CLINICAL CANCER RESEARCH | TRANSLATIONAL CANCER MECHANISMS AND THERAPY

Outcome-Related Signatures Identified by WholeTranscriptome Sequencing of Resectable Stage III/IVMelanoma Evaluated after Starting Hu14.18-IL2 A C

Richard K. Yang1, Igor B. Kuznetsov2, Erik A. Ranheim1, Jun S. Wei3, Sivasish Sindiri3, Berkley E. Gryder3,Vineela Gangalapudi3, Young K. Song3, Viharkumar Patel1, Jacquelyn A. Hank4, Cindy Zuleger5,6,7,Amy K. Erbe4, Zachary S. Morris4, Renae Quale5,6,7, KyungMann Kim8, Mark R. Albertini5,6,7, Javed Khan3,and Paul M. Sondel4,9

ABSTRACT◥

Purpose: We analyzed whole transcriptome sequencing intumors from 23 patients with stage III or IV melanoma from apilot trial of the anti-GD2 immunocytokine, hu14.18-IL2, to iden-tify predictive immune and/or tumor biomarkers in patients withmelanoma at high risk for recurrence.

Experimental Design: Patients were randomly assigned toreceive the first of three monthly courses of hu14.18-IL2 immuno-therapy either before (Group A) or after (Group B) completesurgical resection of all known diseases. Tumors were evaluatedby histology and whole transcriptome sequencing.

Results: Tumor-infiltrating lymphocyte (TIL) levels directlyassociated with relapse-free survival (RFS) and overall survival(OS) in resected tumors from Group A, where early responses tothe immunotherapy agent could be assessed. TIL levels directlyassociated with a previously reported immune signature, which

associated with RFS and OS, particularly in Group A tumors. InGroup A tumors, there were decreased cell-cycling gene RNAtranscripts, but increased RNA transcripts for repair andgrowth genes. We found that outcome (RFS and OS) was directlyassociated with several immune signatures and immune-relatedRNA transcripts and inversely associated with several tumorgrowth–associated transcripts, particularly in Group A tumors.Most of these associations were not seen in Group B tumors.

Conclusions: We interpret these data to signify that bothimmunologic and tumoral cell processes, as measured byRNA-sequencing analyses detected shortly after initiation ofhu14.18-IL2 therapy, are associated with long-term survival andcould potentially be used as prognostic biomarkers in tumorresection specimens obtained after initiating neoadjuvant immu-notherapy.

IntroductionMelanoma is considered an immunogenic and relatively immu-

noresponsive tumor with a relatively high tumor mutation bur-den (1, 2), often harboring many tumor-infiltrating lymphocytes(TIL), which have prognostic value (3, 4). Improved biomarkers ofoutcome following melanoma immunotherapy are needed. Immunecheckpoint inhibition is an effective therapy for some patients withmelanoma (5–8). The mechanisms of melanoma response to check-point inhibition immunotherapy are complex and multifactorial (9).Previous transcriptomic analyses have found adaptive immunesignatures in melanoma tumor biopsies soon after checkpointblockade, which are predictive of response (10). Recently, wholetranscriptomic interrogation of MYCN nonamplified pediatric neu-roblastoma has shown immune signatures, which are associatedwith outcomes in children (11). In contrast, melanoma-deriveddriver mutations (12–14) as well as antiimmunity signaling path-ways (15, 16) also contribute to decreased long-term survival (17).The relative contribution of tumoral versus immunologic para-meters to cancer survival is unclear; in colorectal carcinoma eachside contributes approximately 50% (18).

The hu14.18-IL2 immunocytokine (IC) is a humanized mAb thatbinds to GD2 and is linked, as a fusion protein, to IL-2 at its Fcregion (19, 20). GD2 is a cell membrane disialoganglioside foundin neuroectodermal tumors (melanoma, neuroblastoma, and sarco-mas), but demonstrates only low expression in select normal tissues(cerebellum, peripheral nerves; refs. 21, 22). Hu14.18-IL2 has beenstudied in vitro and in mouse models against melanoma andneuroblastoma (23–25). In mice, the antitumor effects of hu14.18-

1Department of Pathology and Laboratory Medicine, University of Wisconsin–Madison, Madison, Wisconsin. 2Cancer Research Center and Department ofEpidemiology and Biostatistics, University at Albany, Rensselaer, New York.3Oncogenomics Section, Genetics Branch, NCI, NIH, Bethesda, Maryland.4Department of Human Oncology, University of Wisconsin–Madison, Madison,Wisconsin. 5University ofWisconsin Carbone Cancer Center (UWCCC),Madison,Wisconsin. 6Department of Medicine, UW School of Medicine and Public Health,Madison, Wisconsin. 7Medical Service, William S. Middleton Memorial VeteransHospital, Madison, Wisconsin. 8Department of Biostatistics and Medical Infor-matics, University of Wisconsin–Madison, Madison, Wisconsin. 9Departments ofPediatrics and Genetics, and UWCCC, University of Wisconsin–Madison,Madison, Wisconsin.

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

ClinicalTrials.gov identifier: NCT00590824 (http://dx.doi.org/10.13039/501100000272).

R.K. Yang and I.B. Kuznetsov contributed equally as the co-first authors of thisarticle.

J. Khan and P.M. Sondel contributed equally as co-senior authors of this article.

Current address for R.K. Yang: Department of Pathology, MD Anderson CancerCenter, Houston, Texas.

Corresponding Authors: Paul M. Sondel, University of Wisconsin School ofMedicine and Public Health, 1111 Highland Avenue, Madison, WI 53705-2275.Phone: 608-263-9069; Fax: 608-263-4226; E-mail:pmsondel@humonc.wisc.edu; and Javed Khan, khanjav@mail.nih.gov

Clin Cancer Res 2020;26:3296–306

doi: 10.1158/1078-0432.CCR-19-3294

�2020 American Association for Cancer Research.

AACRJournals.org | 3296

IL2 involve cytotoxic T cells and NK cells (26–28). In addition, micewith smaller tumors or with minimal residual disease (MRD) attreatment initiation elicit the most robust responses to single-agentIC therapy (29, 30). Hu14.18-IL2 has undergone phase I and IItesting in adults with metastatic melanoma and children withneuroblastoma (31–35), and shows reproducible antitumor activity,particularly in the MRD setting.

Here we present tumor analyses from a clinical trial (CO05601),wherein 21 of 23 patients with advanced melanoma received surgicalresection of all evident disease to attain a complete response (CR)together with systemic IC administration (36). Patients were random-ized to have their first of three monthly courses of hu14.18-IL2 startedeither just before (Group A, n¼ 13) or just after (Group B, n¼ 8) theircomplete surgical resection. Our primary goal for this report is toidentify the differences in gene expression in these tumors prior to andafter treatment with hu14.18-IL2; our secondary goal is to determinewhether gene expression patterns before treatment, or induced by thetreatment, are associated with outcome. Published data from this trialreported prolonged tumor-free survival in some patients at high riskfor recurrence (36). In this report, we analyze whole transcriptomesequencing from these resected recurrent/refractory melanomatumors and compare results for tumors obtained before versus afterone course of hu14.18-IL2. This is the first transcriptomic analysis oftumors from human patients receiving hu14.18-IL2. The literature isbecoming rich in transcriptomic studies in the context of checkpointinhibition, but it is relatively poor for experimental treatment outsidethis category of drugs. The data presented extend priorwork evaluatingthe predictive ability of immune signatures for long-term outcomeassociated with other forms of immunotherapy (particularly check-point blockade), and indicates the greater predictive power of tran-scriptomic analyses of tumor tissue obtained just after starting thisform of immunotherapy over that obtained prior to initiating thetherapy.

