Serum metabolic signatures of primary biliary cirrhosis and primary sclerosing cholangitis

ORIG INAL ART ICLE

Serummetabolic signatures of primary biliary cirrhosis and primarysclerosing cholangitisLauren N Bell1, Jacob Wulff1, Megan Comerford2, Raj Vuppalanchi2 and Naga Chalasani2

1 Metabolon Inc., Durham, NC, USA

2 Division of Gastroenterology and Hepatology, Indiana University School of Medicine, Indianapolis, IN, USA

Keywords

bile acid metabolism – cholestatic disease –

metabolome – metabolomics

Abbreviations

ALT, alanine aminotransferase; AMA,

antimitochondrial antibody; AST, aspartate

aminotransferase; CYP, cytochrome P450;

Da, dalton; DMA, dimethylarginine; ERCP,

endoscopic retrograde

cholangiopancreatography; FDR, false

discovery rate; GC, gas chromatography;

HETE, hydroxyeicosatetraenoic acid; HODE,

hydroxyoctadecadienoic acid; IDO,

indoleamine 2,3-dioxygenase; INR,

international normalized ratio; MELD, model

for end-stage liver disease; MRCP, magnetic

resonance cholangiopancreatography; MS,

mass spectrometry; PBC, primary biliary

cirrhosis; PSC, primary sclerosing cholangitis;

TDA, tryptophan dioxygenase; UDCA,

ursodeoxycholic acid; UHPLC, ultrahigh

performance liquid chromatography.

Correspondence

Naga Chalasani, MD, FACG

Division of Gastroenterology and Hepatology

Indiana University School of Medicine

702 Rotary Building, Suite 225

Indianapolis, IN 46202, USA

Tel: +317 278 0414

Fax: +317 278 1949

e-mail: [email protected]

Received 23 February 2014

Accepted 26 August 2014

DOI:10.1111/liv.12680

AbstractBackground & Aims: A greater understanding of cholestatic disease is neces-sary to advance diagnostic tools and therapeutic options for conditionssuch as primary biliary cirrhosis (PBC) and primary sclerosing cholangitis(PSC). The purpose of this study was to determine and compare the serummetabolomes of patients with PBC (n = 18) or PSC (n = 21) and healthycontrols (n = 10) and to identify metabolites that may differentiate these twocholestatic diseases. Methods and Results: Using a mass spectrometry-based,non-targeted biochemical profiling approach, we identified 420 serummetabolites, 101 that differed significantly (P ≤ 0.05) between PBC and con-trol groups, 115 that differed significantly between PSC and control groups,and 56 that differed significantly between PSC and PBC groups. Random for-est classification analysis was able to distinguish patients with PBC or PSCwith 95% accuracy with selected biochemicals reflective of protein and aminoacid metabolism identified as the major contributors. Metabolites related tobile acid metabolism, lipid metabolism, inflammation, and oxidative stress/lipid peroxidation were also identified as differing significantly when com-paring the disease groups and controls, with some of these pathways differen-tially affected in the PBC and PSC groups. Conclusion: In this study, weidentified novel metabolic changes associated with cholestatic disease thatwere both consistent and different between PBC and PSC. Validation studiesin larger patient cohorts are required to determine the utility of these bio-chemical markers for diagnosis and therapeutic monitoring of patients withPBC and PSC.

Impaired hepatic production and excretion of bile,resulting in dramatic accumulation of circulating bileacids, form the basis for cholestatic liver disease.Toxins/drugs, gall stones, local tumours, geneticalterations, autoimmune disorders and pregnancy,among other aetiologies, may contribute to reduc-tions in bile flow associated with cholestasis (1–3).

Response to hepatobiliary injury ranges from prolif-eration of cholangiocytes and hepatocytes to devel-opment of periductular fibrosis, biliary fibrosis andcirrhosis (4), ultimately giving rise to end-stageliver disease and the need for liver transplanta-tion. In general, pathogenic mechanisms of cholestat-ic disease are poorly understood; therefore, clinical

Liver International (2014)© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd 1

Liver International ISSN 1478-3223

management of these complex disorders remainsquite limited (5).

Two similar but distinct cholestatic diseases includeprimary biliary cirrhosis (PBC) and primary sclerosingcholangitis (PSC). PBC is an autoimmune disease inwhich progressive lymphocytic inflammation leads todestruction of intrahepatic bile ducts, and is predomi-nantly observed in asymptomatic, middle-aged women(6). PSC is characterized by chronic inflammation ofthe biliary epithelium resulting in destruction of intra-and extrahepatic bile ducts, and is typically diagnosed inmiddle-aged males with a high preponderance of con-current inflammatory bowel disease (7). Currently, dif-ferential diagnosis of PBC and PSC is sometimescomplicated by overlapping clinical, serological, histo-logical and radiological findings and the overall lack ofstandardized diagnostic and treatment criteria (8). Inaddition, ursodeoxycholic acid (UDCA) is currently theonly available therapy for cholestatic liver disease, butseveral new investigational drugs are under evaluationfor either PBC (Obeticholic acid, NCT01473524) orPSC (simtuzumab, NCT01672853). With the regulatoryapproval and eventual availability of these agents, adefinitive diagnosis will be critical. Moreover, identifica-tion of a metabolic profile that would identify respond-ers vs. non-responders may provide an opportunity topersonalize use of these therapeutic products.

Because of their downstream location within the bio-logical spectrum, small molecules (ranging from 99–1000 Da) represent integration of various upstreamprocesses (i.e. genetics, transcriptomics and proteomics)and are often very closely related to phenotype (9). Togain insight into the pathogenesis of PBC and PSC andidentify potential biomarkers to distinguish these tworelated disorders, we utilized a screening approach toprofile the global metabolome (10) of serum samplesfrom patients with PBC or PBC and healthy controls.Apart from a previous report in which 17 serum bileacids were profiled in serum from PBC and PSC patients(11), this is the first study to define and compare theglobal serum metabolomes of patients with PBC and

PSC and to explore utilization of small molecule profilesto differentiate between the two groups.

