doxorubicin stimulated P. aeruginosa biofilm production (21).Therefore, chemotherapy administration may offer selective pressurekilling off other bacterial populations while concurrently enablingPseudomonas outgrowth in the tumor microenvironment. This was

supported by the similar total bacterial counts in each treatment group(untreated versus neoadjuvant chemotherapy) but lower alpha diver-sity scores in the chemotherapy-treated tumors, for example, sameamount of bacteria but decreased variety of strains expressed.

P. aeruginosa secreted factors may also directly impact breastcancer cell proliferative signaling. P. aeruginosa type VI secretion

Figure 5.

Apoptotic and proliferativemolecular signaling pathwaysmodulated by secreted P. aeruginosametabolites on breast cancer cells.A, Treatmentwith low-dose P-CMstimulates prosurvival Akt signaling in MDA-MB-231, MCF7, and ZR-75-1 breast cancer cells (n ¼ 3; � , P < 0.05). B, Administration of P-CM with doxorubicin (DOX)increases cleaved caspase-7 protein expression in MDA-MB-231 and MCF7 breast cancer cells, indicating P-CM enhances apoptosis when combined withchemotherapy in several breast cancer cell lines (n ¼ 3; �, P < 0.05).



phospholipase D effector (PldA) was shown to effect host lungepithelial cell signaling through PI3K/Akt activation (18). P. aerugi-nosa is also a potent producer of LPS (22). LPS-mediated TLR4stimulationwas previously shown to promote breast cancermetastasesthrough Akt activation (23). We show that P. aeruginosa conditionedmedia at low concentrations (5%–10%) stimulate MDA-MB-231, ZR-75-1, andMCF7 breast cancer cell growth and activate protumorigenicAkt signaling, suggesting that the mid-level proportional abundance(20%) of Pseudomonas found in untreated breast tumorsmay promotetumor survival and growth. Treatment ofMDA-MB-231 cells with LPSpromoted proliferation, whereas administration of cancer cells withphospholipase D (albeit derived from Streptomyces) did not affectMDA-MB-231 viability, suggesting the progrowth mediated by P-CMmay be due to LPS secretion.

Literature indicates that hypoxia (low oxygen concentrations oftenfound in solid tumors) decreases the virulence capacity of P. aerugi-nosa, as indicated by decreased pyocyanin secretion (24). Pyocyanin isa potent reactive oxygen species (ROS) inducer. In the cancer field ROSis often considered diphasic; low levels of ROS are protumorigenicwhereas higher ROS concentrations promote cell death. Therefore,intratumoral P. aeruginosa subjected to hypoxia would decreasepyocyanin production, promoting lower levels of ROS and prosurvivalsignaling. The decreased virulence observed by P. aeruginosa subjectedto hypoxia may also give evidence to why elevated tumoral levels ofbacteria do not result in systemic infection. We demonstrated thatpyocyanin administration reduced MDA-MB-231 breast cancer cellgrowth with an observed trend to enhance doxorubicin-killing, sug-gesting a possible ROS-dependent mechanism of P-CM potentiationof chemotherapy observed in the MDA-MB-231 cells.

In the tumor microenvironment, P. aeruginosa may have potentimmunomodulatory effects. P. aeruginosa secretes numerous productsincluding zinc metalloproteases (AprA and LasB to combat hostimmune response; ref. 16), Cif (inhibits CFTR and TAP1, preventingantigen presentation; ref. 20), pyocyanin (inhibits lymphocyte activity;ref. 17), among others that can affect immune cell recruitment andfunction. This suggests that tumor-specific P. aeruginosa producedbioactive compounds may enhance cancer immunoavoidance (25).

Using P-CM, we show that P. aeruginosa secreted factors enhancedoxorubicin killing capacity. Transcriptomic analysis of doxorubicineffects on P. aeruginosa indicate doxorubicin increased pqsH geneexpression (21). PqsH is a FAD-dependent monooxygenase that isrequired for the production of the Pseudomonas quinolone signal(PQS; ref. 26). PQS, while predominately implicated in quorumsensing, has been linked with p38 MAPK signaling (proapoptotic)and NF-kB inhibition (27). Doxorubicin treatment can subsequentlyactivate NF-kB signaling, initiating an inherent pathway promotingtherapeutic resistance (28). Therefore, PQS-mediated inhibition ofNF-kBmay enhance doxorubicin-mediating killing. PQS also serves asa ferric iron chelator, where exogenous PQS treatment results in ironstarvation in the surrounding environment (29). Cancer cells exhibit“iron addiction” and therapies to reduce iron levels were previouslyshown to suppress breast tumor growth (30, 31), suggesting anotherpossible molecular mechanism by which Pseudomonas secreted com-pounds may potentiate chemotherapy responsiveness.

Because of the annotation of breast tumor samples, we were able tocompare primary breast tumor bacteria populations between patientswho displayed tumor recurrence (within 5 years of surgery) and thosepatients who did not have recurrence. We demonstrate elevatedBrevundimonas and Staphylococcus in primary tumors from patientswith breast cancer who later developed metastatic disease. Previoushuman breast tumor microbiome studies have identified elevated

Brevundimonas prevalence in all subtypes (ERþ, HER2þ, and TNBC)of breast tumors when compared with nonmatched control breasttissue (8). However, to our knowledge, this is the first report suggestingelevated Brevundimonas correlates with the development of metasta-ses. Further studies are needed to explore the potential impact of breasttumor Brevundimonas on cancer cell migration and metastatic sig-naling cascades.

Many criticisms for tumor microbiome studies include the possi-bility of sample contamination skewing outcomes (32). Lowmicrobialabundance in tissue, such as mammary gland or breast tumors, wouldbe highly sensitive for sample contamination being detected as real.These outside bacterial contaminationsmay occur in the surgical suite,at the pathologist, in the tumor bank storage, during DNA isolation, orduring the paraffin-embedding process just to name a few possibilities.Our study takes into consideration all of these variables. First, ourbreast tumor sample set was collected over a 9-year period (2004–2013), limiting any possible contaminations that could have occurredeither in the surgical suite or at the pathologist. Having the samebacteria populations at all of these locations over a 9-year periodwouldbe rare. Second, to control for any possible tumor bank storage orDNAisolation contamination, we used a tumor microarray purchased froman outside company to serve as a validation cohort for our significantlyregulated microbes that were detected by 16S sequencing. Further-more, any environmental contamination that could occur duringDNAisolation of the samples would be across all the groups because theseanalysis were performed as batched samples and therefore would notbe differentially expressed depending on treatment group. Our tumorbank samples also serve as a control for any possible contamination inthe paraffin-embedding process in the tumormicroarray cohort. Thesecontingencies for outside contamination overall strengthen our study,thereby demonstrating that breast tumors have their own microbiotapopulation and that these bacterial populations are modified byneoadjuvant chemotherapy which modulates breast cancer cell sig-naling to impact outcome. Furthermore, we also show functionalrelevance of breast tumor–specific P. aeruginosa, modulating breastcancer cell proliferation and doxorubicin sensitivity.

We demonstrate that therapeutic modalities shift the tumor micro-biota, suggesting a role of certain bacteria in potentially increasing theresponse rate to neoadjuvant chemotherapy and improved canceroutcomes. These data may also be extrapolated to suggest tumormicrobiota may impact outcomes such as improved disease-freesurvival and/or overall survival. Our observation that certain primarybreast tumor microbiota populations have increased likelihood ofbreast cancer recurrence suggests that modulating tumor microbiomecould potentially decrease the rate of tumor recurrence. However,further studies with increased samples are needed to more clearlydelineate the role of tumor microbiota in metastases. In this retro-spective analysis of operable breast cancer, breast tumor microbiotamay improve responsiveness to neoadjuvant chemotherapy byincreasing therapeutic efficacy. Further research is warranted toevaluate the possible role of microbiome in improving breast canceroutcomes.


Authors’ ContributionsConception and design: A. Chiba, M. Howard-McNatt, K.L. CookDevelopment of methodology: A. Chiba, K.L. CookAcquisition of data (provided animals, acquired and managed patients, providedfacilities, etc.): A. Chiba, A. Bawaneh, C. Velazquez, E.A. Levine, N. Levi-Polyachenko, S.A. Yates-Alston, D.R. Soto-Pantoja, K.L. Cook

Chiba et al.


Analysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): D.R. Soto-Pantoja, K.L. CookWriting, review, and/or revision of the manuscript: A. Chiba, M. Howard-McNatt,E.A. Levine, N. Levi-Polyachenko, S.A. Yates-Alston, S.P. Diggle, D.R. Soto-Pantoja,K.L. CookAdministrative, technical, or material support (i.e., reporting or organizing data,constructing databases): K.Y.J. Clear, A.S. WilsonStudy supervision: K.Y.J. Clear, K.L. Cook

AcknowledgmentsD.R. Soto-Pantoja was supported by K22CA181274. K.L. Cook was supported by

the Chronic Disease Research Fund, an American Cancer Society Research Scholar

grant (RSG-16-204-01-NEC), and a Career Catalyst grant from the Susan G. KomenFoundation (CCR18547795). Shared Resource services were provided by the WakeForest Baptist Comprehensive Cancer Center's NCI Cancer Center Support GrantP30CA012197.

The costs of publication of this article were defrayed in part by the payment of pagecharges. This article must therefore be hereby marked advertisement in accordancewith 18 U.S.C. Section 1734 solely to indicate this fact.

Received April 28, 2019; revised August 26, 2019; accepted October 15, 2019;published first October 18, 2019.

References1. Urbaniak C, Cummins J, Brackstone M, Macklaim JM, Gloor GB, Baban CK,

et al. Microbiota of human breast tissue. Appl Environ Microbiol 2014;80:3007–14.

2. Urbaniak C, Gloor GB, Brackstone M, Scott L, Tangney M, Reid G. Themicrobiota of breast tissue and its association with breast cancer.Appl Environ Microbiol 2016;82:5039–48.

3. WangH, Altemus J, Niazi F, Green H, Calhoun BC, Sturgis C, et al. Breast tissue,oral and urinary microbiomes in breast cancer. Oncotarget 2017;8:88122–38.

4. Hieken TJ, Chen J, Hoskin TL,Walther-AntonioM, Johnson S, Ramaker S, et al.The microbiome of aseptically collected human breast tissue in benign andmalignant disease. Sci Rep 2016;6:30751.

5. Shively CA, Register TC, Appt SE, Clarkson TB, Uberseder B, Clear KYJ, et al.Consumption of Mediterranean versus Western diet leads to distinct mammaryglandmicrobiome populations: implications for breast cancer. Cell Rep 2018;25:47–56.

6. Aslakson CJ, Miller FR. Selective events in the metastatic process defined byanalysis of the sequential dissemination of subpopulations of amousemammarytumor. Cancer Res 1992;52:1399–405.

7. ThompsonKJ, Ingle JN, Tang X, Chia N, Jeraldo PR,Walther-AntonioMR, et al.A comprehensive analysis of breast cancer microbiota and host gene expression.PLoS One 2017;12:e0188873.

8. Banerjee S, Tian T, Wei Z, Shih N, Feldman MD, Peck KN, et al. Distinctmicrobial signatures associated with different breast cancer types.Front Microbiol 2018;9:951.

9. Rubovszky G, Horvath Z. Recent advances in the neoadjuvant treatment ofbreast cancer. J Breast Cancer 2017;20:119–31.

10. Jones SE, Savin MA, Holmes FA, O'Shaughnessy JA, Blum JL, Vukelja S, et al.Phase III trial comparing doxorubicin plus cyclophosphamide with docetaxelplus cyclophosphamide as adjuvant therapy for operable breast cancer. J ClinOncol 2006;24:5381–7.

11. Kozich JJ,Westcott SL, Baxter NT,Highlander SK, Schloss PD.Development of adual-index sequencing strategy and curation pipeline for analyzing ampliconsequence data on the MiSeq Illumina sequencing platform. Appl EnvironMicrobiol 2013;79:5112–20.

12. Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, et al.Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities.Appl Environ Microbiol 2009;75:7537–41.

13. Sender R, Fuchs S, Milo R. Revised estimates for the number of human andbacteria cells in the body. PLoS Biol 2016;14:e1002533.

14. Iglewski BH. Pseudomonas. In: Baron S, editor. Medical microbiology. 4th ed.Galveston, TX: University of Texas Medical Branch at Galveston; 1996.

15. Fazeli H, Akbari R, Moghim S, Narimani T, Arabestani MR, Ghoddousi AR.Pseudomonas aeruginosa infections in patients, hospital means, and personnel'sspecimens. J Res Med Sci 2012;17:332–7.

16. Casilag F, Lorenz A, Krueger J, Klawonn F, Weiss S, Haussler S. The LasBelastase of Pseudomonas aeruginosa acts in concert with alkaline proteaseAprA to prevent flagellin-mediated immune recognition. Infect Immun 2016;84:162–71.

17. Hall S, McDermott C, Anoopkumar-Dukie S, McFarland AJ, Forbes A,Perkins AV, et al. Cellular effects of pyocyanin, a secreted virulence factorof Pseudomonas aeruginosa. Toxins 2016;8:236.

18. Jiang F, Waterfield NR, Yang J, Yang G, Jin Q. A Pseudomonas aeruginosa typeVI secretion phospholipase D effector targets both prokaryotic and eukaryoticcells. Cell Host Microbe 2014;15:600–10.

19. Nakagami G, Minematsu T, Morohoshi T, Yamane T, Kanazawa T, Huang L,et al. Pseudomonas aeruginosa quorum-sensing signaling molecule N-3-oxododecanoyl homoserine lactone induces matrix metalloproteinase 9expression via the AP1 pathway in rat fibroblasts. Biosci Biotechnol Biochem2015;79:1719–24.

20. Stanton BA. Effects of Pseudomonas aeruginosa on CFTR chloride secretion andthe host immune response. Am J Physiol Cell Physiol 2017;312:C357–C66.

21. Groizeleau J, Rybtke M, Andersen JB, Berthelsen J, Liu Y, Yang L, et al. The anti-cancerous drug doxorubicin decreases the c-di-GMP content in Pseudomonasaeruginosa but promotes biofilm formation. Microbiology 2016;162:1797–807.

22. Pier GB. Pseudomonas aeruginosa lipopolysaccharide: a major virulence factor,initiator of inflammation and target for effective immunity. Int J Med Microbiol2007;297:277–95.

23. Li J, Yin J, Shen W, Gao R, Liu Y, Chen Y, et al. TLR4 promotes breast cancermetastasis via Akt/GSK3b/b-catenin pathway upon LPS stimulation. Anat Rec2017;300:1219–29.

24. Schaible B, Rodriguez J, Garcia A, von KriegsheimA,McClean S, Hickey C, et al.Hypoxia reduces the pathogenicity of Pseudomonas aeruginosa by decreasingthe expression of multiple virulence factors. J Infect Dis 2017;215:1459–67.

25. Sadikot RT, Blackwell TS, Christman JW, Prince AS. Pathogen-host interactionsin Pseudomonas aeruginosa pneumonia. Am J Respir Crit Care Med 2005;171:1209–23.

26. Gallagher LA, McKnight SL, Kuznetsova MS, Pesci EC, Manoil C. Functionsrequired for extracellular quinolone signaling by Pseudomonas aeruginosa.J Bacteriol 2002;184:6472–80.

27. Lin J, Cheng J, Wang Y, Shen X. The Pseudomonas quinolone signal (PQS): notjust for quorum sensing anymore. Front Cell Infect Microbiol 2018;8:230.

28. Kim JY, Jung HH, Ahn S, Bae S, Lee SK, Kim SW, et al. The relationship betweennuclear factor (NF)-kappaB family gene expression and prognosis in triple-negative breast cancer (TNBC) patients receiving adjuvant doxorubicin treat-ment. Sci Rep 2016;6:31804.

29. Diggle SP,Matthijs S,Wright VJ, FletcherMP, Chhabra SR, Lamont IL, et al. ThePseudomonas aeruginosa 4-quinolone signal molecules HHQ and PQS playmultifunctional roles in quorum sensing and iron entrapment. Chem Biol 2007;14:87–96.

30. Manz DH, Blanchette NL, Paul BT, Torti FM, Torti SV. Iron and cancer: recentinsights. Ann N Y Acad Sci 2016;1368:149–61.

31. Hoke EM, Maylock CA, Shacter E. Desferal inhibits breast tumor growth anddoes not interfere with the tumoricidal activity of doxorubicin. Free Radic BiolMed 2005;39:403–11.

32. Eisenhofer R, Minich JJ, Marotz C, Cooper A, Knight R, Weyrich LS. Contam-ination in low microbial biomass microbiome studies: issues and recommenda-tions. Trends Microbiol 2019;27:105–17.



