Combined MEK/PD-L1 Inhibition Alters Peripheral Cytokines and Lymphocyte Populations Correlating with Improved Clinical Outcomes in Advanced Biliary Tract Cancer

Biliary tract cancers (BTC) are a rare group of aggressive malignancies that include gallbladder cancer (GBC), intrahepatic cholangiocarcinoma (ICC), and extrahepatic cholangiocarcinoma (ECC). BTCs collectively have a 5-year median survival rate of less than 10% and are typically diagnosed at advanced stage when curative surgical resection is not possible. These tumors are also refractory to most systemic treatment options. Precision medicine approaches in subsets of patients with tumors harboring FGFR2 fusions or IDH1/2 mutations offer promise, but other targeted therapies have not shown substantial benefit over chemotherapy (1–5). In considering pathways consistently altered in BTC, constitutive activation of Ras/Raf/MEK/ERK signaling is frequent in tumors from patients with BTC (6). Activating KRAS mutations are present in approximately 22% of BTC cases, associating with poor prognosis and uncontrolled cell growth (7). This has prompted investigations with inhibitors of MEK, which limit disease progression as single agents in BTC and have been tested in combination with immunotherapy in patients with other solid tumors, including melanoma and breast cancer (6–11).

Programmed cell death ligand 1 (PD-L1) is expressed on tumor cells heterogeneously among patients with BTC, interacting with its receptor PD-1 to possibly limit cytotoxic T-cell function and antitumor immunity (12). Immune checkpoint inhibitor (ICI) therapies including PD-L1 blockade have limited activity historically against BTC (13, 14), but recent data from the phase III TOPAZ-1 trial show that adding durvalumab, a PD-L1–targeted antibody, to standard of care gemcitabine and cisplatin improves overall survival (OS) in the first-line for patients with advanced BTC (15). As a result, immunotherapy has garnered major attention in the context of BTC, and further investigation of systemic immune features in these patients can inform future therapeutic strategies.

Previous preclinical studies demonstrate that MEK inhibition (MEKi) can enhance CD8⁺ T-cell infiltration into tumors, while PD-L1 blockade invigorates CD8⁺ T-cell mediated antitumor activity. Furthermore, MEKi synergizes with PD-L1 blockade to improve antitumor responses in several solid tumor models (16, 17). Building off these studies, we recently published results from a clinical trial investigating atezolizumab (anti–PD-L1) with or without the MEK inhibitor cobimetinib in patients with advanced BTC (NCT03201458). This trial met its primary endpoint of improved progression-free survival (PFS), with a median PFS of 3.65 months in combination-treated patients, versus 1.87 months in patients receiving atezolizumab monotherapy (18).

In this study, we address the hypothesis that dual MEK/PD-L1 inhibition alters circulating soluble and cellular immune mediators to improve clinical outcomes in advanced BTC. Using a unique collection of peripheral blood samples from patients with advanced BTC undergoing therapy with atezolizumab with or without cobimetinib, we provide a comprehensive analysis of immune features. Our data encompasses cytokines, chemokines, and phenotypically defined immune cell subsets from a large cohort of patients, informing our understanding of how dual MEK/PD-L1 inhibition impacts immune markers during treatment. Finally, using baseline blood samples from patients and a cohort of healthy adult donors, we identify novel immune features prominent in patients with BTC, suggesting potential avenues for future investigations of therapeutic targets.

