Novel technologies and emerging biomarkers for ... · Pia Kvistborg7, Cristina Maccalli8, Holden T. Maecker9, David B. Page10, Harlan Robins11, ... emerging biomarkers for immune

REVIEW Open Access

Novel technologies and emergingbiomarkers for personalized cancerimmunotherapyJianda Yuan1*, Priti S. Hegde2, Raphael Clynes3, Periklis G. Foukas4,5, Alexandre Harari4, Thomas O. Kleen6,Pia Kvistborg7, Cristina Maccalli8, Holden T. Maecker9, David B. Page10, Harlan Robins11, Wenru Song12,Edward C. Stack13, Ena Wang14, Theresa L. Whiteside15, Yingdong Zhao16, Heinz Zwierzina17,Lisa H. Butterfield18 and Bernard A. Fox10*

Abstract

The culmination of over a century’s work to understand the role of the immune system in tumor control has led tothe recent advances in cancer immunotherapies that have resulted in durable clinical responses in patients with avariety of malignancies. Cancer immunotherapies are rapidly changing traditional treatment paradigms andexpanding the therapeutic landscape for cancer patients. However, despite the current success of these therapies,not all patients respond to immunotherapy and even those that do often experience toxicities. Thus, there is agrowing need to identify predictive and prognostic biomarkers that enhance our understanding of the mechanismsunderlying the complex interactions between the immune system and cancer. Therefore, the Society for Immunotherapyof Cancer (SITC) reconvened an Immune Biomarkers Task Force to review state of the art technologies, identify currenthurdlers, and make recommendations for the field. As a product of this task force, Working Group 2 (WG2), consisting ofinternational experts from academia and industry, assembled to identify and discuss promising technologiesfor biomarker discovery and validation. Thus, this WG2 consensus paper will focus on the current status ofemerging biomarkers for immune checkpoint blockade therapy and discuss novel technologies as well ashigh dimensional data analysis platforms that will be pivotal for future biomarker research. In addition, thispaper will include a brief overview of the current challenges with recommendations for future biomarkerdiscovery.

Keywords: Immune checkpoint blockade, Cancer immunotherapy, Biomarkers, Task Force, Immune monitoring,Technology, Bioinformatics

BackgroundThe role of the immune system in cancer control hasbeen debated for over a century. The field of cancer im-munology has progressed with knowledge obtained fromanimal studies, accumulated clinical observations andtranslational research. Interestingly, starting in the 1900sand every 50 years thereafter, three main theories hadbeen proposed to refine our understanding of the impact

of the immune system on cancer. The first theory wassuggested by Paul Ehrlich’s human protective cancer im-munity, followed by Burnet and Thomas’s concept of“cancer immunosurveillance”, and recently by Schreiber,Old and Smyth’s “cancer immunoediting” [1–4]. “Cancerimmunosurveillance” originally implied that the immunesystem was involved at the initial stages of cellular trans-formation and played a solely protective role. Now, theterm “cancer immunoediting” is used to better describethe protective activities, positive and negative sculptingactions of the immune response on developing tumorsin a continuous manner. This process can potentially re-sult in the complete elimination of some tumors, but itcan also generate a non-protective immune state to

* Correspondence: [email protected]; [email protected] Sloan-Kettering Cancer Center, 1275 New York Ave Box 386, NewYork, NY 10065, USA10Earle A. Chiles Research Institute, Providence Cancer Center, 4805 NE GlisanStreet, Portland, OR 97213, USAFull list of author information is available at the end of the article

© 2016 Yuan et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Yuan et al. Journal for ImmunoTherapy of Cancer (2016) 4:3 DOI 10.1186/s40425-016-0107-3

on June 12, 2021 by guest. Protected by copyright.

http://jitc.bmj.com

/J Im

munother C

ancer: first published as 10.1186/s40425-016-0107-3 on 19 January 2016. Dow

nloaded from

http://crossmark.crossref.org/dialog/?doi=10.1186/s40425-016-0107-3&domain=pdfmailto:[email protected]:[email protected]://creativecommons.org/licenses/by/4.0/http://creativecommons.org/publicdomain/zero/1.0/http://jitc.bmj.com/

others that may favor the development of immunologicevasion. Meanwhile, our perception of cancer has chan-ged dramatically. In the past, tumors were thought to bea result of a single, clonal, disordered cell, when in ac-tuality, most resulted from multiple (pre)malignant cells[5]. Tumors are comprised of heterogeneous cell popula-tions, including transformed cells and untransformedcells (such as stromal, endothelial and immune cells),which have indispensable functions in their microenvir-onment [6, 7]. The evasion of immune destruction isnow commonly accepted as a hallmark of cancer [8].Therefore, understanding the status and interaction be-tween cancer and the immune system in the tumormicroenvironment (TME) is of importance for cancerimmunotherapy strategies.T cell infiltration in certain human tumors is associ-

ated with an improved clinical outcome [9, 10]. Theaccumulating evidence suggests that tumors can be clas-sified into two groups: Immunologically-ignorant tumorsand immunologically-responsive tumors (or non-inflamed tumors vs T cell-inflamed tumors), based onthe presence or absence of immune cell infiltrations[11]. Clinically detectable tumors have usually alreadyevolved mechanisms to evade an immune response. Thetumor immune escape mechanism may be different foreach type, which may involve one or multiple steps ofthe cancer-immunity cycle [12]. Immunologically-ignorant tumors may be caused by a low mutation load,immune tolerance against self-antigens and lack of es-sential chemokines and other molecules for T cell hom-ing into tumor sites. In contrast, the progression ofimmunologically-responsive tumors with T cell infiltra-tion indicates an insufficient response that is probablydue to intrinsic T cell immune-inhibition and extrinsictumor-related T cell immunosuppression [13]. IntrinsicT cell immunosuppression involves anergy and exhaus-tion of activated T cells with endogenous immunecheckpoint molecules including cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), programmed cell death1 (PD-1), T cell immunoglobulin mucin-3 (Tim-3) andlymphocyte-activation gene 3 (LAG-3) [14]. The secre-tion of extrinsic inhibitory molecules such as TGF-β, IL-10 and indoleamine 2,3-dioxyenase (IDO) could have adirect negative impact on T cell function in the TMEand on the recruitment of anti-inflammatory cells, in-cluding tolerogenic antigen presenting cells, regulatoryT cells (Treg) and myeloid derived suppressor cells(MDSC). These suppressive immune cells can also in-hibit the actions of cytotoxic T lymphocytes [13].It is a promising approach to block these immunosup-

pressive mechanisms to augment the function of endogen-ous antitumor T cells, which can deliver a robust andeffective clinical response. Blockade antibodies against Tcell checkpoint molecules including CTLA-4 and the PD-

1/PD-L1 axis in mono- or combination therapies havebegun to revolutionize the current standard cancer treat-ment in various cancer types, such as melanoma,non-small cell lung cancer (NSCLC), bladder cancer andHodgkin’s lymphoma [15–22]. Immunologically-responsivetumors are more likely to respond to these checkpointblockade antibody therapies than immunologically-ignoranttumors [11]. Moreover, prior interventions to achieve local,productive inflammation in the TME are required incombination with these therapies to enhance the clinicalresponse for immunologically-ignorant tumors. Thus, it isessential to perform biomarker studies to further characterizethese different classes of tumors and provide guidance fortherapeutic strategies.Improved high-throughput technologies are provid-

ing feasible tools for analyzing the mutation antigenprofile, the gene signature and epigenetic modifica-tion of tumor and immune cells, the breadth of anti-body responses, as well as the magnitude, homingcapacity, cytotoxic function and T cell receptor(TCR) repertoire of T lymphocytes. These noveltechnologies will help advance precision medicine[23]. New technological approaches will enable us toidentify predictive biomarkers such as immunologicsignatures or profiles for the patients who will mostlikely benefit from current immunotherapies. In addition,they will help patients avoid immune-related adverseevents or adverse events of special interest and reducetreatment costs for those unlikely to respond [17]. Fur-thermore, they will enhance our understanding of themechanisms underlying cancer immunotherapies and aidin the development of more appropriate therapies for spe-cific patient populations. In this paper, we will discuss thecurrent progress to identify biomarkers for immunecheckpoint blockade therapies as well as novel technolo-gies and their potential application for future cancer im-munotherapy biomarker discoveries, as illustrated inFig. 1.

