UBC Faculty Research and Publications

A pharmacogenetic signature of high response to Copaxone in late-phase clinical-trial cohorts of multiple… Ross, Colin J; Towfic, Fadi; Shankar, Jyoti; Laifenfeld, Daphna; Thoma, Mathis; Davis, Matthew; Weiner, Brian; Kusko, Rebecca; Zeskind, Ben; Knappertz, Volker; Grossman, Iris; Hayden, Michael R May 31, 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-13073_2017_Article_436.pdf [ 1.32MB ]
JSON: 52383-1.0361805.json
JSON-LD: 52383-1.0361805-ld.json
RDF/XML (Pretty): 52383-1.0361805-rdf.xml
RDF/JSON: 52383-1.0361805-rdf.json
Turtle: 52383-1.0361805-turtle.txt
N-Triples: 52383-1.0361805-rdf-ntriples.txt
Original Record: 52383-1.0361805-source.json
Full Text

Full Text

RESEARCH Open AccessA pharmacogenetic signature of highresponse to Copaxone in late-phaseclinical-trial cohorts of multiple sclerosisColin J. Ross1,2†, Fadi Towfic3†, Jyoti Shankar3† , Daphna Laifenfeld4, Mathis Thoma5, Matthew Davis5,Brian Weiner3, Rebecca Kusko3, Ben Zeskind3, Volker Knappertz5, Iris Grossman4* and Michael R. Hayden4AbstractBackground: Copaxone is an efficacious and safe therapy that has demonstrated clinical benefit for over twodecades in patients with relapsing forms of multiple sclerosis (MS). On an individual level, patients show variabilityin their response to Copaxone, with some achieving significantly higher response levels. The involvement of genes(e.g., HLA-DRB1*1501) with high inter-individual variability in Copaxone’s mechanism of action (MoA) suggests thepotential contribution of genetics to treatment response. This study aimed to identify genetic variants associatedwith Copaxone response in patient cohorts from late-phase clinical trials.Methods: Single nucleotide polymorphisms (SNPs) associated with high and low levels of response to Copaxonewere identified using genome-wide SNP data in a discovery cohort of 580 patients from two phase III clinical trialsof Copaxone. Multivariable Bayesian modeling on the resulting SNPs in an expanded discovery cohort with 1171patients identified a multi-SNP signature of Copaxone response. This signature was examined in 941 Copaxone-treated MS patients from seven independent late-phase trials of Copaxone and assessed for specificity to Copaxonein 310 Avonex-treated and 311 placebo-treated patients, also from late-phase trials.Results: A four-SNP signature consisting of rs80191572 (in UVRAG), rs28724893 (in HLA-DQB2), rs1789084 (in MBP),and rs139890339 (in ZAK(CDCA7)) was identified as significantly associated with Copaxone response. Copaxone-treated signature-positive patients had a greater reduction in annualized relapse rate (ARR) compared to signature-negative patients in both discovery and independent cohorts, an effect not observed in Avonex-treated patients.Additionally, signature-positive placebo-treated cohorts did not show a reduction in ARR, demonstrating thepredictive as opposed to prognostic nature of the signature. A 10% subset of patients, delineated by the signature,showed marked improvements across multiple clinical parameters, including ARR, MRI measures, and higherproportion with no evidence of disease activity (NEDA).Conclusions: This study is the largest pharmacogenetic study in MS reported to date. Gene regions underlying thefour-SNP signature have been linked with pathways associated with either Copaxone’s MoA or the pathophysiologyof MS. The pronounced association of the four-SNP signature with clinical improvements in a ~10% subset of theMS patient population demonstrates the complex interplay of immune mechanisms and the individualized natureof response to Copaxone.Keywords: Pharmacogenetics, Copaxone, Glatiramer acetate, Treatment-response, Inter-individual variability,Multiple sclerosis, Multi-SNP signature, Multivariable Bayesian modeling* Correspondence: Iris.Grossman@teva.co.il†Equal contributors4Teva Pharmaceutical Industries Ltd, Petach Tikva, IsraelFull list of author information is available at the end of the article© The Author(s). 2017 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.Ross et al. Genome Medicine  (2017) 9:50 DOI 10.1186/s13073-017-0436-yBackgroundMultiple sclerosis is a chronic progressive disorder ofthe central nervous system, with a complex pathogenesisand polygenic inheritance [1]. Recent genetic studies ofmultiple sclerosis have found several hundred commonlyoccurring non-coding polymorphic loci to be associatedwith susceptibility to the disease [2–6]. Genetic poly-morphisms within the human leukocyte antigen (HLA)region account for approximately 10% of the genetic riskof multiple sclerosis, with the HLA-DRB1*15:01 alleleassociated with a disproportionately elevated risk of de-veloping the disease [7, 8]. Genetic influences on diseaseprogression and severity remain an active area of re-search [9]. Fourteen disease-modifying therapies (DMTs)are currently approved for management of multiplesclerosis in the USA [10], benefiting patients by reducingrelapses, delaying disability progression and reducingcentral nervous system lesions. These therapies vary intheir mechanism of action (MoA), administration routes,and side effect profiles, with patients demonstrating sub-stantial variability in their responses to each drug [11].This variability, together with the plethora of treatmentoptions, underscores the need for predictive markers ofresponse to optimize treatment selection for individualpatients of multiple sclerosis [11].Copaxone® (glatiramer acetate) is a complex mixtureof numerous polypeptides, each giving rise to severalantigens that beneficially modulate the immune systemthrough mechanisms that have not yet been fully unrav-eled [12–14]. It has consistently demonstrated an annual-ized relapse rate (ARR) reduction of ~30% in Copaxone-treated patients compared to those treated with placebo inclinical trials. It continues to be an efficacious treatmentfor multiple sclerosis with a favorable safety profiledemonstrated over 20 years of clinical use and over twomillion patient-years of exposure [15]. Studies have shownthat a large proportion of Copaxone-treated patients(38 to 56%) demonstrate high response, based on varyingresponse definitions [11]. The involvement of geneswith high inter-individual variability in Copaxone’sMoA [16–18] together with past research findings[19] suggest that genetic determinants may contributeto variability in Copaxone-response [11, 20].To date, pharmacogenetic studies of Copaxone, ran-ging in size from tens to a few hundreds of patients(Additional file 1), have been based on candidate-genespresumed to be associated with its MoA, e.g., productionand activation of Copaxone-specific anti-inflammatoryand regulatory T-cells [16–19, 21]. The presence of vari-ants in the HLA class II genes has been observed to bepositively associated with Copaxone response. Examplesof such variants include the DRB1*1501 allele [16, 17] orthe homozygous presentation of a haplotype derived fromthe DR15 and DQ6 alleles along with absence of the DR17and DQ2 alleles [18]. In contrast, allelic combinations ofHLA DRB1*15, TGFB1*T, CCR5*d, and IFNAR1*G havebeen associated with non-response [22]. Alleles in othernon-HLA genes such as T-cell receptor beta (TRB),Cathepsin S (CTS), Myelin basic protein (MBP), Cluster ofdifferentiation 86 (CD86), Interleukin-1 receptor 1(IL1R1), and IL12RB2 have also been linked to Copaxoneresponse with varying strengths of association [19]. Whilethese candidate-gene studies have increased our under-standing of the pharmacogenetics of Copaxone responseand highlighted the potential importance of immune-response genes in Copaxone therapy, these findings havenot been replicated. Furthermore, a comprehensive andsimultaneous assessment of the contribution of multiplegene variants to Copaxone response has not beenperformed.The current study is the largest pharmacogenetic studyin multiple sclerosis reported thus far (Additional file 1),identifying and independently assessing a genetic signa-ture associated with Copaxone response in patient co-horts from a series of multinational late-phase clinicaltrials. The study design included signature identificationusing an initial exploratory association analysis ofgenome-wide SNP data informed by published research,Bayesian predictive modeling, independent assessmentof the signature for performance and specificity, and fi-nally, clinical characterization of patient subsets delin-eated by the signature.MethodsStudy designA four-stage study design was employed to identify amulti-SNP signature for Copaxone response (Fig. 1). Instage I, genome-wide SNP data were used to identifySNP-by-SNP associations with extreme phenotypes ofCopaxone response in 318 Copaxone-treated patientsfrom the GALA study [23]. Identified SNPs were exam-ined in 196 placebo-treated GALA patients to filter outprognostic markers and then screened for associationwith extreme phenotypes of Copaxone response in 262Copaxone-treated patients from the FORTE study [24].In stage II, multivariable Bayesian modeling wasemployed to identify