Peripheral blood was collected from 77 patients with metastatic, pathologically confirmed ICC, ECC, or gallbladder cancer following informed consent (Table 1). Patients were enrolled in a randomized, national phase II clinical trial (NCT03201458) of atezolizumab with or without cobimetinib (18). Patients randomized to arm A received atezolizumab every 2 weeks, while patients randomized to arm B received cobimetinib daily (21 days on/7 days off) alongside atezolizumab every 2 weeks. This work was carried out under a protocol approved by the NCI Cancer Therapy Evaluation Program (CTEP) and the central and local institutional review boards (IRB). All patients were enrolled between February 2018 and October 2018. Whole blood samples were collected prior to treatment and on treatment at the start of cycle 2 (C2D1), then transported overnight to Winship Cancer Institute (Emory University, Atlanta, GA) for processing and analysis. Due to the aggressive nature of BTC, we collected patient correlative samples at this time point to avoid missing samples from patients with disease progression or treatment intolerance. From a biologic standpoint, prior work suggests that early immune responses (within 4 weeks) are important indicators of favorable clinical outcomes to PD-1/PD-L1–targeted therapies (19). PBMCs and plasma were isolated from whole blood via density gradient centrifugation using Ficoll-Paque (GE Healthcare Bio-Sciences AB). In addition, we obtained samples from two cohorts of de-identified healthy donors, collecting 12 PBMC samples from buffy coats (Sylvan N. Goldman Oklahoma Blood Institute, Oklahoma City, OK) and 16 plasma samples from whole blood (Emory University Hospital), used as comparators for our baseline data. Normal donor blood samples were processed similarly to patient samples by density gradient centrifugation. All patient and normal donor plasma were cryopreserved at −80°C until analysis, and all PBMCs were cryopreserved in liquid nitrogen prior to analysis. Blood samples received more than 2 days after collection were not used for PBMC isolation to ensure consistent quality but were still processed for plasma via centrifugation and used for bioplex analysis. As a result, some patients did not have viable samples at all timepoints, and hence data were collected using samples from 70 patients for PBMC analysis and 51 patients for plasma analysis. Patient sample exclusion criteria is detailed in Fig. 1.

Plasma samples were analyzed using a panel of 45 cytokines, chemokines, and growth factors on a Luminex magnetic bead-based platform according to manufacturer protocol (ProcartaPlex Cytokine/Chemokine/Growth Factor Panel 1, EPX450–12171–901, ProcartaPlex Immunoassays, Invitrogen). Samples were analyzed in duplicate and batch-run on a Luminex 100 instrument and quantified using analyte-specific standard curves for each batch. Only 51 patients who remained on treatment through C2D1 blood collection were analyzed. Quality filtering based on a coefficient of variation of more than 15% was performed for each analyte. Fifteen analytes were excluded from analyses for more than 50% of measurements falling below the detectable assay threshold across patient samples.

Comprehensive phenotypic analysis of peripheral immune cells was conducted via 23-color flow cytometry. Antibodies are detailed in Supplementary Table S1. Cryopreserved PBMCs were thawed at 37°C, washed, centrifuged, and resuspended in FACS buffer (PBS + 3% FBS + 0.05 mmol/L EDTA). Cells were incubated with surface antibody for 30 minutes at 4°C, washed, permeabilized, and fixed overnight using the FoxP3/Transcription Factor Staining Buffer set (00–5523–00, eBioscience). Cells were incubated with intracellular antibodies for 1 hour at room temperature, washed, and resuspended in FACS buffer for analysis. Flow cytometric analysis was conducted on a Cytek Aurora (Cytek Biosciences). Compensation controls were generated using UltraComp eBeads Compensation Beads (01–2222–41, Invitrogen). Data were analyzed using FlowJo software version 10.7.2 (FlowJo, LLC).