ReviewEmerging biomarkers for CTLA-4 immune checkpointblockade immunotherapy

Immune checkpoint blockade has led to durable antitu-mor effects in patients with metastatic melanoma,NSCLC and other tumor types [15, 17, 24–29]. Ipilimu-mab, an antibody that blocks CTLA-4, was approved bythe U.S. Food and Drug Administration (FDA) for pa-tients with advanced melanoma in 2011. However, al-though a subset of patients benefit, it is often withdelayed radiographic response and at the expense ofmechanism-based toxicity [17]. Therefore, it is impera-tive to identify biomarkers in order to elucidate thepharmacodynamic changes, understand the potential

Yuan et al. Journal for ImmunoTherapy of Cancer (2016) 4:3 Page 2 of 25


http://jitc.bmj.com

/J Im

munother C


nloaded from

http://jitc.bmj.com/

mechanisms of action and to find new correlates associ-ated with clinical benefits and/or toxicities.Several serum markers such as lactate dehydrogenase

(LDH), C-reactive protein, vascular endothelial growthfactor (VEGF) and soluble CD25 are associated withclinical outcome in advanced melanoma patients treatedwith ipilimumab [30–34]. In addition, a variety of assaysare available to monitor phenotypic changes in immunecells such as human leukocyte antigen (HLA)-DR andactivated inducible co-stimulator (ICOS) on T cells, tomeasure changes in target immune cell populations suchas MDSC and to assess tumor associated antigen (TAA)specific responses as well as evaluate the functionalityand gene expression profile of antigen-specific T cellpopulations. These assays have led to preliminary find-ings of potential emerging biomarkers for CTLA-4blockade therapy as described in the following section.Ipilimumab augments antitumor immune responses by

activating and increasing the proliferation of T cells [35].Thus, absolute lymphocyte count (ALC) is a potentialpharmacodynamic biomarker for ipilimumab treatmentin patients with melanoma and other solid tumors [36–38].Following treatment with ipilimumab, an ALC ≥1000/μL atweek 7 or an increase in ALC between baseline and week12 was significantly associated with longer overall survival[33, 39, 40]. Because the ALC contains a variable heteroge-neous lymphocyte population as a general biomarker, there

has been strong interest in characterizing changes inspecific T cell subsets during CTLA-4 blockadetherapy. Increased levels of HLA-DR, CD45RO, cen-tral memory markers (CCR7+CD45RA−) and effectormemory markers (CCR7−CD45RA−) on CD4+ andCD8+ T cells were reported after ipilimumab treat-ment in several studies [41–45]. However, the eleva-tion of these T cell markers did not correlate withclinical response to ipilimumab.ICOS is expressed on the cell surface of activated T cells

and plays a role in T cell expansion and survival. The fre-quency of CD4+ICOS+ T cells was shown to increase in adose-dependent manner in patients with bladder cancer,breast cancer and mesothelioma after treatment with ei-ther ipilimumab or tremelimumab [45–49]. In addition, asustained increase in CD4+ICOS+ T cells was observedover 12 weeks after CTLA-4 blockade therapy and corre-lated with improved survival in four independent studies[46, 49–51]. Therefore, an increase in the frequency ofCD4+ICOS+ T cell may be a reproducible pharmacody-namic biomarker to indicate biological activity for CTLA-4 blockade therapy [52]. However, it would be worthwhileto prospectively investigate changes in the frequency ofmultiple T cell subsets in relation to CTLA-4 blockadetherapy in a large cohort of patients.Cancers are immunogenic and express a variety of

TAAs. CTLA-4 blockade was shown to potentiate the

Fig. 1 High-throughput immune assessment for biomarker discovery and personalized cancer immunotherapy. Immunologically-ignorant andimmunologically-responsive tumors are classified by the presence of immune cells in the tumor microenvironment. Potential biomarkers identifiedfrom high-throughput technologies can further differentiate these tumors by the mutation load, gene/protein/antibody signature profile, phenotypeand function of immune cells, and can also provide clinical strategies for personalized cancer immunotherapies. The new and innovative technologiesthat can be utilized to identify potential biomarkers include whole exome sequencing, gene signature, epigenetic modification, protein microarray, B/Tcell receptor repertoire, flow/mass cytometry and multicolor IHC. Arrows indicate a decrease (↓) or increase (↑)



http://jitc.bmj.com

/J Im

munother C


nloaded from


production of TAA-specific antibodies as well as a CD4+

and CD8+ antigen-specific T cell response in patientswith melanoma, ovarian and prostate cancer [45, 53–56].Moreover, melanoma patients seropositive for the cancer-testis antigen NY-ESO-1 were more likely to experienceclinical benefit than those who were seronegative [57]. Incontrast, there was no significant association betweenhumoral response to tumor antigens and clinical benefitin two other studies [45, 58]. However, because of smallsample size, different response criteria and varying dosesof ipilimumab, it is ultimately difficult to make any specificconclusions based on these studies alone. CTLA-4 block-ade has also been shown to actually potentiate a robustspectrum of tumor specific antibody responses. For ad-vanced prostate cancer, it was shown that patients whoclinically responded to CTLA-4 blockade also developedan enhanced antibody response to a greater number of en-dogenous antigens than non-responders. In this study, themajority of antibody responses were patient specific,although there were some shared antibody responsesamong clinical responders [56]. Further prospective valid-ation is warranted to characterize the tumor antigenspecific antibody response as potential biomarker for anti-CTLA-4 therapy.The evaluation of TAA-specific T cell response has also

been an intense focus of immune monitoring for anti-CTLA-4 immunotherapies. A high frequency of Melan-Aor NY-ESO-1 specific CD8+ T cells were detected in mel-anoma and prostate cancer patients who showed a clinicalresponse to anti-CTLA-4 therapy [55, 59]. In these pa-tients, the presence of integrated antibody and the CD8+T cell response to NY-ESO-1 was associated with a signifi-cant survival advantage [57, 60]. In addition, a recentstudy also reported that ipilimumab induced a significantincrease in the number of newly detected melanoma-reactive T cells by enhancing T cell priming [61]. More-over, tumor genetics were shown to be important indefining clinical benefit in ipilimumab treated melanomapatients [62, 63]. The expression of immune-related genesin pretreatment tumor biopsy specimens, especially inter-feron gamma responsive genes, was correlated positivelywith clinical activity in ipilimumab-treated melanoma pa-tients [64]. A recent study showed that two cytolytic genes(granzyme A and perforin) in the TME were significantlyenriched in the ipilimumab clinical benefit cohort com-pared to the cohort that showed no clinical benefit [63].These new findings suggest that the antitumor effect ofCTLA-4 blockade likely involves the amplification of apreexisting or the priming and induction of an immuneresponse against various antigenic targets, especially mu-tant genes.MDSC are a phenotypically heterogeneous cell popula-

tion that is comprised of myeloid-cell progenitors andprecursors of myeloid cells. Interestingly, MDSC can

also function as antigen-presenting cells (APCs). HumanMDSC have been identified in patients with pancreaticcancer, breast cancer, NSCLC and head and neck squa-mous cell carcinoma [65, 66]. Human MDSC have animmature phenotype that is typically lineage negative(Lin−), CD14−, HLA-DR−, CD15+, CD34+, CD11b+,CD33+ and CD13+ [67, 68]. MDSC exert an immuno-suppressive function mainly through the production ofsuppressive molecules, such as ARG1, cytokines, trans-forming growth factor-beta (TGF- β) or IL-10. Thecommon proposed phenotype in humans is CD14+/HLA-DRlow/-, which is based upon this cell popula-tion’s ability to suppress lymphocyte function. The num-ber of CD14+/HLA-DRlow/- cells was shown to beelevated in melanoma patients and this increase corre-lated with melanoma disease activity [69]. Therefore,MDSC have been recently proposed as a potential bio-marker associated with disease progression or survival[70]. Ipilimumab treatment induced an early decrease inthe frequency of MDSC [71]. In addition, a lower base-line MDSC frequency was associated with improvedoverall survival [72, 73].Overall, the biomarkers for CTLA-4 blockade therapy

were mostly identified from small cohort studies. Thus,ongoing efforts are needed to validate these findings in alarger cohort of patients in prospective clinical trials andto determine whether these findings are specific to ipili-mumab treatment compared with other cancer immuno-therapies. Ultimately, the development of robust andvalidated biomarkers that are predictive and/or prognos-tic will help guide future clinical trials. Novel high-throughput technologies, such as exome sequencing,flow-based phenotyping and multifunctional assays andT cell receptor analysis, have advanced recent antigenspecific biomarker discovery and will provide more toolsin order to validate the emerging biomarkers for CTLA-4 blockade therapy. These new technologies and theirpotential application will be discussed in details in sub-sequent sections.

Regulatory T cells as potential biomarkersThe accumulation of Treg and MDSC in human tumorsand their increased frequency in the peripheral circulationof cancer patients have been widely reported [74, 75].Many reports, but not all, link these accumulations ofCD4+FOXP3+CD25hi Treg to poor prognosis due to thesuppression of antitumor immune response by the Treg[74]. However, in human solid tumors such as colorectalcancer or breast carcinomas, which are often richly infil-trated with immune cells, the presence and density ofFOXP3+ Treg have been reported to predict favorable out-come and a better local regional control of the tumor [76].Given recent emphasis on the tumor “immune signature”and emerging correlations of the immunohistology data to



http://jitc.bmj.com

/J Im

munother C


nloaded from


cancer patients’ survival [9, 76], the reliable phenotypic,and especially functional, characterization of Treg in situand in body fluids is of critical importance. To date, mostof the studies examining the association between Tregphenotype and prognosis or therapeutic response are stillbased on the use of FOXP3 as a “specific” Treg marker[77]. Recent data show that FOXP3 is not a reliablemarker of human Treg and that a collection of severalother markers may be a better option [77, 78].In general, human Treg have been difficult to study for

the following reasons: (1) they represent only a minor sub-set of CD4+ T cells (about 5 %) and thus are often limitedin numbers; (2) they lack a specific surface marker, makingtheir isolation and identification questionable; and (3)Treg plasticity has made it difficult to differentiate naïve(nTreg) or thymus-derived Treg (tTreg) from inducible(iTreg) or peripheral Treg (pTreg). Various marker panelsused to phenotypically identify Treg invariably include ex-pression of CD25hi and/or FOXP3+. In addition, the ab-sence of CD127 or CD26 has been useful for Treg typingand isolation [79, 80].The following recommendations for Treg flow cytom-

etry panels have recently been made: (1) a minimal def-inition of Treg should include CD3, CD4, CD25, CD127,FOXP3 markers with Ki67 and CD45RA to clarify theTreg activation status; (2) the sole dependence on any ofthe three most commonly used flow panels for the Tregphenotypic definition [(a) CD25+CD127low FOXP3+Treg;(b) FOXP3+HELIOS+Treg; or (c) FOXP3hiCD45RAneg