CANCER PREVENTION RESEARCH | RESEARCH ARTICLE

Oral Microbiome Profiling in Smokers with andwithout Head and Neck Cancer Reveals VariationsBetween Health and DiseaseAshok Kumar Sharma1, William T. DeBusk2, Irina Stepanov3, Andres Gomez1, andSamir S. Khariwala2

ABSTRACT◥

While smoking is inextricably linked to oral/head andneck cancer (HNSCC), only a small fraction of smokersdevelop HNSCC. Thus, we have sought to identify otherfactors, which may influence the development of HNSCC insmokers including microbiology. To determine microbialassociations with HNSCC among tobacco users, we charac-terized oral microbiome composition in smokers with andwithout HNSCC. 16S rRNA MiSeq sequencing was used toexamine the oral mucosa microbiome of 27 smokers with(cases) and 24 without HNSCC (controls). In addition, wecorrelated previously reported levels of DNA damage withthe microbiome data. Smokers with HNSCC showed lowermicrobiome richness compared with controls (q ¼ 0.012).Beta-diversity analyses, assessed as UniFrac (weighted andunweighted) and Bray–Curtis distances, showed significantdifferences in oral mucosal microbiome signatures between

cases and controls (r2 ¼ 0.03; P ¼ 0.03) and higherinterindividual microbiome heterogeneity in the former(q ≤ 0.01). Higher relative abundance of Stenotrophomonasand Comamonadaceae and predicted bacterial pathwaysmainly involved in xenobiotic and amine degradationwere found in cases compared with controls. The latter, incontrast, exhibited higher abundance of common oral com-mensals and predicted sugar degradation pathways. Finally,levels of DNA damage in the oral cavity were correlatedwith the microbiome profiles above. Oral microbiometraits differ in smokerswith andwithoutHNSCC, potentiallyinforming the risk of eventual HNSCC and sheddinglight into possible microbially mediated mechanisms ofdisease. These findings present data that may be useful inscreening efforts for HNSCC among smokers who areunable to quit.

IntroductionOral/head and neck cancer (HNSCC) represents a group of

tumors strongly associated with tobacco use. While humanpapillomavirus (HPV) has more recently been identified as afrequent cause of oropharyngeal cancers, most head and neckcancers in the oral cavity, larynx, hypopharynx (any many inthe oropharynx) are associated with tobacco use (1). Althoughup to 90% of those with HNSCC use tobacco, there are many

smokers who do not develop HNSCC. To better understandthis issue, we have previously examined the levels of tobacco-related DNA damage occurring in the oral cavity amongsmokers with and without HNSCC (2, 3). In this work, weidentified higher rates of DNA damage in those smokers withHNSCC compared with cancer-free smokers. However, tobac-co carcinogen exposure was similar in both groups such that wesought another explanation for varying levels of DNA damagein the oral cavity.One possible explanation for the varying levels of oral cell

DNA damage found in our prior work pertains to the oralmicrobiome. The impact of the bacterial microbiome onhuman disease has been explored and discussed extensively (4).Specifically, it has been hypothesized that specific microbiomefingerprints in susceptible individuals may influence diseasephenotypes in a variety of infectious, metabolic, and immunedisorders as well as in cancer (5, 6). In cancer, the role ofmicroorganisms in impacting disease onset and progressionhas been deemed as central (7), including such examplesas the role of Helicobacter pylori, Salmonella typhi, andChlamydia pneumoniae in gastric, gallbladder, and lung can-cer, respectively.Because of its multiple risk factors and associated disorders,

multifactorial nature, and the highly diverse oral micro eco-system (second in diversity after the colonic microbiome),studying bacterial carcinogens in the oral cavity poses

1Department of Animal Science and Microbial and Plant Genomics Institute,University of Minnesota, Minneapolis, Minnesota. 2Department of Otolaryngol-ogy-Head and Neck Surgery, University of Minnesota, Minneapolis, Minnesota.3Division of Environmental Health Sciences, School of Public Health, Universityof Minnesota, Minneapolis, Minnesota.

Note: Supplementary data for this article are available at Cancer PreventionResearch Online (http://cancerprevres.aacrjournals.org/).

A.K. Sharma, W.T. DeBusk, and A. Gomez contributed equally to this article.

Corresponding Authors: Samir S. Khariwala, University of Minnesota, MMC 396,420Delaware St SE,Minneapolis, MN55455. Phone: 612-625-0912; Fax: 612-625-201; E-mail: [email protected]; and Andres Gomez, Microbial and Plant Geno-mics Institute, University of Minnesota, 495D AS/VM, 1988 Fitch Ave, St. Paul,MN, 55108; Phone: 612-624-9744; E-mail: [email protected]

Cancer Prev Res 2020;13:463–74

doi: 10.1158/1940-6207.CAPR-19-0459



http://cancerpreventionresearch.aacrjournals.org/cgi/content/full/1940-6207.CAPR-19-0459?ijkey=1JGwGhzUiH9q.&keytype=ref&siteid=aacrjnls

significant challenges. For instance, culture-dependent, andnarrow-range molecular techniques (e.g., checkerboardhybridization) initially showed differences in the abundanceof specific oral commensals in saliva of patients affected withHNSCC compared with healthy subjects (8). In addition,cancerous tissues in HNSCC have shown greater abundanceand diversity of culturable bacteria (aerobes and anaerobes) incontrast with normal tissues (9, 10). However, the use of next-generation sequencing techniques to profile microbiomes hasrevealed novel microbial dynamics in HNSCC. Given thisinformation, we used 16S rRNA MiSeq to profile the oralmicrobiome in a group of smokers with HNSCC who havedemonstrated higher levels of DNA damage than smokerswithout HNSCC.

Materials and MethodsStudy subjectsThis study was approved by the University of Minnesota

Research Subjects’ Protection Programs Institutional ReviewBoard:HumanSubjectsCommittee (IRBStudy# 0903M62203).Patients with HNSCC were identified and enrolled followingwritten informed consent during outpatient visits to the Uni-versity of Minnesota Otolaryngology-Head and Neck SurgeryClinic with a new diagnosis of squamous cell carcinoma of theupper aerodigestive tract. This included tumors of the oralcavity, oropharynx, larynx, and hypopharynx. In some cases,cancers were first identified in our clinic whereas in others,cancers were diagnosed at outside institutions and referred toour clinic for definitivemanagement. Inclusion criteria includedself-report of current daily smoking and having smoked at leastfive cigarettes per day for at least 5 years. A total of 66 cases(smokers with HNSCC) and 51 controls (smokers withoutHSCC) were enrolled in the study from February 2014 to May2017. DNA sufficient for analysis was available in 27 cases and24 controls.Cancer-free controls were recruited in the same outpatient

clinic. The control subjects were visiting the clinic for clinicalevaluation of problems other than cancer (i.e., sinusitis, hearingloss) and were approached for enrollment upon identificationas daily cigarette smokers. All enrolled subjects were smokingcigarettes daily at the time of enrollment. Demographic datacollected included cigarettes per day, duration of use, alcoholicdrinks per day, alcoholic frequency, and tumor-related vari-ables such as subsite and stage.

Buccal cell collectionOral brushings from the buccal mucosa were collected from

enrolled subjects through the Department of Otolaryngology-Head and Neck Surgery at the University of Minnesota (Min-neapolis, MN). Oral cells were collected by brushing the oralmucosa inside one cheek with a clean toothbrush and swirlingthe brush in a sterile polypropylene centrifuge tube with acommercial mouthwash to transfer the collected buccal cellsfrom the brush into the liquid. After the collection, the sampleswere centrifuged at 2,700 rpm to pellet cells; the pellets were

washed with Tris-EDTA buffer (pH 7.4) and stored at �20�Cuntil DNA isolation and analysis.

Microbiome analysesExtraction of genomic DNA from each brushing was per-

formed using the DNA Purification Kit (Qiagen). After DNAintegrity was measured, high-quality DNA of 27 cases and 24controls was used for oral bacteria community profilingthrough 16S rRNA amplicon sequencing, targeting the V4hyper variable region (barcode primer pair 515 f -GTGCC-AGCMGCCGCGGTAA and 806r-GGACTACHVGGGTW-TCTAAT) on the Illumina MiSeq sequencing platform. Rawreads were trimmed to remove primers using cutadapt, andfiltered to remove low-quality reads (less than Q ¼ 30) usingfastx_toolkit. High-quality reads were considered for down-stream analysis using the DADA2 plugin within qiime2 (11),which performs denoising, merging of paired-end reads, andremoval of chimeric sequences to produce unique ampliconsequence variants (ASV). Taxonomic assignment of theseASVs was carried out by the trained na€�ve Bayes classifier onreference sequences (clustered at 99% sequence identity)from Greengenes 13_8, using feature-classifier fit-classifier-naive-bayes, and feature-classifier classify-sklearn pluginswithin qime2. ASV abundances were used for functionalprediction analyses using Phylogenetic Investigation of Com-munities by Reconstruction of Unobserved States (PICRUSt,version 2; maximum nearest sequenced taxon index ¼ 2), andpredicted KEGG pathway relative abundances were used forfurther analysis.

