A Phase I Study of the Combination of Rituximab and Ipilimumab in Patients with Relapsed/Refractory B-Cell Lymphoma

Rituximab as a single agent produces an objective response rate (ORR) in relapsed/refractory indolent lymphoma of approximately 20% to 50% (1). Although the specific mechanism of action of rituximab is likely multifactorial and incompletely understood, most agree that host immune effector mechanisms are critical. To this end, there have been many attempts to utilize agents in combination with rituximab that augment these host immune effector mechanisms, including IL2, IL12, IFN, and cpg (2–5), which have produced modest improvements in efficacy, often with considerable toxicity. Lenalidomide is an immune modulator that is a potent natural killer (NK) and T-cell stimulant and has demonstrated efficacy in NHL and chronic lymphocytic leukemia (CLL) (6). The combination of rituximab and lenalidomide has produced considerable activity in patients with both treatment-naïve and previously untreated mantle cell and follicular lymphoma, with ORR ranging from 57% in rituximab-refractory patients to over 90% in previously untreated follicular lymphoma (7, 8).

Immunotherapeutic approaches to cancer therapy, including immune checkpoint inhibition, have produced exciting results in both solid tumors and hematologic malignancies, reversing T-cell anergy and facilitating an effective T-cell–mediated antitumor response. Cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) is a major negative regulator of the immune system. CTLA-4–blocking monoclonal antibodies like ipilimumab (I) activate antitumor T cells by obstructing their negative regulation, allowing for unopposed T-cell activation (9–11). Ipilimumab may also affect the tumor microenvironment by varied mechanisms, including the depletion of intratumoral CTLA-4–expressing regulatory T cells (12, 13), which has been correlated with response in patients with melanoma and colon cancer. Based on these observations, we hypothesize that ipilimumab may enhance host immune effector mechanisms and thereby augment the efficacy of rituximab.

The primary objective of this study was to evaluate the toxicity associated with adding ipilimumab to rituximab for the treatment of patients with relapsed/refractory histologically confirmed CD20⁺B-cell lymphoma, and to establish an MTD and/or recommended phase II dose (RP2D). Secondary objectives were to conduct mechanistic studies to understand the effect of this combination on the immune system, and to collect clinical data on antitumor response/overall response rates (ORR: complete + partial), and on progression-free survival (PFS).

Patients ≥18 years of age with relapsed or refractory CD20-positive NHL that were ineligible for high-dose chemotherapy and/or hematologic stem cell transplantation (SCT) or any other established curative therapy were eligible for this California Cancer Consortium study. Further inclusion criteria were Karnofsky performance status ≥ 70 and signed informed consent. Patients were excluded from the study in case of central nervous system involvement, prior allogeneic SCT, known HIV or hepatitis B or C virus infection, treatment with steroids or another investigational agent within 4 weeks, previous anti–PD-1 antibody, CD137 agonist, or other immune-activating therapy unless 5 half-lives have intervened (minimum 8 weeks). Patients on steroids or other immune suppressants or patients with autoimmune disease were excluded.

During the 12-week induction, ipilimumab was administered every 3 weeks for four doses, and rituximab was given every week for 4 weeks. In responding patients with acceptable toxicity, induction was followed by maintenance, during which ipilimumab and rituximab were given together every 12 weeks for 1 year, until unacceptable toxicity or disease progression. On days when both drugs were administered, rituximab was given before ipilimumab.

All relevant Institutional Review Boards or ethics committees approved the research methods used in these studies, and all patients provided written informed consent prior to enrolling. The studies were conducted in accordance with general ethical principles outlined in the Declaration of Helsinki, the International Conference on Harmonization guidelines, and Title 21 of the US Code of Federal Regulations and registered at www.clinicaltrials.gov as NCT01729806.

This trial began with a dose escalation, followed by an expansion at the RP2D. Initially, two dose levels of the ipilimumab were planned (3 and 10 mg/kg) during escalation; additional dose levels of 1 and 5 mg/kg were to be included if 3 or 10 mg/kg, respectively, exceeded the MTD (Supplementary Table S1); the dose of rituximab was fixed at 375 mg/m²per dose. Three-plus-three (3+3) rules for dose escalation were used to decide whether to escalate, expand, or de-escalate the dose of ipilimumab. The dose-limiting toxicity (DLT) observation period was defined as the first 6 weeks of induction which included the first two doses of ipilimumab and first four doses of rituximab. The MTD was based on toxicities observed during the DLT observation period and was defined as the highest dose tested in which only zero or one patient of six patients evaluable for toxicity at that dose experienced DLT attributable to the study drugs. To be evaluable for toxicity, a patient must have received at least two doses of ipilimumab and four doses of rituximab and be observed for at least 3 weeks after the second dose of ipilimumab or have experienced a DLT. All patients enrolled were fully followed for toxicity for the duration of the study, but patients who were not evaluable for dose-escalation decisions were replaced.