vs. FOXP3intCD45RA+ to distinguish activated fromnaïve Treg, respectively] leads to an underestimation ofthe Treg frequency ranging from 25 to 65 %. Functionalmarkers, such as CD39 and CTLA-4, denote activatedor iTreg, and thus, may be considered “optional”markers.It has been reported that expression of surface

markers on Treg becomes altered in disease [81] and inpatients undergoing conventional therapies or immunetherapies [82, 83]. Therefore, the selection of a panel ofmarkers for measuring Treg is a critical task that will ul-timately determine its role as a prognostic biomarker incancer and other disease. As in cancer, iTreg are un-doubtedly the predominant Treg subset in situ and inthe peripheral circulation; their number, localization andfunctions are of utmost importance. Thus, “activation”markers, such as CD39, CTLA-4, latency-associatedpeptide (LAP), glycoprotein A repetitions predominant(GARP), PD-1, PD-L1 and others that are often overex-pressed on Treg in cancer, emerge as important surro-gate markers for Treg function and should be includedin the monitoring of Treg in cancer patients. Althoughthese markers are not specific to Treg, they are usefulwhen used in combination with CD25hi and FOXP3+ toassess the functional potential of Treg by flow cytometry

and eliminate the need for Treg isolation that is neces-sary in conventional carboxyfluoresceinsuccinimidylaster (CFSE)-based suppressor assays [84, 85]. Efforts toidentify a specific Treg marker that might distinguishnTreg from iTreg have recently focused on Kruppel-likefactor 2 (KLF2), a transcription factor that regulateschronic inflammation and is necessary for the develop-ment of iTreg but not of nTreg [86]. Although there isstill no consensus on which marker panel (of the severallisted above) is best and which subset of Treg should bemonitored, some investigators prefer to focus on onefunctional subset, e.g., the CD4+CD39+CD25+ adenosine-producing Treg [87]. Based on the principle that functionrather than phenotype determines the biological and clin-ical significance of Treg, this strategy, while limited inscope, offers the advantage of following disease-associatedchanges in a single subset of Treg and correlating thesechanges to disease progression [87].A number of in vitro suppression assays are available

for human Treg [78]. Among these, flow cytometrybased assays to measure the surface expression of LAP/GARP on Treg, the intracellular expression of inhibitorycytokines (TGF-β or IL-10) or the downregulation ofCD69 or CD154 expression in co-incubated respondercells require a short-term, ex vivo activation of Treg. Inthis regard, the flow-based assays are easier and have ahigher throughput potential than the conventional co-culture assays of Treg with CFSE-labeled respondercells.Treg have constitutive expression of FOXP3 and

CTLA-4 on their cell surface and intracellular. Recently,a study illustrated that anti-CTLA-4 antibody depletedTreg in tumor lesions through Fc-dependent mechanismto potentially enhance antitumor immunity in mice [88].Moreover, the number of Treg (CD4+CD25+CD62L+

cells) in peripheral blood decreased at early time pointsbut rebounded to a level at or above baseline value atthe time of next dose [89]. In contrast, several studiesreported that ipilimumab in fact induced the prolifera-tion and expansion of Treg, especially at lower doses,whereas activated effector CD4+ cells were expandedonly at higher ipilimumab doses [43, 90]. Although thedecreasing FOXP3/Treg was associated with a better clin-ical outcome in ipilimumab treated melanoma patients[91], it will be critical to further characterize Treg withnew markers as described above in both peripheral bloodand tumor tissue and explore their correlation with clin-ical outcome in patients treated with immunotherapies.

Emerging biomarkers for PD-1/PD-L1 immune checkpointblockade immunotherapyThe immune checkpoint molecule programmed death-1(PD-1, CD279) is upregulated on activated T lympho-cytes and inhibits T-cell function by binding to its



http://jitc.bmj.com

/J Im

munother C


nloaded from


ligands PD-L1 (B7-H1, CD274) and PD-L2 (B7-DC,CD272) [92–94]. The PD-1/PD-L1 axis plays a pivotalrole in regulating T cell activation and peripheral im-mune tolerance. The PD-1/PD-L1 interaction functionsto prevent bystander tissue damage during inflammation,but it can also maintain an immunosuppressive TMEthat allows tumor cells to evade immune surveillance[95, 96]. Similar to CTLA-4 blockade therapy, monoclo-nal antibodies that block PD-1 on immune effector cellsor PD-L1 on tumor cells and/or APCs have beenemployed to restore immune activation. Several antibodiesagainst PD-1 (nivolumab, pembrolizumab, MED10680,pidilizumab) and PD-L1 (BMS-936559, MED14736, atezo-lizumab, MSB0010718A) are currently under clinical in-vestigation and have demonstrated generally manageablesafety profiles and remarkable anti-tumor responses incancer patients with a wide range of metastatic diseases[17, 18, 97–102]. As a result of positive clinical results inearly studies, the FDA approved pembrolizumab and nivo-lumab for patients with advanced melanoma in 2014 andfor patients with metastatic squamous and non-squamousNSCLC in 2015.Although both are immune checkpoints, CTLA-4 and

PD-1 have distinct roles in regulating immunity. CTLA-4 regulates the amplitude of early activation of naïve andmemory T cells, while PD-1 with the corresponding up-regulation of its ligands PD-L1 and PD-L2 limits the ac-tivity of T cells in the periphery during an inflammatoryresponse [95, 96]. PD-1 plays multiple roles in immuneregulation, and it is expressed on a variety of immunecell types, including T cells (CD8+ and CD4+), B cells,natural killer cells, monocytes and APCs. The PD-1/PD-L1 axis can negatively regulate the activation and func-tion of T and B cells as well as inducing Treg, therebyinhibiting anti-tumor immunity. Moreover, PD-1 isexpressed with other immune checkpoint moleculessuch as LAG-3 and Tim-3 on non-functional CD8+ Tcells, which supports the notion that PD-1 may interactwith other immune checkpoints to control anti-tumorimmunity [99]. In addition to PD-1, PD-L1 also binds toB7.1 (CD80) and can inhibit T cell proliferation [100].These distinct inhibitory interactions act more select-ively within the TME, suggesting that anti-PD-1 andanti-PD-L1 antibodies may have different clinical activ-ities and related toxicities compared with CTLA-4blockade therapy. Research focused on identifying clin-ical biomarkers is necessary to elucidate the mechanismsunderlying the PD-1/PD-L1 mediated blockade and topredict patient outcomes following anti-PD-1/PD-L1therapies. Furthermore, it will also provide key insightsto develop combinatorial therapeutic strategies for fu-ture clinical trials.Studies to identify peripheral blood immune biomarkers

have illustrated that PD-1/PD-L1 blockade increases

effector T-cell proliferation (CD8+/HLA-DR+/Ki67+ Tcells), the production of inducible T-cell alpha chemo-attractant (ITAC), interferon-gamma (IFN-γ) and IL-18.However, in these studies there was no significant correl-ation between these markers and clinical response in pa-tients [28, 29]. Because of the unique expression patternand functional characterization of the PD-1/PD-L1 axismolecules, the local TME may be a key site for identifyingpredictive biomarkers for PD-1 pathway blockade. For ex-ample, the expression of PD-L1 in tumor-infiltrating im-mune cells analyzed by immunohistochemisty (IHC) isassociated with response to atezolizumab in patients withbladder cancer [29]. Additionally, elevated expression ofIFN-γ and IFN-γ-induced genes in pre-treatment tumorsis associated with clinical response in patients with melan-oma, although there was no such association reported inNSCLC or renal cell carcinoma [28]. Another study illus-trated that melanoma patients who had a higher numberof pre-treatment CD8+ T cells and TCR oligoclonality ex-perienced a better clinical response to pembrolizumab[103]. Furthermore, it has been shown that immune PD-1/PD-L1 blockade has the capacity to enhance and sustainendogenous immunity against mutated tumor neoanti-gens, thereby achieving durable tumor control. Proof ofprinciple studies in melanoma and NSCLC have shownthat high mutational load is associated with clinical re-sponse in patients treated with anti-PD-1 antibodies[62, 104]. A recent study also illustrated that tumormismatch repair status predicted the clinical benefitof immune checkpoint blockade with pembrolizumab[105]. Tumors with high mutational load are likelymore immunogenic, which can persistently stimulateneoantigen-specific CD4+ and CD8+ T cells. Thus,these results suggest that tumor mutational loadcould be potential predictive biomarker for PD-1/PD-L1 blockade therapy.PD-L1 is upregulated on many different tumor types

to inhibit the local antitumor T cell response. Because itis functional only through the ligation with its counter-receptor, the cell surface, or membranous, expression ofPD-L1 is essential for its biologic function. Two majorpotential mechanisms, known as adaptive and intrinsicresistance, can regulate the expression of PD-L1 ontumor cells [96, 102]. Adaptive resistance occurs whenthe tumor co-opts the natural physiology of the PD-1pathway to enable its own protection. For example, theexpression of PD-L1 is up-regulated on most epithelialcancers in response to various inflammatory mediatorssuch as cytokine IFN-γ, IL-4, IL-10, LPS, GM-CSF andVEGF. Intrinsic resistance refers to constitutive geneticalternations or the activation of signaling pathways (e.g.,PTEN loss, activation of MEK/ERK or MyD88 signalingor EGFR mutations) that drive the expression of PD-L1on tumor cells [106–109]. Although it has been



http://jitc.bmj.com

/J Im

munother C


nloaded from


described, the prognostic significance of PD-L1 expres-sion on tumor cells remains to be determined.Both prospective and retrospective analyses in large