Materials and MethodsClinical trial design

Twenty-three patients with advancedmelanoma participated in thistrial (UWCCC Protocol CO05601). The UW Human Subjects Com-mittee and the FDA approved the study (ClinicalTrials.gov:NCT00590824; IND-12220). This study was conducted in accordance

with the Declaration of Helsinki. Informed written consent wasobtained from each subject. These human investigations were per-formed after approval by the UW institutional review board and inaccordance with an assurance filed with and approved by the U.S.Department of Health and Human Services. Details of study design,conduct, and clinical results have been reported previously (36). Allpatients had recurrent stage III (recurrent regionalmetastasis), or stageIV (any distant metastasis) melanoma with biopsy-proven (current orprevious) stage III or stage IV disease (36). Eligibility criteria requiredthat at study entry, all patients have recurrent melanoma involving ≤3sites, judged to be totally resectable, where resectionwould be clinicallyrecommended. Patients signed informed consent forms and wererandomized into Group A or B prior to treatment. In addition tohu14.18-IL2 treatment, 3 of these 23 patients also received cilengitide,an antiangiogenicRGD-pentapeptide, as part of the original study (36).After enrolling those 3 patients, the study was amended, removingcilengitide treatment because of toxicity. The three cilengitide-treatedpatients were excluded from the previously reported clinicalreport (36). Two of 10 patients randomized to Group B were nottreatedwith hu14.18-IL2 as they did not achieve the required CR statusfollowing surgery. One of these 2 patients left the protocol due toextensive residual disease after surgery; the second was taken offprotocol because melanoma was detected by PET/CT prior to thefirst hu14.18-IL2 course. The remaining 8 patients were designated asGroup B patients. All 13 Group A patients received one cycle ofhu14.18-IL2 prior to surgery and remained on study after resection.

Our previous report focused on clinical tolerance and treatmentresults for patients receiving only hu14.18-IL2 (36); this article focuseson the biological analyses of the tumors and differences in the tumorsfor patients in Group A versus Group B. To augment statistical powerin this relatively small study, we have included tumors from the 3cilengitide-treated patients. Therefore, this study includes histologicand transcriptomic analyses for all tumors that were resected from all23 patients. Associations with relapse-free survival (RFS) and overallsurvival (OS) were performed for patients that received any amount (1to 3 cycles) of hu14.18-IL2 and included the 3 patients (2 patients fromGroup A and 1 patient from Group B) also given cilengitide.

Tumor processing and histologic analysesEach patient's surgical resection specimen (obtained following IC

course-1 for Group A and prior to IC course-1 for Group B) wasprocessed by theUWHCSurgical PathologyUnit (Madison,WI). Tworepresentative 1 mm punch cores were taken from intratumoralregions within one representative formalin-fixed paraffin-embedded(FFPE) block of each patient's resected melanoma. FFPE cores werecoded and shipped to NCI (Bethesda, MD) on dry ice.

For histologic assessment, representative sections of each patient'stumor were coded and analyzed by our board-certified anatomicpathologist (E.A. Ranheim) for diagnosis confirmation and TILassessment on hematoxylin and eosin slides that were masked fortreatment. Quantitation of TILs (%) was defined as (cross-sectionalarea of TILs/area of tumor) � 100 (see Supplementary Methods).

RNA extraction and transcriptome sequencingRNAwas extracted from FFPE tumor cores using RNeasy FFPE kits

according to the manufacturer's protocol (Qiagen). RNA-sequencing(RNA-seq) libraries were generated using >50 ng of RNA and TruSeqRNA Access Library Prep Kits according to the manufacturer'sprotocol (TruSeq RNA Exome kits; Illumina) and sequenced onNextSeq500 sequencers using 80bp paired-end sequencing method(Illumina). On average, >50 million aligned sequencing reads were

Translational Relevance

This study further advances our understanding of the clinicaland immunologic activity of a novel immunocytokine and dis-cusses plans for subsequent clinical testing. Our data suggest thatboth immune and tumor cell processes, as measured by RNA-sequencing (RNA-seq) analyses, are associated with OS and RFSwhen evaluated in tumors resected approximately 2 weeks afterstarting the first of three scheduledmonthly courses of hu14.18-IL2immunotherapy. We identify specific immune- and tumorresponse–related molecular pathways associated with improvedoutcome after starting immunotherapy. We provide new insightsinto immunologic processes involved in effective melanomaimmunotherapy as well as prognostic information derived fromthe RNA-seq data that can facilitate patient management. Thesefindings may be utilized for functionally similar forms of immu-notherapy for other malignancies.

RNA-seq of Melanoma after Hu14.18-IL2 Predicts Outcome

AACRJournals.org Clin Cancer Res; 26(13) July 1, 2020 3297

generated from each RNA-seq library (>100� coverage of the codingregions in human genome). Quality control data are shown in Sup-plementary Table ST1. Gene expression level was performed asdescribed previously (11).

Gene signature analysisWe determined the immune and stromal scores for each sample by

using the single-sample gene set enrichment analysis (ssGSEA) and theESTIMATE gene sets (37; Supplementary Tables ST2–ST3) asdescribed previously (11). We also utilized 22 CIBERSORT immunecell–specific gene sets (38) and a cytolytic score (average expression ofGZMA, GZMB, GZMH, GZMK, GZMM, and PRF1) to assess theimmune infiltrates in each sample. These signatures underwent val-idation in3,809TheCancerGenomeAtlas transcriptional profiles (37),3,061 human transcriptomes (38), and a cohort of 150 patients withneuroblastoma (11), respectively.

Regarding tumor-associated biomarkers, a more exploratoryanalysis interrogating the data on an individual gene level wasundertaken based on the catalog of “cancer-related” genes under thelist of protein classes on the Human Protein Atlas website (https://www.proteinatlas.org/). A total of 1,713 cancer-related genes werescreened against OS and RFS with the top pathologically relevant(recognized as a neoplastic molecular biomarker aberration inmolecular pathologic diagnosis) hits listed in Fig. 4 and Supple-mentary Fig. SF15. Molecular pathologic relevance was cross-referenced against the known genomic and epigenomic landscapeof cutaneous melanoma (12, 14).

Statistical analysisRFS and OS were defined in the prior report (36), as detailed in

Supplementary Methods. Gene- and gene signature–level survivalanalyses were performed using the univariate Cox proportional hazardregression with the likelihood ratio test (39). This study assumes theproportional hazards requirement holds. The Cox regression wasimplemented in R version 3.2.3 (https://www.R-project.org/). Esti-mated HRs and P values from the Cox regression model were used forscreening in gene- and gene signature–level survival analyses. Survivaldata were also analyzed using the Kaplan–Meiermethodology with thelog-rank test. Descriptive tumor statistics, TIL data analysis, Kaplan–Meier analysis, correlation analysis, and significance tests were per-formed, and HR plots generated using Prism 8, version 8.01, software(GraphPad). Student t tests, Fisher exact tests, Spearman rank andPearson product-moment correlation coefficients, and correspondingP values were calculated inMicrosoft Excel and results confirmed withGraphPad Prism software. Due to the small sample sizes and explor-atory nature of the statistical inference for the TIL data focusedprimarily on comparing Groups A and B across the same set ofgenes/signatures, we report the nominal P values, rather than P valuescorrected for multiple comparisons. To mitigate the issue of multiplecomparisons, we use a P value of < 0.01 to indicate significance in thisstudy. Given small sample size for Group B (n¼ 8), power calculationsare provided in Supplementary Methods.

Na€�ve tumors refer to an aggregate of resected tumors from 8GroupB patients as well as from 2 never-treated patients (these two wererandomized to Group B, but nonevaluable for outcome as they did notachieve CR following resection).

Network- and pathway-level analysis of genes whoseexpression is correlated with survival

The ConsensusPathDB web-server (http://cpdb.molgen.mpg.de;ref. 40) was used to perform two types of analyses: (i) network-level

analysis and (ii) pathway-level enrichment analysis (41, 42) on geneswhose expression levels are correlated withOS time. These are detailedin Supplementary Methods and data are presented in SupplementaryFig. S16.

Data and materials availabilityThe RNA sequencing data are uploaded on theNCBI's GEOwebsite

with Accession Number GSE133713 and with a release date set to beJuly 1, 2020. (NCBI tracking system no.20118519) The raw sequencingdata have been deposited in the dbGaP under the accession phs001947.The processed RNA-seq data can be visualized and explored at https://clinomics.ccr.cancer.gov/clinomics/public/login using the databasenamed “Melanoma_NCT00590824_CCR.”

ResultsDescription of patient treatment timelines tumor characteristics

The study treatment schematic is found in Supplementary Fig. S1.Specific patient treatment timelines and tumor characteristics arefound in SupplementaryTable ST4. Therewas no significant differencein age, sex distribution, largest tumor diameter, or tumor burdenbetween Groups A and B (Supplementary Table ST5). Group Apatients had tumor resection 12.5 days (mean, range 9–17 days) afterstarting hu14.18-IL2 and Group B patients had their resection28.6 days (mean, range 18–43 days) before starting hu14.18-IL2(Supplementary Table ST6). The median diameter of largest tumorfocus for Groups A and B were 0.7 cm and 1.1 cm, respectively(Supplementary Table ST7). The median estimated tumor burden forGroups A and B were 0.25 cm3 and 0.466 cm3, respectively (Supple-mentary Table ST8). None of these parameters were significantlydifferent between Groups A and B.

