Baseline Serum Inflammatory Proteins Predict Poor CAR T Outcomes in Diffuse Large B-cell Lymphoma

Chimeric antigen receptor (CAR) T-cell therapy has changed the treatment paradigm for patients with relapsed/refractory hematologic malignancies leading to the Food and Drug Administration (FDA) approval of several CAR T-cell products (1–5). Despite encouraging efficacy, relapse after therapy is common. Approximately 40% of patients with diffuse large B-cell lymphoma (DLBCL) treated after at least two lines of therapy achieve a durable remission (1, 6, 7). Furthermore, therapy can be complicated by potentially life-threatening immune-mediated toxicities such as cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS), underscoring the need for early identification of patients at risk of CAR T cell treatment failure or development of severe toxicities (4, 5).

We previously reported multiple abnormalities correlated with poor response to CAR T-cell therapy, such as immune dysregulation mediated by inflammation, increased expression of myeloid gene signatures, and extensive chromosomal abnormalities in the tumor compartment (8, 9). These studies collectively suggest a model of CAR T-cell resistance wherein lymphoma cells recruit or induce an inflammatory tumor microenvironment (TME), which mediates poor T-cell function leading to suboptimal CAR T-cell manufacturing in a subset of patients (10). We and others have also described the role of immune cell–derived cytokines in association with the development of CRS and ICANS (11–14). Cytokine measurements as well as assays to comprehensively assess the TME are not readily available, making stratification strategies based on these assays difficult to implement clinically. Therein, in this report, we sought to further develop clinically actionable approaches to identify patients at high risk of CAR T-cell therapy–associated toxicity and/or poor clinical outcomes.

There are limited data on how the identification of these poor risk factors can change the management paradigm of high-risk patients and how they might inform mechanisms of CAR T-cell resistance. Here, we report analyses of patients with relapsed/refractory (R/R) DLBCL treated at the H. Lee Moffitt Cancer Center with standard-of-care axicabtagene ciloleucel (axi-cel) in third line and beyond. We correlate cytokines, peak CAR T-cell expansion, and baseline serum inflammatory markers including C reactive protein (CRP) and ferritin, in addition to patient and tumor characteristics, with safety and efficacy endpoints. We develop a risk stratification model using baseline CRP and ferritin and establish that patients with elevated CRP and ferritin prior to lymphodepleting chemotherapy have inferior response rates.

One hundred and thirty-six patients with R/R DLBCL (after at least two prior lines of therapy) treated with axi-cel are included in initial model building. Baseline patient characteristics are summarized in Table 1. The median follow-up was 36 months at the time of data cutoff, February 18, 2022. Median age was 65 years (range, 19–79 years). Most patients were heavily pretreated with 53% (n = 72) receiving three or more prior lines of therapy. Eighty-six patients (65%) required bridging therapy between the time of leukapheresis and lymphodepleting chemotherapy. Aside from bridging therapy, which was excluded in the Zuma-1 trial, 57 patients (42%) would not have met eligibility criteria for Zuma-1, of whom the most common reason for ineligibility was Eastern Cooperative Oncology Group (ECOG) performance status ≥2 (n = 32).

Most patients developed CRS (n = 126, 93%), 10% of which were severe (grade ≥3). Eighty-three patients (61%) developed ICANS with an incidence of severe ICANS of 28%. Seventy-one patients (52%) were treated with tocilizumab and 66 (49%) required steroids for the management of these toxicities. Nine patients died due to nonrelapse mortality with three deaths attributed to severe CRS and six due to infectious complications. With a median follow-up of 3 years, the ongoing overall response rate was 47%. The median overall survival (OS) and progression-free survival (PFS) were 34.9 and 12.2 months, respectively (Table 2).