Descriptive statistics were used to summarize patients’ demographics and disease characteristics. Biomarker values at baseline and their changes to C2D1 were first summarized and associated with treatment group, then further linked with clinical outcomes (e.g., OS, PFS). To account for data dependency rooted in repeated samples from the same patients, two-way repeated measures ANOVA test, along with Šídák multiple comparisons test, was used to test the interaction effect between percent change and treatment groups. This approach permits determination of whether combination treatment leads to more change in biomarker measurements than single-agent treatment (arm B vs. arm A). All biomarker measurements were compared between patient and healthy donor samples using two-sided Student t test. For either baseline biomarker levels or their changes to C2D1, univariate (UVA) and multivariate (MVA) associations with clinical outcomes (OS, PFS) in each treatment arm were explored using the Cox proportional hazards model, illustrated using the Kaplan–Meier method, and significance was determined by log–rank test. The measures of all biomarkers were dichotomized on the entire sample (<median vs. ≥median) for univariate and multivariate analyses. Soluble factors were dichotomized by percent change, and immune populations were dichotomized by fold change. Multivariate analyses were adjusted simultaneously for age, sex, and anatomic tumor subtype as relevant clinical variables. To correct for multiple testing, the Benjamini and Hochberg method was used to control the FDR on both UVA and MVA, with a cut-off value of P < 0.25 for significance. Significance was adjusted separately for soluble factors and for immune cell populations. Due to limited statistical power from a small sample size, significance for multivariate analyses was determined as P < 0.1, while significance of all other analyses was determined as P < 0.05.

The data generated in this study are available within the article and its supplementary data files or otherwise available upon request from the corresponding author.

To define prominent immune features in patients with BTC, independent of treatment, we compared biomarker data at baseline from this patient cohort with that of healthy donors. We analyzed a panel of 45 soluble factors in peripheral plasma samples, which evaluated cytokines regulating T-cell differentiation and function, inflammatory cytokines, chemokines, and growth factors. Fifteen markers, including a subset of proinflammatory cytokines, were below the detectable assay threshold in more than 50% of all plasma samples and concentrations could not be accurately determined. Therefore, these 15 factors were excluded from further analyses. Of the remaining 30 soluble factors, 21 were significantly elevated among patients with BTC (Fig. 2A; Supplementary Table S2). In particular, platelet-derived growth factor B (PDGF)-BB, IL5, IL17A, and placental growth factor (PlGF)-1 were all elevated five-fold more than what was observed in healthy donors (P ≤ 0.002; Fig. 2B–E). The remaining 9 factors were not present at concentrations significantly different from healthy donors. Independent of treatment effect from the clinical trial, there were no correlations between any soluble factors at baseline with clinical outcomes (data not shown).

Multi-parameter 23-color flow cytometric analysis of PBMCs was used to interrogate unique immune landscapes in blood of patients with advanced BTC (Supplementary Fig. S1). Our analysis encompassed 19 phenotypically distinct populations including lymphocytes, myeloid cells, and cells expressing immune checkpoint proteins. We also used this flow panel to conduct analysis of a separate cohort of 12 healthy adult PBMCs and found that nine phenotypically-defined cell populations were significantly different between patients with BTC and healthy donors (Supplementary Table S3). Circulating total CD8⁺ T cells were lower overall in BTC patient samples (P = 0.003), but of those CD8⁺ T cells, patients with BTC had higher frequencies of cells expressing inhibitory checkpoint ligands. Notably, CD8⁺PD-1⁺ (P = 0.007) and CD8⁺BTLA⁺ (P = 0.016) cells were present at higher frequencies in patients with BTC. In addition, CD4⁺PD-1⁺ (P = 0.001) and CD4⁺BTLA⁺ (P = 0.001) T cells were higher in patients with BTC, as well as PD-1⁺TIM3^– activated CD8⁺ T cells (P = 0.004; Fig. 3A–E; Supplementary Table S3). In contrast, CD4⁺ and CD8⁺ T cells expressing TIM3, LAG3, and VISTA were not present in patients with BTC at frequencies different from healthy donors. We also assessed relationships between distinct immune phenotypes identified and clinical parameters in this cohort of patients with metastatic BTC. We postulated peripheral biomarkers may emerge to predict OS and signify more aggressive disease. Interestingly, the frequency of BTLA⁺CD8⁺ T cells correlated with OS, where patients with above-median population frequencies had longer OS, regardless of treatment received on the clinical trial (P = 0.036; Figs. 3E and F; Supplementary Table S3).