TIL levels and stromal signature are prognostic of RFS andOS inGroup A, but not Group B patients

We found a significant prognostic effect of TILs on survival end-points (RFS, P ¼ 0.0088; OS, P ¼ 0.0307) when separated by medianvalues, irrespective of treatment group, with low TILs associated withworse outcomes (Fig. 1A). When stratified by treatment groups,however, these effects were only seen in Group A patients (RFS,P ¼ 0.0030; OS P ¼ 0.0417), but not in Group B patients (Fig. 1Band C). In addition, the stromal signature (SS; refs. 37, 11) demon-strated significant prognostic value for survival endpoints (RFS, P ¼0.0072; OS, P¼ 0.0036) when separated by the median, irrespective oftreatment group, with low SS demonstrating worse outcomes(Fig. 1D). When stratified by treatment group, however, these effectswere seen in Group A patients (RFS, P¼ 0.0082; OS, P¼ 0.0145), butnot in Group B patients (Fig. 1E and F). In addition, these relation-ships are somewhat similar for the cytolytic and immune signature(Supplementary Fig. S2).

We applied Cox regression analyses to assess whether the geneexpression–based signatures provide adjunct prognostic value whencombined with TILs. In this analysis, the immune and SSs werehandled as continuous variables, while TILs were handled as acategorical variable with “low”/”high” categories (as in Fig. 1). In thecase of RFS time, the univariate Cox regression shows that in Group Aboth TILs and the SS are significantly associated with survival (TILs:HR ¼ 0.073 and P ¼ 0.018, SS: HR ¼ 0.097 and P ¼ 0.006). Theimmune signature in this univariate model shows a trend towardstatistical significance (HR ¼ 0.38 and P ¼ 0.051). The bivariate Coxregression that combines TILs and the SS shows that the overall modelis significant (P ¼ 0.01), with both covariates also being significant

Yang et al.

Clin Cancer Res; 26(13) July 1, 2020 CLINICAL CANCER RESEARCH3298

Figure 1.

Morphologic TILs assessment and the stromal gene signature prognosticate RFS and OS after hu14.18-IL2 IC treatment, but not before IC treatment. A–C, Kaplan–Meier curves of patients stratified by TIL assessment are shown. A, Stratification by median TIL assessment is prognostic for RFS and OS within all patients. Patientswith TIL levels below themedian demonstrate worse outcomes.B andC, This prognostic effect on RFS andOS, respectively, is retained in treatment Group A, but notin treatment Group B. D–F, Kaplan–Meier curves of patients stratified by the stromal signature (37). D demonstrates that the stromal signature is also prognostic ofRFS andOSwithin all patients. Patientswith stromal signature below themedian demonstrate worse outcomes. Similarly, E and F show this prognostic effect on RFSand OS, respectively, retained in treatment Group A, but not in treatment Group B. H&E, hematoxylin and eosin.

RNA-seq of Melanoma after Hu14.18-IL2 Predicts Outcome

AACRJournals.org Clin Cancer Res; 26(13) July 1, 2020 3299

(TILs: HR¼ 0.095 and P¼ 0.038, SS: HR¼ 0.067 and P¼ 0.013). ThebivariateCox regression that combines TILs and the immune signatureshows that the overall model is significant (P¼ 0.02), with TILs beingsignificant (HR¼ 0.095 and P¼ 0.031) and the immune signature notsignificant (HR ¼ 0.405 and P ¼ 0.117). Thus, the results of thisanalysis indicate that, in the case of RFS time in Group A, both TILsand the SS are significantly associated with survival, even aftercontrolling for the other variable (i.e., the SS provides an adjunctprognostic value). The results of a similar analysis performed for OStime in Group A indicate that only the SS is significantly associatedwith survival after controlling for the other variable (HR ¼ 0.027and P ¼ 0.04). In Group B, for both OS and RFS time, no significantassociations are observed in either univariate or bivariate Coxregression models.

Positive correlations betweenmyeloid and lymphoid cells genesignatures in Group A (pretreatment) tumors

Spearman correlation matrices of the 25 whole transcriptomics–based immunologic gene signatures were created on treatment-na€�vetumors (n ¼ 10; Supplementary Fig. S3A) as well as on Group Atumors previously treated with hu14.18-IL2 (n ¼ 13; SupplementaryFig. S3B). Significance levels in the form of log10 (1/P) estimates areshown in Supplementary Fig. S3A and S3B. While na€�ve tumorsdemonstrated high levels of correlation between lymphocytic signa-tures (B, T, and NK cells) in Supplementary Fig. S3A, these were lesspronounced correlations than seen with treated tumors betweenlymphocytic and myeloid signatures (monocytes, macrophages, den-dritic cells, eosinophils, and neutrophils) in Supplementary Fig. S3B.Subtraction of these Spearman correlation coefficients betweenhu14.18-IL2 treated and na€�ve tumors demonstrates amarked increase

in correlation of myeloid signatures with lymphoid signatures (Sup-plementary Fig. S3C) in tumors treated with hu14.18-IL2; this impliesthat the hu14.18-IL2 treatment caused parallel changes in lymphoidand myeloid compartments. A comparison using alternative methodsof determining correlation (Spearman and Pearson) confirmed thesefindings (Supplementary Fig. S4C–S4D).

Immunologic gene signatures prognosticate favorable patientsurvival in hu14.18-IL2–treated (Group A) tumors, but not inuntreated (Group B) tumors

The 25 immunologic gene signatures consistently demonstratedfavorable HRs for RFS and OS (Fig. 2A and B). For Group A patients,10 (40%) of these signatures showed statistically significant favorableHRs of RFS (and 10 others showed a trend, 0.01 < P ≤ 0.1); in addition,6 (24%) were statistically significant favorable HRs of OS (and 8 othersshowed a trend, 0.01 < P ≤ 0.1). In contrast, none of these relationshipsbetween gene signatures and survival demonstrated statistical signif-icance (or showed a trend, 0.01 <P≤ 0.1) withinGroup B patients (RFSin Fig. 2A and OS in Fig. 2B). The sample size of Group B is smallerthan Group A (n ¼ 8 vs. n ¼ 13, respectively), but the difference inP values between Groups A and B is large, and thus is unlikely to onlybe attributed to the smaller sample size. Another observation empha-sizing the key difference between Groups A and B independently ofsample size is that in Group A the HRs for all 25 immunologicsignatures in both the OS and RFS analyses are favorable(<1, Fig. 2), whereas in Group B in the OS analysis, the HRs for 20of 25 signatures are unfavorable (>1) and in the RFS analysis the HRsfor 9 of 25 signatures are unfavorable for OS. Thus, the immunologicsignatures do not show significant associations with improved out-come inGroup B because theHRs forGroup B are larger than those for

-5 0 5 10

B.cells_naiveB.cells_memory

T.cells_regulatoryT.cells_follicular_helper

Macrophages_M0T.cells_CD4_memory_activated

T.cells_CD4_naivePlasma_cells

NeutrophilsT.cells_CD8

T.cells_CD4_memory_restingNK.cells_activated

Dendritic_cells_activatedT.cells_gamma_delta

NK.cells_restingMacrophages_M2

Mast_cells_restingEosinophils

Dendritic_cells_restingMast_cells_activated

MonocytesMacrophages_M1

Sign

atur

e

*

*

***

*

********

***

******

*

***

****#

#

-5 -4 -3 -2 -1 0 1 2 3 4 5

T.cells_regulatoryT.cells_CD4_memory_activated

T.cells_follicular_helperB.cells_memory

B.cells_naiveMacrophages_M0

T.cells_gamma_deltaNK.cells_activated

T.cells_CD8Plasma_cells

Dendritic_cells_activatedT.cells_CD4_naive

NK.cells_restingT.cells_CD4_memory_resting

Macrophages_M1Mast_cells_activated

Dendritic_cells_resting

NeutrophilsMast_cells_resting

EosinophilsMonocytes

Macrophages_M2Stromal signature

Sign

atur

e

***

*

**

**

#

**

**#

*

*

***

#

A B-ln of HR RFS Group A vs. B

-ln (HR w/ 95%CI of RFS)

Group A patientsGroup B patients Group A patientsGroup B patients-ln of HR OS Group A vs. B

-ln (HR w/ 95%CI of OS)

Immune signature

Stromal signatureCytolytic.signature

Immune signature

Cytolytic.signature

Figure 2.