Patients and methods

Patients

Serum samples were obtained from a local blood bank(healthy controls) or from an ongoing liver bio bank atIndiana University. Liver bio bank was approved by theIndiana University Institutional Review Board and allparticipants signed an informed consent that allowsarchived serum and liver tissue samples to be used forfuture research purposes. Serum samples were preparedimmediately after their acquisition and were frozen at�80°C until the metabolomic studies were conducted.

Metabolic profiling

Sample preparation

The non-targeted metabolic profiling platform employedfor this analysis combined three independent platforms:ultrahigh performance liquid chromatography/tandemmass spectrometry (UHPLC/MS/MS) optimized for basicspecies, UHPLC/MS/MS optimized for acidic species,and gas chromatography/mass spectrometry (GC/MS).Samples were processed essentially as described previ-ously (10, 12). For each sample, 100 ll of serum was usedfor analyses. Using an automated liquid handler (Hamil-ton LabStar, Salt Lake City, UT, USA), protein was pre-cipitated with methanol that contained four standards toreport on extraction efficiency. The resulting supernatantwas split into equal aliquots for analysis on the three plat-forms. Aliquots, dried under nitrogen and vacuum-desic-cated, were subsequently either reconstituted in 50 ll0.1% formic acid in water (acidic conditions) or in 50 ll6.5 mM ammonium bicarbonate in water, pH 8 (basicconditions) for the two UHPLC/MS/MS analyses orderivatized to a final volume of 50 ll for GC/MS analysisusing equal parts bistrimethyl-silyl-trifluoroacetamideand solvent mixture acetonitrile:dichloromethane:cyclo-hexane (5:4:1) with 5% triethylamine at 60°C for 1 h. Inaddition, three types of controls were analysed in concertwith the experimental samples: aliquots of a well-charac-terized human plasma pool served as technical replicatesthroughout the data set, extracted water samples servedas process blanks, and a cocktail of standards spiked intoevery analysed sample allowed instrument performancemonitoring. Experimental samples and controls wererandomized across two platform run days.

LC/MS/MS and GC/MS

For UHLC/MS/MS analysis, aliquots were separated usinga Waters Acquity UPLC (Waters, Millford, MA, USA)and analysed using an LTQ mass spectrometer (ThermoFisher Scientific, Inc., Waltham, MA, USA) which

Key Points

● This study compares the global serum metabolo-mes of patients with PBC and PSC.● Metabolites reflective of protein metabolismappear to have the greatest power in differentiatingbetween PBC and PSC.● Inflammatory signature (elevations in the trypto-phan-derived metabolite kynurenine and reductionsin the lipooxygenase-derived eicosanoid 12-HETE)was observed in both the PBC and PSC.● Levels of several free amino acids (serine, aspar-tate, glutamate, phenylalanine and ornithine) andnine dipeptides were significantly (P ≤ 0.05) lower inpatients with PSC as compared to PBC.

Liver International (2014)© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd2

Metabolomics of PBC and PSC Bell et al.

consisted of an electrospray ionization (ESI) source andlinear ion-trap (LIT) mass analyser. The MS instrumentscanned 99–1000 m/z and alternated between MS andMS2 scans using dynamic exclusion with approximately 6scans per second. Derivatized samples for GC/MS wereseparated on a 5% phenyldimethyl silicone column withhelium as the carrier gas and a temperature ramp from60°C to 340°C and then analysed on a Thermo-FinniganTrace DSQ MS (Thermo Fisher Scientific, Inc.) operatedat unit mass resolving power with electron impact ioniza-tion and a 50–750 atomic mass unit scan range.

Metabolite identification

Metabolites were identified by automated comparisonof the ion features in the experimental samples to a ref-erence library of chemical standard entries that includedretention time, molecular weight (m/z), preferred ad-ducts and in-source fragments as well as associated MSspectra, and were curated by visual inspection for qual-ity control using software developed at Metabolon (13).At the time of this writing, more than 4000 commer-cially available purified standard compounds wereincluded in the reference library.

Statistical analysis

For statistical analyses and data display purposes, anymissing values were assumed to be below the limits ofdetection and these values were imputed with the com-pound minimum (minimum value imputation). Statis-tical analysis of log-transformed data was performedusing ‘R’ (http://cran.r-project.org/), which is a freelyavailable, open-source software package. Welch’s two-sample t-tests were used to compare differences amonggroups. Classification by principal component analysisand unsupervised hierarchal clustering was carried outusing Array Studio (V5.0). Classification by the unbi-ased and supervised random forest technique, providingan estimate of the ability to classify individuals, was car-ried out between groups using the R-package ‘Random-Forest.’ Briefly, a set of classification trees, based oncontinual sampling of the experimental units and com-pounds, was created and each observation was classifiedbased on the majority votes from all classification trees.Identified drug metabolites contained within the xeno-biotic pathway were not included in the RF classificationscheme. P-values ≤ 0.05 were considered statisticallysignificant and P-values < 0.10 were reported as trends.Multiple comparisons were accounted for by estimatingthe false discovery rate (FDR) using q-values (14).

Results

Patient characteristics

A total of 49 subjects were included in this serum meta-bolic profiling study. Characteristics of the 18 patients

with PBC, 21 patients with PSC and 10 healthy controlsubjects, including demographics and liver biochemist-ries, are summarized in Table 1. Consistent with theestablished gender predispositions, the PBC group waspredominantly female (94%), while the PSC group waspredominantly male (76%). Additional characteristicssuch as age, Caucasian race and body mass index (BMI)were generally well-matched across the cholestatic dis-ease and healthy control groups. Statistical comparisonsof liver biochemistries and additional biochemicalparameters revealed significantly higher (P-value <0.05) levels of alkaline phosphatase and serum total pro-tein in the PBC and PSC groups as compared to healthycontrols, with non-significant elevations in alanineaminotransferase (ALT), aspartate aminotransferase(AST) and total bilirubin. Additional parameters such

Table 1. Clinical characteristics and liver biochemistries of studyparticipants

Primarybiliarycirrhosis(n = 18)

Primarysclerosingcholangitis(n = 21)

Control(n = 10)