To assess the effect of both single-agent treatment with atezolizumab and dual therapy with cobimetinib, we calculated percent change of plasma concentrations of soluble factors from baseline to C2D1 (Fig. 4A). We stratified all patients by median percent change for each soluble factor, then evaluated correlations between percent change and clinical outcomes within each treatment group (Supplementary Table S4). For patients in arm B, dual treatment with atezolizumab and cobimetinib significantly decreased plasma concentrations of PDGF-BB (P = 0.0456), BDNF (P = 0.0036), and PlGF-1 (P < 0.0001) from baseline to C2D1, while patients in arm A receiving atezolizumab monotherapy had no significant changes in plasma concentration for these analytes (Fig. 4B–D). Of these factors, the decrease in plasma concentrations of PDGF-BB in arm B was associated with improved OS (UVA P = 0.023, FDR = 0.345; MVA P = 0.084, FDR = 0.921) and PFS (UVA P = 0.040, FDR = 0.360; MVA P = 0.067, FDR = 0.395) on exploratory univariate analysis and in multivariable analysis. Given the limited sample size however, these trends were no longer statistically significant upon FDR adjustment for multiple comparisons (Fig. 4E; Supplementary Table S4). Changes in concentration of other soluble factors correlated with improved clinical outcomes for patients in arm A and arm B, but could not be directly associated with treatment effects following two-way ANOVA (Supplementary Table S4). We next evaluated baseline plasma concentrations of soluble factors as predictors of response to cobimetinib and atezolizumab combination therapy (Supplementary Table S5). Baseline levels were stratified by median concentration of each analyte as a whole, then evaluated for correlations with clinical outcomes within each treatment group. Higher baseline concentrations of IL23 in arm B patients correlated with improved OS (UVA P < 0.001, FDR = 0.03; MVA P = 0.02, FDR = 0.60), while higher baseline concentrations of BDNF (UVA P = 0.063, FDR = 0.330; MVA P = 0.017, FDR = 0.170), IL8 (UVA P = 0.005, FDR = 0.075; MVA P = 0.013, FDR = 0.170), and RANTES (UVA P < 0.001, FDR = 0.030; MVA P = 0.001, FDR = 0.030) indicated better PFS (Supplementary Table S5), both in univariate exploratory analysis and after adjusting for multiple comparisons. Baseline plasma concentrations of other soluble factors in both arm A and arm B had relationships to clinical outcomes that trended toward significance by UVA and/or MVA (Supplementary Table S5). Measurement of these factors at baseline may be predictive of how patients with advanced BTC respond to dual blockade of MEK and PD-L1.

From baseline to C2D1, combination therapy was associated with an increase in CD4⁺TIM3⁺ T cells, as evidenced by a notable interaction via two-way ANOVA (P_interaction = 0.032; Fig. 5B). However, this increase correlated with worse OS for these patients (UVA P = 0.015, FDR = 0.285; MVA P = 0.016, FDR = 0.219; Fig. 5C; Supplementary Table S6). Both OS and PFS were improved for single-agent–treated patients with increasing populations of CD4⁺BTLA⁺ (UVA_OSP = 0.006, FDR = 0.057; MVA_OSP = 0.013, FDR = 0.247; UVA_PFSP = 0.003, FDR = 0.029; MVA_PFSP = 0.002, FDR = 0.038) and CD8⁺BTLA⁺ T cells (UVA_OSP = 0.005, FDR = 0.057; MVA_OSP = 0.060, FDR = 0.418; UVA_PFSP = 0.003, FDR = 0.029; MVA_PFSP = 0.02, FDR = 0.019). In addition, arm A patients with an increase in CD4⁺LAG3⁺ T cells from baseline to C2D1 also had improved OS (UVA P = 0.014, FDR = 0.089; MVA P = 0.066, FDR = 0.418), while decreasing CD8⁺VISTA⁺ T cells correlated with improved PFS for arm B patients (UVA P = 0.001, FDR = 0.019; MVA P = 0.003, FDR = 0.057). Changes in other immune populations from baseline to C2D1 had correlations with clinical outcomes trending toward significance (Supplementary Table S6).