Gene signatures prognosticate favorable patient survival in hu14.18-IL2–treated tumors, but not in Group B (untreated) tumors. Tree plots of the 25 immunologicsignatures consistently demonstrated favorable HRs for RFS and OS.A, For Group A patients, 10 (40%) of these signatures showed statistically significant favorableHRs of RFS (and 10 others showed a trend, 0.01 <P ≤0.1) but nonewere significant (or showed a trend) for Group Bpatients.B, For GroupApatients, 6 (24%) of thesesignatures showed statistically significantly favorableHRs ofOS (and 8 others showed a trend, 0.01 <P≤0.1), but nonewere significant (or showed a trend) for GroupB patients. HRs are –ln (natural log) transformed such that a point estimate of 1.0 denotes an HR of 0.368 and a point estimate of �1.0 denotes an HR of 2.72, thuspoints plotted to the right of 0.0 indicate favorable outcome and those to the left indicate unfavorable outcome. The P values are from the likelihood ratio test (itperforms better when sample size is small), whereas the 95% confidence intervals are based on the Wald test (��� , P < 0.001; �� , P < 0.01; � , P < 0.05; #, P < 0.1; CI,confidence interval).

Yang et al.

Clin Cancer Res; 26(13) July 1, 2020 CLINICAL CANCER RESEARCH3300

Group A, not because of a smaller sample size. A more granularrepresentation of the data as a heatmap with patient-level tumor genesignatures is found in Supplementary Fig. S5.

Immunologic gene expression levels demonstrate favorablesurvival and increased significance in Hu14.18-IL2–treated(Group A) tumors, but not in untreated (Group B) tumors

The transcriptomic data were interrogated on a more granularlevel by analyzing individual expression levels of 644 immune-related genes (Supplementary Table ST9) within Groups A and B.These relationships are represented in volcano plots between GroupsA and B for associations with RFS and for OS (Fig. 3). We foundseveral immune-related genes that demonstrated significantly favor-able prognostic values for both OS and RFS in Group A tumors(Fig. 3A and C). In Fig. 3A, there were 59 genes associated with RFSversus 4 genes associated with decreased RFS in Group A patients(Supplementary Tables ST10–ST11). In Fig. 3C, there were 67 genesassociated with OS versus 6 genes associated with decreased OS inGroup A patients (Supplementary Tables ST12–ST13). Interestingly,expression levels of CCR3 (highlighted in red), a chemokine receptoron eosinophils and Th cells, stood out as highly significant andassociated with a very skewed HR, prognostic of favorable RFS andOS in Group A (Fig. 3A and 3C). While not as significant orpotently influencing HR, expression of VNN1 (a membrane proteininfluencing hematopoietic and T-cell trafficking, highlighted in red)is also significantly associated with a favorable HR for both RFS andOS in Group A (Fig. 3A and C). In contrast, we found very fewimmunologic genes that demonstrated statistically significant favor-able prognostic values for OS or RFS in Group B tumors (Fig. 3Band D). Associations between expression levels of individual immu-nologic genes and RFS and OS are presented (SupplementaryFig. S6). Again, these relationships were most robustly seen inGroup A, but not in Group B patients.

We also examined the association of HRwith the gene expression ofclass-based immune molecules and RFS and OS times. Heatmaps ofHR and P values for RFS and OS times for TLRs (Toll-like receptors),KIRs (Killer-like Ig receptors), FcRs (antibody fragment constantregion receptors), TNFSFs (tumor necrosis factor superfamily proteinsand receptors), ILs (interleukins), and immune-related CD (clusters ofdifferentiation) markers are shown in Supplementary Figs. S7–S12,respectively. The above gene sets were taken from existing literature,showing these classes of genes having significant contribution towardantitumor immunity (43). Consistent with our observations describedabove, the gene expression of many of these were more significantlyassociated with improved HR (green) in Group A tumors, but not inGroup B tumors.

Melanoma-related gene expression increases HR andapoptosis-related gene expression decreases HR of RFS andOS in Group A, but not Group B patients

Next, we interrogated transcriptomic data of 1,713 cancer-relatedgenes (found at https://www.proteinatlas.org/search/protein_class:Cancer-relatedþgenes), which may have an influence on survivaltimes. We found a number of clinically relevant melanoma-relatedgenes that demonstrated significantly increased (unfavorable) HRsfor patients whose tumors express these genes (Fig. 4, top) in GroupA patients but not in Group B. These include MLANA (Melan A),SOX10, S100A, S100B, MITF, and PMEL (HMB-45). These markersare used regularly as melanoma markers for clinical pathologicidentification and assessment by IHC. In contrast, apoptosis-relatedgenes [including: BCL10, CASP8 (Caspase 8), CASP10 (Caspase10), FAS, TNFSF10 (TRAIL), and others shown] revealed signifi-cantly decreased (favorable) HR in Group A patients, but not GroupB (Fig. 4, bottom). A more granular, individual patient level,breakdown of these melanoma-related and apoptosis-related genesmay be found in Supplementary Fig. S13.

Figure 3.

Granular immunologic gene expres-sion demonstrates a predominance ofparameters with favorable prognosticvalue rather than unfavorable prog-nostic value. Volcano plots of granularimmunologic gene expression areplotted with HRs against significancelevels (A–D). A and C represent theGroup A patients’ (n ¼ 13) RFSand OS relationships, respectively,against these immunologic genes.B and D represent the Group Bpatients’ (n ¼ 8) RFS and OS relation-ships, respectively, against theseimmunologic genes. Several genesshow significantly favorable HRs (redpoints) inA andC. Among these, CCR3(a chemokine receptor on eosinophilsand Th cells) and VNN1 (a membraneprotein felt to influence hematopoieticand T-cell trafficking, both labeled inred) stood out as highly significant,and associatedwith a very skewedHR,prognostic of favorable RFS and OS inGroup A. The horizontal dotted linerepresents a P value of 0.01. The ver-tical dotted line represents anHRof 1.0.

RNA-seq of Melanoma after Hu14.18-IL2 Predicts Outcome

AACRJournals.org Clin Cancer Res; 26(13) July 1, 2020 3301

Hu14.18-IL2–treated tumors demonstrate increased levels ofgrowth factors and repair, overall gene expression, butdecreased levels of melanoma-related genes

Next, we examined the global gene expression alterations result-ing from one course of Hu14.18-IL2 treatment by comparing genessignificantly different in Group A versus Group B and untreated(Na€�ve) tumors; such individual genes may shed light on hu14.18-IL2 effects on the tumor microenvironment (Fig. 5). There are52 genes (at P < 0.01) and 610 genes (at P < 0.05) that showsignificant fold difference between treated and hu14.18-IL2 na€�vetumors (Supplementary Table ST14). In general, the data demon-strate a marked preponderance of genes that are significantlyincreased in expression in 13 Group A versus 10 na€�ve (8 GroupB and 2 never-treated) tumors, indicating an overall increase ingene expression of many genes after treatment with hu14.18-IL2. A

more granular assessment indicates that Hu14.18-IL2–treatedtumors show increased levels of repair (growth factor and DNAdamage repair) genes and immunologic genes, but decreased levelsof melanoma, Warburg metabolism, and cell cycling–related genes(Supplementary Fig. S14).