DemographicsAge (years) 49 � 10 52 � 16 45 � 15Female (%) 94 24 40Caucasian (%) 89 95 90BMI (kg/m2) 30 � 7 34 � 8 30 � 6Hypertension (%) 28 19 N/ADiabetes (%) 22 5 N/AIBD (%) 0 71 N/A

MedicationsAntihyperlipidaemics(%)

33 10 N/A

Ursodeoxycholic acid(%)

78 76 N/A

Liver BiochemistriesALT (U/L) 96 � 154 65 � 48 17 � 7AST (U/L) 136 � 327 57 � 28 23 � 3Alk P (U/L)* 231 � 117 249 � 234 58 � 15Total Bilirubin (mg/dl) 1.5 � 2.3 1.4 � 1.1 0.5 � 0.2

Other parametersSerum albumin (mg/dl) 3.7 � 0.6 3.6 � 0.5 4.0 � 2.3Serum total protein(mg/dl)*

7.5 � 1.4 7.6 � 0.7 6.3 � 0.3

INR 1.0 � 0.1 1.2 � 0.5 N/ASerum creatinine (mg/dl) 0.8 � 0.2 0.9 � 0.3 0.9 � 0.1MELD score 8.1 � 2.8 10.6 � 3.8 N/A

*P-value < 0.05.

All values are expressed in mean � standard deviation unless otherwise

mentioned.

BMI, body mass index; IBD, inflammatory bowel disease; ALT, alanine

aminotransferase; AST, aspartate aminotransferase; Alk P, alkaline

phosphatase; INR, international normalized ratio; MELD, model for end-

stage liver disease; N/A, not applicable.


Bell et al. Metabolomics of PBC and PSC

as serum albumin and serum creatinine were not differ-ent between the liver disease and healthy control groups,and the international normalized ratio (INR) and modelfor end-stage liver disease (MELD) scores were not dif-ferent when comparing the PBC and PSC groups. It isimportant to note that the majority of PBC and PSCpatients were receiving UDCA therapy (78% and 76%,respectively), and that a small subset of cholestaticpatients was also taking antihyperlipidaemic medica-tions (33% of the PBC group and 10% of the PSCgroup). Administration of these medications has thepotential to affect metabolic pathways highlighted inthis report, including bile acid, cholesterol and lipidmetabolism.

Metabolic profiling

Findings from the global serum metabolomic analysisare summarized in Fig. 1 and a list of all identifiedmetabolites, grouped by super- and subpathways, theirchanges when comparing among groups, and P- and q-values are provided as Table S1. Overall, 420 metaboliteswere identified and, of these, 101 differed significantly(P ≤ 0.05) when comparing PBC and control groups,115 differed significantly when comparing PSC and con-trol groups, 56 differed significantly when comparingPSC and PBC groups, and 4 differed significantly acrossall three comparisons (Fig. 1). Directionality of thesechanges (i.e. biochemicals that were increased ordecreased) is reported in Table 2. In addition to signifi-cant changes, trending (0.05 < P < 0.10) changes werealso considered and a summary of the numbers anddirection of these changes is included in Table 2.Although the number of subjects taking lipid-loweringmedications was too low for meaningful comparisons,patients with PBC or PSC who were taking UDCA werecompared to those not receiving therapy. In general, the

number of statistically significant changes when com-paring patients receiving UDCA to those not takingUDCA within the PBC (D = 23) and PSC (D = 29)groups was roughly equivalent to the number of changesthat would be expected by random chance alone (5changes for every 100 metabolites). Results from thesecomparisons are included within Table S1.

Classification analyses

Utilization of global metabolic profiles for classificationof control subjects and patients by disease type wasdetermined by principal component analysis (PCA),random forest analysis (RF) and unsupervised hierar-chal clustering. As shown in Fig. 2A, results from thePCA demonstrate that control subjects were quite dis-tinguishable from those with cholestatic disease, butthat a substantial amount of overlap was present whendifferentiating patients with PBC and PSC. However,control subjects and patient groups were readily identi-fied by RF analysis, which gave an overall predictiveaccuracy of 94% when differentiating between all threestudy groups (Fig. 2B). Similarly, classification of onlythe PBC and PSC groups by RF gave an identical overallpredictive accuracy of 95% as well (Fig. 2C). RF pro-duces a list of biochemicals ranked by their importanceto the classification scheme, and the top 30 metabolitesidentified for each analysis are displayed in Figs 2B andC. Metabolites involved in amino acid/protein metabo-lism (including fibrinogen cleavage peptides), lipidmetabolism (including eicosanoids, sterol metabolismand bile acids), glucose metabolism, nucleotide metabo-lism and xenobiotics were identified as contributing tothe separation of groups and many of these biochemi-cals are discussed in more detail in this report. Of note,22 common biochemicals were included in both RFimportance plots. As shown in Fig. S1, unsupervisedhierarchal clustering resulted in limited grouping ofhealthy control samples with extensive overlap observedbetween the two liver disease groups.

Pathway analyses

Further consideration of significant biochemicalchanges identified several metabolic pathways that weredifferentially affected in patients with PBC or PSC as

Table 2. Summary of significant and trending biochemicalchanges observed when comparing serum from patients with pri-mary biliary cirrhosis (PBC; n = 18), primary sclerosing cholangitis(PSC; n = 21) and healthy controls (n = 10). The directionality ofchanges is also indicated in red (increase) or green (decrease)

Control(n = 10)

PBC(n = 18)

PSC(n = 21)

115

101

56

4

Fig. 1. Summary of significant biochemical changes ( of 420 identi-fied metabolites) when comparing among groups. Numbers of sig-nificant changes (P ≤ 0.05) when comparing across the primarybiliary cirrhosis (PBC), primary sclerosing cholangitis (PSC) andhealthy control groups, in addition to the numbers of subjectswithin each of the study groups, are displayed.