DNA adduct analysisThe process for DNA adduct quantification has been

described previously, and values for these samples werereported separately (2, 3). Briefly, DNA was isolatedfrom the collected samples by using the commercialDNA Purification Kit (Qiagen). The isolated DNA sampleswere subjected to acid hydrolysis to release 4-hydoxy-1-(3-pyridyl)-1-butonanone (HPB) and purified on 25-mgHyperSep Hypercarb Artridges (Thermo Fisher Scientific).The analysis of HPB in the purified samples was carried outon an LTQ Orbitrap Velos Instrument (Thermo Fisher Scien-tific) interfaced with a Nano2D-LC HPLC (Eksigent) systemwith nanoelectrospray ionization.

Statistical analysesAll microbial community ecology analyses were performed

within the R statistical interface (12). Relative abundances ofeach ASV on complete and rarefied data (depth ¼ 1,000sequences) along with the rooted tree (generated using thealignment mafft plugin on qiime2 and representativesequences) were used for the calculation of distance matrices(Bray–Curtis and UniFrac), ordination analysis [principalcoordinates analysis (PCoA)], and alpha diversity indices(observed taxa and Shannon's H indices) using the R phyloseqpackage (13). Permutational multivariate analysis of variance(PERMANOVA) was calculated using the adonis function

Sharma et al.

Cancer Prev Res; 13(5) May 2020 CANCER PREVENTION RESEARCH464

within the R vegan package (14). Random forest (RF)classifi-cation models at ntree ¼ 500 and cv ¼ 10 folds were con-structed in the random Forest R package (15) to assess clas-sification accuracy of cases and controls based on microbialtaxonomic features and predicted KEGG pathways. These RFclassification models were used to select potential discriminat-ing taxa and predicted pathways based on their respectivemeandecrease in accuracy observations. In addition, indicator spe-cies analyses within the labdsv package (16) and Wilcoxonrank-sum test, always correcting for FDR with an acceptablethreshold of 0.05, were applied to check for discriminatorypower and statistical significance of all discriminant features.Strength and direction of association between (i) significantlydiscriminating taxa, (ii) between HPB-releasing adductslevels and significantly discriminating taxa, and (iii) betweenpredicted pathways and significantly discriminating taxa wasmeasured using Spearman correlations coefficients withinthe R psych package (http://www.test.personality-project.org/r/psych/ psych-manual.pdf; ref. 17). All graphs were plottedusing the ggplots R package (18), whereas correlation networkswere generated and visualized using Cytoscape v3.7.1 (19). Allanalysis codes along with the data files are available on https://github.com/ashoks773/Oral-microbiome-16S.

ResultsDemographic data and tobacco useOur study population consisted of 24 cigarette smokers

(controls) and 27 smokers who had developed tumors of theupper aerodigestive tract (cases). Demographics data are sum-marized in Table 1. Mean age of cases was 58 years, and themean age of controls was 48 years. Eighty-three percent of caseswere male, and 57% of controls were male. Tumor site distri-bution was oropharynx (12/27, 44%), larynx (6/27, 22%), oralcavity (6/27, 22%), hypopharynx (2/27, 7%), and other (1/27,3%). Univariate analysis revealed both groups reported thesame level of cigarette exposure (14 cigarettes/day). Meanduration of cigarette use was 22.3 years for cases and 23.6 yearsfor controls.

Oral microbiome profiles in smokers with and withouthead and neck cancerA total of 1,544,643 16S rRNA sequence reads were obtained,

1,317,658 of which remained after quality filtering, reflecting1,463 unique ASVs, each representing a unique taxon. Aftertaxonomic assignment, we obtained an average sequencingdepth of 22,082 reads per sample (range ¼ 537–62532, SD ¼15482; Supplementary Table S1). Lower microbiome richness(number of observedASVs)was observed in smokerswith headand neck cancer, compared with controls (Wilcoxon rank-sumtests, q ¼ 0.012), and using both all reads (Fig. 1A) or a set of1,000 reads randomly selected in each sample to control forsequencing depth differences (Supplementary Fig. S1).Although these trends were maintained, no differences wereobserved in terms of the Shannon's H diversity index (q >0.05; Fig. 1A).

Beta-diversity analyses assessed as UniFrac (weightedand unweighted) and Bray–Curtis distances, showed signif-icant differences in oral mucosal microbiome signaturesbetween cases and controls (PERMANOVA, weightedUniFrac: r2 ¼ 0.04, P ¼ 0.05; unweighted UniFrac: r2 ¼0.05, P ¼ 0.001; and Bray–Curtis distances: r2 ¼ 0.03, P ¼0.03), along axis 1 or 2 of a PCoA ordination. Ordinationscores differed significantly depending on distance chosen(Fig. 1B; q < 0.01, Wilcoxon rank-sum test), with unweight-ed UniFrac and Bray–Curtis distances showing the mostdiscriminant patterns. In addition, the potential effect ofalcohol consumption on oral mucosal microbiome signa-tures was explored, which was found not significant (Sup-plementary Table S2). To overcome potential sequencingdepth biases across samples, we rarefied the data at 1,000reads per sample, observing the same beta diversity patternsas when using all reads generated (Supplementary Fig. S2A–S2C). Beta diversity analyses also showed significantlyhigher interindividual microbiome variation among smo-kers with head and neck cancer compared with controls,demonstrating high oral mucosal microbiome heterogeneityunder disease conditions (Fig. 1C; q ≤ 0.01, Wilcoxon rank-sum test).A RF classification model was used to identify the discrim-

inatory taxonomic features (using mean decrease accuracy

Table 1. Demographics and characteristics of subjects classifiedas case and control.

Cases Controls

N 27 24Male/Female 22 (81%)/5 (19%) 14 (58%)/10 (42%)Age (mean) 58.1 years 48.6 years

Tumor siteOral cavity 6 (22%) N/AOropharynx 12 (44%) N/ALarynx 6 (22%) N/AHypopharynx 2 (7%) N/AN/A 1 (4%) N/A

AJCCstageStage I 3 N/AStage II 5 N/AStage III 4 N/AStage IV 15 N/A

Alcohol useNever 7 8Monthly 2 42-4�/month 3 5>4�/month 13 7N/A 2 0

Cigarettes per day (mean) 13.9 13.8Smoking duration years (mean) 22.3 23.6

Mean total urinary cotinine(ng/mL)

1,609.8 1,817.4

Oral Microbiome in Smokers with Head/Neck Cancer

AACRJournals.org Cancer Prev Res; 13(5) May 2020 465

http://www.test.personality-project.org/r/psych/



https://github.com/ashoks773/Oral-microbiome-16S



index) between the oral microbiome of head and neck cancersmokers and healthy controls. The classification performanceof the RFmodel constructed on the training set (n¼ 36, 70% ofdata) was accessed on a test set (n ¼ 16, 30% of the data). Theresults showed that cases and controls could be moderatelyclassified on the basis of their oral mucosal microbiometaxonomic features (with error rate of 23.52% for cases and27.77% for controls and AUC of 0.75 indicating moderateaccuracy and sensitivity of the model, at 500n trees; Fig. 2A).The top 60 taxonomic features, according to mean decrease inaccuracy (>1.14), were selected and then checked using indi-cator species analyses (indicator value > 0.5) and statisticalsignificance according to FDR-adjusted Wilcoxon rank-sumtest, q < 0.05. These analyses revealed 11 significant taxonomicmarkers distinguishing cases from controls, with relative abun-dances of Stenotrophomonas andRuminococcus, and the familyComamonadaceae higher in smokers with head and neckcancer (Fig. 2B; Supplementary Table S3). In contrast, theremaining eight taxonomic markers were highly abundant incontrols; among them, common oral commensals such asTannerella, Capnocytophaga, Selenomonas, Veillonella, andKingella were highlighted.