The predictive value of baseline populations of peripheral immune cells on clinical outcomes was next evaluated (Supplementary Table S7). Baseline populations of immune cells were stratified by median percentage of parent population. In arm B patients, higher peripheral CD8⁺ T cells (UVA P = 0.016, FDR = 0.304; MVA P = 0.138, FDR = 0.836) and CD8⁺ cells that were BTLA⁺ (UVA P = 0.061, FDR = 0.5795; MVA P = 0.092, FDR = 0.836) correlated with improved OS. Patients in this arm with fewer peripheral CD8⁺VISTA⁺ T cells also had better OS (UVA P = 0.036, FDR = 0.228; MVA P = 0.125, FDR = 0.497) and improved PFS (UVA P = 0.018, FDR = 0.342; MVA P = 0.09, FDR = 0.656). Above-median populations of intermediate monocytes in single-agent–treated arm A patients also correlated with better OS (UVA P = 0.027, FDR = 0.228; MVA P = 0.013, FDR = 0.247). These findings were significant upon exploratory UVA and MVA analyses, but due to small sample size were no longer significant when adjusted for multiple comparisons.

This study provides a comprehensive profile of the systemic immune landscape of advanced BTC and evaluates how MEK inhibition and anti–PD-L1 therapy modulate systemic immune factors and their relationship to clinical outcomes. We assess peripheral immune biomarkers in pretreatment samples to profile the immune landscapes of a large cohort of patients with advanced BTC. We identified several soluble factors, including PDGF-BB, IL5, IL17A, and PlGF-1, that are elevated in plasma of patients with BTC when compared with healthy donors, as well as CD4⁺ and CD8⁺ T cells expressing PD-1 and BTLA. In our clinical trial, dual MEK/PD-L1 blockade altered production of soluble factors in patient plasma compared with PD-L1 blockade alone, including a decrease in PDGF-BB that correlated with improved PFS. Further, changes in several T-cell populations expressing immune checkpoint markers were observed in patients receiving combined MEK/PD-L1 inhibitors and correlated with improved clinical outcomes.

The use of MEKi to enhance PD-1/PD-L1–targeted immunotherapy has been investigated in various preclinical tumor models and in clinical trials. While we observed correlations between important immune factors and survival outcomes, our overall clinical experience with dual MEK/PD-L1 inhibition in patients with advanced BTC revealed a modest extension of PFS compared with PD-L1 blockade alone. The low objective response rate in both treatment arms in this study reflect the immune resistant nature of BTC and the need to understand mechanisms of response and resistance to immunotherapy in patients with BTC. Although MEKi has been shown to promote CD8⁺ T-cell infiltration and limit T-cell receptor (TCR)-mediated exhaustion, previous work from our group shows systemic MEK administration can also inhibit T-cell activation and function. In particular, the addition of MEKi to PD-L1 blockade in this trial limited expansion of T-cell populations with activated phenotypes, whereas anti–PD-L1 alone increased these cells (20). Fortunately, subsequent preclinical studies showed that this could be overcome through co-treatment with antibodies that act as agonists to T-cell function, such as 4–1BB (20). Inspired by these results, we have initiated a follow-up randomized trial investigating the combination of cobimetinib and atezolizumab with varlilumab, an agonistic mAb for CD27, to restore T-cell activation (NCT04941287). MEKi and PD-L1 blockade have distinct functions in controlling the immune response to solid tumors, which may explain why specific correlations to OS and PFS were evident that were not related to overall response rates. We hope our work interrogating other immune features associated with clinical benefit will guide development of future treatment options for these patients (18).