–20 –10 0 10 20 30

Comp.1 [12.10%]

–15

–10

–5

0

5

10

15

20

Com

p.2

[9.5

7%]

HCPBCPSC

15 20 25 30 35 40 45 50 55

gamma-glutamylalanine

propionylcarnitine

creatine

ribose

leucylalanine

threonate

cysteine-glutathione disulfide

guanosine

tryptophylglutamate

mannose

pyroglutamine*

pyroglutamylglycine

leucylleucine

hyocholate

bilirubin (Z Z)

bilirubin (E E)*

glycerate

5-oxoproline

glycylvaline

aspartylphenylalanine

ADpSGEGDFXAEGGGVR*

7-beta-hydroxycholesterol

inosine

glycylphenylalanine

HWESASXX*

histidyltryptophan

lactate

12-HETE

DSGEGDFXAEGGGVR*

leucylphenylalanine

PREDICTED GROUP

ACTUAL GROUP

Healthy Controls

PBC Patients

PSC Patients

Healthy Controls 10 0 0

PBC Patients 1 16 1

PSC Patients 0 1 20

Predictive accuracy = 94%

Mean decrease accuracy

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

10 20 30 40 50 60

lactate

glycylglycine

fructose

propionylcarnitine

leucylalanine

ornithine

serine

1-palmitoylplasmenylethanolamine*

gamma-tocopherol

creatine

ribose


ADpSGEGDFXAEGGGVR*

hypoxanthine

pyroglutamylglycine

phenylalanylleucine

bilirubin (E E)*

threonate

tryptophylglutamate

glycylphenylalanine

guanosine

glycerate

inosine

bilirubin (Z Z)

5-oxoproline

glycylvaline

12-HETE

histidyltryptophan

aspartylphenylalanine

DSGEGDFXAEGGGVR*

PREDICTED GROUP

ACTUAL GROUP

PBC Patients

PSCPatients

PBC Patients 17 1

PSCPatients 1 20

Predictive accuracy = 95%

Mean decrease accuracy

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

#

(A)

(B) (C)

Fig. 2. Utilization of global metabolic profiles for classification of control subjects and patients by disease type by principal componentanalysis (PCA) and random forest analysis (RF). Results from the PCA (Panel A) demonstrated that healthy control subjects (HC; clear circles)were quite distinguishable from those with cholestatic disease, but that a substantial amount of overlap was present when differentiatingpatients with PBC (light grey circles) and PSC (dark grey circles). Classification of all three study groups (Panel B) or the two liver diseasegroups (Panel C) by RF gave overall predictive accuracies of 94% and 95% (inset confusion matrices), respectively, and 22 common metabo-lites (designated by#) were identified within the biochemical importance plots generated from the RF analyses. Importance plots display bothrank of metabolites according to the contribution of each to the classification scheme (mean decrease accuracy) and biochemical pathwayidentifications (red = amino acid, orange = carbohydrate/energy, yellow = cofactors and vitamins, green = lipid, blue = nucleotide andpink = peptide).



compared to healthy controls. As shown in Fig. 3, sig-nificant elevations in several bile acids were observed inthe liver disease groups as indicated by both P ≤ 0.05and q ≤ 0.10 when comparing the PBC or PSC groupsand healthy controls. For example, higher levels of theconjugated primary bile acids glycocholate, taurocho-late, glycochenodeoxycholate and taurochenodeoxycho-late were present in patients with PBC or PSC. Similarly,significant elevations were noted in the unusual trihy-droxy bile acid hyocholate, as well as its conjugatedderivatives, glycohyocholate and taurohyocholate, inpatients with liver disease. Finally, while the secondarybile acid deoxycholate and its glycine- and taurine-con-jugated derivatives were not different when comparingpatients with liver disease and controls, greater than 23-fold elevations were observed in glycoursodeoxycholateand tauroursodeoxycholate in the PBC and PSC groups.Although UDCA is synthesized endogenously, theselarge increases in PBC and PSC patients also reflectexogenous administration of this bile acid as a treatmentfor cholestatic disease. Indeed, higher levels of glyco-ursodeoxycholate and tauroursodeoxycholate wereidentified in PBC and PSC patients taking UDCA ascompared to those not receiving this therapy. As bileacids are the primary route for excretion of cholesterolfrom the body, it is also important to note the signifi-

cant elevation (P ≤ 0.05 and q ≤ 0.10) in circulatingcholesterol in both PBC and PSC patients as comparedto controls.

Alterations in biochemicals related to lipid metabo-lism were also observed in patients with cholestaticdisease as compared to healthy controls (Fig. 4).Despite unchanged levels of the lipolysis marker glyc-erol, consistent elevations (P ≤ 0.05 and q ≤ 0.10) inmany free fatty acids, including the essential fattyacids linoleate (18:2n6) and linolenate (18:3n3 or 6)and the long-chain unsaturated fatty acids palmitol-eate (16:1n7) and oleate (18:1n9), were noted in thePBC and PSC groups. As summarized in Table 3, boththe number of trending (0.05 < P < 0.10) or signifi-cant (P ≤ 0.05) changes as well as the magnitude ofthose changes was slightly more pronounced inpatients with PBC as compared to PSC. In addition tochanges in free fatty acids, significant elevations in theketone bodies acetoacetate and 3-hydroxybutyrate(BHBA), as well as several acylcarnitines, includingstearoylcarnitine, were also noted in the PBC and PSCgroups, respectively, as compared to controls. It isimportant to note that administration of UDCA had aprofound effect on circulating free fatty acid andketone body levels, but this was only observed withinthe PSC group (Table S1).

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Glycocholate

Ctrl PBC PSC05

101520253035

Taurocholate

Ctrl PBC PSC0

1020304050607080

Glycochenodeoxycholate

Ctrl PBC PSC0123456789

Taurochenodeoxycholate

Ctrl PBC PSC0

10

20

30

40

50

Hyocholate

Ctrl PBC PSC02468

10121416

Glycohyocholate

Ctrl PBC PSC0

20406080

100120140160

Taurohyocholate

Ctrl PBC PSC02468

10121416

Glycoursodeoxycholate

Ctrl PBC PSC0

10

20

30

40

50

60Tauroursodeoxycholate

Ctrl PBC PSC0

10

20

30

40

50

60

Fig. 3. Differential bile acid levels in patients with primary biliary cirrhosis (PBC) or primary sclerosing cholangitis (PSC) as compared tohealthy controls. Boxplots convey the spread of the data with the middle 50% of the data represented by shaded box and the maximumand minimum values (unless there are extreme values) shown by the whiskers. Extreme values are defined as those that exceed, in eitherdirection, 1.5 times the interquartile range. The solid bar across the box represents the median value, while the + shows the mean. For eachbiochemical, data are median scaled with the median value across all samples set equal to 1.0; thus, the y-axis reflects scaled intensity foreach metabolite. All displayed bile acids were significantly greater (as indicated by both P ≤ 0.05 and q ≤ 0.10) in the PBC and PSC groupsas compared to healthy controls.