Univariate logistic regression was used to test the association of each baseline cytokine to severe grade ≥3 toxicities as previously reported (13). At the time of the previous publication, baseline samples were available only in 51 patients. Here we report a total of 80 patients with available baseline serum analysis, which includes the 51 patients we previously reported (13). Baseline levels of IL6 (P = 0.037), CRP (P = 0.03), and ferritin (P = 0.003) were associated with severe ≥3 CRS (Supplementary Fig. S1A–S1C). Baseline IL6 (P = 0.036), CRP (P = 0.008), ANG2/ANG1 (P = 0.004), and lactate dehydrogenase (LDH; P = 0.028) were associated with severe ≥3 ICANS (Supplementary Fig. S1D–S1G). In a univariate logistics model, baseline CRP and ferritin as continuous variables were significantly associated with grade ≥2 ICANS (OR 1.4; 95% CI, 1.2–1.6; P < 0.05; OR 1.3; 95% CI, 1.1–1.6; P < 0.05) but were not associated with ≥2 CRS. In a multivariable model adjusting for age, ECOG, and bridging therapy, only baseline CRP was significantly associated with ≥2 ICANS (OR 1.3; 95% CI, 1–1.6; P < 0.05; Supplementary Table S1).

Patients were stratified into three risk groups based on baseline CRP and ferritin. Considering that IL6 levels correlate with poor clinical outcomes in patients treated with CD19-targeted CAR T-cell (13), we associated pre-lymphodepletion CRP and ferritin with baseline IL6. We initially determined the cutoffs based on clinical criteria. Kaplan–Meier OS curves were used to confirm optimal cutoffs for baseline CRP of 4.08 mg/dL and ferritin of 421 ng/mL (Supplementary Fig. S2A and S2B). Baseline IL6 significantly associated (P < 0.001 across groups, Fig. 1) with risk categories based on CRP and ferritin levels defined as low (CRP <4 mg/dL and ferritin <400 ng/mL), intermediate (not meeting criteria for high risk or low risk), and high (CRP ≥4 mg/dL and ferritin ≥400 ng/mL) in the subset of patients with all three serum protein measurements at baseline (n = 79). In the entire cohort, the intermediate group encompassed 40 patients with CRP <4 mg/dL and ferritin ≥400 ng/mL and only 7 patients with CRP ≥4 mg/dL and ferritin <400 ng/mL. Patients in the high-risk group were more likely to have ECOG ≥2, higher baseline metabolic tumor volume, IPI score ≥3, and to have received bridging therapy (P < 0.001 across variables; Table 1). Seven patients (26%) in the high-risk group developed severe CRS as compared with six patients and one patient in the intermediate- and low-risk groups, respectively (P = 0.001 across variables). Severe ICANS was observed in 52%, 30%, and 16% in the high-, intermediate-, and low-risk groups, respectively (P = 0.002 across variables; Table 2).

Due to the low rate of grade ≥3 toxicities, univariate and multivariable analyses were done to test the association between the risk categories and grade ≥2 CRS or ICANS. Although there were no statistically significant differences in CRS across the risk groups, grade ≥2 ICANS differed across the groups with rates of 29%, 44%, and 67% in the low-, intermediate, and high-risk groups, respectively (P = 0.004). In a multivariable analysis controlling for age, ECOG, and bridging therapy, high-risk patients were more likely to develop grade ≥2 ICANS (HR 4.2; 95% CI, 1.4–13.1; P = 0.01; Supplementary Table S2).

Patients within the low-, intermediate-, and high-risk groups had a median follow-up of 27.6, 32.4, and 27.7 months, respectively. The median OS differed in the high-risk group at 6.9 months compared with 14.9 months and not reached in the intermediate- and low-risk groups, respectively (P < 0.001; Table 2; Fig. 2A). Median PFS was 3, 7.9 months, and not reached in the high-, intermediate-, and low-risk groups, respectively (P < 0.001; Table 2; Fig. 2B).