Phase I, II and III trials in NSCLC and melanoma pa-tients have demonstrated the association betweentumor PD-L1 expression and response to anti-PD-1agents [17, 98]. In addition, a correlation between PD-L1expression in the TME and clinical responsive to PD-1blockade has also been shown [28, 29, 110]. Interestingly,PD-L1 expression on tumor-infiltrating immune cells wassignificantly associated with clinical response in NSCLCpatients treated with atezolizumab [28]. However, due tothe complexities of PD-L1 IHC, further studies are neededto carefully validate these observation and other TMEcharacteristics in either mono or combinatorial therapysettings [110]. In addition, due to the temporal changes inPD-L1 expression during the treatment, the immune pro-file and tumor signature need to be assessed at baseline.Moreover, measuring alterations in the TME during treat-ment will also be critical for future biomarker studies forPD-1/PD-L1 targeted immunotherapy. Despite these ca-veats, the current landscape for archival or pre-dosetumor PD-L1 as a predictive marker of efficacy for PD-L1/PD-1 targeted agents looks promising. Pembrolizumabwas the first checkpoint inhibitor to get approved with acompanion diagnostic test to measure tumor PD-L1 ex-pression in 2nd line + NSCLC. Patients with high tumorPD-L1 expression as defined by a proportional score of≥50 % (PS ≥50 %) demonstrated improved objective re-sponse rates of 45 % compared with 19 % in all enrolledNSCLC patients [111]. Moreover, although it is not re-quired for patient selection, nivolumab was approved fornon-squamous NSCLC with the first complementarydiagnostic test to measure tumor PD-L1 expression inorder to help identify patients with a greater likelihood ofimproved survival [112]. However, because patients withPD-L1 negative tumors had comparable activity to doce-taxel with a favorable tolerability profile and the overallintention-to-treat population in the trial was positive,nivolumab was approved for 2nd line NSCLC, regardlessof PD-L1 status [113]. PDL1 was not predictive ofoutcome for squamous NSCLC patients treated withnivolumab.Currently, there are a number of emerging biomarkers

for CTLA-4 and PD-1/PD-L1 immune checkpoint block-ade therapies. However, progress to identify new and val-idate current biomarker candidates has been limited bythe use of unstandardized assays that provide limited dataand variable results. Recent technological advances inhigh-throughput techniques will not only allow potentialbiomarkers to be validated across large prospective stud-ies, but will also facilitate the discovery of novelbiomarkers and enhance our understanding of the mecha-nisms underlying cancer immunotherapies. Therefore, the

remainder of this paper will focus on novel technologiesand highlight the potential impact of each technology oncurrent biomarker validation and future biomarkerdiscovery.

Whole exome sequencing for neoantigen discoveryCancer is a genetic disease. The accumulation of geneticmutations in a tumor leads to a change in its proteome.A “cancer anti-genome” generated during this processcan be recognized by T cells [114, 115]. Somatic muta-tions in cancer may give rise to mutated proteins thatare degraded into peptides (neoepitopes) presented inthe complex with major histocompatibility complex(MHC) molecules on the cell surface as neoantigens.There is a long standing interest in mutated antigens asdiscussed in a landmark review by Dr. Gilboa in 1999[116]. Heroic efforts were made by multiple groups toassess reactivity against such antigens using DNA libraryscreens. Although these studies illustrated proof ofprinciple, it was not practically feasible to assess thisclass of antigen in a systemic manner. Only a minorityof mutations are shared between patients; thus, the vastmajority of mutated antigens are patient-specific. There-fore, the assessment of neoantigens needs to be basedon the genome of individual tumors. The revolution innext-generation sequencing technology at affordablecosts along with the progress in bioinformatics has nowmade it feasible to describe the full mutation load (i.e.,the ‘genetic landscape’) of human tumors [117–120].Specifically, a comparison of the genomic sequence ofcancer tissue to that of non-transformed cells from thesame patient can be used to reveal the full range of gen-omic alterations within a tumor, including nucleotidesubstitutions, structural rearrangements and copy num-ber alterations [117].Several preclinical and clinical reports underscore the

importance of understanding the immunogenicity ofneoantigens and their potential application in cancer im-munotherapies. Two studies in mouse models providedthe first evidence that cancer exome based approaches canbe utilized to identify neoantigens recognized by CD8+ Tcells [121, 122]. Moreover, a recent study showed thattumor specific mutant antigens are important targets ofimmune checkpoint blockade therapy [123]. Subsequently,it has likewise been demonstrated that similar approachescan be utilized in the clinical setting to identify immuno-genic neoantigen specific CD8+ T cells in patients treatedwith tumor infiltrating lymphocyte (TIL) therapy andcheckpoint targeting therapies [124, 125]. Two humanstudies reported that neoantigens were recognized byintratumoral CD4+ T cells in patients with epithelial can-cer and melanoma [126, 127]. This accumulating evidencesuggests that the immune response to mutant neoepitopesplays a dominant role in tumor rejection. Due to the



http://jitc.bmj.com

/J Im

munother C


nloaded from


uniqueness of neoantigens, research into tumor immuno-genicity has shifted the interest from TAAs (differentiationantigen, cancer/testis antigen and overexpressed self-antigen) to patient-specific mutation antigens.The studies in both mouse models and human mater-

ial used exome sequencing, computer algorithm-guidedepitope prediction and the tandem minigene library ap-proach to identify MHC Class I- or II-binding neoepi-topes that were processed and presented by APCs andrecognized by neoantigen specific CD8+ and CD4+ Tcells. A tumor harbors hundreds of putative neoepitopesper the analysis of the current TCGA database. It is im-perative to differentiate and identify actual tumor pro-tective neoepitopes from the putative neoepitopesdefined in silico. There are two major factors that can besubject to variability when identifying tumor specificmutated antigens using these novel approaches. First,multiple computational tools to identify tumor specificmutations have been developed simultaneously. Differ-ent mutation calling tools such as EBcall, JointSNVMix,MuTect, SomaticSniper, Strelka and VarScan 2 havebeen developed to compare tumor samples with normaltissue samples at each variant locus in order to increasethe accuracy of somatic single nucleotide variant (sSNV)calling. These tools used to identify mutations have ahigh degree of overlap [128, 129]. As a next step to iden-tify neoepitopes, algorithms to predict binding affinity topatient specific HLA alleles can be used together withpredictions on proteasomal processing. The accuracy ofthe prediction algorithms mostly depends upon calculat-ing the score of binding to the MHC complex. Recentstudies showed that combined use of multiple tools gavea better prediction [130–132]; however, more work isneeded to accurately assess the immunoprotective prop-erties of mutation-derived neoepitopes. Second, it hasbeen demonstrated by unbiased screens that not all mu-tations result in neoantigens that are recognized by au-tologous T cells. Therefore, it would be valuable to haverobust pipelines to filter whole exome data, especiallyfor tumors with high mutation loads. Multiple groupshave made significant efforts to establish such pipelines.The filtering steps that have been applied are based onthe expression level of the mutations, e.g., RNA sequen-cing data, and the likelihood that a given mutated epi-tope will be processed by the proteasome and presentedby patient specific MHC molecules [123, 125, 131, 133].The two latter filtering steps can be assessed using algo-rithms that are already established to identify pathogen-derived epitopes. Currently, the data is still too sparse toknow which of these filters is most relevant and how toaccurately apply thresholds these filters to include im-munogenic and exclude non-immunogenic neoepitopes.However, the most significant improvement in these pre-dictions might be on the T cell side; the establishment of

algorithms that can identify the subset of epitopes thatare most likely to be recognized by TCR repertoire.The development of robust in vitro T cell culture proto-

cols, high-throughput combinatorial encoding of MHCmultimer flow staining and high-throughput TCR genecapture allows us to assess the frequency, phenotype andpolyfunctionality of the particular neoantigen specific T cellresponse [134–136]. These high-throughput technologiesfurther reduce the large number of potential neoepitopesto a small number of real immunogenic neoepitopes.Therefore, these technologies will help us reevaluate theaccuracy of computational tools as well as select candidateneoepitopes for vaccines and subsequently monitor theneoepitope specific T cell response during therapy. We willdiscuss the potential application of these high-throughputassays in the corresponding section. Overall, this approach,despite being in its early stages, has shown that the level ofmutation load as a potential biomarker can correlate withclinical outcome to checkpoint blockade therapy in pa-tients with advanced melanoma, colorectal cancer andNSCLC [62, 63, 104, 105, 137]. Patients with highly muta-genized tumors are most likely to respond to ipilimumabtreatment. However, some melanoma patients with lowmutation load have also experienced long-term clinicalbenefit. In addition, similar observations were reported inpatients with NSCLC treated with anti-PD-1 antibody[104].