As displayed in Fig. 5, several metabolites that serveas biochemical markers of inflammation and oxidativestress were altered in patients with PBC or PSC as com-pared to controls. For example, serum levels of the tryp-tophan degradation product kynurenine and thearachidonate-derived eicosanoid 12-hydroxyeicosatetra-enoic acid (HETE) were altered in the PBC and PSCgroups, suggestive of an inflammatory signature in thesepatients. Similarly, increased levels of the linoleate-derived product of lipid peroxidation 13-hydroxyocta-decadienoic acid (HODE) + 9-HODE, the marker ofoxidative stress and pro-apoptotic metabolite 7b-hy-droxycholesterol, and the free radical scavenger biliver-din were observed in PBC patients and increased levels

of dimethylarginine (SDMA + ADMA) and biliverdin,along with altered antioxidant (vitamin C and vitaminE) metabolism, was observed in PSC patients.

Finally, substantial changes with regard to proteinand amino acid metabolism were observed in patientswith PBC and PSC (Table 4). In particular, levels of sev-eral free amino acids (serine, aspartate, glutamate, phen-ylalanine and ornithine) and nine dipeptides weresignificantly (P ≤ 0.05) lower in patients with PSC ascompared to PBC. Moreover, significant elevations inthree fibrinogen cleavage peptides, produced whenfibrinogen is cleaved by thrombin for blood clot forma-tion, were also observed in the PSC as compared to PBCgroup.

Acetyl-CoA

Malonyl -CoA

Fatty acid synthesis

TCA cycleFatty acidβ-oxidation

Acyl-CoA

Acyl-carnitine

Acyl-CoA

Cytosol

Mitochondria

Free fatty acid

Triacylglyceridesmembrane phospholipids

Citrate

acetyl-CoA

Dietaryuptake

Ketogenesis

Acetoacetate

BHBA

Fasting

LysolipidsGlycerol

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

**

*

*

Stearoylcarnitine

Ctrl PBC PSC0.0

0.5

1.0

1.5

2.0

2.5

3-hydroxybutyrate (BHBA)

Ctrl PBC PSC0

1

2

3

4

5

6Acetoacetate

Ctrl PBC PSC0

1

2

3

4

Glycerol

Ctrl PBC PSC0.0

0.5

1.0

1.5

2.0

2.5

Fig. 4. Changes in metabolites related to lipid metabolism in the primary biliary cirrhosis (PBC) and primary sclerosing cholangitis (PSC)groups. In addition to increased serum-free fatty acid levels (Table 3), significant (P ≤ 0.05) elevations in ketone bodies and acylcarnitineswere also observed in the cholestatic disease groups as compared to controls. *P ≤ 0.05 vs. control



Discussion

Although metabolomic profiling has been used exten-sively for biomarker identification and informationalstudies involving many other liver diseases, particularlyhepatocellular carcinoma (15) and drug-induced liverinjury (16), to the best of our knowledge this study isthe first to investigate the global serum metabolome ofPBC and PSC. Here, we observed important differencesin the serum metabolic profiles of patients with PBC orPSC as compared to healthy controls and, as a result,potential biomarker candidates for differentiating PBCfrom PSC were also identified. While clinical diagnosisof PBC is relatively straightforward, PSC presents agreater challenge without the use of endoscopic retro-grade cholangiopancreatography (ERCP) or magneticresonance cholangiopancreatography (MRCP). In addi-tion, a serological hallmark that is present in PBC butnot PSC includes the antimitochondrial antibody(AMA), an autoantibody that is found in 90–95% ofPBC patients and less than 1% of normal controls (17–19). In this study, differences in the serum metabolomewere identified when comparing PBC and PSC patientsand classification analyses demonstrated that, in general,the two disease states were quite distinguishable. In par-ticular, predictive accuracy of the RF analyses, includingclassification of the PBC and PSC groups only, was highat 94–95% and analysis of the biochemical importanceplots generated from both RF analyses revealed thatmetabolites reflective of protein metabolism (i.e. dipep-tides and fibrinogen cleavage peptides) appear to have

the greatest power in differentiating groups. Additionaldifferentially altered pathways with metabolites able todistinguish among groups were also identified whencomparing patients with PBC or PSC and these findingsare discussed in more detail below.

By definition, circulating levels of bile acids areincreased in cholestatic disorders such as PBC and PSCas a result of biliary injury and hepatocyte accumulationof bile acids (3). In this study, elevations in unconjugat-ed, conjugated, and sulphated bile acids were observedin both liver disease groups, with minor changes in sec-ondary bile acids derived from the gut microbiome. Thispattern of change is mostly in agreement with the previ-ous report by Trottier et al. demonstrating that thepathology of PBC and PSC mainly affects primary bileacids (11). One finding of particular interest included asignificant elevation in serum hyocholate, an unusual tri-hydroxy bile acid synthesized by hydroxylation of chen-odeoxycholate (20), in both PBC and PSC patients ascompared to controls. Hyocholate has been found to beelevated in urine from bile acid-loaded adults (20) andchildren with Byler disease (21) and may therefore act asa biomarker of cholestasis and the ability of the liver tometabolize bile acids to their secretory forms. Indeed,hyocholate was the only bile acid included within therandom forest importance plot as contributing to theseparation of the control and liver disease groups.