Spearman correlation analyses were used to explore coa-bundance and exclusion patterns between discriminating taxa.This analysis revealed that the markers that characterized cases—Ruminococcus, Stenotrophomonas, and Comamonadaceae,tended to coabund together (Fig. 2C), while showing exclusionpatterns with Weeksellaceae (r ¼ �0.55, and q ¼ 0.0001) andCapnocytophaga (r ¼ �0.51, and q ¼ 0.0002), markers char-acterizing controls. These two control-specific taxa showedsignificant coabundance with other oral commensals moreabundant in controls such as Haemophilus, Kingella, Tanner-ella, Selenomonas, and Veillonella (Fig. 2B; SupplementaryTable S4).

Correlation of HPB-releasing DNA adducts withmicrobiome profilesHigh levels of HPB-releasing DNA adducts were observed in

buccal cells of cases compared with the controls (Fig. 3A);therefore, we explored whether these adducts were associatedwith distinct microbiome markers. For instance, an increase inHPB-releasingDNAadduct levels in buccal cells was negativelyassociated with the number of observed bacterial taxa on thesame mucosal site (Fig. 3B; Spearman correlation, r ¼ �0.38,

Figure 1.

Microbial diversity of healthy controls and individuals with oral/head and neck cancer (cases). A, The boxplots show differences in each alpha diversity measures(observed ASVs, left and Shannon's index, right) between cases and controls. Double asterisks show differences (q < 0.01) according to FDR-corrected Wilcoxonrank-sum tests; ns, no significant difference. B and C, Beta-diversity analysis in oral microbiome signatures between cases and controls. PCoA analysis on weightedUniFrac distances, unweighted UniFrac distances, and Bray-Curtis distances, respectively, calculated using normalized ASV abundances (B; left to right), andweightedUniFrac distances, unweightedUniFrac distances, and Bray-Curtis distances (C; left to right) showhigher inter individual variations in cases versus controls.Double asterisks show differences (q < 0.01) according to FDR-corrected Wilcoxon rank-sum tests.

Sharma et al.


q ¼ 0.006),. In addition, it was observed that the adduct levelstended to increase along with the cumulative abundance (sumof relative abundances) of Stenotrophomonas, Ruminococcus,and Comamonadaceae, the three taxonomic markers distin-guishing cases from controls (Spearman correlation, r ¼ 0.32,q ¼ 0.02; Fig. 3C). The cumulative abundance of taxonomic

signatures characterizing controls, and that are mostly com-mon oral commensals, tended to decrease with increasingDNA adduct levels; indicating that increased tobacco-induced DNA damage in buccal cells may be associated withlower abundance of this commensals; although this relation-ship did not show significance at alpha 0.05 (Spearman

Figure 2.

Taxonomic signatures of the oralmicrobiomeof healthy controls and individualswith oral/head and neck cancer.A,Performance of RF classificationmodel at 10-foldcross-validation used for the identification of discriminating taxa among cases and controls. B, Boxplots show specific differences in the relative abundance ofsignificantly discriminating taxa between the oral microbiome of cases and controls (RF mean decrease in accuracy > 1.14, Wilcoxon rank-sum tests, q values < 0.05,and Indval > 0.5). C, Correlation network plot constructed using significant correlations (Spearman r > 0.4, q < 0.05) between discriminant taxonomic features, andindicating cooccurrence or exclusion patterns of case and control-specific taxa. Edges, Spearman correlation coefficients; nodes, the discriminant taxa.



correlation, r ¼ �0.24, q ¼ 0.08; Fig. 3C). However, whenanalyzing correlations between the abundance of individualtaxa and HPB-releasing adduct levels, Stenotrophomonas(enriched in cases) showed positive correlations (Spearman r¼ 0.36, q ¼ 0.008), whereas Capnocytophaga, Kingella, Veillo-nella, andWeeksellaceae, markers of healthy controls, showed anegative relationship with the levels of these DNA adducts(Spearman r ¼ �0.27 to �0.33, and q < 0.05; Fig. 3D).Remaining taxa enriched in cases and controls also showednonsignificant (q > 0.05) positive and negative correlationswith HPB-releasing DNA adduct levels and hence were notincluded in the network (Supplementary Table S5).

Correlation between predicted pathways andmicrobiome profilesResults from the functional prediction analysis via PICRUSt

were used to detect predicted discriminatory pathways betweenthe oral microbiome of head and neck cancer smokers andhealthy controls through an RF classification model (ntrees ¼500). The top 100 pathway features, according to meandecrease in accuracy (>1.08), were selected and then checkedusing statistical significance according to the Wilcoxon rank-sum test, q < 0.05. This analysis revealed a higher number ofpredicted functional pathways enriched in cases (16 pathwaysout of total 26), mainly involved in the degradation of

Figure 3.

Correlation between levels of HPB-releasing adductlevels and oral microbiome profiles of healthy con-trols and individuals with oral/head and neck cancer.A, Box plot shows differences in HPB-releasing DNAadduct levels between cases and controls. Doubleasterisks show differences (q < 0.01) according toFDR-corrected Wilcoxon rank-sum tests. B, Correla-tions between alpha diversity measures and adductlevels shows a negative association between thenumber of observed ASV and the adduct levels(Spearman correlation, r ¼ 0.38, q ¼ 0.006). C,correlations between the cumulative abundances ofsignificantly discriminating taxa in controls andHPB-releasing DNA adduct levels (Spearman correlation,R¼�0.24, q¼0.08) and significantly discriminatingtaxa in cases and adduct levels (Spearman correla-tion, R ¼ 0.32, q ¼ 0.02). Correlation network plotconstructed using significant correlations (Spear-man r > 0.27, q < 0.05) between significantly dis-criminating taxa and DNA adducts.

Sharma et al.


xenobiotics (e.g., vanillin, hydroxyacetophenone, phenolics,and toluene) amino acids and amines (arginine, histidine,polygenic amines, and putrescine), antibiotic resistance (poly-myxin), and other biosynthetic and metabolic pathways (Sup-plementary Fig. S3; Supplementary Table S6). These selectedpathways showed positive correlations (Spearman r > 0.5,q < 0.05) with all three taxa enriched in cases (Fig. 4; Supple-mentary Table S7). In contrast, fewer pathways were enrichedin controls (10 pathways out of total 26), andmost of themwereassociated with the biosynthesis of sugars, lipids and aminoacids and degradation of sugars such as xylose, galacturonate,and arabinose (Supplementary Fig. S3; SupplementaryTable S6). These predicted pathways showed positive correla-

tions (Spearman r > 0.5, q < 0.05) with Tannerella, Veillonella,Weeksellaceae, Heamophilus parainfluenzae, and Lachnospir-aceae (Fig. 4; Supplementary Table S7); all taxonomic markerscharacterizing controls.

DiscussionIn this study, we examined the oral microbiome profiles of

cigarette smokers, evaluating for a propensity toward devel-opment of HNSCC. Broadly, microbiome profiling of 27smokers affected with HNSCC and 24 controls, via MiSeqamplicon sequencing of the bacterial 16S rRNA gene, indicatedsignificant differences between the cohorts based on bacterial

Figure 4.

Correlation network showing associations between discriminating taxa and functional predicted pathways of themicrobiomeof healthy controls and individualswithoral/head and neck cancer. The correlation network plot was constructed using significant correlations (Spearman r > 0.5, q < 0.05) between significantlydiscriminating taxa and significantly discriminating predicted pathways between cases and controls. Edges, Spearman correlation coefficients; nodes, thediscriminant taxa and predicted pathways.



community composition. Likewise, we report higher interin-dividual variability and lower bacterial richness in the oralmucosa of patients with HNSCC, as well as specific taxonomicand predicted functional markers distinguishing diseased fromhealthy states, likely correlated with DNA damage levels. Theseresults add to growing evidence linking oral microbial com-munities to HNSCC status, highlighting the value of microbialmarkers for disease diagnosis and shedding light on themicrobial mechanisms likely associated with disease risk.A handful of reports have previously described associations

between the oral microbiome and HNSCC also using next-generation sequencing techniques. These reports have focusedon multiple traits such as tumor stage, different sources in theoral cavity (e.g., tissue, saliva, and oral rinses), healthy versustumor tissues in the same subjects and comparisons of patientswith cancer with healthy controls. For instance, reports focus-ing only on a limited number of patients with HNSCC anddisease stage show variation in bacterial diversity and taxo-nomic profiles depending on mutational signatures (20), andsite sampled within the oral cavity, with significant differencesbetween saliva and tumor tissue samples (21) However, focus-ing only on patients with HNSCC did not show clear stage-dependent differences based on community composition (betadiversity) analysis.In line with our findings, studies in larger cohorts including