As of the data cutoff on February 18, 2022, the ongoing response rate in the low-risk group was 71% compared with 30% and 22%, respectively, in the intermediate- and high-risk groups (P < 0.001 across variables; Table 2). The 2-year nonrelapse mortality was 0% in the low-risk group as compared with 13% (n = 6) and 12% (n = 3) in the intermediate- and high-risk groups, respectively (P < 0.001 across variables; Table 2). Utilizing multivariable Cox regression analysis (n = 99, reduced sample size due to metabolic tumor volume missing in 37 patients), we adjusted for variables that are known to be associated with inferior efficacy to assess the differences in OS and PFS among the risk groups. This included age, ECOG, bridging therapy, and metabolic tumor volume (MTV). ECOG was found to be significantly associated with both OS and PFS. Patients in the high-risk category had inferior OS relative to low-risk (HR 3.0; 95% CI, 1.2–8.3, P = 0.0161) as shown in Supplementary Table S3A. After adjustment, no association between risk category and PFS was identified (Supplementary Table S3B).

We sought to validate our findings using two independent international cohorts. The first cohort included 139 adult patients receiving standard-of-care axi-cel (n = 49) or tisagenlecleucel (tisa-cel, n = 90) for R/R DLBCL. Patients were treated between May 2018 and June 2021 across six European Union (EU) CAR-T treatment centers (15). The median OS was significantly inferior in the high-risk group at 5.4 months compared with 10.9 months and not reached in the intermediate- and low-risk groups, respectively (P < 0.001; Fig. 2C). The median PFS was 2.4, 3.2, and 10.3 months in the high-, intermediate-, and low-risk groups, respectively (P < 0.001; Fig. 2D). A second independent validation cohort included adults with DLBCL who were refractory or relapsed to at least two prior lines of therapy and who were enrolled and treated with axi-cel in cohorts 1, 2, 4, and 6 of the Zuma-1 clinical trial; patients in this validation included the 83% of patients with nonmissing baseline clinical CRP and ferritin (n = 151). There was no significant difference in OS among the three risk groups (Fig. 2E; P = 0.072). PFS was found to differ across the groups with a median PFS of 27.8 months versus 15.7 months versus 3.9 months in low-, intermediate-, and high-risk patients, respectively (Fig. 2F, P = 0.039). Baseline characteristics for the validation cohorts are described in Supplementary Table S4.

Data from cohort 6 of Zuma-1 resulted in an updated FDA label to consider prophylactic dexamethasone in recipients of axi-cel considering individual patient risk and benefit (16). Therefore, using the CRP and ferritin criteria to identify high-risk patients, institutional clinical standards were revised in February 2021 to administer prophylactic corticosteroids to all patients consecutively who met high-risk criteria February to December 2021 (n = 10). These patients were not included in the initial model. Baseline characteristics of patients who met high-risk criteria and treated with prophylactic dexamethasone were compared with the high-risk patients treated as per institutional standards (n = 27; Supplementary Table S5). There were no significant differences in median baseline CRP, ferritin, IL6, or other variables (Supplementary Table S5).

Although most high-risk patients in both cohorts deve­loped CRS, none developed severe CRS in the prophylactic dexamethasone group which compares favorably to the 7 patients (0 vs. 26%, P = 0.16) with severe CRS in the historical high-risk cohort (Supplementary Table S6). The rates of ICANS were similar among the prophylactic dexamethasone cohort and the historical high-risk cohort (70 vs. 78%, P = 0.68). Severe ICANS was lower (30%) in those that received prophylactic steroids as compared with 52% in the historical high-risk cohort but did not reach statistical significance (P = 0.29). The median PFS of three months was poor in both groups (P = 0.46) with a median OS of 6 months in the prophylactic dexamethasone group as compared with 7 months (P = 0.9) in the high-risk group (Supplementary Table S6; Supplementary Fig. S3A and S3B). There were no differences in median CAR T-cell expansion among patients in the historical high-risk cohort (n = 14) with median peak CAR T-cell expansion of 6.57 × 10⁶ CAR copies/μg as compared with 8.96 × 10⁶ CAR copies/μg in the prophylactic steroid cohort (n = 8; Supplementary Fig. S3C; P = 0.9).