Gene signature/patternTumor immune biology is a complex interplay of manyimmunosuppressive and immune stimulatory componentsinvolve in connected pathways that define the inflamma-tory state of the TME. Single molecule perturbation inquantity and quality can induce a ripple effect under agiven condition at a cellular and more importantly mo-lecular level. Treatment induced coordinated changes orthe natural course of the tumor can only be appreciatedwhen all key components are examined simultaneously asa whole temporally and longitudinally. Evaluation of thiscomplex interaction with respect to treatment outcomesin cancer have led to the identification of novel cell typesthat drive or contribute to the efficacy of or resistance totherapy and biomarkers that can predict clinical outcomeor drive mechanisms of rejection [138–140]. A holisticevaluation of immune intervention in tumors can beachieved by a system biology approach using gene expres-sion technology with high dimensional data analysis. Asthe technology become more affordable and reproduciblewith minimal input material, the type of sample that canbe used for gene expression analysis ranges from fine nee-dle aspirates, punch biopsies and needle sticks to lasercapture microdissected (LCM) samples and archived sam-ples with degraded RNAs. In addition, it makes the typesof tissue usually available from clinical trials, such as



http://jitc.bmj.com

/J Im

munother C


nloaded from


formalin fixed paraffin embedded (FFPE) tumors, applic-able. Combined with multiplexing IHC technologies, whichallow for the comprehensive interrogation of multiple celltypes as well as their location within the same tumor, geneexpression is frequently used as a reasonable surrogate toidentify the immune status of tumors [141–143].However, reproducible and reliable expression data

can only be achieved when the type of sample collection,timing of collection, sample processing and storage,standard laboratory procedures and platform selectionare carefully planned. Several critical factors need to beconsidered when using gene expression as a tool for im-mune profiling. Two such factors are described below,including the platform technology and definition of genesignatures that signify the various immune cell subsets.

Sample quality and quantityBecause they most accurately represent directly ex vivotissues, fresh samples should be the first choice for geneexpression analysis. Because of advances in sample col-lection tools, the collection of fresh samples is nowmuch less challenging. In order to monitor alterations inthe peripheral blood immune profile, PAXgene RNAtubes (PreAnalytix GMbH, Hombrechtikon, Switzerland)or other similar products are ideal due to the directpreservation of the sample without the need for process-ing. For large volume blood collection, peripheral bloodmononuclear cells (PBMC) should first be isolatedfollowed by direct lysis into an RNA isolation reagent,such as QIAzol or Buffer RLT (QIAGEN, Venlo,Netherlands), depending on the isolation kit used. Alter-natively, a cell pellet can be directly frozen in an RNAstabilization reagent. Ideally, tissue samples should becollected at bedside, immediately processed and sub-merged in an RNA stabilization reagent (e.g., RNAlater,AMBION, Inc.). Time is a critical factor in sample col-lection, and this common step often introduces bias.

The technologiesFor excisional biopsies, depending the tumor type andlesion, immune infiltrating cells are often quite diverseand comprise only a fraction of the total tumor mass.Consequently, RNA from immune infiltrating cells isgenerally poorly represented when enrichment methodssuch as LCM are not applied. Therefore, technologiesthat are capable of reliably detecting low abundancetranscripts are the most suitable for the immune tran-scriptional profiling of human tumors.RNA from FFPE tissues is often degraded and thus,

poses a challenge for gene expression analysis. Recentadvances in technology circumvent this challenge withthe development of methods specialized for degradedRNA analysis. Of the various technologies available, thethree most frequently used are digital PCR, single cell

real-time PCR (using the Fluidigm BioMark or Nano-string nCounter analysis systems or Cytoseq technology),or whole transcriptome RNA sequencing. Each has itslimitations and advantages [144–146]. The cDNA ampli-fication step makes PCR the most sensitive techniquelisted above for measuring gene expression in immunecells. However, multiplex PCR is a tedious process andmay consume significant amounts of RNA from precioustumor material. In contrast, digital PCR utilizes nano-droplet technology and makes multiplexing effortless.Alternatively, the Fluidigm BioMark system and Cytoseqtechnology utilize either a nanofluid approach or a com-binatorial library of beads bearing cell- and molecular-barcoding capture probes and makes semi high-throughputreal-time PCR possible and single cell profiling achievable.Although RNA or PCR amplification have been extensivelyused in molecular biology, amplification bias introducedduring the multiple steps and enzymatic reactions can stillaffect data reproducibility. Given the lack of an amplifica-tion step, the Nanostring nCounter platform is the closestto representing the true copy number of mRNA and per-forms well for detection of RNA derived from FFPE tissues[147]. However, the lack of amplification may affect thesensitivity of the platform in detecting key immune celltranscripts particularly from cells that are poorly repre-sented in tumors. The emerging front of high resolutionwhole transcriptome RNA sequencing is rapidly becomingthe platform of choice for RNA profiling due to the afford-able cost and in-depth resolution of data. It provides notonly transcript copy number information, but also poly-morphism information as well as transcript splicing variantinformation. The importance of splicing variants is becom-ing more and more appreciated and understood in terms offunctional diversity and in relation to pathophysiology.Consequently, RNA sequencing of tumor samples is in-creasingly being used to identify neoantigens presented byMHC Class I molecules [148].

The definition of gene signaturesThe power of gene expression platforms is in the abilityto analyze genes in multiple cell types within a single ex-periment and to identify intrinsic immunosuppressivemolecules and extrinsic inhibitory signatures, which maybe predictive biomarkers and the targets for future im-munotherapies. A simple strategy to evaluate multiplecell types is to incorporate lineage markers as represen-tatives of the individual cell types. For example, CD20and CD8 transcripts adequately represent B cell andCD8+ T cell densities, respectively. Cell types includingtumor-associated macrophages, Th2 cells, and Tregamong others are constantly changing both temporallyand in response to changes in the microenvironment.These characteristics may be best studied using a genesignature approach. However, the ever changing



http://jitc.bmj.com

/J Im

munother C


nloaded from


microenvironment and cell dynamics makes generatingreproducible gene signatures a moving target. Transcriptanalysis is a single snapshot of molecular activity at thepoint of sample collection and determined by multiplefactors including the host’s genetic makeup, somaticgenetic alterations, comorbidities, treatment procedures,protocol and time. Despite the evolving nature of theTME, common patterns of up or down regulated genesets have been identified and validated in independentstudies [140]. Gene analysis has evolved from pure genesignatures to identifying expression patterns based onpathway connections and molecules that are coordinatedand associated with specific cellular phenotypes. Genesthat are abundantly expressed in cells of interest tend tocluster together, thereby providing a surrogate readoutfor those cells. These gene signatures are often derivedfrom the expression analysis of distinct individual celltypes, representing the phylogeny of immune cells interms of their differentiation [149]. One of the chal-lenges of this approach in human tumors is the promis-cuous expression of these genes in multiple cell types.For example, markers like perforin and eomesodermin(eomes) may represent both activated T cells as well asNK cells. Therefore, the preferred analysis method is tointegrate data from multiple assays by correlating theresults from different technologies, such as complement-ing gene expression analysis with flow cytometry stain-ing and T and B cell receptor deep sequencing withmultiplex IHC. Thus, this integrated approach yields apowerful method to accurately evaluate the immuneprofile of human tumors. The standardization of genesignatures that represent distinct immune cell types maybe an important step in ensuring consistent interpret-ation of data from gene expression. Meta-analyses basedon similar diseases and treatment regimens using publicdatabases have been very fruitful in data validation andconfirmation. Although a greater understanding of can-cer biology and molecular immunology has beenachieved, developing biomarkers to use in clinical prac-tice would require further testing in large clinical studiesand a broader database available for public access. To doso, extensive cross validation is necessary not only at atechnological level, but also in interdisciplinary clinicaltrials.

Epigenetic-differentiation based measurement of immunecell and other cell frequencies in blood and tissue usingquantitative real-time PCR assisted cell countingEpigenomics investigates key functional componentsthat regulate gene expression in a cell, by providinginformation about patterns in which molecules suchas methyl groups label DNA and histones. Epige-nomic modifications provide a common set of in-structions to achieve a cell type specific identity,

despite sharing the same DNA sequence with allother cells in the body. Therefore, comprehensive epi-genomic analyses can provide the missing link be-tween genomic variation and cellular phenotype [150].Epigenomic organization with cell type specificity is amajor determinant of the cancer mutation landscape[151]. The National Institutes of Health (NIH) Road-map Epigenomics Consortium established global mapsof regulatory elements and defined regulatory mod-ules of coordinated activity together with their likelyactivators and repressors. Thus explaining how cell-specific programs of gene expression are achieved andtranscriptional and translational control is ensured.These data are a valuable resource for understandingthe relationships between cells and tissues and inter-preting the molecular basis of human disease [152].One of these epigenetic modifications, the methylation

status of either actively expressed or silenced genes, isthe basis of a novel cell identification and quantificationtechnology. Selective addition and removal of a methylgroup to the 5′-carbon of the cytosine base occurs ex-clusively in the dinucleotide cytosine phosphate guanine(CpG). DNA methylation is a non-random event andoften associated with inactive gene expression, if the tar-get CpGs are located in the proximity of coding regions.In contrast, demethylation of CpG in regulatory ele-ments is commonly accompanied by activation of geneexpression. Recent discovery of cell type specific epigen-etic CpG demethylation markers permits precise and ro-bust quantification of immune cells from only smallamounts of human blood or tissue samples.These epigenetic biomarkers located on genomic DNA

are stably associated with a cell type of interest. The cellquantification methodology is based on quantitativereal-time PCR (qPCR), targeting differentially demethy-lated CpG marker regions in the genomic DNA revealedafter a bisulfite conversion (BSC) step. During initialassay development, the cell type and subtype-specificepigenetic marker regions are identified through genomewide differential CpG demethylation analysis of highlypurified reference cell populations of interest. The re-gions are selected based on specific DNA sequences withdigitally differential BSC properties between differentcell types. During BSC, unmethylated cytosines are con-verted to uracil, while methylated cytosines do notchange. Therefore, respective CpG dinucleotides mustbe fully demethylated in the cell type of interest andmethylated in all other cell types. Resulting determin-ation of a cell type specific demethylation status in rele-vant loci is the basis for the development of segregatingprimer and probes. The readout technology is qPCR ofthe bisulfite converted DNA. Cell type specific qPCR as-says are designed so that only the demethylated DNA isamplified. This facilitates subsequent fast quantification