oral rinses in cancer-free subjects or healthy tissue swabs in thesame affected individuals have shown significant composi-tional differences between healthy and HNSCC sub-jects (22, 23). In these studies, it is also shown that bacterialrichness and diversity tend to increase in oral rinses and tumortissues of affected individuals, which is not concordantwith ourobservations of decreased alpha diversity in HNSCC cases.Nonetheless, the aforementioned studies, and others showingthe same diversity trends in the saliva of a few HNSCC andhealthy subjects (24) did not stratify the microbial signalsobtained on the basis of smoking status, or other lifestyle riskfactors, which can add major confounders. Indeed, smoking isreported to have significant effects on oral microbiome com-position (25, 26) and possibly explain the discrepancies in alphadiversity found between ours and those previous studies.Thus, the results presented herein should be first considered

in the context of a tissue microenvironment already predis-posed by smoking and the harmful constituents present incigarette smoke, characteristics that make our study unique incomparison to all the aforementioned reports. For example, it iswell known that tobacco smoke results in exposure to a myriadof toxic and carcinogenic constituents, resulting in DNAdamage and subsequent development of several malignancies,including oral cancer. The tobacco-specific nitrosamines N'-nitrosonornicotine (NNN) and 4-(methylnitrosamino)-1(-3-pyridyl)-1-butanone (NNK) are two carcinogens that havespecifically been shown to have high malignant potential.Exposure to NNN and NNK triggers macromolecular altera-tions, which interfere with replication, transcription, and nor-mal DNA repair mechanisms. Primary among the effects of

nitrosamines is the generation of HPB-releasing DNA adductsfollowing metabolic activation. Measurement of DNA adductsoffers a direct assessment of DNA damage. Recent studies haveshown that the amount of NNN/NNK exposure, as assessed byurinary biomarkers, does not necessarily correlate to the level ofHPB-releasing DNA adducts; therefore, there are likely to beother factors, besides the carcinogen doses, contributing tothe variations in the levels of DNA damage amongsmokers (27–32). The oral microbiome can disrupt host'sdefense mechanisms, inducing chronic inflammatory changesresulting in a cascade of events capable of causing extensiveDNA damage (32–35). It has recently been shown that NNKcarcinogenicity, and likely that of NNN, can be substantiallyincreased in the presence of inflammatory agents (36). It is alsoplausible to expect that certain bacteria are capable of meta-bolically activating tobacco carcinogens. This raises the ques-tion whether specific smokers are at a higher risk of developingoral cancer given a higher baseline level of inflammation or ahigher level of carcinogen activation in the oral cavity, thussparking our interest in investigating the bacterial microbiomeof the oral cavity.Along these lines, we report that predicted bacterial func-

tions associated with the degradation of xenobiotics (e.g.,toluene, phenyl compounds) and amines (e.g., aromatic bio-genic amines) characterize HNSCC, in contrast with healthysmoker tissues, who exhibited greater abundance of predictedcarbohydrate metabolism pathways. Increased abundance ofpredicted genes involved in xenobiotic degradation pathways,at the expense of carbohydrate metabolism, have also beenreported in oral rinses of smokers versus never-smokers in alarge cohort (26), which is in line with our comparisons of casesversus healthy controls. As such, one important questionfocuses on investigating whether exposure to cigarette smokemodulates the oral microbiome to increase the presence ofxenobiotic-metabolizing bacteria, or whether the individuals atrisk for HNSCC inherently have higher abundance of suchbacteria; and on determining why this is not the case in cancer-free tobacco users. Moreover, as patterns of increased HPB-releasing DNA adducts were correlated with specific bacterialsignatures in HNSCC smokers, whether such bacteria contrib-ute to metabolic activation of NNN and/or NNK should beinvestigated.Notably, some microbial markers in patients with HNSCC

that positively correlated with HPB-releasing adduct levels,have been previously characterized as pathogenic, xenobioticdegrader, and/or multidrug-resistant taxa. For instance, Steno-trophomonas is reported to be a nosocomial pathogen, partic-ularly associated with immunosuppressed individuals, andresistance to a broad range of antibiotics (37–39). This obser-vation is in line with our findings of increased abundanceof predicted polymyxin resistance in cases, correlated withabundances of this taxon. Stenotrophomonas has been associ-ated with a variety of infections in respiratory tract, blood,bones and joints, urinary tract, soft tissues, and lung cancer(reviewed in ref. 40.) Cytotoxic and protease activities by

Sharma et al.


Stenotrophomonas may also degrade tissue and cause celldeath (41). Furthermore, in concordance with an increase inpredicted aromatic xenobiotic and amine degradation in cases,this taxon has the capacity to process a wide range of toxicants,including phenolics (42), toluene (43), and phenanthrene (44);with metabolites from the latter commonly reported in smo-kers at lung cancer risk (45).Although not commonly abundant in the human oral cavity,

unclassifiedComamonadaceae have been detected before in theoral washes of immunocompromised (HIVþ) smokers (46), inhigh mean abundance (25.5%) on the laryngeal tissues ofsmokers (47) and in the saliva of obese individuals (48).However, the influence of Stenotrophomonas and Comamao-nadaceae in HNSCC progression, their ecological interactions,and their potential role in metabolic activation of tobacco-related carcinogens or DNA damage is still unclear and war-rants further investigation.This study also showed that the abundance of common oral

commensals (e.g., Tannerella, Capnocytophaga, Selenomonas,Veillonella, and Kingella) that are usually associated withcarbohydrate metabolism in the oral cavity (49–52) (as con-firmed by the functional predictions presented) are altered ordiminished in smokerswithHNSCC. Thus, our hypothesis thathealthy smokers still exhibit markers of a homeostatic mucosalmicrobiome, which exerts protective effects against carcinogenactivators (e.g., through competitive exclusion), should beexplored further. Indeed, partially in line with our results, arecent study analyzing oral washes of a larger cohort (n ¼ 129cases and n¼ 245 controls) shows no bacterial taxa associatedwith HNSCC risk, while identifying higher abundance of somecommonoral commensals (e.g.,Kingella) in controls comparedwith disease subjects (53). These observations are also con-nected with our finding that patients with HNSCC showedincreased interindividual variation and heterogeneity in theabundance of common oral commensals (higher beta diversi-ty), and lower alpha diversity (reduced richness), inverselycorrelated with higher HPB-releasing DNA adduct levels. Assuch, further studies should focus on elucidating the ecologicaland molecular underpinnings of microbiome-mediated pro-tection against smoke carcinogens in healthy smokers, andmicrobiome-mediated toxicity in affected smokers, in thecontext of taxonomic and functional diversity of the oralmucosal microbiome.

LimitationsThe main limitations of this study are the high microheter-

ogeneity of the oral cavity (51) and the difficulty in obtainingmarkers of disease that coincide with previous studies. How-ever, previous studies have focused on other specimencollection techniques (e.g., saliva, tissue/fluid type, etc.),varying proximity to tumors and other underlying risk factors(e.g., HPV, diet), hence, showing different disease-associatedtaxa [e.g., Fusobacterium and Lactobacillus, Prevotella,Streptococcus (8, 54, 55)]. Furthermore, while our sample ismade up primarily of oral and oropharyngeal tumors, our

cohort also included some tumors in the larynx/hypopharynx,which also potentially differentiates our dataset from thosealready present in the literature.In addition, interpretation of our results requires the recog-

nition that we have investigated an association between diseasestatus (oral/head and neck cancer) and the oral microbiome.Therefore, we cannot make inferences on the direction ofcausality, as the microbiome findings may be the result oftumor development rather than a precursor to tumor forma-tion. However, given our prior data that DNA adduct forma-tion is greater in those with oral/head and neck cancer, we areworking toward elucidating the nature of several potentialmutagenic variables in this patient population, which willeventually contribute to an understanding of factors influenc-ing tobacco-induced carcinogenesis. Another limitation of ourstudy is the low sample size as a total of 51 subjects wereexamined. Although it represents the first and largest sample ofpatients studied for both DNA adduct formation and oralmicrobiome of smokers, we nonetheless desire an analysis ina larger cohort as soon as possible. However, the fact that wefocus exclusively on smokers is a potential and novel strength ofthis report compared with other previous efforts. A slightdifference in the average age between cases and controls couldbe perceived as another limitation; however, there is no data inthe literature on the potential effect of age on oral microbiome,while research on gut microbiome suggests no effect in therelevant age range (between 40 and 60 years; refs. 56, 57).In addition, our cohort did not contain specific oral

health data, was studied via buccal cells for characterizationof the oral microbiome and included a small percentage ofnonoral/oropharyngeal cancers. Given that the role of the oralmicrobiome and cancer is still in the early stages, it is difficult toknow if this is impactful in our dataset. Oral health practicesmay certainly impact our findings but given the dichotomousresults between cases and controls relating to species variabilityand interindividual variation, oral health habits are less likely tobe an explanation for our results. While there is likely to bevariability in the nature of the microbiome at different sites inthe upper aerodigestive tract, the same can be said of differentsites in the oral cavity. That is, the microbial makeup of thegingiva, palate, and buccalmucosa are known to be distinct (50)and the impact of this on oral microbiome research, in thecontext of disease, to date is unknown. Thus, while it would beideal to adequately capture the microbiology of the entire oralcavity, or even the upper aerodigestive tract, this would behighly cumbersome and represents a hurdle that has not yetbeen overcome in the medical literature.Finally, although we aimed for increased accuracy of our

functional predictions (maximum nearest taxon index ¼ 2),our functional analyses are based on comparing the 16S rRNAsequences obtainedwith those of bacterial genomes reported indatabases (58). Therefore, these predicted functional analysesshould not be considered as substitutes of metagenomicapproaches and actual functional analyses must be conductedto confirm the trends reported in a larger cohort, especially to



determine potential HNSCC associations with the metabolismof xenobiotics by the mucosal oral microbiome.