Cytokine signatures have been surveyed to predict responders to immunotherapy in other tumor models and may have utility in patients with advanced BTC (21). Several factors, including PlGF-1 and PDGF-BB, that are markedly higher in plasma from patients with BTC compared with healthy individuals, also have an established relationship with the biology of BTC tumors (22–24). For example, PlGF-1 is present at high levels in blood from patients with ICC and associated with desmoplasia and disease progression (23). Collectively this factor promotes aggressive disease, and its inhibition limits progression in hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA) tumor models (23, 25). Like PlGF-1, PDGF-BB is prominently produced by myofibroblasts in hepatobiliary tumors and promotes prosurvival signaling in BTC (26). Plasma levels of PDGF-BB were substantially reduced in patients with BTC receiving cobimetinib with atezolizumab versus atezolizumab alone, and this decrease correlated with improved PFS in combination-treated patients. PDGF-BB signals exclusively through the PDGFRβ receptor, activating downstream MEK/ERK and PI3K/AKT signaling (24, 27, 28). PDGFRβ is associated with epithelial-to-mesenchymal transition (EMT) to promote invasion and metastasis in colorectal cancer (29). To further support a relationship between MEK signaling and PDGF-BB, this growth factor has a critical mitogenic role in fibroblasts and other stromal cells in the tumor microenvironment, promoting stromal cell activation and angiogenesis (30, 31). Since peripheral PDGF-BB was also elevated at baseline in patients with BTC compared with healthy donors, and PDGFRβ is expressed in human cholangiocarcinoma tumors, targeting of the PDGF-BB/PDGFRβ axis likely has a role in controlling BTC. Previous studies have demonstrated genetic and pharmacologic inhibition of PDGFRβ promotes apoptosis and reduces tumor growth in in vivo models of BTC (26). However, further investigation is necessary to elucidate mechanistic relationships between PDGF-BB and BTC for potential use in future therapeutic strategies.

Analysis of peripheral immune populations showed notable changes in both CD4⁺ and CD8⁺ T cells expressing immune checkpoint proteins associating with clinical outcomes in patients receiving combination therapy. Atezolizumab alone led to higher LAG3⁺ and BTLA⁺ T-cell populations correlating with better OS and PFS, while atezolizumab and cobimetinib combination therapy led to fewer CD8⁺VISTA⁺ T cells that correlated with improved PFS. Conversely, combination therapy significantly increased the frequency of CD4⁺TIM3⁺ T cells, correlating with worse PFS. These data were consistent with prior studies in melanoma whereby MEKi also increased TIM3 expression on lymphocytes, and higher TIM3 expression frequently correlates with poor clinical outcomes in several other cancer models (32, 33). Further investigations following this clinical trial may benefit from addition of agents targeting TIM3, which could be more effective in combination with other checkpoint inhibitors (33). In addition to treatment-related changes in T-cell populations, analysis of peripheral immune cells at baseline revealed higher frequency of CD8⁺BTLA⁺ T cells compared with healthy donors, which correlated with increased OS in this BTC patient cohort. These results prompt further questions related to the role of BTLA on peripheral T cells as a reflection of BTC progression. BTLA is a co-stimulatory molecule in the CD28 immunoglobulin superfamily that harbors classical inhibitory signaling motifs (34, 35). However, the role of BTLA as a regulator of T-cell–mediated immune responses to cancer is not straightforward. Regulation of immune checkpoint expression and function is key to controlling the immune response in many cancers, but there has been little investigation into associations between MEKi and BTLA, LAG3, or VISTA expression. Indeed, there are documented molecular connections between activation of ERK with downstream transcription factors that control expression of genes encoding checkpoint proteins, though further investigation is required to fully characterize the relationship between MEKi and immune checkpoint expression on immune cells.