http://jitc.bmj.com

/J Im

munother C


nloaded from


of various leukocyte and other cell populations in agiven DNA containing sample by simple qPCR.Representative examples are the complete demethylation

of the Treg cell-specific demethylated region (TSDR) inTreg [153, 154], the demethylated region in the intergenicregion of CD3D/CD3G in T cells [155] and the demethyla-tion within the CCR6 locus in CCR6-positive cells [156].Epiontis has identified, characterized and validated variousepigenetic immune cell biomarkers, including those forTreg, Th17, Tfh, CD4+, CD8+ and CD3+ T cells, B cells,monocytes, NK cells and granulocytes. Results have anintra-assay coefficient of variation (CV) ≤15 % and inter-assay CV ≤20 %. Available assays cover the major leukocytetypes and can be evaluated in regulated, clinical studies re-quiring a total of 2 ml or less whole blood for all markerscombined. The amount of sample required for such studiesis being lowered in the near future by a factor of 20 toabout only 100 μl for all available assays together. Frozentissue material requires 250 μg to 1 mg tissue to measure12 markers, less for fewer markers. The exact materialneeded is less defined compared to blood because thenumber of cells and therefore DNA content in tissue byweight or volume varies significantly.Due to the inherent stability of DNA and its markers,

epigenetic assays have a distinctive advantage over assaysthat require intact or viable/functional cells in blood andtissue samples. This allows for a significantly broaderrange of acceptable sample conditions collected by clinicalsites. By simply freezing and shipping the collected wholeblood or tissue samples without any other additional steps,it allows for routine monitoring of patients and immune-monitoring during clinical trials, multicenter studies andretrospective studies as well as the comparison of resultsacross different studies [157, 158]. A common applicationof this standardized, epigenetic-based, immune diagnostictechnology is monitoring cell-mediated immunity duringimmune-modulatory clinical trials for cancer patients orinflammatory diseases. Because these tests can be appliedon both blood and tissue, standardized measurements anda comparison of circulating and tissue-infiltrating immunecells can be obtained as an alternative to flow cytometryfor peripheral blood samples and IHC for solid tissues. Fu-ture publication and extensive clinical studies are neededto validate the potential application of this novel technol-ogy for disease diagnosis and biomarker discovery for can-cer immunotherapy. A Phase II immune modulationstudy has shown utility of Treg and CD3 cell monitoringin peripheral blood of patients in the SELECT trial [159].

Protein microarray (seromics)Proteomics, analogous to genomics, is the large-scalestudy of proteins, such as their structure, interactionsand functions [160, 161]. Immunoproteomics is an ex-tension of the proteomics field that studies immune

related proteins and peptides. The release of tumor-derived proteins initiates an immune response that in-volves antigen-specific T and B lymphocyte targetingpeptides binding to self-MHC molecules and generatingspecific antibodies to corresponding proteins. An auto-antibody is an antibody that recognizes one or moreproteins from an individual’s own cells. Autoantibodiesin the peripheral blood are associated with autoimmunedisorders, infectious diseases and cancer. An effectivecancer immunotherapy would destroy tumor tissues,thereby releasing proteins and consequently priming Tand B cells against additional tumor antigens that werenot a part of the original therapy. This phenomenon isreferred to as antigen spreading, or also known as epi-tope or determinant spreading [162, 163]. Therefore, themagnitude and spectrum of autoantibodies and their in-tegration into the T cell response may be a feasible sur-rogate marker for measuring the adaptive immuneresponse to cancer and a potential promising clinicalbiomarker.Several immunoproteomics approaches, such as Sero-

logic Proteome Analysis (SERPA), Serological analysis ofrecombinant cDNA expression libraries (SEREX) and pro-tein microarrays, have been investigated to identify TAAsand their cognate antibodies [164, 165]. SERPA is a clas-sical immunoproteomic approach that provides a robustway of screening an antibody reactivity profile in sera frompatients with various diseases. SEREX was used to dis-cover tumor specific antigens that elicit a high titer im-munoglobulin G (IgG) antibody in sera from patients withdifferent types of cancer [166]. NY-ESO-1 was the firstcancer testis antigen discovered by SEREX technology[167]. However, the application of SERPA and SEREXtechnologies is limited due to the assay specificity and thecomplexity of the assay preparation and procedure.With the development of microarray techniques and

thousands of purified proteins immobilized on a solidsurface, protein microarrays have been employed toidentify proteins, detect various protein binding proper-ties, study protein posttranslational modifications anddefine potential biomarkers in a high-throughput man-ner. There are three major types of protein microarraysthat are classified based on their technology and applica-tion: analytical, reverse-phase and functional protein mi-croarrays [168]. The broader application of the first twotypes of protein microarrays is restricted by the specifi-city and availability of antibodies. Therefore, we willfocus on the functional protein microarrays, especiallycommercial protein microarrays, to describe the advan-tages, current application and drawbacks of this technol-ogy in basic research and clinical studies.Protein microarrays have several advantages compared

with other techniques, including a reduction in samplevolume used, high sensitivity and specificity and high-



http://jitc.bmj.com

/J Im

munother C


nloaded from


dimensional data generation. For example, Gnjatic et al.[169] reported that the sensitivity and specificity of a329 full-length protein microarray had a 94 % concord-ance with a standard ELISA. Another commercial pro-tein microarray, ProtoArray® (Life Technologies) offers aunique way to assay the serological response againstthousands of proteins (~9,000) simultaneously. ProtoAr-ray® does not have full coverage of the proteome, butserological analysis of this protein microarray representsa substantial portion of the human proteome (seromics).Integration of these types of high-dimensional data bet-ter representing the dynamic processes of the immuno-logic response associated with the development of thedisease, related toxicity and clinical outcome to cancerimmunotherapies. Specific autoantibodies have beenshown to correlate with the status and tumor progres-sion in patients with prostate, lung, ovarian and breastcancer [170–172]. In addition, humoral antigen spread-ing induced by Sipuleucel-T therapy was associated withimproved overall survival [173]. Moreover, CTLA-4blockade induced a broader antibody response in pros-tate cancer patients who responded to therapy comparedwith non-responders [56]. However, in another report, aprostate cancer patient who experienced a sustainedcomplete response to CTLA-4 blockade mounted a stronghumoral response against a small number of proteins, in-cluding one that is mutated in 5.5 % of prostate cancers[27]. Thus, further research using advanced methods willbe necessary to fully understand the role of autoantibodiesas a biomarker for cancer immunotherapy.Similar to DNA microarrays, proper serum collec-

tion, sample storage and careful standard lab proced-ure for protein microarray analyses are required toavoid the inter- and intra-assay variation and improvedata reproducibility. In addition, bioinformatics is crit-ical for handling and processing the large datasetsarising from these experiments. The analysis of pro-tein microarray data involves six steps: data acquisi-tion, pre-processing, visualization, differential analysis,result verification and computational feature annota-tion and network analysis [174]. Several softwarepackages and computational tools have been devel-oped for signal detection, data preprocessing, qualitycontrol and data normalization (see details in Table 1).A recent study proposed appropriate improvementson the default data analysis workflow [175]. Despitethe ability of protein microarrays to generate sub-stantial amounts of data for immune profiling in in-dividual patients, the limitations of this technology,such as large-scale protein and antibody production,lack of label-free detection systems and relativelyhigh cost, need to be overcome in the future. Inaddition, the target antibody/antigen identified fromprotein microarrays need to be validated by other

technologies such as Western blot, ELISA, Luminexassays or mass spectrometry.

Flow cytometry and mass cytometrySince its inception, flow cytometry has been a powerfultechnique for the field of immunology because of itsunique ability to analyze large numbers of single cellswith multiple parallel probes. This allows for the identi-fication as well as the deep phenotypic and functionalanalysis of rare subpopulations of cells. The lack ofstandardization in flow cytometry has traditionally hin-dered its application in multicenter clinical trials. How-ever, there have been recent efforts to recommend bestpractices for such multicenter studies [176]. Standard-ized panels have also been published for PBMC or wholeblood immunophenotyping [177, 178], and for leukemiaand lymphoma diagnoses [179]. Moreover, the numberof available fluorochromes for flow cytometry has stead-ily increased. Recent years have seen the advent of so-called “Brilliant” dyes, or π-conjugated polymers, whichare bright (due to cooperative energy transfer) and havetunable emission wavelengths [180]. These dyes havefostered a quantum leap in the ability to do multicolorflow cytometry, making 12–15 colors not only feasible,but routine.During the same timeframe, mass cytometry (or