a multi-SNP signature correlatedwith response from among the SNPs selected in stage I.A combined cohort of 1171 patients from the GALAand FORTE studies was employed for modeling. Thesignature was tested in 311 placebo-treated patientsfrom the GALA study to confirm its non-prognostic na-ture. In stage III, the identified multi-SNP signature wasassessed in seven independent late-phase trial cohorts,as well as in a cohort treated with Avonex (IFN-β) totest for specificity to Copaxone. In stage IV, patientsubsets defined by the multi-SNP signature wereRoss et al. Genome Medicine  (2017) 9:50 Page 2 of 15characterized to identify trends in clinical measures indi-cative of disease progression.Study populationsDiscovery cohortsThe discovery cohorts included patients with relapsing-remitting multiple sclerosis (RRMS) from two largephase III double-blind (DB) clinical trials of Copaxone(Fig. 1, Table 1).GALA DB [23]: The Glatiramer Acetate Low-frequencyAdministration study (ClinicalTrials.gov: NCT01067521)compared a three-times-a-week regimen of 40 mg/mLCopaxone with placebo. The duration of the DB phasewas 12 months. It was conducted at 142 sites in 17countries, including Bulgaria, Croatia, Czech Republic,Estonia, Georgia, Germany, Hungary, Israel, Italy,Lithuania, Poland, Romania, Russia, South Africa,Ukraine, United Kingdom and United States. In stage I,318 Copaxone-treated patients from the DB phase of theGALA study were analyzed (Fig. 1); 196 placebo-treatedpatients from the DB phase of the GALA study wereassessed to filter out prognostic markers. Subsequently,639 total patients from the study were genotyped instage II. Only the Copaxone arms were used in multi-SNP modeling.FORTE DB The FORTy mg Efficacy of glatiramer acet-ate study (ClinicalTrials.gov: NCT00337779) comparedonce-daily doses of 20 mg/mL to 40 mg/mL Copaxone[24]. The duration of the DB phase was 12 months. Itwas conducted at 136 sites in 20 countries includingArgentina, Belgium, Canada, Czech Republic, Estonia,Finland, France, Germany, Hungary, Israel, Italy, Latvia,Lithuania, Netherlands, Poland, Romania, Russia, Spain,the UK, and the US. The study concluded that the meannumber of relapses were equivalent at both doses [24].Both arms were thus pooled for analysis. In stage I, 262Copaxone-treated patients from the study were analyzed(Fig. 1). Subsequently, a total of 532 patients from thestudy were genotyped in stage II.Independent assessment cohortsThe independent assessment cohorts included patientsfrom an additional independent set of RRMS clinical tri-als and one study in clinically isolated syndrome (CIS).Data from these cohorts were utilized to assess themulti-SNP signature in stage III.GALA OL The open-label (OL) phase of the GALAstudy (ClinicalTrials.gov: NCT01067521) comprised 311patients who were on placebo treatment in the DB phaseand were switched to three times-a-week Copaxone40 mg/mL once the DB phase was completed (“delayedstart cohort”) [25]. The OL phase of the study is ongoingat the same sites and countries as the GALA DB phase(see “GALA DB” section above). The placebo-responsedata from these patients, prior to the switch to activetreatment, was used to filter out prognostic markers instage I and also confirm the non-prognostic nature ofthe multi-SNP signature in stage III.GA-9001 Thirty-eight Copaxone-treated patients fromthe placebo-controlled DB phase (ClinicalTrials.gov:NCT00004814) and 74 delayed-start Copaxone-treatedpatients from the OL phase (ClinicalTrials.gov:Fig. 1 Study design. The four stages of the study design are shown in sequence along with the sample sizes of each of the trial cohorts utilizedin the study. DB double-blind, OL open-label; RRMS relapsing-remitting multiple sclerosisRoss et al. Genome Medicine  (2017) 9:50 Page 3 of 15Table1DemographicsandbaselineclinicalcharacteristicsofthestudypopulationDiscoveryIndependentassessmentSpecificityassessmentGALADBFORTEDBGALAOLGA-9001DBGA-9001OLGA-9003DBGA-9003OLPreCISeDBPreCISeOLBRAVOAvonexarmTypeofmultiplesclerosisRRMSRRMSRRMSRRMSRRMSRRMSRRMSCISCISRRMSPhaseoftrialPhaseIIIPhaseIIIPhaseIVPhaseIIIPhaseIVPhaseIIIPhaseIVPhaseIIIPhaseIVPhaseIIINationalityMultinationalMultinationalMultinationalUSUSMultinationalMultinationalMultinationalMultinationalMultinationalNumberofpatients63953233338744084132240310Durationoffollow-up1year1year~3yearsa~3years~20yearsa0.75years0.75years3years5years2yearsAge(mean±SD)37.59±9.3336.19±8.7638.5±9.2535.89±5.4036.35±5.9233.33±7.7533.46±7.6531.71±7.2532.08±7.1538.17±9.28Sex(percentagefemale)69.01%72.37%67.57%68.42%71.62%70.00%76.19%63.64%63.75%68.06%Caucasian(%)97.81%100.00%99.70%92.11%90.54%97.50%97.62%96.21%97.50%98.71%BaselineEDSS2.79±1.222.14±1.112.86±1.322.79±1.342.66±1.622.14±0.992.31±1.320.99±0.821.27±1.082.62±1.15BaselineARR0.93±0.450.98±0.440.88±0.501.49±0.561.05±0.781.21±0.781.25±0.78NANA0.94±0.45Copaxonedoses:GALADB,40mg/mLthrice-a-week;ForteDB,both20mg/mLand40mg/mLadayarmswereincluded;GALAOL,40mg/mLaday;GA-9001,GA-9003,andPreCISe,20mg/mLaday.a Thefollow-upisongoingandthevaluesrepresentthetime-pointatwhichthedataweresummarizedforthisstudy.Avonexinterferonβ-1a,BaselineARRindividualARRfortwoyearspriortostudy,CISclinicallyisolatedsyndrome,EDSSKurtzkeexpandeddisabilitystatusscale,OLopen-label,RRMSrelapsing-remittingmultiplesclerosis.Onlypatientswhogavetheirinformedconsenttobeinggenotypedwereincludedinthestudy.GenotypedpatientswererepresentativeofthestudypopulationintheparenttrialRoss et al. Genome Medicine  (2017) 9:50 Page 4 of 15NCT00203021) were included [26]. The duration of theDB phase of the study was 35 months (initially24 months, and later extended 11 additional months)[27]. It was conducted in 11 sites in the US. The OLphase is ongoing at the same sites as the DB phase.GA-9003 For the GA-9003 cohort [28] eighty-fourCopaxone-treated patients from the placebo-controlledDB phase of the European-Canadian Copaxone trialand 40 delayed-start Copaxone-treated patients from theOL phase [29] of the study were included. The durationof the DB and OL phases of the study was 9 monthseach, and both were conducted at 29 sites in sevencountries including Belgium, Canada, France, Germany,Italy, Netherlands, and the UK.PreCISe The early glatiramer acetate in patients Pre-senting with a Clinically Isolated Syndrome study(ClinicalTrials.gov: NCT00666224) [30] demonstratedthe efficacy of Copaxone in delaying the progression ofunifocal CIS to clinically defined RRMS. The durationof the DB phase was 36 months and that of the OLphase 60 months. Both were conducted at 80 sites in16 countries including Argentina, Australia, Austria,Denmark, Finland, France, Germany, Hungary, Italy,New Zealand, Norway, Romania, Spain, Sweden, theUK, and the US.The study cohort comprised 132 Copaxone-treated pa-tients from the placebo-controlled DB phase and 240delayed-start Copaxone-treated patients from the OLphase [31].BRAVO The Laquinimod DB placebo controlled studywith a rater-Blinded Reference Arm of interferon β-1a(AVOnex®) (ClinicalTrials.gov: NCT00605215) [32] com-pared the effect of Laquinimod with that of Avonex forRRMS patients. The duration of the DB phase was24 months. It was conducted in 155 sites in 18 countriesincluding Bulgaria, Croatia, Czech Republic, Estonia,Georgia, Germany, Israel, Italy, Lithuania, Macedonia,Poland, Romania, Russia, Slovakia, South Africa, Spain,Ukraine, and the US. The Copaxone specificity of themulti-SNP signature was evaluated using response datafrom 310 patients from the Avonex arm of the trial. Inclu-sion criteria and assessment frequency were similar to theother Copaxone studies.Baseline demographics as well as clinical character-istics of the patient cohorts included in this study(Table 1) are representative of parent trial populationsand are within one standard deviation for continuousmeasures and had similar percentages for categoricalmeasures [23–32].Response definitionsExtreme phenotypes of Copaxone responseFor each patient, ARR-reduction was calculated as thedifference between ARR during the study and ARR forthe two years prior to the study. For patients with fewerthan two years of recorded clinical history, time sincethe first symptom (in years) was used when calculatingthe pre-study ARR. To determine the extreme pheno-types of Copaxone response, the distribution of ARRreduction was examined in the discovery cohorts andcut-off thresholds were selected to define high and lowresponse categories (Fig. 1). An ARR reduction >1 wasdefined as high and patients meeting this definition wereclassified as high responders. The highest responders werepatients with high ARR reduction and, additionally, nonew T2-weighted brain MRI lesions (T2 lesions). Pa-tients with an ARR reduction of ≤0 (i.e., no ARR reduc-tion or a worsening of ARR) were classified as lowresponders. The lowest responders were patients with lowARR reduction and, additionally, one or more new T2lesions. Patients with an ARR reduction between 0 and 1inclusive were considered intermediate responders andnot genotyped in stage I.Relapse-free