Pathway-level analysis reveals overrepresentation of genesinvolved in immunologic pathways in Group A, but not inGroup B

A more integrative pathway-level enrichment analysis performedby using the ConsensusPathDB web-server indicates that the elevatedexpression of genes involved in innate immune response pathways islinked to favorable survival in the hu14.18-IL2–treated tumors (GroupA), but not in the yet untreated tumors (Group B; Fig. 6, green left-sided panels). A more extensive catalogue of 53 pathways associated

0.000.200.400.600.801.001.021.041.061.081.10

Hazard Ra�o

P value Group A OS Group B OS

Gene HR P HR P

AKT3 1.1263 0.0010 0.971 0.235S100A1 1.0040 0.0020 0.999 0.367ERBB3 1.0048 0.0081 0.996 0.232SOX10 1.0339 0.0127 0.970 0.162DDIT3 1.1904 0.0144 0.975 0.553

CDKN1A 1.0458 0.0265 0.978 0.304S100B 1.0015 0.0447 1.000 0.824MDM2 1.0109 0.0496 0.996 0.720

CTNNB1 1.0017 0.0654 1.000 0.668CSPG4 1.0062 0.0697 0.984 0.180BCL2 2.0287 0.0817 0.758 0.321

MLANA 1.0023 0.0844 1.000 0.901TYR 1.0007 0.1002 1.000 0.861

Group A RFS Group B RFS

Gene HR P HR P

CDKN1A 1.1207 0.00023 1.006 0.590ERBB3

Melanomagenes andoncogenes

Apoptoticgenes

1.0045 0.00244 0.999 0.799AKT3 1.0753 0.00267 1.004 0.835

MLANA 1.0033 0.00536 1.000 0.894CDK4 1.0140 0.00965 1.014 0.167SOX10 1.0224 0.01198 1.001 0.971PMEL 1.0004 0.01668 1.000 0.420MITF 1.0087 0.01823 0.997 0.463TYR 1.0009 0.01975 1.000 0.810

CTNNB1 1.0015 0.02009 1.000 0.622S100A1 1.0011 0.04177 1.000 0.897MDM2 1.0091 0.04591 1.010 0.269DDIT3 1.1172 0.04620 1.026 0.415

Group A OS Group B OS

Gene HR P HR P

TNFSF10 0.7242 0.0001 0.993 0.899

BCL10 0.7042 0.0010 0.982 0.912

FAS 0.9263 0.0022 1.010 0.496

CASP8 0.9508 0.0077 1.009 0.550

CASP10 0.8023 0.0109 1.001 0.987

FASLG 0.6364 0.0136 1.048 0.783

TNFRSF1B 0.9003 0.0477 1.042 0.355

TNFRSF10A 0.9230 0.0656 1.033 0.297

TNFRSF8 0.2719 0.1192 1.663 0.297

CASP5 0.6852 0.1379 0.839 0.502

Group A RFS Group B RFS

Gene HR P HR P

TNFSF10 0.9036 0.00068 0.973 0.522

BCL10 0.7708 0.00136 0.898 0.532

CASP10 0.8423 0.00302 0.959 0.505

TNFRSF1B 0.8897 0.01133 0.997 0.940

TNFRSF10A 0.8885 0.01344 1.001 0.980

FAS 0.9615 0.02171 1.007 0.550

CASP8 0.9745 0.03980 0.994 0.650

TNFRSF8 0.2443 0.04366 0.958 0.912

TNFRSF1A 0.9858 0.05691 1.033 0.161

FASLG 0.8337 0.08202 0.919 0.561

0.10.090.080.070.060.050.040.030.020.01

0.0050.0020.001

Figure 4.

Expression of melanoma markers and oncogenes portends unfavorable RFS and OS in Group A but not Group B. Expression of apoptotic marker genesportends favorable RFS and OS in Group A but not Group B. Gene expression level relationships between melanoma-related (top) and apoptosis-related(bottom) expression and RFS and OS are shown in heatmap form. Melanoma-related genes were selected from molecular markers routinely used in diagnosticanatomic pathology. The green to red legend demonstrates either favorable prognosis (HR <1, green) or unfavorable prognosis (HR >1, red), respectively. Theblue coloring represents level of significance, starting with light blue at P ¼ 0.10 and becoming deeper blue with increasing significance. These melanoma andapoptotic parameters tend to prognosticate unfavorable or favorable survival, respectively, in Group A more than they do in Group B.

Yang et al.

Clin Cancer Res; 26(13) July 1, 2020 CLINICAL CANCER RESEARCH3302

with favorable OS (HR < 1) also demonstrated innate immunologicprocesses (Supplementary Fig. S15). Conversely, elevated expression ofgenes involved in glycolysis pathways was linked to unfavorablesurvival in the hu14.18-IL2–treated tumors (Group A), but not inthe treatment na€�ve tumors (Group B) (Fig. 6, pink right-sided panels).In addition, the elevated expression of genes involved in deubiquitina-tion and processing of ubiquitinated proteins appears to be linked tounfavorable survival in Group B (Fig. 6, pink right-sided panels), butnot in Group A.

Network-level analysis of genes correlated with the OS time inGroup A indicates that ELAVL1 is a hub protein that interactswith the products of many of these genes

Intriguingly, the network-level analysis of genes whose expression iscorrelated with the OS time in Group A revealed a complex network ofphysical interactions (Supplementary Fig. S16A) in which ELAVL1 is ahub protein that interacts with the products of 25 of these genes. Themajority of these genes (20 of 25) were genes whose expressionpositively correlated with the OS time for Group A in this study. Incontrast to Group A, the network-level analysis of genes whoseexpression was correlated with the OS time in Group B produced avery limited network of physical interactions that did not includeELAVL1 and did not contain any hub proteins (SupplementaryFig. S16B). This implies that hu14.18-IL2 treatment may result in acoordinated immune response centered with ELAVL1 which mayimpact on patient's outcome.

DiscussionWe analyzed histology and tumor mRNA whole transcriptomes

from stage III and IV patients with melanoma to screen forpredictive biomarkers for patients participating in this hu14.18-IL2 immunotherapy trial. Patients were scheduled to receive threecourses of hu14.18-IL2 and were randomized to have course 1 ofthe hu14.18-IL2 begin just before (Group A) or just after (GroupB) surgical resection of all disease. No significant differences inRFS or OS between Groups A and B were seen in the OS (36).However, when evaluating tumor immune microenvironmentusing immune scores generated from ssGSEA (11), we found thatTIL levels, as measured by histologic assessment, strongly anddirectly correlate with immune scores for all patients. Patients’TIL levels in their resected tumors directly correlated with RFSand OS in the tumors from Group A patients who had been

Figure 5.

IC treatment reveals a predominance of increased rather than deceased geneexpression. A, Gene expression for individual genes are plotted with log2 foldchanges fromna€�ve [8 inGroupBand 2untreated (No IC)] patients to 13GroupAhu14.18-IL2–treated patients (y-axis) against significance levels (x-axis). There isa marked preponderance of genes that are significantly increased (top rightquadrant), indicating an overall increase in gene expression of transcripts aftertreatmentwith hu14.18-IL2.B,Gene expression plot for individual cancer-relatedgenes (annotated from the Human Protein Atlas) with log2 fold changes from 10na€�ve patients to 13 Group A patients demonstrates a preponderance of

significantly increased (top right quadrant), but not decreased, expression ofgenes in Group A versus na€�ve tumors. C, Gene expression plot for individualssGSEA CIBERSORT immune genes with log2 fold changes from 10 na€�vepatients to 13 Group A patients demonstrates a preponderance of significantlyincreased, but not decreased, expression of genes in Group A versus na€�vetumors. There are 52 genes (at P < 0.01) and 610 genes (at P < 0.05) thatshow significant fold difference between treated and hu14.18-IL2 na€�vetumors. The vast majority of these were upregulated rather than down-regulated. Fold change is log2 transformed such that a point estimate of 1.0denotes a fold change of 2.0 and a point estimate of �1.0 denotes a foldchange of 0.5. In A, B, and C, the vertical dotted lines represent P values of0.05 and 0.01. The horizontal dotted line represents a ratio of 1.0 forexpression of the individual gene in the tumors from 13 Group A patientsversus in the 10 na€�ve tumors. Immunologic genes were taken from all645 immunologically related genes that are included in our ssGSEA analysis.For A–C, to the right of the dot plots for gene expression ratio versusP value are box plots and density histograms of those points that aresignificant (P < 0.01), together with those that have 0.01 < P < 0.05.

RNA-seq of Melanoma after Hu14.18-IL2 Predicts Outcome

AACRJournals.org Clin Cancer Res; 26(13) July 1, 2020 3303

treated with hu14.18-IL2 prior to resection but no such correlationwas seen in Group B patients who had not yet received hu14.18-IL2 prior to resection (Group B). In addition, SS is associated withimproved HR of RFS and OS, in tumors that were resected after thefirst course of hu14.18-IL2 treatment (Group A), but not observedin tumors that were resected before initiating the hu14.18-IL2(Group B).