Several studies have investigated how the host immune system and/or the infused cell product factors are associated with severe toxicity and/or inferior CAR T-cell efficacy (7, 17, 18). We and others reported the role of cytokines and the TME on the development of severe toxicities and CAR T-cell treatment resistance (9, 13). However, there is no widely available lab test or biomarker that can rapidly identify patients at the highest risk for poor outcomes prior to CAR T-cell infusion. In our prior work, we observed that baseline elevated levels of IL6 were associated with higher rates of severe CRS and/or death (13). Rapid quantitative detection of IL6 is limited to research settings, hindering its utility in clinical practice. In this report, we aimed to establish novel clinically actionable cutoffs using readily available labs and sought to validate our findings in two independent cohorts. We establish that baseline levels of CRP ≥4 mg/dL and ferritin ≥400 ng/mL correlated with baseline levels of IL6, thereby making it easier to identify high-risk patients prior to the start of lymphodepleting chemotherapy. We demonstrate that patients in the high-risk group had higher rates of severe grade ≥3 CRS and ICANS, and significantly inferior PFS and OS.

Various groups have developed predictive models that evaluate outcomes post CAR T-cell therapy including the EASIX score (19, 20). However, in these models, patient scoring and stratification relied on different ferritin cutoffs for CRS and ICANS and were not designed to predict efficacy outcomes in terms of OS, PFS, or nonrelapse mortality. Additionally, correlative studies including cytokines and expansion of target cell populations were not included (19). The validation cohorts utilized in our study included patients treated in real-world settings as well as patients treated in the Zuma-1 clinical trial. One limitation of using the Zuma-1 data is that patients from cohort 6 were included who received prophylactic corticosteroids which may have biased clinical outcomes. The model reported here was also validated in cohorts that utilized both axi-cel and tisa-cel, demonstrating the potential utility of this model irrespective of the CD19 CAR T-cell product infused.

Patients categorized as high-risk group per CRP/ferritin criteria had well-described poor risk features at baseline such as poor performance status, higher MTV, and were more likely to require bridging therapy. Baseline tumor burden as assessed by high MTV has been associated with decreased CD19 CAR T-cell efficacy (17). However, manual calculation of MTV is cumbersome and is not readily available. In multivariable analyses wherein we adjusted for key prognostic factors such as ECOG and MTV, we demonstrate the association between CRP/ferritin risk category and OS and grade ≥2 ICANS. We determined that ECOG is significantly associated with both OS and PFS, as has been previously shown by our group and others (7). Measures such as ECOG rely on subjective assessments and therefore are prone to bias when used as predictive markers. In a multivariable analysis adjusting for ECOG and MTV, CRP/ferritin risk stratification was significant for OS. However, only ECOG was significantly associated with inferior PFS in this multivariable model. One explanation is that patients with preserved ECOG (0–1) and high-risk features may have received maintenance therapies and/or went on to receive consolidative allogeneic stem cell transplantation. Future studies are needed to better understand patterns of relapse and the role of maintenance therapies in high-risk patients.

Early intervention strategies utilizing cytokine-blocking agents and/or steroids to mitigate severe toxicity associated with CAR T-cell therapy have been explored (11, 12, 16, 21). By using our novel preinfusion score, we show that low-risk patients may not need prophylactic or preemptive toxicity management strategies. In our report, only patients stratified as high-risk received prophylactic corticosteroids. As observed in cohort 6 of the Zuma-1 pivotal trial, none of the patients who received prophylactic steroids developed severe CRS and proportionally fewer patients experienced severe ICANS compared with the historical high-risk cohort (16). There were no statistically significant differences in CAR T-cell expansion between the two groups, albeit this analysis included a small number of patients. These findings do not rule out the possibility that CAR T cells may undergo poor expansion in high-risk patients and do not exclude the possibility that steroids may further impair CAR T-cell expansion. Identifying high-risk patients may inform CAR T-cell therapy patient selection and enable the use of novel preemptive strategies for toxicity mitigation and management for those who do go on to receive standard CAR T-cell products.