Also of interest with regard to bile acidmetabolism, cir-culating levels of cholesterol were significantly elevated inserum from both PBC and PSC patients. Perturbations incholesterol homoeostasis have been commonly associ-

Table 3. Alterations in free fatty acids when comparing patients with primary biliary cirrhosis (PBC), primary sclerosing cholangitis (PSC)and healthy controls. Trending (0.05 < P < 0.10) or significant (P ≤ 0.05) elevations are denoted by pink and red shading, respectively



ated with chronic cholestatic diseases (22). Changes inserum cholesterol are related to altered bile acid pro-duction through mechanism(s) that involve, at least tosome extent, genetic, transcriptional, translational andpost-translational modifications of hepatic cytochromeP450 7A1 (CYP7A1) expression/activity (23–25). On arelated note, serum levels of the oxidized cholesterolderivatives 7-a-hydrocholesterol (bile acid precursor)and 7-b-hydroxycholesterol (marker of lipid oxidationand cytotoxic, pro-apoptotic lipid mediator) and thesteroid hormone cortisol were elevated in only thePBC group. Although 7-a-hydroxycholesterol is notan accurate systemic indicator of hepatic bile acid syn-thesis (26), and cortisol can be derived from exoge-nous administration, these differential changes may beindicative of greater lipid peroxidation and potentialabnormalities of the pituitary–adrenal axis in patientswith PBC. These key differences observed when com-paring the two cholestatic groups may point to spe-cific metabolic alterations that contribute to thepathogenesis of PBC but not PSC.

While abnormalities in lipoprotein and cholesterolmetabolism have been well-described in patients withcholestatic disease (22), investigation into changes incirculating free fatty acids and hepatic b-oxidation isquite limited (27). In this study, although consistent ele-vations in serum-free fatty acids were observed in bothPBC and PSC patients as compared to controls,increases were slightly more pronounced in the PBC ascompared to the PSC group. This finding may be attrib-utable to the fact that PSC patients receiving UDCAtherapy (76% of the PSC group) exhibited lower levelsof circulating free fatty acids as compared to those nottaking UDCA. As serum levels of the lipolysis markerglycerol were unchanged in the PBC and PSC groups,the observed changes in free fatty acids may be reflectiveof alterations in hepatic lipid metabolism as opposed toincreased release from adipose tissue stores. Indeed,similar elevations in the ketone bodies acetoacetateand BHBA, along with modest accumulation of severalmedium- and long-chain acylcarnitines, were observedin both liver disease groups despite lower circulating

12-HETE

Ctrl PBC PSC0

2

4

6

8

10

12

14

Sca

led

inte

nsity

Sca

led

inte

nsity

* *

*

Kynurenine

Ctrl PBC PSC0.0

0.5

1.0

1.5

2.0

2.5

#

Threonate

Ctrl PBC PSC0

1

2

3

4

5

13-HODE + 9-HODE

Ctrl PBC PSC012345678

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

Sca

led

inte

nsity

*

*#

*

*

*


Ctrl PBC PSC0.0

0.5

1.0

1.5

2.0

2.5

3.0Biliverdin

Ctrl PBC PSC0

1

2

3

4

5

6

7

Delta-tocopherol

Ctrl PBC PSC0.00.51.01.52.02.53.03.54.0

Dimethylarginine (SDMA + ADMA)

Ctrl PBC PSC0.0

0.5

1.0

1.5

2.0 #

(A)

(B)

Fig. 5. Alterations in biochemical markers of inflammation and oxidative stress in patients with primary biliary cirrhosis (PBC) or primarysclerosing cholangitis (PSC). Metabolites related to inflammation (Panel A) and oxidative stress (Panel B) that exhibited trending(0.05 < P < 0.10) or significant (P ≤ 0.05) changes when comparing PBC and/or PSC patients to controls are displayed. *P ≤ 0.05 vs.control; #0.05 < P < 0.10 vs. control



levels of ketone bodies in PSC patients taking UDCA ascompared to those not receiving therapy. Ketones aresynthesized within the mitochondrial matrix from ace-tyl-CoA in a pathway that shares several steps with thecholesterol biosynthetic pathway. Therefore, thesechanges may be indicative of a general increase in activ-ity of the mevalonate and ketogenesis pathways in bothliver disease groups. However, mitochondrial dysfunc-tion is known to play a key role in cholestatic disease(28) and is typically associated with impaired fatty acidb-oxidation. Thus, accumulation of acylcarnitines andketone bodies in patients with PBC and PSC mayinstead be suggestive of reduced fatty acid oxidation andketogenesis, resulting in increased circulating free fattyacids as well as decreased peripheral utilization ofketones by target tissues such as brain and cardiac mus-cle. Support for this hypothesis comes from previousstudies demonstrating a central role for impaired keto-genesis in altered hepatic lipid metabolism observed incholestatic rats following bile duct ligation (29, 30).

Inflammation and oxidative stress are known to con-tribute to the pathogenesis of cholestatic diseases

including PBC and PSC (7, 28, 31). In this study, a simi-lar inflammatory signature was observed in both thePBC and PSC groups, with elevations in the trypto-phan-derived metabolite kynurenine and reductions inthe lipooxygenase-derived eicosanoid 12-HETE. Themajority of tryptophan is degraded in the liver by theenzyme tryptophan dioxygenase (TDO), but this aminoacid can also be metabolized by the highly inducibleenzyme indoleamine 2,3-dioxygenase (IDO) that isstimulated by pro-inflammatory cytokines such as inter-feron-c and tumour necrosis factor-a (32). Arachido-nate-derived products of the lipid-peroxidating enzyme12/15-lipoxygenase, including 12-HETE, generally pos-sess anti-inflammatory properties to counteract pro-inflammatory responses (33). Therefore, higher levels ofkynurenine and reductions in 12-HETE are suggestiveof a systemic inflammatory signature that is similar inpatients with PBC and PSC. Conversely, while elevationsin various biochemical markers of oxidative stress wereobserved in both the PBC and PSC groups, differentialchanges in specific metabolites reflective of an oxidativeenvironment were noted when comparing the twogroups. For example, the oxysterol 7b-hydroxycholes-terol, which is formed by autooxidation of cholesteroland is a specific marker of endogenous oxidative stressand lipid peroxidation, and the lipoxygenase-derivedmetabolite 13-HODE + 9-HODE, which also serves as alipid peroxidation marker, were only elevated in thePBC group. In contrast, levels of asymmetric dimethy-larginine (ADMA), which are elevated when oxidativestress increases the activity of arginine-methylatingenzymes (i.e. protein arginine N-methyltransferases)and decreases the activity of the ADMA-degradingenzyme dimethylarginine dimethylaminohydrolase (34),were increased only in patients with PSC. These findingssuggest that while greater oxidative stress is present inboth cholestatic diseases, the source(s) and pathwaysbehind this change may differ between PBC and PSC.