We found a number of other immune signatures associated withdecreased (favorable) HR of RFS and OS for Group A patients, butthese findings were not found in Group B tumors. Specifically,several individual immune gene transcripts were associated withdecreased HR of RFS and OS for Group A patients (but not forGroup B patients). Among these transcripts associated withimproved outcome for Group A patients were those related tocytotoxic T cells (Granzymes), NK cells (KIRs), and innate immunecells (TLRs). In addition, there appeared to be a number of tumor-related genes where increased gene expression was significantlyassociated with increased HR of RFS/OS (melanoma marker tran-scripts) or decreased HR of survival (apoptosis transcripts) inGroup A patients; no similar associations were seen for GroupB. These findings suggest that the associations with outcome forthese tumor transcripts were the result of changes in transcriptexpression induced by the hu14.18-IL2 therapy that were associated

with RFS and OS. The inability of the immunotherapeutic inter-vention to induce these changes in a portion of the evaluablepatients in Group A strongly predicted poorer clinical outcomesin these patients.

In addition, the data revealed that compared with na€�ve (untreat-ed) tumors, hu14.18-IL2–treated tumors (Group A) demonstrateddecreased levels of cell-cycling transcripts (MKI67, CDK1), mela-noma markers [MLANA (Melan A), MAGED1],and glycolysistranscripts [LDHA, HK1 (Supplementary Fig. S14)]. Conversely,Group A tumors demonstrated increased levels of growth factor[PDGFRA, FGFR1, PDGFB, TGFB3 (Supplementary Figs. S14)] andrepair transcripts (XPC) compared with na€�ve tumors. We takerelationships between tumor-related transcripts as evidence ofdifferences in tumor biology (or differences in tumor biology dueto immune effects) that may indicate a change from a tumorreplicative phenotype to an inflamed/healing-wound phenotypefollowing hu14.18-IL2 therapy.

When evaluating Group A tumors, we found that expression of anumber of individual genes were strongly and significantly associatedwith outcome (Fig. 3; Supplementary Figs. S7–S12). Some of thesemayprovide clues to mechanisms involved in improved outcome usinghu14.18-IL2 like therapy and could potentially indicate additionalmolecules/pathways for targeting effective immunotherapy. We

q-value Log (1 / q-value) Pathway Source q-value Log (1 /

q-value) Pathway Source

Group A pa�ents, HR < 1, increased overall survival (OS) Group A pa�ents, HR > 1, decreased overall survival (OS)0.000031 4.506 Innate immune system Reactome 0.015880 1.799 Ac�va�on of SMO Reactome0.000188 3.726 Toll-like receptors cascades Reactome 0.016262 1.789 Ectoderm differen�a�on Wikipathways0.000391 3.407 Toll-like receptor 4 (TLR4) cascade Reactome 0.019784 1.704 Asymmetric localiza�on of PCP proteins Reactome

Group A pa�ents, HR < 1, increased overall survival (RFS) Group A pa�ents, HR > 1, decreased overall survival (RFS)

0.020 1.691 Platelet-mediated interac�ons with vascular and circula�ng cells Wikipathways 0.014 1.859 SREBF and miR33 in cholesterol and lipid homeostasis Wikipathways

0.020 1.691 Eicosanoid ligand-binding receptors Reactome 0.014 1.859 Glycolysis Reactome0.031 1.508 Class A/1 (Rhodopsin-like receptors) Reactome 0.014 1.844 Glycolysis pathway D (2) Wikipathways

0.031 1.508 Immunoregulatory interac�ons between a Lymphoid and a non-Lymphoid cell Reactome 0.014 1.844 Glucose metabolism Reactome

0.037 1.429 GPCRs, Class A Rhodopsin-like Wikipathways 0.019 1.715 Factors and pathways affec�ng insulin-like growth factor (IGF1)-Akt signaling Wikipathways

0.037 1.429 Hematopoie�c stem cell differen�a�on Wikipathways 0.021 1.672 Carboxyterminal pos�ransi�onal modifica�ons of tubulin Reactome

0.037 1.429 Splicing factor NOVA-regulated synap�c proteins Wikipathways 0.021 1.672 Metabolism of steroids Reactome0.037 1.429 Chemokine receptors bind chemokines Reactome 0.022 1.651 Intraflagellar transport Reactome0.037 1.429 IL-3 Signaling pathway Wikipathways 0.023 1.633 Glycolysis and gluconeogenesis Wikipathways

0.037 1.429 Protein alkyla�on leading to liver fibrosis Wikipathways 0.023 1.633 Transcrip�onal regula�on of white adipocyte differen�a�on Reactome

0.037 1.429 GPCR ligand binding Reactome 0.024 1.611 Sudden infant death syndrome (SIDS) suscep�bility Pathways Wikipathways

0.037 1.429 Photodynamic therapy-induced AP-1 survival signaling. Wikipathways

Group B pa�ents, HR < 1, increased overall survival (OS) Group B pa�ents, HR > 1, decreased overall survival (OS)0.007074 2.150 tRNA processing Reactome 0.014434 1.841 Ub-specific processing proteases Reactome0.013113 1.882 Focal adhesion Wikipathways 0.014434 1.841 Deubiqui�na�on Reactome

0.018355 1.736 Focal adhesion -PI3K-Akt-mTOR-signaling pathway Wikipathways

q-value Log10 (1 / q-value)

0.100000 1.00.031623 1.50.010000 2.00.003162 2.50.001000 3.00.000316 3.50.000100 4.00.000032 4.50.000010 5.0

Figure 6.

Pathway-level enrichment analysis of gene sets whose expression correlates with survival. The pathway-level enrichment analysis indicates that genes involved ininnate immune response are significantly overrepresented in patients with improvedOS (HR < 1) in the IC-treated tumors (n¼ 13, Group A, green top left panels), butnot in the untreated tumors (n ¼ 8, Group B). A complete list of 53 pathways significantly associated with favorable OS in Group A patients may be found inSupplementary Fig. S15. In contrast, only a few pathways including ectodermal differentiation and activation of SMO were found to associate with Group A patientswith decreased OS (pink top right panel). Pathways associated with improved RFS (HR < 1) in Group A patients include chemokine and cytokine signaling (greenmiddle left panels). In contrast, pathways associatedwith unfavorable RFS (HR > 1) in Group A patients include glycolysis and other glucosemetabolism (pinkmiddleright panels). In Group B patients, pathways associated with focal adhesion associated with favorable OS in Group B (green bottom left panel). In contrast, genesinvolved in deubiquitination and processing of ubiquitinated proteins are overrepresented in patients with unfavorable OS (HR > 1) in Group B (pink bottom rightpanels). Pathways are ranked by their log (1/q-value) values and colored blue when the q-value <0.05, with darkening blue representing increasing statisticalsignificance. The q-value is the FDR-adjusted P value and provides a correction for multiple tests (i.e., takes into account the total number of pathways used in theanalysis).The q-value is commonly used in genome-wide studies.

Yang et al.

Clin Cancer Res; 26(13) July 1, 2020 CLINICAL CANCER RESEARCH3304

believe that the differences in transcript expression reflect changes intumor-biology induced by the antitumor immune effects of hu14.18-IL2 treatment that was initiated only 12.5 days (mean) prior to thetumor resection for Group A. These changes in tumor biology aresomewhat similar to that seen after a subacute (�2 weeks prior)injurious event (44).

Recently, an immuno-predictive score (IMPRES), a predictor ofimmune checkpoint blockade (ICB) response in melanoma whichencompasses 15 pairwise transcriptomics parameters betweenimmune checkpoint genes, has been described (45). The authorsdeveloped this gene expression predictor to indicate whethermelanoma in a specific patient is likely to respond to treatmentwith ICB. We have done an analysis of IMPRES scores for these 13Group A and 8 Group B patients; all patients have somewhatsimilar scores (ranging from 9 to 13). We see no statisticalassociation of RFS or OS with IMPRES score within Group A orwithin Group B (data not shown). In contrast, we demonstrate theutility of the immune score as a predictor for response to hu14.18-IL2 when evaluated after treatment in this study of hu14.18-IL2.Checkpoint blockade is considered a mechanism to “release thebrakes” on an immune response that is already underway. Thus,one would hypothesize that pretreatment scores indicating preex-isting immune-activation (like the IMPRES) might be associatedwith clinical benefit from checkpoint blockade. In contrast, theinduction of innate immune recognition by antibody-dependentcell-mediated cytotoxicity, together with IL2 induced activation-proliferation of immune cells, as shown for hu14.18-IL2 (46),might be considered an approach toward “starting the engine andproviding some gas” to the immune response (47). From thisperspective, one might expect that increased expression of immunegenes seen following treatment initiation with this form of therapy,rather than gene expression seen before treatment, would be abetter predictor of outcome, as observed in this report.