Accumulating evidence demonstrates that the presence of inflammatory mediators in the host immune system can significantly affect CAR T-cell expansion and function (9, 10). Bailey and colleagues demonstrated that the inflammatory status of patients can affect the development of inflammatory macrophage responses which exacerbate CAR T-cell therapy-associated toxicity (22). We previously reported that systemic inflammation as demonstrated in addition to elevated expression of macrophage gene signatures is associated with nondurable response to axi-cel in DLBCL (9, 13). Future mechanistic studies are needed to examine if and how systemic inflammation in high-risk patients supports an immunosuppressive phenotype that may affect CAR T-cell function.

This is a single-center retrospective observational study to evaluate factors associated with the development of immune-mediated toxicities and treatment outcomes in patients with R/R DLBCL treated with axi-cel at the H. Lee Moffitt Cancer Center. The institutional review board reviewed and approved the protocol. Clinical investigation was conducted according to the Declaration of Helsinki principles. All patients signed written informed consent. Since this was a retrospective study, accurate information regarding race and ethnicity was not available and therefore not included in this analysis. Between November 2017 and February 2021, 161 patients were treated with commercial axi-cel at our institution. Patients were excluded if they had second CAR T product (n = 4) and/or if they had missing baseline inflammatory markers (n = 21). One hundred thirty-six patients treated with axi-cel between November 2017 and February 2021 in third line and beyond who had baseline inflammatory markers are included in this initial model building. Patients received lymphodepletion with fludarabine 30 mg/m²/d × 3 days and cyclophosphamide 500 mg/m²/d × 3 days (1). Patients received bridging treatment between the time of leukapheresis and lymphodepleting chemotherapy at the discretion of the treating physician (1).

We validated findings in an independent, deidentified data set of 139 patients treated per standard of care with either axi-cel (n = 49) or tisa-cel (n = 90) for R/R DLBCL as previously reported across European centers (15, 23). We further validated results using Zuma-1 clinical trial data (n = 151), including cohorts 1, 2, 4, and 6, which were restricted to patients who had available baseline CRP and ferritin for analysis (1, 16).

CRS and ICANs were defined and graded using the ASTCT consensus grading guidelines (4). Management of CRS and ICANS was as per institutional guidelines, which were adapted from the CARTOX working group (5).

Based on our analysis of 136 patients detailed in the Results section, patients with elevated baseline pre-lymphodepletion (day −6) CRP ≥4 mg/dL and ferritin ≥400 ng/mL were classified as having inferior PFS and OS, and higher rates of severe toxicities. During the study period, institutional clinical standards were revised to administer prophylactic corticosteroids consecutively to all patients meeting high-risk criteria starting in February 2021 as per Zuma-1 cohort 6 (16).

Patients underwent response assessment using standard-of-care positron emission tomography and/or computed tomography scans at baseline prior to lymphodepleting chemotherapy. Tumor response was determined by the treating physician per the Lugano 2014 classification (24). PFS was defined as the time from CAR T-cell infusion to disease progression or death. Baseline MTV (n = 99) was measured as previously described by Dean and colleagues (17).