CyTOF, for Cytometry by Time of Flight) has emergedas a competitive platform for high-dimensional single-cell analysis [181, 182]. This technology uses probes thatare labeled with heavy metal ions via covalently coupledchelation polymers, rather than fluorescent probes. Thesubsequent readout by mass spectrometry allows for thesimultaneous detection of many more unique probes,with little or no spillover between detector channels[183]. The current state of the art is about 40 parametersper cell, with both phenotypic and functional assays de-veloped [184–189].The main drawbacks of mass cytometry include slow

collection speed (about 300 events/s), low recovery ofcells in the instrument (typically 30 %), and expense. Tosome extent, these drawbacks are mitigated by the abilityto stain a single tube rather than create a panel of sev-eral tubes for conventional flow cytometry, which re-quires more cells, time, and reagents. While thesensitivity depends on the choice of label, instrumentsetup, and other factors, there are limitations to the sen-sitivity of CyTOF, as no channel can provide an equiva-lent resolution sensitivity to the brightest conventionalfluorophores. This, too, may be mitigated by the abilityto resolve populations in many more dimensions, but itremains a limitation at the single-marker level.One application of highly multiparameter cytometry,

especially mass cytometry, is for the broad analysis ofimmune competence in cancer patients. Given the



http://jitc.bmj.com

/J Im

munother C


nloaded from


Table 1 Summary of novel technologies

Technology Suggestions and potential biomarkers Sample preparation Bioinformatic tools References and recommended reading

Whole exomesequencing forneoantigendiscovery

• Mutation load for CTLA-4and PD-1 blockade therapy

• Neoantigen-specific T cell response

DNA from tumor and normalcells

EBcall, JointSNVMix, MuTect,SomaticSniper, Strelka, VarScan 2, BIMAS,RNAKPER SYFPEITHI, IDEB, NetMCHpan,TEPITOPEpan, PickPocket, Multipred2,MultiRTA

Van Buuren et al., 2014 [248]; Duan et al., 2014 [130];Snyder et al., 2014 [62]; Snyder et al., 2015 [131]; Rizviet al., 2015 [104]; Le et al., 2015 [105]; Van Allen et al.,2015 [63]

Gene signatureand pattern

• MAGE-A3 gene signature• Chemokine

expression in melanoma• Neoantigen signature

DNA and RNA from tumor,lymph node and PBMCs

BRB-ArrayTools, LIMMA, SAM, PAM, Partek,Genomic Suite, GSEA, Ingenuity IPA

Quackenbush et al., 2002 [231]; Simon et al., 2013[249]; Simon et al., 2007 [250]; Subramanian et al.,2005; Smyth et al., 2005; Tusher et al., 2001 [251];Tibshirani et al., 2002 [252]; Leek et al., 2010 [243];Gaujoux et al., 2013 [245]; Ulloa-Montoya et al., 2013[142]; Brown et al., 2014 [148]

Epigenetic-differentiationbased immunecellquantification

• Immune cell lineage specific epigeneticmodification

• Leukocyte ratios in blood and tissue

Genomic DNA from fresh orfrozen whole blood, PBMC,lymph node and fresh tissue orFFPE tissue and blood clots

HOMER package Motif Finder algorithmfindMotifGenome.pl, MatInspector(Genomatix), Mendelian randomization

Wieczorek et al., 2009 [154]; Sehouli et al., 2011 [155];Schildknecht et al., 2015 [253]; Steinfelder et al., 2011[156]; Lavin et al., 2014; Gosselin et al., 2014; Liang etal., 2015

Proteinmicroarray(seromics)

• TAA antibody response• Broad antibody signature• New antigen discovery

Fresh or frozen serum andplasma

Prospector, LIMMA package, PAA package,Spotfire package

Gnjatic et al., 2009 [254]; Kwek et al., 2012 [56];Turewicz et al., 2013 [175]; Graff et al., 2014 [27]

Flow Cytometryand MassCytometry

• Use best flow practices and recommendedflow panels

• Multimers for T cell epitope screening• TAA-specific T cell response for CTLA-4

blockade therapy• CD4+ICOS+ T cells for CTLA-4 blockade

therapy• Baseline MDSC for CTLA-4 blockade therapy

Whole blood; Fresh or frozenPBMCs and TILs; Fresh orfrozen cells from ascites orpleural effusion

Computational algorithm-driven analysisfor MDSC, Cytobank, FlowJo, SPADE, Phe-noGraph, PCA, viSNE, Citrus, ACCENSE, Iso-map, 3D visualization

Maecker et al., 2010 [176]; Maecker et al., 2012 [177];Streitz et al., 2013 [178]; Kvistborg et al., 2012 [255];Chang et al., 2014 [190]; Yuan et al., 2011 [57];Carthon et al., 2010 [50]; Kitano et al., 2014 [72];Levine et al.,2015 [189]

T and B cellreceptor deepsequencing

• CD3 T cell count• T Cell clonotype stability for CTLA-4 blockade

therapy• Baseline T cell clonality in tumor in PD-1

blockade therapy

DNA from FFPE; Frozen cellsfrom tumor, lymph node orPBMCs; Fresh or frozen cellsfrom ascites or pleural effusion

Shannon Entropy, Morisita’s distance,Estimated TCR gene rearrangements perdiploid genomes, Clonality, ImmuneID,Adaptive ImmunoSeq software

Cha et al., 2014 [205]; Tumeh et al., 2014 [103];Howie et al., 2015 [202]

Multicolor IHCstaining

• CD3 Immune score• CD8/FOXP3 ratio for tumor necrosis• PD-L1 expression on tumor in PD-1

blockade therapy

FFPE tissue; Fresh or frozentissue

TissueGnostic system, PerkinElmer system Galon et al., 2006 [10]; Hodi et al., 2008 [54];Taube et al., 2014 [110]

Abbreviations: PBMC peripheral blood mononuclear cells, TAA tumor associated antigen, MDSC myeloid derived suppressor cells, TILs tumor infiltrating lymphocytes, IHC immunohistochemical staining, TCR T cell receptor, FFPEformalin-fixed, paraffin-embedded, PD-1 programmed cell death-1, PD-L1 programmed cell death ligand −1

Yuanet

al.JournalforIm

munoTherapy

ofCancer

(2016) 4:3 Page

13of

25


http://jitc.bmj.com

/J Im

munother C


nloaded from


present rise in use of immunotherapy and its reliance onthe immune system to provide a response, it is surpris-ing that comprehensive and standardized measures ofimmune competence are not more frequently per-formed. By simultaneously probing the phenotypes andfunctions of multiple immune cell subsets in a singlePBMC sample, mass cytometry can provide a “finger-print” of immune responsiveness, which may eventuallyyield correlates of responsiveness to therapy [190]. Even-tually, this concept could be expanded from measure-ment of mitogen-stimulated responses to parallelmeasurement of specific T cell responses to multipletumor antigens. It could also be used to measure im-mune cell phenotypes and functions in tumor biopsies,which are likely to be more informative than PBMC forpredicting patient responsiveness.The delineation of T cells into distinct functional pop-

ulations defines the quality of immune response which iscrucial to disease outcome [191]. Multiple functional pa-rameters including cytokine, chemokine and degranula-tion in response to antigen specific stimulation can besimultaneously detected by flow cytometry. Polyfunc-tional T cell responses have been demonstrated to cor-relate with clinical response in patients with infectiousdisease and in cancer patients treated with immunother-apy [192–194]. Another attractive application of multi-parameter cytometry is the use of peptide-MHCmultimers to identify T cells of a given specificity or toscreen for multiple specificities. To date, combinatorialmultimer analyses have been performed using both flowand mass cytometry and allowed screening for 145 and109 T-cell specificities, respectively [135, 195]. The strat-egy was key to demonstrate that treatment with ipilimi-mab enhanced the priming of new T-cell responsesrather than boosting preexisting responses [61]. Simi-larly, the usage of MHC multimers is also likely to be-come a powerful tool for epitope discovery, allowinghigh-throughput identification, enumeration and profil-ing of neoantigen-specific T cells.

T and B cell receptor deep sequencingAdvances in high-throughput sequencing have enabledthe development of a powerful new technology for prob-ing the adaptive immune system called immunosequen-cing [196–199]. Millions of B or T cell receptor (BCR orTCR) sequences can be read in parallel from a singlesample. Each B or T cell clone has a unique (or nearlyunique) adaptive immune receptor generated through ahighly regulated process of somatic DNA rearrangement.When the BCR or TCR of a clone binds its target anti-gen as part of an immune response, the clone dividesrapidly, called clonal expansion. Unlike sequencing hu-man (or human cancer) genomes, immunosequencingmust be quantitatively accurate, because the adaptive

immune system works by the principle of clonalexpansion.Despite the new challenge of quantitatively sequencing

a highly variable, complex locus, the field has developedquite rapidly. A few techniques have been developed toaccomplish the quantitative sequencing of the loci andstandardize the methods, even between laboratories[200–202]. In the next year or two, the use of immuno-sequencing for diagnosis and monitoring of lymphoidmalignancies is expected to achieve FDA approval andCE mark [203, 204].There are many potential applications of immunose-

quencing in immunotherapy. Because each clone has anearly unique sequence, T cell clones can be trackedover time, between tissues and between phenotypic sub-sets. This technology is helping researchers understandthe mode of action and differences between therapeuticagents. For example, when comparing melanoma tumorsamples before and after anti-PD-1 therapy, the primaryT cell clonal expansions are from clones present prior totherapy [103]. This suggests that anti-PD-1 therapy pri-marily enhances and/or unblocks a pre-existing immuneresponse in the tumor as opposed to inducing a new re-sponse. Although, this was suggested as a mode of ac-tion, immunosequencing was required for solidevidence.A set of potential immunosequencing biomarkers in