Finally, a network level analysis of genes associated with OS inthis study revealed a complex interaction of many moleculesassociated with outcome for Group A (Supplementary Fig. S16A),with no similar complex network identified for Group B (Supple-mentary Fig. S16B). The network for Group A indicates a keyposition for ELAVL1, which is known to bind 30-UTR regions ofmRNAs with A-U rich elements and stabilize them, thus augment-ing gene expression (48). ELAVL1 expression is known to beassociated with better prognosis in stage II melanoma, increasedC-reactive protein and inflammation, increased TNF and TLR4, aswell as regulation of GATA-3 and Th2 cytokine genes (49). Thesefunctions may put ELAVL1 in a key position for further analysesand possible targeting.

In summary, these data suggest that both immune and tumor cellprocesses, as measured by RNA-seq analyses, are associated with OSand RFS when evaluated in tumors resected approximately 2 weeksafter starting the first of three scheduled monthly courses of hu14.18-IL2 immunotherapy. These resultsmay help point to specific immune-related and tumor response–related molecular pathways associatedwith improved outcome after immunotherapy. Furthermore, if thisapproach is validated, it may enable analyses of RNA-seq immune and

tumor cell signatures obtained after initiating immune therapy topotentially predict which patients should continue their currentregimen and which might need a change or addition to their therapy,due to the absence of a favorable histologic or RNA-seq response.

Disclosure of Potential Conflicts of InterestZ.S. Morris is an employee/paid consultant for Archeus Technologies and

Seneca Therapeutics, reports receiving speakers bureau honoraria from ViewRay,and holds ownership interest (including patents) in WARF, Archeus Technol-ogies, and Seneca Therapeutics. No potential conflicts of interest were disclosed bythe other authors.

Authors’ ContributionsConception and design: R.K. Yang, E.A. Ranheim, J.A. Hank, Z.S. Morris, J. Khan,P.M. SondelDevelopment of methodology: R.K. Yang, I.B. Kuznetsov, E.A. Ranheim, J.S. Wei,Y.K. Song, Z.S. Morris, J. KhanAcquisition of data (provided animals, acquired and managed patients, providedfacilities, etc.): R.K. Yang, Y.K. Song, J.A. Hank, M.R. Albertini, J. Khan, P.M. SondelAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): R.K. Yang, I.B. Kuznetsov, E.A. Ranheim, J.S. Wei,S. Sindiri, B.E. Gryder, V. Gangalapudi, V. Patel, K. Kim, J. Khan, P.M. SondelWriting, review, and/or revision of the manuscript: R.K. Yang, I.B. Kuznetsov,J.S. Wei, S. Sindiri, V. Patel, J.A. Hank, C. Zuleger, A.K. Erbe, Z.S. Morris, K. Kim,M.R. Albertini, J. Khan, P.M. SondelAdministrative, technical, or material support (i.e., reporting or organizing data,constructing databases): R.K. Yang, J.S. Wei, S. Sindiri, R. Quale, J. Khan,P.M. SondelStudy supervision: E.A. Ranheim, J. Khan, P.M. SondelOther (hypothesis generation for RNA-seq): I.B. Kuznetsov

AcknowledgmentsThe authors thank the UWCCC Melanoma clinical research and laboratory

support team for the clinical research involved. They acknowledge contributions bythe surgical oncologists participating in this study, including Drs. Sharon Weber,Heather Neuman, Greg Hartig, Tracey Weigel, and David Mahvi, and also contribu-tions by Mary Beth Henry, NP. This study was facilitated by the high-performancecomputational capabilities of the Biowulf Linux cluster at the National Institutes ofHealth (https://hpc.nih.gov/docs/userguide.html), and was enabled by funding fromthe Division of Intramural Research Program of the National Institutes of Health,National Cancer Institute, Center for Cancer Research. This work was supported byNIH grants CA032685, CA166105, CA197078 (to P.M. Sondel), GM067386,Clinical and Translational Science Award 1TL1RR025013-01 to the Universityof Wisconsin Institute for Clinical and Translational Research (to E.A. Ranheim,Z.S. Morris, M.R. Albertini, K. Kim, and P.M. Sondel), Cancer Center SupportGrant, P30 CA014520, to the University of Wisconsin Carbone Cancer Center (toE.A. Ranheim, Z.S. Morris, M.R. Albertini, J.A. Hank, K. Kim, and P.M. Sondel),The Midwest Athletes for Childhood Cancer Fund (to J.A. Hank and P.M.Sondel), Ann's Hope Foundation (to M.R. Albertini), the Tim Eagle Memorial(to M.R. Albertini), the Jay Van Sloan Memorial from the Steve Leuthold Family(to M.R. Albertini), and The Stand Up To Cancer – St. Baldrick's Pediatric DreamTeam Translational Research Grant (SU2C-AACR-DT1113). Stand Up To Canceris a division of the Entertainment Industry Foundation administered by theAmerican Association for Cancer Research.

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

Received October 7, 2019; revised January 24, 2020; accepted March 4, 2020;published first March 9, 2020.

References1. Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SA, Behjati S, Biankin

AV, et al. Signatures of mutational processes in human cancer. Nature 2013;500:415–21.

2. LawrenceMS, Stojanov P, Polak P, KryukovGV,Cibulskis K, SivachenkoA, et al.Mutational heterogeneity in cancer and the search for new cancer-associatedgenes. Nature 2013;499:214–18.

RNA-seq of Melanoma after Hu14.18-IL2 Predicts Outcome

AACRJournals.org Clin Cancer Res; 26(13) July 1, 2020 3305

3. Lee N, Zakka LR, MihmMC Jr, Schatton T. Tumour-infiltrating lymphocytes inmelanoma prognosis and cancer immunotherapy. Pathology 2016;48:177–87.

4. Saldanha G, Flatman K, Teo KW, BamfordM. A novel numerical scoring systemfor melanoma tumor-infiltrating lymphocytes has better prognostic value thanstandard scoring. Am J Surg Pathol 2017;41:906–14.

5. Albertini MR. The age of enlightenment in melanoma immunotherapy.J Immunother Cancer 2018;6:80.

6. Wolchok JD, KlugerH, CallahanMK, PostowMA, Rizvi NA, LesokhinAM, et al.Nivolumab plus ipilimumab in advanced melanoma. N Engl J Med 2013;369:122–33.

7. Larkin J, Chiarion-Sileni V, Gonzalez R, Grob JJ, Cowey CL, Lao CD, et al.Combined nivolumab and Ipilimumab ormonotherapy in untreatedmelanoma.N Engl J Med 2015;373:23–34.

8. SchadendorfD,Hodi FS, Robert C,Weber JS,Margolin K,HamidO, et al. Pooledanalysis of long-term survival data from phase II and phase III trials ofipilimumab in unresectable or metastatic melanoma. J Clin Oncol 2015;33:1889–94.

9. Andrews MC, Reuben A, Gopalakrishnan V, Wargo JA. Concepts collide:genomic, immune, and microbial influences on the tumor microenvironmentand response to cancer therapy. Front Immunol 2018;9:946.

10. ChenPL, RohW,ReubenA, Cooper ZA, Spencer CN, Prieto PA, et al. Analysis ofimmune signatures in longitudinal tumor samples yields insight into biomarkersof response and mechanisms of resistance to immune checkpoint blockade.Cancer Discov 2016;6:827–37.

11. Wei JS, Kuznetsov IB, Zhang S, Song YK, Asgharzadeh S, Sindiri S, et al.Clinically relevant cytotoxic immune cell signatures and clonal expansion ofT cell receptors in high-risk MYCN-not-amplified human neuroblastoma.Clin Cancer Res 2018;24:5673–84.

12. Dahl C, Guldberg P. The genome and epigenome of malignant melanoma.APMIS 2007;115:1161–76.

13. Ibrahim N, Haluska FG. Molecular pathogenesis of cutaneous melanocyticneoplasms. Annu Rev Pathol 2009.4:551–79.

14. Zhang T, Dutton-Regester K, Brown KM, Hayward NK. The genomic landscapeof cutaneous melanoma. Pigment Cell Melanoma Res 2016;29:266–83.