definitionSubsequent to identification of genetic variants corre-lated with extreme phenotypes of Copaxone response,predictive models were built on the most clinically usefulresponse definition as indicated by treating physicians,i.e., being relapse-free. Only patients with at least one re-lapse at baseline or a baseline ARR ranging from 0.5 to1.0, given one to two years of available clinical history,were included in the parent trials. Therefore, beingrelapse-free during the trial was assumed to be a treat-ment effect. A patient was considered relapse-free or aresponder if he or she did not experience any relapseswithin one year of starting treatment. As a result, therelapse-free definition captured all patients with an ARRreduction between 0.5 and 1 inclusive as well as thoseclassified as intermediate responders (see section above).Ninety-seven patients experiencing a relapse within thefirst 47 days after starting therapy were excluded becauseDMTs, and specifically Copaxone, do not reach full effi-cacy until after this period [33]. Therefore, relapses inthe first 47 days were not considered as failures of drugresponse. Sensitivity analysis indicated that results werenot affected by this exclusion.Genotyping and quality controlThe Illumina OMNI-5M genome-wide array covering4,301,331 SNPs was utilized for genotyping the patientswith extreme phenotypes of Copaxone response in stageI. Genotypes were called with the Illumina Genome Stu-dio software and their quality checked with evaluationsRoss et al. Genome Medicine  (2017) 9:50 Page 5 of 15for call rate, cluster separation, mean normalized inten-sity, proximity of heterozygote clusters to a homozygotecluster, heterozygous excess, false homozygosity, andreproducibility-related errors. SNPs with call rates of≥95% were retained and those with <95% were either re-clustered or removed. Deviation of genotype distribu-tions from Hardy–Weinberg equilibrium (HWE) wastested in placebo arms of the discovery cohorts. SNPswith a p value <1.0 × 10−4 for Fisher’s exact test forHWE were excluded. A total of 4,296,423 SNPs wereretained after the quality control steps outlined above.SNPs identified at the end of stage I were genotypedusing TaqMan SNP genotyping assays in 1171 patientsfrom GALA DB and FORTE DB cohorts, and in an add-itional 941 patients from GALA OL, GA-9001, GA-9003,PreCISe, and BRAVO cohorts. Cluster plots for theseSNPs were visually inspected. A small number of samples(n = 16) underwent confirmatory Sanger sequencing forall SNPs. Within these samples, there was 100% con-cordance between the genotypes called by the OMNI-5M genome-wide array, the TaqMan assay, and Sangersequencing, regardless of minor allele frequency.Statistical methodsAssociation analysisStage I was an initial exploratory analysis to identify can-didate SNPs for follow-up in later stages. For each SNP,a logistic regression model was built using response vari-ables based on the extreme phenotypes of Copaxone re-sponse. Models were estimated using SVS software,version 8.3.0. A four-step analysis (Table 2) incorporateda priori evidence in the SNP selection process. Step 1employed regression models to select SNPs from a set of35 candidate variants supported by prior literature (Add-itional file 2). Subsequent steps expanded the set ofSNPs tested in a non-overlapping manner, ending in abroad genome-wide analysis (Table 2). Since the purposeof this stage was hypothesis generation, lenient thresh-olds were adopted at each step to capture SNPs basedon both strong biological plausibility and pre-existing lit-erature evidence. Candidate variants and genes analyzedin steps 1 and 2 are listed in Additional file 2.SNP encoding and inheritance modelsThe inheritance model for each SNP was determinedusing PLINK [34]. Each SNP was coded either as a con-tinuous covariate with values 0, 1, and 2, denoting anadditive inheritance model, or as a binary variable withtwo levels, 0 and 1, denoting a dominant inheritancemodel that specified whether or not a patient had twocopies of the minor allele. The frequencies for eachgenotype of the four SNPs in the four-SNP model areshown in Additional file 3.Table 2 Association analysis of genome-wide SNP data in patients with extreme-phenotypes of Copaxone-responseAnalysis steps and inclusion thresholds Selected SNPs Copaxone-treated patientsGALA DB FORTE DBGene SNP rsID Odds ratio P value Odds ratio P valueStep 1. Replicated variants from 35 prioritizedcandidate variants. Inclusion threshold: p value<0.05 GALA, p value <0.05 FORTEHLA-DRB1*1501 rs3135391 0.66 0.040 0.64 0.0499Step 2. Priority list of 4012 variants in 30 genes.Inclusion threshold: p value <0.05 GALA, p value<0.05 FORTEHLA-DQB2/DOB rs28724893 0.53 0.00060 0.46 0.00037HLA-DOB/TAP2 rs1894408 1.72 0.0030 1.82 0.0093MBP rs1789084 0.70 0.036 0.57 0.01Step 3. Broad genome-wide analysis. Inclusionthreshold: p value <0.01 GALA, p value <0.05 FORTEPTPRT rs117602254 0.21 0.0037 0.28 0.016ALOX5AP rs10162089 1.56 0.0078 1.58 0.032MAGI2 rs16886004 2.15 0.0023 5.56 3.3E-05ZAK(CDCA7) rs139890339 0.05 3.4E-05 0.14 0.011SLC5A4(RFPL3) rs73166319 * 0.0060 * 0.015Step 4. Secondary genome-wide screen in patientswith highest Copaxone response (relapse-free withno new T2 lesions). Inclusion threshold: p value<0.01 GALA, p value <0.05 FORTEUVRAG rs80191572 0.20 0.0024 0.12 3.4E-05SLC1A4 rs759458 3.31 4.4E-05 1.86 0.049The 35 prioritized candidate variants and the 30 genes analyzed in steps 1 and 2, respectively, are presented in Additional file 2. SNPs selected at each analysisstep met the indicated threshold of significance in the SNP-by-SNP logistic regression models built separately in the GALA DB and the FORTE DB cohorts. Thesemodels estimated the odds ratios of high versus low response. The SNPs that were selected at each step were not associated with the extreme phenotype ofresponse in patients treated with placebo. *Odds ratios were not informative since the rare allelic variant of SLC5A4(RFPL3) was only present in high responders ofCopaxone treatment and not in low responders. DB double-blind phase, MAF minimum allelic frequencyRoss et al. Genome Medicine  (2017) 9:50 Page 6 of 15Predictive modelingIn stage II, logistic regression models were built, bothwith and without disease-related baseline covariates, todetermine which of the SNPs identified in stage I weremost predictive of relapse-free status. Covariates con-sisted of the baseline Kurtzke Expanded Disability StatusScale (EDSS) [34], Log (number of relapses for past twoyears + 1), baseline T2 lesion volume, and gadolinium-enhancing T1-weighted MRI lesion (T1 lesion) status (0for no lesion, 1 for at least one lesion at baseline). Thelogistic regression models were estimated using Bayesianmodel averaging (BMA) [35–39] with a spike-and-slabprior distribution [40], as implemented in the BoomSpi-keSlab R package [41]. The sparsity-inducing spike-and-slab prior in this model embodied the expectation thatnot all identified SNPs from stage I were important forpredicting relapse-free status. Model convergence diag-nostics for BMA were inspected to determine that50,000 iterations of the Markov chain Monte Carlo(MCMC) sampler were sufficient to explore the space ofall possible SNP combinations and, hence, estimate thefinal BMA model [40–42]. For each SNP, BMA providedthe posterior probability of inclusion in the model andthe 95% Bayesian confidence interval (CI) of the poster-ior distribution of its regression coefficient. SNPs weredeemed statistically significant if the 95% CI of their re-gression coefficients did not include zero, both with andwithout the inclusion of disease-related baseline covari-ates in the model. The strength of evidence for eachSNP’s effect was assessed quantitatively by computing itsposterior probability of inclusion in the model and quali-tatively by examining the width of the 95% CI of its re-gression coefficient. Potential interactions were assessedbetween those SNPs whose main effects were significant.The final model was obtained by refitting the statisticallysignificant SNPs without baseline covariates. The model-building process and the rationale behind the choiceof the Bayesian framework are described in detail inAdditional file 4.Classification performanceStage III evaluated the classification performance ofthe multi-SNP signature resulting from stage II ana-lyses in each of the independent cohorts using sensi-tivity, specificity, and the area under the receiveroperating characteristic (ROC) curve (AUC). To clas-sify patients as either relapse-free or relapsing, an op-timal threshold on the predicted probabilities fromthe multi-SNP logistic regression model was deter-mined. This threshold maximizes the sensitivity andspecificity of the signature and corresponds to thepoint on the ROC curve closest to the top left corner(“top-left" threshold). Signature-positive patients werethose who either met or exceeded the predicted prob-ability that corresponded to the “top-left” threshold inthe multi-SNP model.Clinical characterizationIn stage