Differential changes in metabolites related to proteinand amino acid metabolism were also observed whencomparing patients with PBC and PSC and several ofthese were identified as key biochemicals with regard tothe RF classification scheme. In particular, levels ofmany free amino acids and dipeptides were lower in thePSC as compared to the PBC group. Reports in the liter-ature corroborate findings from this study, demonstrat-ing perturbations in protein and amino acidmetabolism in cholestatic disease, including differentialchanges in the plasma amino acid profile of patientswith PBC and PSC (35). For example, administration ofalanine (significantly reduced in both the PBC and PSCgroups in this study) was shown to reduce bilirubin lev-els and improve symptoms in three patients with end-stage PBC (36), while reductions in circulating tyrosinelevels (which were not significantly altered in this study)may be related to greater fatigue and reduced quality oflife in patients with PBC (35). In addition, increasedprotein degradation and inefficient protein synthesis are

Table 4. Differential alterations in free amino acids, dipeptidesand fibrinogen cleavage peptides when comparing patients withprimary biliary cirrhosis (PBC), primary sclerosing cholangitis (PSC)and healthy controls. Trending (0.05 < P < 0.10) or significant(P ≤ 0.05) elevations are denoted by pink and red shading, respec-tively, while trending or significant reductions are indicated by lightgreen and dark green shading respectively

*completely confident in the identity of the metabolite but a commercially-

available standard was not available for purchase.



among the mechanisms associated with malnutrition incholestatic liver disease (37), although it is important tonote that in this study levels of serum total protein weresignificantly greater in the PBC and PSC groups as com-pared to healthy controls. Together with this study,findings from these previous investigations highlight animportant role for changes in protein and amino acidmetabolism in the cholestatic disease process.

Despite the small number of subjects profiled in thisdescriptive study and the inclusion of patients takingUDCA and/or antihyperlipidaemic medications, novelperturbations in the global serum metabolome wereidentified in patients with PBC and PSC and this infor-mation was successfully utilized to classify cholestaticpatients and healthy controls. In addition to commonchanges in patients with cholestatic disease, findingsfrom this study also identified alterations in lipid metab-olism, oxidative stress/lipid peroxidation, stress hor-mones and protein/amino acid metabolism that differedwhen comparing the PBC and PSC groups. These differ-ential changes may not only have diagnostic value withregard to identifying patients with PBC vs. PSC, but alsoprovide insight into differential metabolic pathways thatcontribute to the pathogenesis of these similar condi-tions. Moreover, identification of a novel therapeuticeffect of UDCA on the serum-free fatty acid profile waslimited to the PSC group and is worthy of additionalinvestigation. Metabolic profiling of an independent setof serum samples from cholestatic patients and healthycontrols will be necessary to validate findings from thispilot study and determine the true diagnostic utility ofbiomarker candidates identified in this study.

Acknowledgements

Financial Support: This work is in part supported byNIH K24 DK069290A (NC).

Conflict of interest: LNB and JW are employees ofMetabolon, Inc. and, as such, have affiliations with orfinancial involvement with Metabolon, Inc. The authorshave no other relevant affiliations or financial involve-ment with any organization or entity with a financialinterest in or financial conflict with the subject matteror materials discussed in the manuscript apart fromthose disclosed. Dr. Chalasani has consultant agree-ments and research grants from several pharmaceuticalcompanies but none represent a potential conflict forthis paper. Dr. Vuppalanchi serves as a member of thespeaker’s bureau, research grants from pharma andserves as a consultant but none represent a potentialconflict for this paper. Ms. Megan Comerford has nofinancial conflicts of interests to declare.

References

1. Woolbright BL, Jaeschke H. Novel insight into mecha-nisms of cholestatic liver injury. World J Gastroenterol2012; 18: 4985–93.

2. Kim WR, Ludwig J, Lindor KD. Variant forms of chole-static diseases involving small bile ducts in adults. Am JGastroenterol 2000; 95: 1130–8.

3. Poupon R, Chazouilleres O, Poupon RE. Chronic chole-static diseases. J Hepatol 2000; 32: 129–40.

4. Hirschfield GM, Heathcote EJ, Gershwin ME. Pathogene-sis of cholestatic liver disease and therapeutic approaches.Gastroenterology 2010; 139: 1481–96.

5. Hirschfield GM, Chapman RW, Karlsen TH, et al. Thegenetics of complex cholestatic disorders. Gastroenterology2013; 144: 1357–74.

6. Czul F, Peyton A, Levy C. Primary biliary cirrhosis: thera-peutic advances. Clin Liver Dis 2013; 17: 229–42.

7. Eaton JE, Talwalkar JA, Lazaridis KN, Gores GJ, LindorKD.Pathogenesis of primary sclerosing cholangitis andadvances in diagnosis and management. Gastroenterology2013; 145: 521–36.

8. Trivedi PJ, Hirschfield GM. Review article: overlap syn-dromes and autoimmune liver disease. Aliment PharmacolTher 2012; 36: 517–33.

9. Eckhart AD, Beebe K, Milburn M. Metabolomics as akey integrator for “omic” advancement of personalizedmedicine and future therapies. Clin Transl Sci 2012; 5:285–8.

10. Evans AM, DeHaven CD, Barrett T, Mitchell M, MilgramE. Integrated, nontargeted ultrahigh performance liquidchromatography/electrospray ionization tandem massspectrometry platform for the identification and relativequantification of the small-molecule complement of bio-logical systems. Anal Chem 2009; 81: 6656–67.