15. Spranger S, Bao R, Gajewski TF. Melanoma-intrinsic b-catenin signallingprevents anti-tumour immunity. Nature 2015;523:231–5.

16. Hu-Lieskovan S, Homet Moreno B, Ribas A. Excluding T cells: is b-catenin thefull story? Cancer Cell 2015;27:749–50.

17. Gonzalez-Cao M, Mayo de Las Casas C, Jordana Ariza N, Manzano JL, Molina-

Vila M�A, Soriano V, et al. Spanish Melanoma Group. Early evolution ofBRAFV600 status in the blood of melanoma patients correlates with clinicaloutcome and identifies patients refractory to therapy. Melanoma Res 2018;28:195–203.

18. Pag�es F, Mlecnik B, Marliot F, Bindea G, Ou FS, Bifulco C, et al. Internationalvalidation of the consensus immunoscore for the classification of colon cancer: aprognostic and accuracy study. Lancet 2018;391:2128–139.

19. Gillies SD, Reilly EB, Lo KM, Reisfeld RA. Antibody-targeted interleukin 2stimulates T-cell killing of autologous tumor cells. ProcNatl Acad SciU SA1992;89:1428–32.

20. Reisfeld RA, Gillies SD. Antibody-interleukin 2 fusion proteins: a new approachto cancer therapy. J Clin Lab Anal 1996;10:160–6.

21. Mujoo K, Cheresh DA, Yang HM, Reisfeld RA. Disialoganglioside GD2 onhuman neuroblastoma cells: target antigen for monoclonal antibody-mediatedcytolysis and suppression of tumor growth. Cancer Res 1987;47:1098–104.

22. Saleh MN, Khazaeli MB, Wheeler RH, Dropcho E, Liu T, Urist M, et al. Phase Itrial of the murine monoclonal anti-GD2 antibody 14G2a in metastatic mela-noma. Cancer Res 1992;52:4342–7.

23. Hank JA, Surfus JE, Gan J, Jaeger P, Gillies SD, Reisfeld RA, et al. Activationof human effector cells by a tumor reactive recombinant anti-gangliosideGD2 interleukin-2 fusion protein (ch14.18-IL2). Clin Cancer Res 1996;2:1951–9.

24. Lode HN, Xiang R, Varki NM, Dolman CS, Gillies SD, Reisfeld RA. Targetedinterleukin-2 therapy for spontaneous neuroblastoma metastases to bone mar-row. J Natl Cancer Inst 1997;89:1586–94.

25. ImbodenM,Murphy KR, Rakhmilevich AL, Neal ZC, Xiang R, Reisfeld RA, et al.The level of MHC class I expression on murine adenocarcinoma can change theantitumor effector mechanism of immunocytokine therapy. Cancer Res 2001;61:1500–7.

26. Becker JC, Pancook JD, Gillies SD, Furukawa K, Reisfeld RA. T cell-mediatederadication of murine metastatic melanoma induced by targeted interleukin 2therapy. J Exp Med 1996;183:2361–6.

27. Yang RK, Kalogriopoulos NA, Rakhmilevich AL, Ranheim EA, Seo S, Kim K,et al. Intratumoral hu14.18-IL-2 (IC) induces local and systemic antitumoreffects that involve both activated T and NK cells as well as enhanced ICretention. J Immunol 2012;189:2656–64.

28. Morris ZS, Guy EI, Francis DM, Gressett MM, Carmichael LL, Yang RK, et al.In situ tumor vaccination by combining local radiation and tumor-specificantibody or immunocytokine treatments. Cancer Res 2016;76:3929–41.

29. Yang RK, Kalogriopoulos NA, Rakhmilevich AL, Ranheim EA, Seo S, Kim K,et al. Intratumoral treatment of smaller mouse neuroblastoma tumors with arecombinant protein consisting of IL-2 linked to the hu14.18 antibody increasesintratumoral CD8þ T and NK cells and improves survival. Cancer ImmunolImmunother 2013;62:1303–13.

30. Neal ZC, Yang JC, Rakhmilevich AL, Buhtoiarov IN, LumHE, ImbodenM, et al.Enhanced activity of hu14.18-IL2 immunocytokine against murine NXS2neuroblastoma when combined with interleukin 2 therapy. Clin Cancer Res2004;10:4839–47.

31. KingDM,AlbertiniMR, SchalchH,Hank JA,Gan J, Surfus J, et al. Phase I clinicaltrial of the immunocytokine EMD 273063 in melanoma patients. J Clin Oncol2004;22:4463–73.

32. Osenga KL, Hank JA, Albertini MR, Gan J, Sternberg AG, Eickhoff J, et al. Aphase I clinical trial of the hu14.18-IL2 (EMD273063) as a treatment for childrenwith refractory or recurrent neuroblastoma and melanoma: a study of theChildren's Oncology Group. Clin Cancer Res 2006;12:1750–9.

33. Shusterman S, London WB, Gillies SD, Hank JA, Voss SD, Seeger RC, et al.Antitumor activity of hu14.18-IL2 in patients with relapsed/refractory neuro-blastoma: aChildren'sOncologyGroup (COG)phase II study. J ClinOncol 2010;28:4969–75.

34. Yang RK, Sondel PM. Anti-GD2 strategy in the treatment of neuroblastoma.Drugs Future 2010;35:665.

35. Albertini MR, Hank JA, Gadbaw B, Kostlevy J, Haldeman J, Schalch H, et al.Phase II trial of hu14.18-IL2 for patients with metastatic melanoma.Cancer Immunol Immunother 2012;61:2261–71.

36. Albertini MR, Yang RK, Ranheim EA, Hank JA, Zuleger CL,Weber S, et al. Pilottrial of the hu14.18-IL2 immunocytokine in patients with completely resectablerecurrent stage III or stage IV melanoma. Cancer Immunol Immunother 2018;67:1647–58.

37. Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-GarciaW, et al. Inferring tumour purity and stromal and immune cell admixture fromexpression data. Nat Commun 2013;4:2612.

38. Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y, et al. Robustenumeration of cell subsets from tissue expression profiles. Nat Methods 2015;12:453–7.

39. Zhou M, Zhao H, Wang Z, Cheng L, Yang L, Shi H, et al. Identification andvalidation of potential prognostic lncRNA biomarkers for predicting survival inpatients with multiple myeloma. J Exp Clin Cancer Res 2015;34:102.

40. Herwig R, Hardt C, Lienhard M, Kamburov A. Analyzing and interpretinggenome data at the network level with ConsensusPathDB. Nat Protoc 2016, 11:1889–907.

41. Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P, et al.The reactome pathway knowledgebase. Nucleic Acids Res 2018;46:D649–55.

42. Slenter DN, Kutmon M, Hanspers K, Riutta A, Windsor J, Nunes N, et al.WikiPathways: amultifaceted pathway database bridgingmetabolomics to otheromics research. Nucleic Acids Res 2018;46:D661–67.

43. Johansson M, Denardo DG, Coussens LM. Polarized immune responses differ-entially regulate cancer development. Immunol Rev 2008;222:145–54.

44. Gurtner GC, Werner S, Barrandon Y, Longaker MT. Wound repair andregeneration. Nature 2008;453:314–21.

45. Auslander N, Zhang G, Lee JS, Frederick DT, Miao B, Moll T, et al. Robustprediction of response to immune checkpoint blockade therapy in metastaticmelanoma. Nat Med 2018;24:1545–49.

46. Sondel PM, Gillies SD. Current and potential uses of immunocytokines as cancerimmunotherapy. Antibodies 2012;1:149–71.

47. Tirapu I, Mazzolini G, Rodriguez-Calvillo M, Arina A, Palencia B, Gabari I, et al.Effective tumor immunotherapy: start the engine, release the brakes, step on thegas pedal, and get ready to face autoimmunity. Arch Immunol Ther Exp 2002;50:13–8.

48. Takeuchi O. HuR keeps interferon-b mRNA stable. Eur J Immunol 2015;45:1296–9.

49. Stellato C, Gubin MM, Magee JD, Fang X, Fan J, Tartar DM, et al. Coordinateregulation of GATA-3 and Th2 cytokine gene expression by the RNA-bindingprotein HuR. J Immunol 2011;187:441–9.

Clin Cancer Res; 26(13) July 1, 2020 CLINICAL CANCER RESEARCH3306

Yang et al.