Conclusions and future directionsWith this current report, we are now able to examine our

DNA adduct results in the context of information pertaining tothe oral microbiome in the same patient sample. This analysisrevealed that the increase in DNA adducts seen among casescorresponded to, and is associated with a reduction in bacterialrichness present in the microbiome. In addition, we foundconcordance between increasing adduct levels and the numberof taxa present among cases. The association between specifictaxa of the oral microbiome, specifically the genera Stenotro-phomonas, and the presence of DNA damage in mucosal tissueis readily apparent in our analysis and should be furtherinvestigated focusing on direct functional approaches (meta-genomics, metabolomics, transcriptomics) to shed more lighton the functions of HNSCC-associated taxa and the potentialrole of the microbiome on metabolic activation of smokecarcinogens among affected versus healthy smokers. Further-more, we cannot assign cause or effect modification of cancerrisk to the microbial characteristics identified here, and thisassociation requires further study to better understand the roleof bacterial richness, commensal homeostasis, and specificbacterial profiles in tobacco-induced oral/head and neck can-cer. Finally, an analytical approach that also considers how themucosal epithelia in tumors associates with themicrobiome onsite, through immune and metabolic related pathways in thehost, should be considered.In summary, we have examined and characterized the oral

microbiome and quantified the presence of DNA adduct levelsin smokers with oral/head and neck cancer compared withcancer-free smokers. Not only have we identified specificbacterial taxa characterizing both cancer status and those whoare cancer free, but we found that there are specific bacterial

taxa (e.g., Stenotrophomonas) directly correlated to increasinglevels of DNA adducts and the increased abundance of xeno-biotic-processing functions of the oral microbiome in affectedversus healthy smokers. We therefore conclude that our datasuggest an association between bacterial richness, bacterialdiversity, and specific bacterial taxa and tobacco-inducedoral/head and neck cancer.


Authors’ ContributionsConception and design: I. Stepanov, A. Gomez, S.S. KhariwalaAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): I. Stepanov, S.S. KhariwalaAnalysis and interpretation of data (e.g., statistical analysis,biostatistics, computational analysis): A. Kumar-Sharma, I. Stepanov,A. Gomez, S.S. KhariwalaWriting, review, and/or revision of the manuscript: A. Kumar-Sharma,W.T. DeBusk, I. Stepanov, A. Gomez, S.S. KhariwalaAdministrative, technical, or material support (i.e., reporting ororganizing data, constructing databases): A. Kumar-Sharma, A. Gomez

AcknowledgmentsThis work is supported by NIDCR/NIH K23DE023572 (to S.S. Khar-

iwala) and NCI/NIH R01CA180880 (to I. Stepanov).The authors acknowledge the Minnesota Supercomputing Institute

(MSI) at the University of Minnesota for providing resources that con-tributed to the research results reported within this paper. URL: http://www.msi.umn.edu.


Received October 1, 2019; revised January 8, 2020; accepted February 14,2020; published first February 18, 2020.

References1. Marur S, D’Souza G, Westra WH, Forastiere AA. HPV-associated

head and neck cancer: a virus-related cancer epidemic. Lancet Oncol2010;11:781–9.

2. Khariwala SS,Ma B, Ruszczak C, Carmella SG, Lindgren B,HatsukamiDK, et al. High level of tobacco carcinogen–derived DNA damage inoral cells is an independent predictor of oral/head and neck cancer riskin smokers. Cancer Prev Res 2017;10:507–13.

3. Ma B, Ruszczak C, Jain V, Khariwala SS, Lindgren B, Hatsukami DK,et al. Optimized liquid chromatography nanoelectrospray–high-resolution tandem mass spectrometry method for the analysis of4-hydroxy-1-(3-pyridyl)-1-butanone-releasing DNA adducts inhuman oral cells. Chem Res Toxicol 2016;29:1849–56.

4. Cho I, BlaserMJ.The humanmicrobiome: at the interface of health anddisease. Nat Rev Genet 2012;13:260–70.

5. Round JL, Mazmanian SK.The gut microbiota shapes intestinalimmune responses during health and disease. Nat Rev Immunol2009;9:313–23.

6. Mazmanian SK, Round JL, Kasper DL. A microbial symbiosis factorprevents intestinal inflammatory disease. Nature 2008;453:620–5.

7. Schwabe RF, Jobin C.The microbiome and cancer. Nat Rev Cancer2013;13:800–12.

8. Mager DL, Haffajee AD, Devlin PM, Norris CM, Posner MR,Goodson JM. The salivary microbiota as a diagnostic indicator oforal cancer: a descriptive, non-randomized study of cancer-freeand oral squamous cell carcinoma subjects. J Transl Med 2005;3:27.

9. Nagy KN, Sonkodi I, Sz€oke I, Nagy E, Newman HN. The micro-flora associated with human oral carcinomas. Oral Oncol 1998;34:304–8.

10. Bolz J, Dos�a E, Schubert J, Eckert AW. Bacterial colonization ofmicrobial biofilms in oral squamous cell carcinoma. Clin Oral Investig2014;18:409–14.

11. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet C, Al-GhalithGA, et al. QIIME 2: reproducible, interactive, scalable, and extensiblemicrobiome data science. PeerJ Preprints 2018. doi: 10.7287/peerj.preprints.27295v2.

12. RCoreTeam. R: a language and environment for statistical computing.Vienna, Austria: R Foundation for Statistical Computing; 2013. Avail-able: https://www.R-project.org/.

13. McMurdie PJ, Holmes S. phyloseq: an R package for reproducibleinteractive analysis and graphics of microbiome census data.PLoS One 2013;8:e61217.

Sharma et al.


http://www.msi.umn.edu



https://www.R-project.org/

14. Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’Hara RB,et al. vegan: community ecology package. R package version 2011;117–118.

15. Liaw A, Wiener M. Classification and regression by randomForest.R news 2002;2:18–22.

16. Roberts DW. labdsv: ordination and multivariate analysis for ecology.R package version 2007. Available: http://ecology.msu.montana.edu/labdsv/R.

17. RevelleW,Revelle MW.Package “psych.” the comprehensive r archivenetwork. 2015. Available: http://www.test.personality-project.org/r/psych/psych-manual.pdf.

18. Wickham H. ggplot2: elegant graphics for data analysis. New York,NY: Springer; 2016.</bib

19. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al.Cytoscape: a software environment for integrated models of biomo-lecular interaction networks. Genome Res 2003;13:2498–504.

20. Yang S-F, Huang H-D, Fan W-L, Jong Y-J, Chen M-K, Huang C-N,et al. Compositional and functional variations of oral microbiotaassociated with the mutational changes in oral cancer. Oral Oncol2018;77:1–8.

21. Zhang Z, Yang J, Feng Q, Chen B, Li M, Liang C, et al. Compositionaland functional analysis of the microbiome in tissue and saliva of oralsquamous cell carcinoma. Front Microbiol 2019;10:1439.

22. Yang C-Y, Yeh Y-M, Yu H-Y, Chin C-Y, Hsu C-W, Liu H, et al. Oralmicrobiota community dynamics associated with oral squamous cellcarcinoma staging. Front Microbiol 2018;9:862.

23. ZhaoH,ChuM,HuangZ, YangX, Ran S, HuB, et al. Variations in oralmicrobiota associated with oral cancer. Sci Rep 2017;7:11773.