While this study lends meaningful insight into systemic immune features in patients with BTC, it does have limitations. Since the trial was open at multiple sites and patients had unresectable metastatic disease, the feasibility of obtaining tissue to probe intratumoral immune features was limited, with a greater sample size of peripheral blood specimens. Our data suggest peripheral blood is useful for understanding systemic immune alterations that accompany advanced cancer and may reveal provocative differences for further study. Although our patient population was substantial for biliary tract cancer studies, sample sizes in our analyses were much smaller than other clinical trials. In particular, data in our analyses of treatment effects was dichotomized by the median value of each analyte before stratification by treatment arm for correlation with clinical outcomes. Furthermore, to make our conclusions more succinct, we describe changes in immune factors from baseline to C2D1 as either increases or decreases depending on which side of the median they fall on. This simplifies some quantitative details where the median is not precisely at no fold or percent change, but makes our conclusions more concise overall. We also acknowledge this study was only conducted in the United States, and therefore does not encapsulate the pathological manifestations of BTC common in other regions such as liver fluke infection. In addition, our comparison of baseline samples to healthy donors does not account for inflammatory diseases of the liver and biliary tract, including hepatitis, primary sclerosing cholangitis, and gallbladder disease, that could influence the peripheral immune factors in our patient population. Finally, targeted therapy against FGFR2 fusions and isocitrate dehydrogenase (IDH) alterations are among the most promising areas of clinical development for BTC, but we did not have extensive genome-level data on tumors from patients enrolled in this trial to consider for advancing this line of study (36, 37). Despite these understandable limitations, our comprehensive analysis of immune cell and soluble factor signatures in the blood of patients with advanced BTC is of value to the field.

Overall, this study investigates how MEKi combines with anti–PD-L1 to modulate clinically-relevant, systemic soluble factors and immune populations in a large cohort of patients with advanced metastatic BTC. We have also delineated differences in peripheral immune markers of patients with BTC compared with healthy individuals. This work advances our understanding of how MEKi synergizes with immune checkpoint blockade and has uncovered factors that merit additional study for a disease with few effective treatment options.

M. Yarchoan reports grants and personal fees from Genentech and Exelixis; grants from Bristol Myers Squibb, Incyte, and Eisai; and personal fees from AstraZeneca, Replimune, and Hepion outside the submitted work. B.F. El-Rayes reports grants from Merck, Roche, Novartis, AstraZeneca, and BMS outside the submitted work. N.S. Azad reports grants from NCI during the conduct of the study. G.B. Lesinski reports other support from Merck and Co., Bristol Myers Squibb, Boehringer Ingelheim, and Vaccinex, as well as personal fees from ProDa Biotech, LLC outside the submitted work. No disclosures were reported by the other authors.

A.N. Ruggieri: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. M. Yarchoan: Conceptualization, supervision, funding acquisition, investigation, writing–review and editing. S. Goyal: Resources, software, formal analysis, visualization, writing–review and editing. Y. Liu: Resources, software, formal analysis, supervision, methodology, writing–review and editing. E. Sharon: Conceptualization, resources, funding acquisition. H.X. Chen: Conceptualization, resources, funding acquisition, writing–review and editing. B.M. Olson: Methodology, writing–review and editing. C.M. Paulos: Resources, methodology, writing–review and editing. B.F. El-Rayes: Conceptualization, resources, supervision, funding acquisition. S.K. Maithel: Conceptualization, resources. N.S. Azad: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. G.B. Lesinski: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing.

We would like to acknowledge the cores at Winship Cancer Institute and Emory University that made this research possible including the Pediatric/Winship Flow Cytometry Core, the Winship Biostatistics and Bioinformatics Core, under NIH/NCI award number P30CA138292. In addition, sample collection at enrolling sites was supported by the NCI Experimental Therapeutics Clinical Trials Network (ETCTN) UM1 grants. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This work is supported by NIH grants (grant nos. R01CA228414, R01CA208253, R01CA228406, and P30CA006973). The clinical study from which samples were derived was coordinated by the NCI ETCTN, and supported by F. Hoffmann-La Roche, Ltd. and the NCI of the NIH under the award number UM1CA186691.

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