IV, patients in the discovery cohorts were di-vided into five similar-sized bins (Fig. 2) based on quin-tiles of the predicted probabilities from the multi-SNPmodel. The distribution of key clinical measures wasassessed across these bins. For each bin, descriptivesummaries, including the mean and standard deviationfor continuous variables and the percentage of patientsin the category of interest for categorical variables, werecalculated for baseline measures (EDSS score, number ofT1 lesions, T2 lesion volume, ARR) and for on-treatment measures (on-trial number of T1 lesions,change in volume of T2 lesions, on-trial ARR, change inARR, time to first relapse (in days), percentage of pa-tients who were classified as non-relapsing (responders)and percentage of patients meeting the two definitionsof no evidence of disease activity (NEDA3 and NEDA4))[43], computed at 12 months after initiation of treat-ment. NEDA3 consisted of three criteria: (a) no relapse;(b) no confirmed disease progression defined as a 1-point increase of EDSS from baseline for patients withbaseline EDSS between 0 and 5, or a 0.5 increase for pa-tients with baseline EDSS higher than 5, confirmed3 months later; and (c) no T1-weighted gadolinium-enhancing lesions or new or enlarging T2 lesions mea-sured by MRI during the study. NEDA4 additionallyincluded brain volume loss of ≥0.4% as a criterion. Clin-ical characteristics of signature-positive and signature-negative patients were also compared.ResultsEleven SNPs were associated with extreme phenotypes ofCopaxone responseEleven SNPs were associated with high versus low re-sponse to Copaxone based on the initial exploratoryanalysis of genome-wide SNP data from the patients inboth the GALA DB and FORTE DB discovery cohorts(“Methods”, Table 2). These SNPs were not associatedwith response in the GALA DB placebo arm.Briefly, in step 1 of the analysis, out of 35 candidatevariants tested, two SNPs in complete linkage disequilib-rium (LD) (rs3135391, rs3135388) tagging the HLA-DRB1*15:01 allele met the threshold for selection inboth GALA DB and FORTE DB. rs3135391 was selectedfor all subsequent analyses. In step 2, out of 4012 vari-ants in 30 candidate genes, 36 variants were selected inboth discovery cohorts. Of these variants, three SNPsnot in LD with each other and located in the HLA-DQB2/DOB, HLA-DOB/TAP2, and MBP gene regions,respectively, were selected. In step 3, a broad genome-Ross et al. Genome Medicine  (2017) 9:50 Page 7 of 15wide analysis identified five SNPs located in the PTPRT,ALOX5AP, MAGI2, ZAK/CDCA7, and RFPL3/SLC5A4gene regions, respectively. In step 4, a broad genome-wide analysis limited to the patients with the highest re-sponse (defined as relapse-free with no new T2 lesions)identified two SNPs in the UVRAG and SLC1A4 gene re-gions, respectively.A four-SNP signature was associated with the binaryrelapse-free response definitionFrom among the 11 SNPs associated with extreme phe-notypes of Copaxone response (stage I), Bayesian pre-dictive modeling in 1171 patients comprising thebroader discovery cohort identified a subset of fourSNPs (Table 3) that distinguished relapse-free patients(responders) from relapsing patients (non-responders).Each of these four SNPs attained a >80% posteriorprobability of inclusion in the model with statisticallysignificant effects (i.e., the 95% CI of the posterior distri-bution of their regression coefficients did not includezero). In contrast, the remaining seven SNPs which werenot selected had posterior inclusion probabilities smallerthan 60% and non-significant effects. Additional file 5presents the regression coefficients and the posterior in-clusion probabilities of the 11 SNPs. Interactions be-tween the top four SNPs had posterior inclusionprobabilities of <1% and were thus not included in thefinal model.Signature-positive patients showed better clinicalcharacteristics than signature-negative patientsEach of the 1171 patients in the discovery cohort wasclassified as either relapse-free (signature-positive) or re-lapsing (signature-negative) by applying the “top-left”ab[0.26, 0.81] (0.81, 0.83] (0.83, 0.89] (0.89, 0.9] (0.9, 0.94]Bin 1 Bin 2 Bin 3 Bin 4 Bin 5Fig. 2 Clinical characterization of patients in the discovery cohort. a Proportion of relapsing and non-relapsing patients across bins in the discoverycohort. b Clinical characterization of patients within each bin in the discovery cohorts. Panels a and b show descriptive summaries of clinicalcharacteristics relevant to disease progression across five patient bins. These bins were constructed using the logistic regression modelwhich predicted the probability of being relapse-free conditional on the four SNPs. In a, each tick on the x-axis corresponds to a bin basedon a quintile of the predicted probability from the logistic regression model and is labeled with the lowest and the highest predictedprobability for the bin. As we move from left to right along the x-axis, the predicted probability of being relapse-free (or being a responder) increases.In each bar, the observed percentages of non-responders and responders are shown using two colors. For a good model, the predicted probabilitiesshould be close to the observed percentages. The figure confirms that this is indeed the case for the logistic regression model. The bars in the graphin a and the columns of the table in b are lined up to show the one-to-one correspondence between the graph and the table. Panel b illustrates thatthe trends of several alternative clinical response definitions which were not used to construct the four-SNP model align well with the predictedprobabilities from the four-SNP model. This suggests that the predictive value of the four-SNP genotype extends beyond the clinical responsedefinitions used to build it. T1 lesions are gadolinium-enhancing T1-weighted lesions on MRI; T2 lesions are T2-weighted MRI lesions. ARRannualized relapse rate, NEDA3 no evidence of disease activity (version 3), NEDA4 no evidence of disease activity (version 4). Percentages ofpatients meeting the NEDA3 and NEDA4 definitions are shown. The discovery cohorts consisted of the patients from GALA DB and FORTE DBRoss et al. Genome Medicine  (2017) 9:50 Page 8 of 15threshold (see “Methods”) on the predicted probabilitycalculated by the four-SNP model. When compared withsignature-negative patients, signature-positive patientshad a 54–64% reduction in mean ARR (Table 4). Seven-teen and 40% of signature-positive patients were able tomaintain NEDA4 and NEDA3 status, respectively, for upto 12 months on Copaxone treatment. In contrast, alower percentage, 12 and 32%, of the signature-negativepatients were able to maintain NEDA4 and NEDA3 sta-tus on Copaxone treatment, respectively. Signature-positive patients had, on average, a longer time to firstrelapse (mean = 344.3 days, standard deviation (SD) =57.06 days) when compared with signature-negative pa-tients (mean = 316.9 days, SD = 92.10 days). In addition,signature-positive patients had fewer T1 lesions (mean =1.63, SD = 4.25) and a lower volume of T2 lesions (mean= 14.18, SD = 15.92) at baseline when compared withsignature-negative patients who had a higher number ofT1 lesions (mean = 2.21, SD = 6.04) and a higher volumeof T2 lesions (mean = 16.26, SD = 17.52) at baseline.EDSS was similar between signature-positive andsignature-negative patients (mean = 2.48, SD = 1.21 andmean = 2.51, SD = 1.22, respectively). Overall, signature-positive patients showed consistent and clinically mean-ingful improvements over signature-negative patientsacross multiple clinical measurements that were notemployed in developing the signature.Independent assessment of the four-SNP signatureTable 5 shows the classification performance of the four-SNP signature in the discovery cohorts (AUC = 0.66) andin the independent assessment cohorts (AUC = 0.45 to0.65). ARR reductions in signature-positive patients fromthe RRMS independent cohorts ranged from 13 to 53%(Table 4). CIS cohorts were inconclusive, with thesignature-positive DB cohort showing a 5% lower ARRand the signature-positive OL cohort showing a 14%higher ARR compared to signature-negative patients.Table 4 Summary of ARR change based on predicted responseCohort Type of MS Number of patients Follow-upduration (years)Mean ARR change:Sig + versus Sig−Total Sig+ Sig−DiscoveryGALA DB RRMS 639 323 316 1 −54%FORTE DB 532 268 264 1 −64%Independent assessmentGALA OL RRMS 333 190 143 ~3 −14%GA-9001 DB 38 21 17 ~3 −13%GA-9001 OL 74 35 39 ~20 −22%GA-9003 DB 40 21 19 0.75 −53%GA-9003 OL 84 41 43 0.75 −49%PreCISe DB CIS 132 69 63 3 −5%PreCISe OL 240 129 111 5 +14%Specificity assessmentBRAVO – Avonex RRMS 310 176 134 2 +10%Sig + (signature-positive) and sig − (signature-negative) indicate patients classified as relapse-free and relapsing, respectively, after applying the “top-left” thresholdon the predicted probabilities from the four-SNP logistic regression model (see “Methods”). Mean ARR was calculated by dividing the total number of relapses inSig + (or Sig−) patients