immunotherapy are presently being explored. These in-clude predictors of response to therapy as well as moni-toring of pharmacodynamics changes, drug efficacy andside effects. The predictive biomarkers can be dividinginto two groups. The first group is measuring immuno-sequencing of tumor infiltrating lymphocytes (TIL). Arecent study has shown that both the number of TILand degree of specific clonal expansions (a telltale signof an adaptive immune response) in pre-treatment mel-anoma samples is predictive of response to anti-PD-1therapy [103]. There is ongoing work to confirm thefindings in larger cohorts and other tumor types. Add-itionally, similar biomarkers are being evaluated for dif-ferent immunotherapeutic agents [205]. Importantly,such biomarkers have the potential to help guide com-bination therapies or dose regimens [206]. If a particularTIL signature is needed for efficacy of anti-PD-1 ther-apy, then other therapies that can generate the TIL sig-nature would be likely combination candidates. Theadvantage of immunosequencing is that the activation ofinfiltrating killer T cells are thought to be the mode ofaction for checkpoint inhibitor therapy, so TILs are acausative biomarker, not just a correlative biomarker.Many of the practical issues associated with tumor sam-ples have been (or are presently being) addressed.Tumor heterogeneity (sampling) does not appear to be asignificant issue relating to TIL [207, 208]. With the



http://jitc.bmj.com

/J Im

munother C


nloaded from


proper controls, FFPE can be readily utilized in thisassay.Less direct, although potentially practical and broadly

applicable, is the second group of biomarkers, which areblood based. There is increasing evidence that the distri-bution of T cell clones in the blood is a correlate of im-mune competence. If a patient’s immune system is notfunctioning properly, an immunotherapy is unlikely tobe successful. In addition, unregulated responses, suchas autoimmune reactions are more likely [209].For other types of immunotherapy, such as adoptive T

cell transfer or chimeric antigen receptor T cell (CAR-T)therapy, immunosequencing is used to monitor the ther-apy itself by tracking the injected T cells. Moreover,when the target is a lymphoid malignancy, immunose-quencing is commonly utilized to monitor minimal re-sidual disease post treatment [210]. The breadth ofpotential immunosequencing biomarkers in immuno-therapy is very large. Immunosequencing is a molecular,reproducible tool for evaluating the adaptive immunesystem in humans, which has opened many avenues ofbiomarker research.

Multicolor IHC stainingThe detection of structural and functional proteins alongwith their spatial localization within cells or extracellularcompartments in tissue samples is achieved by immuno-labeling with specific antibodies. The antibody binding isthen detected with the application of either an enzymaticreaction that induces chromogen precipitation at the siteof antibody-antigen binding (immunoenzyme method)or by using fluorescent dyes (fluorochromes, fluorescentquantum dot nanocrystals), conjugated either to primaryor secondary antibodies (direct or indirect immunofluor-escence, respectively).For routine histopathological diagnosis, the immu-

noenzyme methods offer the advantage of permanentstaining by using panels of antibodies (1 reaction/sec-tion) specific for a tumor subtype, while optimal morph-ology is maintained. Rarely, double staining protocolsare applied to evaluate the ratio of two different celltypes or to evaluate pairs of antigens localized in differ-ent intracellular compartments, e.g., the cytoplasm ornucleus. However, in order to maximize the informationthat can be acquired from the intact tumor anatomy anddelineate any spatial and temporal (pre- versus post-treatment) heterogeneity that influence tumor biologyand could be used as a biomarker, multiplexed stainingapproaches and imaging systems are required. Practic-ally, multiplex approaches constitute repetitions of indi-vidual immunolabeling methods, either applied in onestep or in sequential rounds [211]. These methodologiessuffer from inherent problems that have to be overcome,such as cross-reactions between individual stains and

limitations regarding the color (chromogens or fluoro-chromes) combinations. Aside from using primary anti-bodies from different species, cross-reactions are preventedin immunoenzyme methods by heat-induced removal ofprimary antibodie(s) from the previous staining round,whereas the heat-stable chromogen remains and tags theantigen target. For these chromogenic methods, 3–4 colorsappear to be the multiplex limit.Alternatively, for multiplexed immunofluorescence one

can perform the tyramide signal amplification technique,in which fluorophore-labeled tyramide, upon activation byhorseradish peroxidase, covalently binds to tyrosine resi-dues of proteins adjacent to HRP-conjugated antibodies.Following the heat-induced removal of antibodies fromthe previous round, a new individual stain can be applied,with a minimal risk of cross-reaction. This technique isimportant for fluorescent multiplexing. After the comple-tion of the multiplexed staining procedure, up to 5 fluor-escent dyes can be evaluated using common fluorescentmicroscopes with standard optical filter cubes or up to 8fluorescent dyes can be evaluated using microscopesequipped with a multispectral camera [212].Other multiplexed methods that could circumvent the

limitations of overlapping emission spectral of fluores-cent microscopy are based on successive cycles of anti-body tagging, imaging and removal/bleaching of thefluorophore(s) [213, 214]. With dedicated software, thesehyperplexed approaches have been used to characterizemultiple analytes (>100) in the same section. Interest-ingly, recently described multiplexed methodologies havebeen performed with antibodies labeled with metal iso-topes and analyzed by mass cytometry-based approaches[215, 216].Overall, these in situ multiplexed methods add greater

depth to our understanding of tumor pathogenesis. Fur-thermore, when applying these methods to immunity-related analytes in the TME by combining phenotypicand functional markers with specific spatial point-pattern analyses, these methods will increase our know-ledge of underlying mechanisms that could be used tooptimize the efficacy of immunotherapy protocols.

Three-dimensional (3D) cell culture modelsTraditionally, cell-based assays to explore cell biologyand drug efficacy have been aimed at growing cellson two-dimensional (2D) plastic surfaces or in singlecell suspensions [217]. However, the cellular biologyis profoundly influenced by the microenvironment.Thus, cell based assays are needed that reflect the ef-fects of factors such as the extracellular matrix, cell-cell contacts, and cell-matrix interactions [218–221].Not only the cell morphology but also the drug sensi-tivity of cancer cells in 2D systems have been shownto be different compared with 3D cell cultures [6, 7].



http://jitc.bmj.com

/J Im

munother C


nloaded from


Cells cultivated on plastic surfaces usually exhibit anincreased sensitivity to cytotoxic drugs, while com-pounds targeting cell-cell adhesions, cell maturation,epithelial-mesenchymal transition and stemness oftenshow a decreased efficacy in 3D cell culture systems.Thus, 3D cell culture models reflect in vivo tumorgrowth more reliably and may provide better readoutsfor drug testing [222–224].The hanging drop technique is a well-established

cell culture method to form spherical microtissuesfrom immortalized and primary cell lines [225–227].In contrast to most liquid overlay technologies, the hang-ing drop method allows the precise control over the initialcell composition in each microtissue [228, 229]. Inaddition, to generate multicellular co-culture microtissues,neither additional supplements nor artificial scaffoldsmimicking extracellular matrix components (e.g., collagenmatrigel) are required. A 3D hanging drop system hasbeen established for cancer cell lines that are cultivated ei-ther alone or together with a lung fibroblast cell line to in-vestigate tumor stroma interaction [230]. In this model,IHC, flow cytometry, epifluorescence, confocal and scan-ning electron microscopy can be performed to assess al-terations not only in protein expression and viability butalso in microtissue aggregation. The addition of endothe-lial cells to these co-cultures of cancer cells and fibroblastscan further mimic an in vivo environment.Upregulation of mesenchymal markers and down-

regulation of adhesion molecules can be observed inmulticellular microtissues compared with 2D mono-cultures. Moreover, a difference in Ki67 expression in-dicates different states of high metabolic activity. Inaddition, changes in the morphology of tumor cellscan be achieved, particularly in microtissues culturedwith endothelial cells. Immune cells from freshly iso-lated PBMCs can be added to microtissues in the 3Dhanging drop technology. After short or long termco-incubation, the microtissues are “harvested” andimmune cell penetration into the spheroids is ana-lyzed. In this model, cytokine stimulation (e.g., IL-2)of immune cells leads to a significantly increased abil-ity of immune cell migration into the microtissue andinduction of cellular cytotoxicity.The 3D cell model represents a promising approach

for defining and analyzing immune-based biomarkers,as it provides an ex vivo approach to measure im-mune cell function in an organotypic culture model.In addition, this method is uniquely advantageous inthat primary tumor cells and autologous immune cellsfrom individual patients can be used in this model.Although there is a certain limitation to this systemas it does not fully model a whole organism, the 3Dculture system could be combined with studies inmouse models to overcome this disadvantage.

Bioinformatic tools and data analysis for high-throughputdataAs high-throughput technologies become more stan-dardized and widely available, bioinformatics tools toanalyze, interpret, and visualize data will be in high de-mand. Several tools and software packages have been de-veloped to mine data and extract useful informationfrom large datasets. A list of commonly used bioinfor-matics tools for different high-throughput technologiesis summarized in Table 1.There are many key steps involved in the analysis of

high-throughput data. For example, data normalizationis a critical step to reduce the intra- and inter-array vari-ability caused by systematic artifacts without losing use-ful biological information. Different normalizationapproaches have been extensively studied for data gener-ated from various high-throughput platforms, e.g., geneexpression arrays, miRNA arrays, protein microarrays,and RNA-sequencing experiments. For each high-throughput platform, comparisons have been made be-tween the different normalization methods, such as glo-bal normalization, Lowess normalization, quantilenormalization or conditional quantile normalization,variance stabilizing normalization, Z-Scorenormalization and robust linear model normalization[231–241]. For epigenomic data such as the epigeneticregulation of immune cells, data preprocessing andnormalization includes inverse normal transformation orZ-score

Documents

Novel technologies and emerging biomarkers for ... · Pia Kvistborg7, Cristina Maccalli8, Holden T. Maecker9, David B. Page10, Harlan Robins11, ... emerging biomarkers for immune