24. Pushalkar S, Mane SP, Ji X, Li Y, Evans C, Crasta OR, et al. Microbialdiversity in saliva of oral squamous cell carcinoma. FEMS ImmunolMed Microbiol 2011;61:269–77.

25. Wu J, Peters BA,Dominianni C, ZhangY, Pei Z, Yang L, et al. Cigarettesmoking and the oral microbiome in a large study of American adults.ISME J 2016;10:2435–46.

26. Mason MR, Preshaw PM, Nagaraja HN, Dabdoub SM, Rahman A,Kumar PS. The subgingival microbiome of clinically healthy currentand never smokers. ISME J 2015;9:268–72.

27. Stepanov I, Muzic J, Le CT, Sebero E, Villalta P, Ma B, et al. Analysis of4-hydroxy-1-(3-pyridyl)-1-butanone (HPB)-releasing DNA adductsin human exfoliated oral mucosa cells by liquid chromatography–electrospray ionization–tandem mass spectrometry. Chem Res Tox-icol 2013;26:37–45.

28. Foiles PG, Akerkar SA, Carmella SG, Kagan M, Stoner GD, Resau JH,et al. Mass spectrometric analysis of tobacco-specific nitrosamine-DNA adducts in smokers and nonsmokers. Chem Res Toxicol 1991;4:364–8.

29. Schl€obe D, H€olzle D, Hatz D, von Meyer L, Tricker AR, Richter E. 4-Hydroxy-1-(3-pyridyl)-1-butanone-releasing DNA adducts in lung,lower esophagus and cardia of sudden death victims. Toxicology 2008;245:154–61.

30. Hecht SS. Tobacco smoke carcinogens and lung cancer. J Natl CancerInst 1999;91:1194–210.

31. PryorWA,Hales BJ, Premovic PI, ChurchDF. The radicals in cigarettetar: their nature and suggested physiological implications. Science1983;220:425–7.

32. NairU, BartschH,Nair J. Lipid peroxidation-inducedDNAdamage incancer-prone inflammatory diseases: a review of published adducttypes and levels in humans. Free Radic Biol Med 2007;43:1109–20.

33. Bartsch H, Nair J.Accumulation of lipid peroxidation-derived DNAlesions: potential lead markers for chemoprevention of inflammation-driven malignancies. Mutat Res 2005;591:34–44.

34. Bullon P, Newman HN, Battino M. Obesity, diabetes mellitus, ath-erosclerosis and chronic periodontitis: a shared pathology via oxida-tive stress and mitochondrial dysfunction? Periodontol 2000 2014;64:139–53.

35. Johannsen A, Susin C, Gustafsson A. Smoking and inflammation:evidence for a synergistic role in chronic disease. Periodontol 20002014;64:111–26.

36. Melkamu T, Qian X, Upadhyaya P, O’Sullivan MG, Kassie F. Lipo-polysaccharide enhances mouse lung tumorigenesis: a model forinflammation-driven lung cancer. Vet Pathol 2013;50:895–902.

37. Denton M, Kerr KG.Microbiological and clinical aspects of infectionassociated with stenotrophomonas maltophilia. Clin Microbiol Rev1998;11:57–80.

38. Baumrin E, Piette EW,Micheletti RG. Stenotrophomonasmaltophilia:an emergingmultidrug-resistant opportunistic pathogen in the immu-nocompromised host. BMJ Case Rep 2017;2017:pii: bcr-2017-221053.

39. Chang Y-T, Lin C-Y, Chen Y-H, Hsueh P-R. Update on infectionscaused by Stenotrophomonas maltophilia with particular attention toresistancemechanisms and therapeutic options. FrontMicrobiol 2015;6:893.

40. Brooke JS. Stenotrophomonas maltophilia: an emerging global oppor-tunistic pathogen. Clin Microbiol Rev 2012;25:2–41.

41. Figueiredo PMS, Furumura MT, Santos AM, Sousa ACT, Kota DJ,Levy CE, et al. Cytotoxic activity of clinical stenotrophomonas mal-tophilia. Lett Appl Microbiol 2006;43:443–9.

42. Gunasundari D, Muthukumar K. Simultaneous Cr(VI) reduction andphenol degradation using stenotrophomonas sp. isolated from tanneryeffluent contaminated soil. Environ Sci Pollut Res Int 2013;20:6563–73.

43. Olapade OA, Ronk AJ. Isolation, characterization and communitydiversity of indigenous putative toluene-degrading bacterial popula-tions with catechol-2,3-dioxygenase genes in contaminated soils.Microb Ecol 2015;69:59–65.

44. Gao S, Seo J-S, Wang J, Keum Y-S, Li J, Li QX. Multiple degradationpathways of phenanthrene by Stenotrophomonas maltophilia C6.Int Biodeterior Biodegradation 2013;79:98–104.

45. Patel YM, Park SL, Carmella SG, Paiano V, Olvera N, StramDO, et al.Metabolites of the polycyclic aromatic hydrocarbon phenanthrene inthe urine of cigarette smokers from five ethnic groups with differingrisks for lung cancer. PLoS One 2016;11:e0156203.

46. Mukherjee PK, Chandra J, Retuerto M, Tatsuoka C, Ghannoum MA,McComsey GA. Dysbiosis in the oral bacterial and fungal microbiomeof HIV-infected subjects is associated with clinical and immunologicvariables of HIV infection. PLoS One 2018;13:e0200285.

47. Jett�eME,Dill-McFarlandKA,HanshewAS, SuenG, Thibeault SL. Thehuman laryngeal microbiome: effects of cigarette smoke and reflux.Sci Rep 2016;6:35882.

48. Wu Y, Chi X, Zhang Q, Chen F, Deng X. Characterization of thesalivary microbiome in people with obesity. PeerJ 2018;6:e4458.

49. Duran-Pinedo AE, Frias-Lopez J. Beyond microbial community com-position: functional activities of the oral microbiome in health anddisease. Microbes Infect 2015;17:505–16.

50. Aas JA, Paster BJ, Stokes LN, Olsen I, Dewhirst FE. Defining thenormal bacterial flora of the oral cavity. J Clin Microbiol 2005;43:5721–32.

51. Dewhirst FE, Chen T, Izard J, Paster BJ, Tanner ACR, Yu W-H, et al.The human oral microbiome. J Bacteriol 2010;192:5002–17.

52. Marsh PD, Martin MV, Lewis MAO, Williams D. Oral microbiologye-book. Philadelphia, PA: Elsevier Health Sciences; 2009.

53. Hayes RB, Ahn J, Fan X, Peters BA, Ma Y, Yang L, et al. Association oforal microbiome with risk for incident head and neck squamous cellcancer. JAMA Oncol 2018;4:358–65.

54. Guerrero-Preston R, White JR, Godoy-Vitorino F, Hilario AR,Navarro K, Gonz�alez H, et al. High-resolution microbiome profilinguncovers Fusobacterium nucleatum, Lactobacillus gasseri/johnsonii,and Lactobacillus vaginalis associated to oral and oropharyngealcancer in saliva fromHPV positive and HPV negative patients treatedwith surgery and chemo-radiation. Oncotarget 2017;8:110931–48.



http://ecology.msu.montana.edu/labdsv/R



http://www.test.personality-project.org/r/psych/psych-manual.pdf



55. Hernandez BY, Zhu X, Goodman MT, Gatewood R, Mendiola P,Quinata K, et al. Betel nut chewing, oral premalignant lesions, and theoral microbiome. PLoS One 2017;12:e0172196.

56. Odamaki T, Kato K, Sugahara H, Hashikura N, Takahashi S, Xiao J-Z,et al. Age-related changes in gut microbiota composition from new-born to centenarian: a cross-sectional study. BMC Microbiol 2016;16:90.

57. de la Cuesta-Zuluaga J, Kelley ST, ChenY, Escobar JS,MuellerNT, LeyRE, et al. Age- and sex-dependent patterns of gutmicrobial diversity inhuman adults. mSystems 2019;4:e00261–19.

58. Langille MGI, Zaneveld J, Caporaso JG, McDonald D, Knights D,Reyes JA, et al. Predictive functional profiling of microbial commu-nities using 16S rRNA marker gene sequences. Nat Biotechnol 2013;31:814–21.


Sharma et al.

Documents

TUMOR MICROBIOME · 1/1/2020 · Cancer Epidemiol Biomarkers Prev Oct 1, 2019 28:10 1687–93; doi: 10.1158/1055-9965.EPI-19-0008 Chemoprevention of Colorectal Cancer by Anthocyanidins