by the total sum of exposure to Copaxone (in years). The difference between mean ARR of Sig + and Sig − patients is presented in thelast column. ARR annualized relapse rate, CIS clinically isolated syndrome, DB double-blind phase, OL open-label phase, RRMS relapsing-remitting multiple sclerosisTable 3 The four-SNP model coefficients and odds ratios with 95% Bayesian confidence intervalsSNP rsID Gene Regression coefficient (95% CI) Odds Ratio (95% CI)rs80191572 UVRAG −0.68 (−1.06, −0.29) 0.50 (0.35, 0.75)rs28724893 HLA-DQB2 −0.52 (−0.75, −0.29) 0.59 (0.47, 0.75)rs1789084 MBP −0.61 (−0.98, −0.25) 0.54 (0.38, 0.78)rs139890339 ZAK(CDCA7) −1.46 (−2.31, −0.63) 0.23 (0.10, 0.53)Coefficients of the four-SNP model were obtained by fitting a logistic regression model on data from the patients treated with Copaxone in the GALA DB andFORTE DB studies. The SNP from MBP was coded according to a dominant inheritance model. All the other SNPs were coded according to an additive inheritancemodel. The logistic regression model estimated the log odds of being relapse-free conditional on the four SNPs. A negative regression coefficient for a given SNPimplies that the major allele (coded as the reference level in the logistic regression model) was associated with increased odds of being relapse-free while theminor allele was associated with increased odds of relapsesRoss et al. Genome Medicine  (2017) 9:50 Page 9 of 15Additional file 6 plots the mean ARR change (signature-positive versus signature-negative) against the samplesize of the cohort. The plot shows that the sample sizeof the cohorts is not a determinant of the mean ARRchange. Additional file 7 summarizes the performancemetrics of all of the three-SNP subsets. None of thesemodels outperformed the four-SNP model. Additionalfile 8 shows the results of pairwise test of differences be-tween the AUC in the discovery cohort relative to eachof the independent cohorts.The four-SNP signature was specific to Copaxoneresponse and not to Avonex or to placeboSignature-positive patients in the Avonex-treated arm ofthe BRAVO cohort did not show an ARR reduction. Onthe contrary, they showed an increase in ARR of 10% rela-tive to signature-negative patients (Table 4). Furthermore,the four-SNP signature was not associated with responsein the placebo-treated arm of the GALA DB cohort. Boththese findings provided complementary evidence that thefour-SNP signature is specific to Copaxone-response.Clinical characterization of patient subsetsTo identify the subset of patients in whom the four-SNPsignature was associated with clinically meaningful im-provements in response, patients in the combined dis-covery cohort were split into five bins based on thequintiles of predicted probabilities generated by thefour-SNP logistic regression model (Fig. 2). Patients ineach bin were characterized using a set of relevant base-line and on-treatment clinical measures that were notused in the discovery of the four-SNP signature (Fig. 2b).Additional file 9 shows the results of a principal compo-nents analysis on all of these clinical response variables.The pattern of the principal component loadings indicatedthat the clinical response variables that were not used inbuilding the four-SNP model were orthogonal to the onesthat were used to train the model. Several of these alterna-tive clinical measures showed, on average, steady improve-ments that corresponded with the quintiles of predictedprobabilities from the four-SNP model. Patients in the binwith the highest predicted probabilities (0.90 to 0.94) ofbeing relapse-free had the highest proportion of observedrelapse-free Copaxone responders (95.3%), the highestmean group-level ARR reduction (93%), the longest meantime to first relapse (351.2 days), and the greatest percent-age of patients who met the NEDA3 and NEDA4 defini-tions at 12 months after initiation of treatment (Fig. 2b).In contrast, the bin with the lowest predicted probabilities(0.26 to 0.81) of being relapse-free had the greatest pro-portion of Copaxone non-responders (29%), the lowestmean group-level ARR reduction (58%), the shortest timeto first relapse (300.8 days), and the lowest percentage ofpatients who met the NEDA3 and NEDA4 definitionsafter initiation of treatment (Fig. 2b).DiscussionInter-individual variability in patient response to each ofthe available therapies for multiple sclerosis, combinedwith the variable course of disease, emphasizes the needfor tools that help guide treatment choice in multiplesclerosis. The current study on Copaxone, a first-lineDMT with a well-established favorable efficacy andsafety profile, constitutes the largest pharmacogeneticTable 5 Model performance summary on all the cohortsCohort Number of patients Follow-upduration (years)Specificity Sensitivity AUCTotal Sig+ Sig−DiscoveryGALA DB 639 323 316 1 66% 54% 0.65FORTE DB 532 268 264 1 71% 54% 0.68GALA DB + FORTE DB 1171 591 580 1 68% 54% 0.66Independent assessmentGALA OL 333 190 143 ~3 47% 58% 0.54GA-9001 DB 38 21 17 ~3 41% 52% 0.45GA-9001 OL 74 35 39 ~20 48% 45% 0.49GA-9003 DB 40 21 19 0.75 67% 61% 0.65GA-9003 OL 84 41 43 0.75 67% 54% 0.59PreCISe DB 132 69 63 3 48% 52% 0.49PreCISe OL 240 129 111 5 49% 54% 0.50Sig + (signature-positive) and Sig − (signature-negative) indicate patients classified as relapse-free and relapsing, respectively, after applying the “top-left” thresholdon the predicted probabilities from the four-SNP logistic regression model (see “Methods”). AUC is a threshold-independent metric that computes the overallperformance of the model at all possible thresholds on the predicted probabilities. All performance metrics are rounded to two decimal places.DB double-blind phase, OL open-label phaseRoss et al. Genome Medicine  (2017) 9:50 Page 10 of 15study in multiple sclerosis reported to date (Additionalfile 1). A four-SNP signature was identified as associatedwith treatment response. Signature-positive Copaxone-treated RRMS patients demonstrated better response inmultiple clinically meaningful measures, including ARR,MRI, and NEDA in two discovery RRMS cohorts. Im-proved ARRs were also observed in five independentRRMS cohorts but not in either placebo- or Avonex-treated RRMS patients, demonstrating the predictive asopposed to prognostic nature of the signature, and itsspecificity to Copaxone. The signature identified a ~10%subset of Copaxone-treated RRMS patients with thehighest clinical improvements.Copaxone is a synthetic heterogeneous mixture of upto 1029 variant antigenic polymers [44]. Its MoA is com-plex [12–14] and not completely elucidated. Knownmechanisms include suppression of autoimmune inflam-matory processes by inducing type II monocytes, activa-tion of HLA type I CD8+ T-cells, and an increase in thenumber of T-regulatory cells [14]. Copaxone protectsthe myelin sheath by competing with MBP, which it wasdesigned to mimic [44]. It binds to HLA class II sites onAPCs which present the antigen to naïve T-cells, result-ing in the production of Copaxone-specific Th2 cells(Additional file 10). These cells migrate into the centralnervous system, cross-react with MBP, and induce localsecretion of anti-inflammatory cytokines [14]. Copaxoneis also known to promote neurotrophic factors and in-duce B-cell activation [14, 45, 46]. Gene regions span-ning the identified four-SNP signature, HLA-DQB2,MBP, UVRAG, and ZAK(CDCA7), are known to berelated to either the MoA of Copaxone or the patho-physiology of multiple sclerosis (Additional file 10):HLA-DQB2 is involved in antigen processing and pres-entation, central to Copaxone’s MoA. Other HLA classII variants have been linked with response to Copaxonein prior candidate-gene studies [16–18, 22] but have notbeen reliably replicated. The HLA class II variant DRB1has been associated with treatment response to bothIFN-β and Copaxone [47]. MBP, whose gene product ismimicked by Copaxone, has been shown to be associ-ated with Copaxone response in at least one previouscandidate-gene study [19]. Novel genetic associationsidentified in this study include UVRAG andZAK(CDCA7). UVRAG was recently identified as a regu-lator of naïve peripheral T-cell homeostasis [48], and isin keeping with Copaxone’s effect on T-cells and with aprevious candidate-gene study that reported an associ-ation between TRB and Copaxone response [17]. ZAK, amember of the MAP3K family, is known to be activatedby stress and inflammation [49], while CDCA7 variantsare associated with cell division and brain lesion forma-tion in multiple sclerosis [50]. Thus, the signature spansa multitude of mechanisms which are consistent withCopaxone’s complex MoA and are supported by gene-expression [13, 14] and physicochemical studies [51, 52].Collectively, findings from the current study as well asother studies [13, 14, 51, 52] suggest that the associationof the signature to treatment response is unique toCopaxone’s MoA, which depends on its physicochemicalproperties and distinguishes it from other glatiramoidsand follow-on products.Prior pharmacogenetic studies of Copaxone