CRP, LDH, and ferritin were collected prior to lymphodepletion on day −6 for all patients included (n = 146) in this analysis. Data from 136 patients was included for model building. Ten additional patients were treated with risk-adapted steroid prophylaxis. The Department of Pathology and Laboratory Medicine used standard laboratory tests to measure serum CRP (range, 0–0.5 mg/dL), LDH (range, 135–225 U/L), and ferritin levels (range, 30–400 ng/mL) using the Roche Cobas analyzer system. Serum samples for cytokine analysis were collected in 80 patients prior to lymphodepleting chemotherapy (between days −30 and −6). Fifty-six patients did not have baseline serum samples available for cytokine analysis. Cytokines analyzed include GM-CSF, IL1β, IL2, IL6, IL15, IFNγ, TNFα, and angiopoietin 1 and 2. Serum was analyzed using the Ella automated simple plex immunoassay system (ProteinSimple) in a Moffitt research laboratory as previously described (13).

qPCR analysis (n = 22) was performed on 10 ng samples of genomic DNA using SYBR Green PCR master Mix and the 7900HT Fast Real-Time PCR System to detect the integrated SFG CAR-Transgene. Each sample was analyzed using albumin (as a reference) and SFG as primers. Samples were drawn at day 1 (±3), day 7 (±3), and day 14 (±3) to determine peak CAR-T expansion as previously described (9).

The data generated in this study are available upon request from the corresponding author.

Cytokine levels, patient characteristics, and other clinical outcomes were described by summary statistics (median and range for continuous measure and proportions and frequencies for categorical measures). The association between continuous variables and risk groups was assessed using Kruskal–Wallis tests. Wilcoxon rank sum test was applied when comparing the association between continuous variables and binary outcomes. The associations between categorical variables and endpoints were evaluated using Chi-squared tests or Fisher exact tests when the cell sizes were less than 5. The patient survival outcomes across risk groups were compared using Kaplan–Meier curves and subsequent log-rank tests. A multivariable Cox proportional hazards regression model with adjustment for clinical covariates was used to estimate hazard ratios for survival outcomes (PFS and OS). No adjustments were made for multiple hypothesis testing; a nominal P value of 0.05 was considered statistically significant.

R.G. Faramand reports personal fees from Kite/Gilead, grants and other support from Kite/Gilead during the conduct of the study, and other support from Novartis outside the submitted work. M.D. Jain reports grants and personal fees from Kite/Gilead, personal fees from Myeloid Therapeutics, Novartis, and BMS, and grants from Incyte, Loxo@Lilly, FACCA, Mark Foundation, and Bankhead-Coley State of Florida outside the submitted work. K. Rejeski reports grants and other support from Kite/Gilead and other support from Novartis, BMS/Celgene, and Pierre-Fabre outside the submitted work. M. Subklewe reports grants from Amgen, Novartis, BMS/Celgene, Gilead/Kite, Miltenyi, Avencell, Seattle Genetics, Molecular Partners, and Roche, and personal fees from Ichnox outside the submitted work. N.Y. Saini reports other support from Panbela Therapeutics outside the submitted work. E.A. Dean reports personal fees from the Clinical Education Alliance outside the submitted work; in addition, E.A. Dean has a patent for methods of enhancing CAR T-cell therapy pending to Moffitt Cancer Center. B.D. Shah reports personal fees from Amgen, Novartis, BMS, Precision Biosciences, Beigene, Adaptive, Century Therapeutics, Autolus, Deciphera, Lilly/Loxo, Takeda, and AstraZeneca, grants, personal fees, and nonfinancial support from Kite/Gilead, grants and personal fees from Jazz Pharmaceuticals, grants from Servier, and nonfinancial support from Pepromene Bio during the conduct of the study. A. Lazaryan reports consultancy/honoraria/scientific advisory bureau for Sanofi. J. Chavez reports personal fees from Novartis, Kite/Gilead, BMS, GenMab, ADC Therapeutics, BeiGene, Cellectar, and Lilly, grants from Merck and AstraZeneca, and grants and personal fees from Janssen outside the submitted work. J.A. Pinilla-Ibarz reports personal fees from Janssen, AbbVie, Beigene, Lilly, Novartis, and Takeda outside the submitted work. C.A. Bachmeier reports other support from Kite Pharma during the conduct of the study. K. Speth reports a patent for US-2023-0296610-A1 issued. M. Mattie reports other support from Kite, a Gilead company outside the submitted work. F.L. Locke reports grants from Kite a Gilead company during the conduct of the study, personal fees from A2, Allogene, Amgen, Bluebird Bio, BMS, Calibr, Caribou, Cowen, EcoR1, Gerson Lehrman Group (GLG), Iovance, Kite Pharma, Janssen, Legend Biotech, Novartis, Sana, Umoja, and Pfizer, personal fees and other support from Aptitude Health, ASH, BioPharma, Communications CARE Education, Clinical Care Options Oncology, Imedex, and Society for Immunotherapy of Cancer outside the submitted work; in addition, F.L. Locke has patents 10,414,810 and 11,149,072 issued, a patent for variant survivin vaccine for treatment of cancer issued, a patent for methods, systems, and computer-readable media for predicting a cancer patient's response to immune-based or targeted therapy issued, a patent for methods of enhancing CAR T-cell therapy issued, a patent for CAR T cells with enhanced metabolic fitness issued, a patent for method of enhancing immunotherapy using endoplasmic reticulum stress pathway issued, a patent for circulating tumor DNA and methods of use thereof issued, a patent for differential alternative splicing in relapsed and refractory diffuse large B-cell lymphoma patients receiving CAR T therapy issued, a patent for BCL-XL variants for enhancing cellular immunotherapy function issued, a patent for mutated and wild-type humanin proteins to improve CAR T or tumor infiltration lymphocytes (TIL) adoptive cell therapy issued, a patent for PGC1a and mutant PGC1a for TIL therapy improvement issued, and a patent for survivin mRNA vaccine issued. M.L. Davila reports grants from Kite during the conduct of the study, and personal fees from Adicet, Autolus, Caribou, and Lyell outside the submitted work. No disclosures were reported by the other authors.