11. Trottier J, Bialek A, Caron P, et al. Metabolomic profilingof 17 bile acids in serum from patients with primary bili-ary cirrhosis and primary sclerosing cholangitis: a pilotstudy. Dig Liver Dis 2012; 44: 303–10.

12. Ohta T, Masutomi N, Tsutsui N, et al. Untargeted meta-bolomic profiling as an evaluative tool of fenofibrate-induced toxicology in Fischer 344 male rats. Toxicol Pathol2009; 37: 521–35.

13. Dehaven CD, Evans AM, Dai H, Lawton KA. Organizationof GC/MS and LC/MS metabolomics data into chemicallibraries. J Cheminform 2010; 2: 9.

14. Storey JD, Tibshirani R. Statistical significance forgenomewide studies. Proc Natl Acad Sci USA 2003; 100:9440–5.

15. Wang X, Zhang A, Sun H. Power of metabolomics in diag-nosis and biomarker discovery of hepatocellular carci-noma. Hepatology 2013; 57: 2072–7.

16. Hawkins MT, Lewis JH. Latest advances in predicting DILIin human subjects: focus on biomarkers. Expert Opin DrugMetab Toxicol 2012; 8: 1521–30.

17. Lindor KD, Gershwin ME, Poupon R, et al. Americanassociation for study of liver D Primary biliary cirrhosis.Hepatology 2009; 50: 291–308.

18. Kaplan MM, Gershwin ME. Primary biliary cirrhosis. NEngl J Med 2005; 353: 1261–73.

19. Gershwin ME, Mackay IR, Sturgess A, Coppel RL. Identifi-cation and specificity of a cDNA encoding the 70 kd mito-chondrial antigen recognized in primary biliary cirrhosis. JImmunol 1987; 138: 3525–31.

20. Nakashima T, Sano A, Seto Y, et al. Unusual trihydroxybile acids in the urine of patients treated with chen-odeoxycholate, ursodeoxycholate or rifampicin and thosewith cirrhosis. Hepatology 1990; 11: 255–60.



21. Jacquemin E, Dumont M, Bernard O, Erlinger S, Hadcho-uel M. Evidence for defective primary bile acid secretionin children with progressive familial intrahepatic cholesta-sis (Byler disease). Eur J Pediatr 1994; 153: 424–8.

22. Kowdley KV. Lipids and lipid-activated vitamins in chroniccholestatic diseases. Clin Liver Dis 1998; 2: 373–89.

23. Bertolotti M, Carulli L, Concari M, et al. Suppression ofbile acid synthesis, but not of hepatic cholesterol 7alpha-hydroxylase expression, by obstructive cholestasis inhumans. Hepatology 2001; 34: 234–42.

24. Inamine T, Higa S, Noguchi F, et al. Association of genesinvolved in bile acid synthesis with the progression of pri-mary biliary cirrhosis in Japanese patients. J Gastroenterol2013; 48: 1160–70.

25. Sauter G, Berr F, Beuers U, Fischer S, Paumgartner G.Serum concentrations of 7alpha-hydroxy-4-cholesten-3-one reflect bile acid synthesis in humans. Hepatology 1996;24: 123–6.

26. Hahn C, Reichel C, von Bergmann K. Serum concentra-tion of 7 alpha-hydroxycholesterol as an indicator of bileacid synthesis in humans. J Lipid Res 1995; 36: 2059–66.

27. Mortiaux A, Dawson AM. Plasma free fatty acid in liverdisease. Gut 1961; 2: 304–9.

28. Arduini A, Serviddio G, Tormos AM, Monsalve M, SastreJ. Mitochondrial dysfunction in cholestatic liver diseases.Front Biosci (Elite Ed) 2012; 4: 2233–52.

29. Lang C, Schafer M, Serra D, et al. Impaired hepatic fattyacid oxidation in rats with short-term cholestasis: charac-terization and mechanism. J Lipid Res 2001; 42: 22–30.

30. Lang C, Berardi S, Schafer M, et al. Impaired ketogenesisis a major mechanism for disturbed hepatic fatty acidmetabolism in rats with long-term cholestasis and afterrelief of biliary obstruction. J Hepatol 2002; 37: 564–71.

31. Hirschfield GM, Gershwin ME. The immunobiology andpathophysiology of primary biliary cirrhosis. Annu RevPathol 2013; 8: 303–30.

32. Dantzer R, O’Connor JC, Freund GG, Johnson RW, KelleyKW. From inflammation to sickness and depression: when

the immune system subjugates the brain. Nat Rev Neurosci2008; 9: 46–56.

33. Conrad DJ. The arachidonate 12/15 lipoxygenases. Areview of tissue expression and biologic function. Clin RevAllergy Immunol 1999; 17: 71–89.

34. Sydow K, Munzel T. ADMA and oxidative stress. Atheros-cler Suppl 2003; 4: 41–51.

35. ter Borg PC, Fekkes D, Vrolijk JM, van Buuren HR. Therelation between plasma tyrosine concentration and fati-gue in primary biliary cirrhosis and primary sclerosingcholangitis. BMC Gastroenterol 2005; 5: 11.

36. Nishiguchi S, Habu D, Kubo S, et al. Effects of alanine inpatients with advanced primary biliary cirrhosis: prelimin-ary report. Hepatol Res 2003; 25: 8–13.

37. Munoz SJ. Nutritional therapies in liver disease. SeminLiver Dis 1991; 11: 278–91.

Supporting information

Additional Supporting Information may be found in theonline version of this article:

Fig. S1. Heat map representation of unsupervisedhierarchal clustering of all 420 measured metabolites.Green and red shading represent reductions and eleva-tions, respectively, relative to the median metabolite lev-els across all patients. Individual samples from the threestudy groups [healthy controls (HC), primary biliarycirrhosis (PBC) and primary sclerosing cholangitis(PSC)] are displayed on the x-axis and biochemicals aregrouped from top to bottom according to metabolicsuper pathway.

Table S1. List of all identified metabolites, groupedby super and subpathways, their changes when compar-ing among groups, and P- and q-values.



Documents

Serum metabolic signatures of primary biliary cirrhosis and primary sclerosing cholangitis