responsehave utilized candidate-gene approaches in cohortslargely drawn from observational and hospital-based pa-tient populations [16–19, 21]. This has resulted in lim-ited reproducibility, potentially due to variable responsecriteria and small sample sizes. In contrast, the currentstudy assessed patient cohorts from two large phase IIIclinical trials in the discovery phase, with a combinedsample size of 1171 patients. Subsequently, the identifiedfour-SNP signature was assessed in five additional inde-pendent late-phase clinical trial cohorts with RRMS, twoCopaxone-treated cohorts with CIS, as well as Avonex-and placebo-treated cohorts. Additionally, the studyemployed a comprehensive genome-wide SNP-chip witha coverage of around five million SNPs combined with amulti-step association analysis that selected SNPs withthe maximum a priori evidence. This was followed bya Bayesian predictive modeling approach that system-atically explored all possible SNP combinations andsimultaneously evaluated the probability of inclusionof each of the SNPs in the signature. Adopting theBayesian approach allowed efficient identification of aminimal set of SNPs with the greatest potential togeneralize to newer populations and avoided the needfor multiple hypotheses testing while reducing falsediscoveries.Relapses are the primary target phenotype of DMTs inRRMS patients. Therefore, this study employed responsedefinitions that incorporated relapses both for the initialidentification of extreme-phenotypes of Copaxone re-sponse as well as for Bayesian modeling to identify thefour-SNP signature. The presence of the signature wascorrelated with higher ARR reduction as well as in-creased time to first relapse even in patients who hadbeen treated with Copaxone for up to 20 years. Whencompared with the 20-year cohorts, those with ~3 yearsof follow-up had a lower mean ARR reduction. However,it is challenging to interpret ARR patterns over time be-cause they are dependent on several factors. For ex-ample, it is well known that RRMS patients eventuallydevelop a secondary progressive type of disease which isaccompanied by a decrease in ARR [53]. In enrichmentclinical trial designs, patients with a higher baseline ARRexperience a lower mean number of relapses as the dur-ation of follow-up in the trial increases [54–56], aphenomenon termed “regression to the mean”. BothRoss et al. Genome Medicine  (2017) 9:50 Page 11 of 15these factors result in a decrease in the absolute ARRwith time in trial cohorts.In comparison to signature-negative patients, signature-positive patients showed a larger reduction in ARR uponswitching to Copaxone treatment in the OL phase in al-most all of the trial cohorts studied (Additional file 11).Additionally, signature-positive patients showed betterMRI parameters (T1 and T2 lesions), reflecting improve-ment in inflammatory disease activity and burden as wellas increased NEDA3/4 that provide an overall assessmentof disease progression. The consistent association of thesignature with improvement in multiple clinical parame-ters, together with its specificity to Copaxone therapy ver-sus placebo and Avonex, demonstrate its robustness.Notwithstanding the comprehensive approach taken inthis study, only a limited proportion of observed hetero-geneity in response phenotypes could be explained bygenetic variation. Specifically, the AUC, which quantifiesthe classification performance of the four-SNP signaturein the overall Copaxone-treated population, ranged from0.45 to 0.67, demonstrating insufficient discriminatorypower for clinical practice. Nevertheless, the signaturewas able to identify a genetically homogeneous ~10%subset of the multiple sclerosis patient population in thediscovery cohort with a 93% reduction in ARR versusbaseline (Fig. 2) and substantially improved response inmultiple clinical measures. In the four-SNP signature,ZAK(CDCA7) had a low MAF and imbalanced samplesizes in the patient groups with the major and minor al-lele (Additional file 3), which resulted in a wider confi-dence interval for its regression coefficient (and OR)(Table 3). However, the posterior inclusion probabilityfor this SNP remained high (Additional file 5), indicatingthat the potential bias introduced by the imbalanced al-lelic groups had little impact on the identified signature.It is also important to note that the clinical trial co-horts employed in the discovery of the signature hadonly a year of patient follow-up. In the context of achronic disease such as multiple sclerosis that affects pa-tients over several years of their lives, a year of follow-upmight not be sufficient to observe consistent patterns ofresponse to treatment. Analysis of additional clinicallyrelevant response definitions in the context of the four-SNP signatures in Fig. 2 is a challenging task, given thediversity of response definitions employed by clinicianstreating multiple sclerosis patients. Nevertheless, it isimportant to strive for consensus signatures and validatethe performance of these signatures in geneticallydefined subsets of RRMS patients, in additional inde-pendent cohorts with larger sample sizes, and in non-Caucasian multiple sclerosis patients using a variety ofclinical response definitions.Studies examining the pharmacogenetics of responseto IFN-β (Avonex) [57] share commonalities with thisstudy in terms of both methodology, such as a multi-stage study design, and an emerging trend towards iden-tifying multi-SNP signatures [22, 57]. Furthermore,genetic signatures detected in the IFN-β studies werespecific to IFN-β and not generalizable to Copaxone[58], paralleling the Copaxone-specific nature of thefour-SNP signature detected in this study. Interestingly,a recent assessment of the pharmacogenetics of IFN-βnon-response identified genotypic patient subsets com-prising ~17% of the cohort who were not likely to re-spond at all to IFN-β therapy [59]. These findings areanalogous to the results in the current study, albeit wepursued identification of high-response rather than non-response. Overall, findings from pharmacogenetic stud-ies on the two major DMTs in multiple sclerosis, Copax-one and IFN-β, demonstrate genetic associations thatare DMT-specific but confined to a small subset of theRRMS population. This suggests that while geneticsalone cannot fully account for drug response variabilityin the overall multiple sclerosis patient population, diag-nostic tools that incorporate genetics or other factorsthat enable the definition of more homogeneous diseasesubtypes may aid in guiding treatment choices in mul-tiple sclerosis.ConclusionsThe findings from this study emphasize the need forrigorous, large-scale studies with multiple independentcohorts to fully understand the contribution of geneticsto multiple sclerosis drug response. For the first time, aCopaxone-specific multi-SNP signature identifies pa-tients with higher response to treatment in multiple, in-dependent cohorts and over extended periods oftreatment, lending more evidence to the contribution ofgenetic variation to drug response in patients with mul-tiple sclerosis. The pronounced association of the signa-ture with clinical improvements in a small subset of thepatient cohort demonstrates the complex interplay ofimmune mechanisms and the individual nature of re-sponse to Copaxone.Additional filesAdditional file 1: A comprehensive summary of sample sizes andcohorts in pharmacogenetics studies in the field of multiple sclerosis.(DOCX 23 kb)Additional file 2: Candidate variants and genes from stage I of analysis.(DOCX 17 kb)Additional file 3: Frequencies for each genotype of the four SNPs inthe four-SNP model. (DOCX 14 kb)Additional file 4: Notes on the model-building procedure used in thestudy. (DOCX 14 kb)Additional file 5: The eleven-SNP model coefficients, odds ratios with95% Bayesian confidence intervals and posterior inclusion probabilities.(DOCX 15 kb)Ross et al. Genome Medicine  (2017) 9:50 Page 12 of 15Additional file 6: Funnel-plot visualization of mean change in ARR(signature-positive versus signature-negative) versus the sample size ofthe cohort. Discovery as well as the independent cohorts are shown.(DOCX 58 kb)Additional file 7: Performance metrics of each of the possible three-SNPmodels. (DOCX 16 kb)Additional file 8: Results from the bootstrap version of the pairwise testof differences in AUC between the discovery cohort and each of theindependent cohorts, originally described by Hanley and McNeil.(DOCX 15 kb)Additional file 9: Principal component loadings of clinical responsevariables in discovery cohorts. (DOCX 118 kb)Additional file 10: Biology of the four-SNP signature: A schematicillustration of the relationship of the identified four-SNP signature tothe known components of Copaxone’s complex mechanism of action.