R.G. Faramand: Conceptualization, data curation, supervision, investigation, methodology, writing–original draft, writing–review and editing. S.B. Lee: Conceptualization, formal analysis, methodology, writing–original draft. M.D. Jain: conceptualization, supervision, methodology, writing–review and editing. B. Cao: Formal analysis. X. Wang: Formal analysis, methodology. K. Rejeski: Data curation, validation, writing–review and editing. M. Subklewe: Data curation, validation, writing–review and editing. J.F. Fahrmann: Resources, data curation, writing–review and editing. N.Y. Saini: Resources, data curation, writing–review and editing. S.M. Hanash: Resources, data curation, writing–review and editing. Y.P. Kang: Formal analysis. D. Chang: Data curation, writing–review and editing. P.C. Rodriguez: Supervision, methodology. E.A. Dean: Data curation, writing–review and editing. T. Nishihori: Writing–review and editing. B.D. Shah: Writing–review and editing. A. Lazaryan: Writing–review and editing. J. Chavez: Writing–review and editing. F. Khimani: Writing–review and editing. J.A. Pinilla-Ibarz: Writing–review and editing. M. Dam: Writing–review and editing. K.M. Reid: Formal analysis, methodology. S.A. Corallo: Data curation. M. Menges: Data curation, methodology. M. Hidalgo Vargas: Methodology, writing–review and editing. J.K. Mandula: Methodology, writing–review and editing. B.A. Holliday: Data curation. C.A. Bachmeier: Data curation, Writing–review and editing. K. Speth: Resources, formal analysis, methodology, writing–original draft. Q. Song: Resources, formal analysis, writing–review and editing. M. Mattie: Resources, formal analysis, supervision, funding acquisition, validation, writing–original draft. F.L. Locke: Resources, supervision, methodology, writing–original draft, writing–review and editing. M.L. Davila: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing.

This work is supported by an NIH NHLBI R01-HL167232-01 (M.L. Davila) and the Rustum Family Endowed Chair in Translational Research (M.L. Davila). This work has been supported in part by the Biostatistics and Bioinformatics Shared Resource at the H. Lee Moffitt Cancer Center and Research Institute, an NCI-designated Comprehensive Cancer Center.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.