(DOCX 1050 kb)Additional file 11: Change in ARR among placebo patients who wereswitched to Copaxone treatment in the OL phase. (DOCX 12 kb)Additional file 12: Details on participating institutional or clinical sitesat which Institutional Review Boards or Ethics Committees approved theclinical trials included in the study. (DOCX 59 kb)AbbreviationsARR: Annualized relapse rate; AUC: Area under the receiver operatingcharacteristic curve; BRAVO: The Laquinimod DB placebo controlled studywith a rater-Blinded Reference Arm of interferon β-1a (Avonex®);BMA: Bayesian model averaging; CI: Confidence interval; CIS: Clinicallyisolated syndrome; DB: Double-blind; DMT: Disease-modifying therapy;EDSS: Expanded Disability Status Scale; FORTE: The FORTy mg Efficacy ofglatiramer acetate study; GA: Glatiramer acetate; GALA: The GlatiramerAcetate Low-Frequency Administration study; HWE: Hardy–Weinbergequilibrium; IFN: Interferon; LD: Linkage disequilibrium; MBP: Myelin basicprotein; MoA: Mechanism of action; MS: Multiple sclerosis; NEDA: Noevidence of disease activity; OL: Open-label; PreCISe: The early glatirameracetate in patients Presenting with a Clinically Isolated Syndrome study;ROC: Receiver operating characteristic; RRMS: Relapsing-remitting multiplesclerosis; SD: Standard deviation; SNP: Single nucleotide polymorphism;TRB: T-cell receptor betaAcknowledgementsWe thank all the patients who participated in the Copaxone clinical trials andthus made this research possible. We thank Ronen Mansuri and Pippa Loupefor assisting with information retrieval. We thank Sarah Kolitz and Tamar Erezfor reviewing this manuscript and contributing helpful comments.FundingThis study was supported by funding from Teva PharmaceuticalIndustries Ltd.Availability of data and materialsThe datasets generated and/or analyzed during the current study are notpublicly available due to the terms of informed consent forms signed bypatients providing the DNA samples.Authors’ contributionsCJR, FT, JS, MT, DL, and IG wrote the manuscript. CJR, FT, JS, MT, MD, BW,and RK analyzed the data. CJR, FT, JS, DL, MT, MD, BZ, IG, and MRHinterpreted the analysis. All authors read and approved the final manuscript.Competing interestsJS, BW, RK, and BZ are employees and stockholders of ImmuneeringCorporation, which is majority-owned by Teva Pharmaceutical Industries,the maker of Copaxone. DL, MT, MD, VK, IG, and MRH are employees ofTeva Pharmaceutical Industries. CJR has received grants from Teva PharmaceuticalIndustries. Since performing the work described, FT has become an employeeand equity holder of Celgene Corporation and declares no competing interests.Consent for publicationNot applicable.Ethics approval and consent to participateThis study was approved by either the Ethics Committees or InstitutionalReview Boards of each clinical institution that participated in the parenttrials. Details for each trial and participating institution are provided inAdditional file 12. In accordance with the principles of the Declaration ofHelsinki, patients included in the study provided informed consent forcollection of their DNA samples and use of their clinical and therapeuticdata.Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.Author details1Faculty of Pharmaceutical Sciences, University of British Columbia,Vancouver, BC, Canada. 2BC Children’s Hospital, Department of MedicalGenetics, University of British Columbia, Vancouver, BC, Canada.3Immuneering Corporation, Cambridge, MA, USA. 4Teva PharmaceuticalIndustries Ltd, Petach Tikva, Israel. 5Teva Pharmaceutical Industries, Frazer, PA,USA.Received: 8 December 2016 Accepted: 8 May 2017References1. International Multiple Sclerosis Genetics Consortium (IMSGC), Bush WS,Sawcer SJ, de Jager PL, Oksenberg JR, McCauley JL, et al. Evidence forpolygenic susceptibility to multiple sclerosis–the shape of things to come.Am J Hum Genet. 2010;86:621–5. http://dx.doi.org/10.1016/j.ajhg.2010.02.027.2. International Multiple Sclerosis Genetics Consortium (IMSGC), Beecham AH,Patsopoulos NA, Xifara DK, Davis MF, Kemppinen A, et al. Analysis ofimmune-related loci identifies 48 new susceptibility variants for multiplesclerosis. Nat Genet. 2013;45:1353–60. http://dx.doi.org/10.1038/ng.2770.3. International Multiple Sclerosis Genetics Consortium, Hafler DA, CompstonA, Sawcer S, Lander ES, Daly MJ, et al. Risk alleles for multiple sclerosisidentified by a genomewide study. N Engl J Med. 2007;357:851–62. http://dx.doi.org/10.1056/NEJMoa073493.4. International Multiple Sclerosis Genetics Consortium, Wellcome Trust CaseControl Consortium 2, Sawcer S, Hellenthal G, Pirinen M, Spencer CCA, et al.Genetic risk and a primary role for cell-mediated immune mechanisms inmultiple sclerosis. Nature. 2011;476:214–9. http://dx.doi.org/10.1038/nature10251.5. Bahlo M, Booth DR, Broadley SA, Brown MA, Foote SJ, Griffiths LR, et al.Genome-wide association study identifies new multiple sclerosissusceptibility loci on chromosomes 12 and 20. Nat Genet. 2009;41:824–8.http://dx.doi.org/10.1038/ng.396.6. Lill CM, Luessi F, Alcina A, Sokolova EA, Ugidos N, de la Hera B, et al.Genome-wide significant association with seven novel multiple sclerosis riskloci. J Med Genet. 2015;52:848–55. http://dx.doi.org/10.1136/jmedgenet-2015-103442.7. Isobe N, Keshavan A, Gourraud P-A, Zhu AH, Datta E, Schlaeger R, et al.Association of HLA Genetic Risk Burden With Disease Phenotypes inMultiple Sclerosis. JAMA Neurol. 2016;73:795–802. http://dx.doi.org/10.1001/jamaneurol.2016.0980.8. Hollenbach JA, Oksenberg JR. The immunogenetics of multiple sclerosis: Acomprehensive review. J Autoimmun. 2015;64:13–25. http://dx.doi.org/10.1016/j.jaut.2015.06.010.9. Jokubaitis VG, Butzkueven H. A genetic basis for multiple sclerosis severity:Red herring or real? Mol Cell Probes. 2016;30:357–65. http://dx.doi.org/10.1016/j.mcp.2016.08.007.10. Cree BAC, Hartung H-P. Steering through complexity: managementapproaches in multiple sclerosis. Curr Opin Neurol. 2016;29:263–71. http://dx.doi.org/10.1097/WCO.0000000000000332.11. Grossman I, Knappertz V, Laifenfeld D, Ross C, Zeskind B, Kolitz S, et al.Pharmacogenomics strategies to optimize treatments for multiple sclerosis:Insights from clinical research. Prog Neurobiol. 2016. http://dx.doi.org/10.1016/j.pneurobio.2016.02.001.12. Towfic F, Funt JM, Fowler KD, Bakshi S, Blaugrund E, Artyomov MN, et al.Comparing the Biological Impact of Glatiramer Acetate with the BiologicalRoss et al. Genome Medicine  (2017) 9:50 Page 13 of 15Impact of a Generic. PLoS One. 2014;9:e83757. https://doi.org/10.1371/journal.pone.0083757.13. Kolitz S, Hasson T, Towfic F, Funt JM, Bakshi S, Fowler KD, et al. Geneexpression studies of a human monocyte cell line identify dissimilaritiesbetween differently manufactured glatiramoids. Sci Rep. 2015;5:10191.https://doi.org/10.1038/srep10191.14. Hasson T, Kolitz S, Towfic F, Laifenfeld D, Bakshi S, Beriozkin O, et al.Functional effects of the antigen glatiramer acetate are complex and tightlyassociated with its composition. J Neuroimmunol. 2016;290:84–95. http://dx.doi.org/10.1016/j.jneuroim.2015.11.020.15. Boster AL, Ford CC, Neudorfer O, Gilgun-Sherki Y. Glatiramer acetate: long-termsafety and efficacy in relapsing-remitting multiple sclerosis. Expert RevNeurother. 2015;15:575–86. http://dx.doi.org/10.1586/14737175.2015.1040768.16. Fusco C, Andreone V, Coppola G, Luongo V, Guerini F, Pace E, et al. HLA-DRB1*1501 and response to copolymer-1 therapy in relapsing-remittingmultiple sclerosis. Neurology. 2001;57:1976–9. http://www.ncbi.nlm.nih.gov/pubmed/11739812.17. Gross R, Healy BC, Cepok S, Chitnis T, Khoury SJ, Hemmer B, et al.Population structure and HLA DRB1 1501 in the response of subjects withmultiple sclerosis to first-line treatments. J Neuroimmunol. 2011;233:168–74.http://dx.doi.org/10.1016/j.jneuroim.2010.10.038.18. Dhib-Jalbut S, Valenzuela RM, Ito K, Kaufman M, Ann Picone M, Buyske S.HLA DR and DQ alleles and haplotypes associated with clinical response toglatiramer acetate in multiple sclerosis. Mult Scler Relat Disord. 2013;2:340–8.http://dx.doi.org/10.1016/j.msard.2013.02.005.19. Grossman I, Avidan N, Singer C, Goldstaub D, Hayardeny L, Eyal E, et al.Pharmacogenetics of glatiramer acetate therapy for multiple sclerosisreveals drug-response markers. Pharmacogenet Genomics. 2007;17:657–66.https://www.ncbi.nlm.nih.gov/pubmed/17622942.20. Tsareva E, Kulakova O, Boyko A, Favorova O. Pharmacogenetics of multiplesclerosis: personalized therapy with immunomodulatory drugs.Pharmacogenet Genomics. 2016;26:103–15. https://www.ncbi.nlm.nih.gov/pubmed/26678572.21. Tsareva EI, Kulakova OG, Makarycheva OI, Boĭko AN, Shchur SG, Lashch NI,et al. Pharmacogenomics of multiple sclerosis: association of immuneresponse genes polymorphism with copaxone treatment efficacy. Mol Biol.2011;45:963–72. https://doi.org/10.1134/s0026893311060185.22. Tsareva EY, Kulakova OG, Boyko AN, Shchur SG, Lvovs D, Favorov AV, et al.Allelic combinations of immune-response genes associated with glatirameracetate treatment response in Russian multiple sclerosis patients.Pharmacogenomics. 2012;13:43–53. http://dx.doi.org/10.2217/pgs.11.136.23. Khan O, Rieckmann P, Boyko A, Selmaj K, Zivadinov R, GALA Study Group.Three times weekly glatiramer acetate in relapsing-remitting multiplesclerosis. Ann Neurol. 2013;73:705–13. http://dx.doi.org/10.1002/ana.23938.24. Comi G, Cohen JA, Arnold DL, Wynn D, Filippi M, FORTE Study Group. PhaseIII dose-comparison study of glatiramer acetate for multiple sclerosis. AnnNeurol. 2011;69:75–82. http://dx.doi.org/10.1002/ana.22316.25. Khan O, Rieckmann P, Boyko A, Selmaj K, Ashtamker N, Davis MD, et al. Efficacyand safety of a three-times-weekly dosing regimen of glatiramer acetate inrelapsing-remitting multiple sclerosis patients: 3-year results of the GlatiramerAcetate Low-Frequency Administration open-label extension study. Mult Scler.2016;1352458516664033. http://dx.doi.org/10.1177/1352458516664033.26. Johnson KP, Brooks BR, Cohen JA, Ford CC, Goldstein J, Lisak RP, et al.Copolymer 1 reduces relapse rate and improves disability in relapsing-remitting multiple sclerosis: results of a phase III multicenter, double-blindplacebo-controlled trial. The Copolymer 1 Multiple Sclerosis Study Group.Neurology. 1995;45:1268–76. http://dx.doi.org/10.1212/WNL.45.7.1268.27. Johnson KP, Brooks BR, Cohen JA, Ford CC, Goldstein J, Lisak RP, et al.Extended use of glatiramer acetate (Copaxone) is well tolerated andmaintains its clinical effect on multiple sclerosis relapse rate and degree ofdisability. Copolymer 1 Multiple Sclerosis Study Group. Neurology. 1998;50:701–8. https://doi.org/10.1212/wnl.50.3.701.28. Comi G, Filippi M, Wolinsky JS. European/Canadian multicenter, double-blind, randomized, placebo-controlled study of the effects of glatirameracetate on magnetic resonance imaging-measured disease activity andburden in patients with relapsing multiple sclerosis. European/CanadianGlatiramer Acetate Study Group. Ann Neurol. 2001;49:290–7. https://doi.org/10.1002/ana.64.29. Wolinsky JS, Comi G, Filippi M, Ladkani D, Kadosh S, Shifroni G, et al. Copaxone’seffect on MRI-monitored disease in relapsing MS is reproducible and sustained.Neurology. 2002;59:1284–6. https://doi.org/10.1212/wnl.59.8.1284.30. Comi G, Martinelli V, Rodegher M, Moiola L, Bajenaru O, Carra A, et al. Effectof glatiramer acetate on conversion to clinically definite multiple sclerosis inpatients with clinically isolated syndrome (PreCISe study): a randomised,double-blind, placebo-controlled trial. Lancet. 2009;374:1503–11. http://dx.doi.org/10.1016/S0140-6736(09)61259-9.31. Comi G, Martinelli V, Rodegher M, Moiola L, Leocani L, Bajenaru O, et al.Effects of early treatment with glatiramer acetate in patients with clinicallyisolated syndrome. Mult Scler. 2013;19:1074–83. http://dx.doi.org/10.1177/1352458512469695.32. Vollmer TL, Sorensen PS, Selmaj K, Zipp F, Havrdova E, Cohen JA, et al. Arandomized placebo-controlled phase III trial of oral laquinimod for multiplesclerosis. J Neurol. 2014;261:773–83. http://dx.doi.org/10.1007/s00415-014-7264-4.33. Davis M, Ashtamker N, Steinerman J, Knappertz V. Time course of glatirameracetate efficacy in relapsing-remitting multiple sclerosis patients in theGlatiramer Acetate Low-Frequency Administration (GALA) Study (S51.006).Neurology. 2016;86. http://www.neurology.org/content/86/16_Supplement/S51.006.abstract.34. Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expandeddisability status scale (EDSS). Neurology. 1983;33:1444–52. https://doi.org/10.1212/wnl.33.11.1444.35. Hoeting JA, Madigan D, Raftery AE, Volinsky CT. Bayesian model averaging:a tutorial (with comments by M. Clyde, David Draper and E. I. George, and arejoinder by the authors). Stat Sci. 1999;14:382–417. http://projecteuclid.org/euclid.ss/1009212519.36. Stephens M, Balding DJ. Bayesian statistical methods for genetic associationstudies. Nat Rev Genet. 2009;10:681–90. http://dx.doi.org/10.1038/nrg2615.37. Fridley BL. Bayesian variable and model selection methods for geneticassociation studies. Genet Epidemiol. 2009;33:27–37. http://dx.doi.org/10.1002/gepi.20353.38. Wu X-L, Gianola D, Rosa GJM, Weigel KA. Bayesian model averaging forevaluation of candidate gene effects. Genetica. 2010;138:395–407.http://dx.doi.org/10.1007/s10709-009-9433-4.39. Peterson C, Swartz M, Shete S, Vannucci M. Bayesian model averaging forgenetic association studies. In: Kim-Anh D, Qin ZS, Vannucci M, editors.Advances in Statistical Bioinformatics: models and integrative inferencefor high-throughput data. Cambridge University Press; 2013. p. 208.https://goo.gl/84u6zm.40. George EI, Mcculloch RE. Approaches for Bayesian variable selection. StatSin. 1997;7:339–73. http://www3.stat.sinica.edu.tw/statistica/j7n2/j7n26/j7n26.htm. Accessed 21 May 2017.41. Scott SL. BoomSpikeSlab: MCMC for spike and slab regression. 2014.http://CRAN.R-project.org/package=BoomSpikeSlab. Accessed 21 May 2017.42. Scott SL, Varian HR. Predicting the present with Bayesian structural timeseries. 2013. http://dx.doi.org/10.2139/ssrn.2304426. Accessed 8 Mar 2016.43. Kappos L, De Stefano N, Freedman MS, Cree BA, Radue E-W, Sprenger T,et al. Inclusion of brain volume loss in a revised measure of “no evidence ofdisease activity” (NEDA-4) in relapsing-remitting multiple sclerosis. MultScler. 2015;22(10):1297–305. http://dx.doi.org/10.1177/1352458515616701.44. Teitelbaum D, Arnon R, Sela M. Cop 1 as a candidate drug for multiplesclerosis. J Neural Transm Suppl. 1997;49:85–91. https://doi.org/10.1007/978-3-7091-6844-8_9.45. Comi G, Amato MP, Bertolotto A, Centonze D, De Stefano N, Farina C, et al.The heritage of glatiramer acetate and its use in multiple sclerosis. MultipleSclerosis Demyelinating Disorders. 2016;1:1–13. http://dx.doi.org/10.1186/s40893-016-0010-2.46. Aharoni R. Immunomodulation neuroprotection and remyelination - thefundamental therapeutic effects of glatiramer acetate: a critical review.J Autoimmun. 2014;54:81–92. http://dx.doi.org/10.1016/j.jaut.2014.05.005.47. Kulakova OG, Tsareva EY, Lvovs D, Favorov AV, Boyko AN, Favorova OO.Comparative pharmacogenetics of multiple sclerosis: IFN-β versusglatiramer acetate. Pharmacogenomics. 2014;15:679–85. http://dx.doi.org/10.2217/pgs.14.26.48. Afzal S, Hao Z, Itsumi M, Abouelkheer Y, Brenner D, Gao Y, et al. Autophagy-independent functions of UVRAG are essential for peripheral naive T-cellhomeostasis. Proc Natl Acad Sci U S A. 2015;112:1119–24. http://dx.doi.org/10.1073/pnas.1423588112.49. Kyriakis JM, Avruch J. Mammalian MAPK signal transduction pathwaysactivated by stress and inflammation: a 10-year update. Physiol Rev. 2012;92:689–737. http://dx.doi.org/10.1152/physrev.00028.2011.50. Gourraud P-A, McElroy JP, Caillier SJ, Johnson BA, Santaniello A, Hauser SL,et al. Aggregation of multiple sclerosis genetic risk variants in multiple andRoss et al. Genome Medicine  (2017) 9:50 Page 14 of 15single case families. Ann Neurol. 2011;69:65–74. http://dx.doi.org/10.1002/ana.22323.51. Conner JB, Bawa R, Michael Nicholas J, Weinstein V. Copaxone in the era ofbiosimilars and nanosimilars. In: Bawa R, Audette GF, Rubinstein I, editors.Handbook of clinical nanomedicine: nanoparticles, imaging, therapy, andclinical applications. Boca Raton: CRC Press; 2014. p. 783–826. http://www.crcnetbase.com/doi/abs/10.1201/b19915-32.52. Weinstein V, Schwartz R, Grossman I, Zeskind B, Michael Nicholas J.Glatiramoids. In: Crommelin DJA, de Vlieger JSB, editors. Non-biologicalcomplex drugs. Springer International Publishing; 2015. p. 107–48. https://doi.org/10.1007/978-3-319-16241-6_4.53. Scalfari A, Neuhaus A, Daumer M, Muraro PA, Ebers GC. Onset ofsecondary progressive phase and long-term evolution of multiplesclerosis. J Neurol Neurosurg Psychiatry. 2014;85:67–75. http://dx.doi.org/10.1136/jnnp-2012-304333.54. Martínez-Yélamos S, Martínez-Yélamos A, Martín Ozaeta G, Casado V,Carmona O, Arbizu T. Regression to the mean in multiple sclerosis. MultScler. 2006;12:826–9. http://dx.doi.org/10.1177/1352458506070820.55. Nicholas R, Straube S, Schmidli H, Schneider S, Friede T. Trends inannualized relapse rates in relapsing-remitting multiple sclerosis andconsequences for clinical trial design. Mult Scler. 2011;17:1211–7. http://dx.doi.org/10.1177/1352458511406309.56. Lavery AM, Verhey LH, Waldman AT. Outcome measures in relapsing-remitting multiple sclerosis: capturing disability and disease progression inclinical trials. Mult Scler Int. 2014;2014:1–13. https://doi.org/10.1155/2014/262350.57. Mahurkar S, Moldovan M, Suppiah V, Sorosina M, Clarelli F, Liberatore G,et al. Response to interferon-beta treatment in multiple sclerosis patients:a genome-wide association study. Pharmacogenomics J. 2016;2016:1–7.http://dx.doi.org/10.1038/tpj.2016.20.58. Mahurkar S, Suppiah V, O’Doherty C. Pharmacogenomics of interferon betaand glatiramer acetate response: a review of the literature. Autoimmun Rev.2014;13:178–86. http://dx.doi.org/10.1016/j.autrev.2013.10.012.59. Esposito F, Sorosina M, Ottoboni L, Lim ET, Replogle JM, Raj T, et al. Apharmacogenetic study implicates SLC9a9 in multiple sclerosis diseaseactivity. Ann Neurol. 2015;78:115–27. http://dx.doi.org/10.1002/ana.24429.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:Ross et al. Genome Medicine  (2017) 9:50 Page 15 of 15


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items