In the expansion cohort, an additional 20 patients were to be treated at the RP2D to further evaluate safety/toxicity, to obtain preliminary estimates of the objective response/remission rate and PFS, and to compare the immune response to rituximab with and without concurrent ipilimumab, as measured by immune subset analysis, antibody-dependent cell-mediated cytotoxicity (ADCC) based on the kinetics and magnitude of B-cell depletion (BCD). These additional patients were randomized 1:1 to one of two schedules: on arm A, ipilimumab was given on day 1 together with the first dose of rituximab, and on arm B, ipilimumab was first given on day 15 together with the third dose of rituximab (Supplementary Fig. S1). This permitted the assessment of whether or not I enhanced R-mediated ADCC during the first 15 days of treatment. The sample size of 20 (10 per schedule) evaluable patients was selected to ensure at least 80% power, using a one-sided 0.10-level two-sample t test, when the true difference, when comparing R+I versus R, in the change (increase in activated T cells and ADCC, or decrease in B cells) exceeded one standard deviation—where the standard deviation is intrinsic variability between patients in terms of the change. Blood for correlative analysis was drawn before treatment on days 1, 8, 15, 60, and 90. During the expansion, safety boundaries using a modified sequential probability ratio test were used to flag an excessive number of DLTs.

Objective response was assessed according to the revised response criteria for malignant lymphoma (14). To be evaluable for response, a patient must have received at least two doses of ipilimumab and four doses of rituximab. Imaging to assess tumor burden and response to treatment was done prior to treatment and every 8 weeks after start of treatment.

Toxicities were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE, Version 4.0, http://ctep.cancer.gov). Hematologic DLT was defined as any of the following adverse events (AEs) considered by the investigator to be related or possibly related to one of the study drugs: grade 3 thrombocytopenia with hemorrhage and/or requiring transfusion, grade 4 thrombocytopenia regardless of duration, grade 3 thrombocytopenia without hemorrhage lasting ≥ 28 days, grade 3 thrombocytopenia with potentially life-threatening morbidity, grade 3 or 4 neutropenia lasting ≥ 28 days (despite the use of growth factors), or grade 3 or 4 neutropenia with potentially life-threatening morbidity. Nonhematologic DLT was defined as any grade 3 or greater nonhematologic toxicity, except immune-related AEs (irAE). Grade 3 irAEs, which included inflammatory autoimmune events involving the gastrointestinal, skin, and nervous systems, were not DLT if they resolved to grade 1 or baseline with adequate steroid treatment tapered to maintenance or replacement doses (typically ≤ 10 mg/day) within 28 days and did not involve severe events such as bowel perforation or events requiring life-saving interventions. The inability to complete at least two doses of ipilimumab and four doses of rituximab for reasons of toxicity or lack of tolerability was a DLT unless caused by the 28-day resolution period for irAEs that eventually resolved.

Standard descriptive statistics were used to summarize clinical results. PFS was calculated as the time from start of treatment to the date of progression or death, whichever came first; patients who were alive and free of progression were censored at the date that their status was last documented; patients who started another therapy prior to progression were censored at that time. The Kaplan–Meier product limit method was used to display the PFS pattern over time. Median PFS and PFS probabilities were based on Kaplan–Meier plots; associated standard errors for the probabilities were based on Greenwood's formula (15).

Two analyses were undertaken: (1) to compare the two schedules in terms of increase in activated T-cell subsets and ADCC (assessed based on BCD), and (2) to explore whether any immune measures had the potential to identify patients who were more or less likely to respond to R+I. Establishing an association between a biomarker and response to treatment is the first step in identifying potential predictive biomarkers; further studies will be necessary to determine whether this association translates into a useful predictive biomarker of response.

Graphical methods were used to display the results: means or medians, as appropriate, were plotted along with confidence intervals (CIs) or ranges or interquartile ranges; boxplots were also plotted. Samples for analysis of plasma cytokine levels and subset populations of T cells were collected repeatedly over time. To quantitatively evaluate the patterns over time, a mixed-effects linear regression model was constructed using log cytokine concentration (pg/mL) and log T-cell subset counts as the dependent variables. Time (as a categorical variable) and response to therapy were set as fixed effects, as was the interaction term, and patient was set as random effect. In these mixed-effect linear regression models, none of the interaction terms was statistically significant, indicating that differences between responders and nonresponders, if they existed, tended to be constant over time. These interaction terms were dropped from the final models. To control for multiple testing, the Bonferroni method adjustment was used.

Between November of 2012 and June of 2016, 33 patients with relapsed or refractory CD20-positive NHL were enrolled on this multi-institutional phase I trial. The data cutoff for this article was May 1, 2018. The median age was 62, and the median number of prior treatment regimens was four. The majority (60%) had follicular lymphoma or diffuse large B-cell lymphoma, and 33% had failed an autologous SCT. Overall, 87% were considered refractory to anti-CD20 therapy including 92% of follicular lymphoma patients. Patients were considered refractory to anti-CD20 therapy if they had progressed on or within 6 months of being treated with an anti–CD20 therapy-containing regimen. A full summary of patient demographics and characteristics can be found in Table 1.

The first three patients treated at dose level 1 (3 mg/kg of ipilimumab) were evaluable for DLT, and one patient experienced DLT (prolonged diarrhea not successfully managed by steroid treatment); per the 3+3 rules, this dose level was expanded to enroll three more evaluable patients. Five more patients were treated with three evaluable for DLT and two inevaluable for DLT (one patient died in less than 6 weeks due to disease progression, and another patient did not receive the second dose of ipilimumab which was held for non-DLT grade 2 diarrhea); none of these five patients experienced DLT. With only one of the six evaluable patients experiencing DLT, consideration was given to escalating the dose of ipilimumab. After review of all toxicities during induction and maintenance, as well as the results of other trials comparing 3 mg with 10 mg/kg in patients with melanoma and discussion with the NCI/Cancer Therapy Evaluation Program, a decision was made to consider 3 mg/kg the RP2D and to use this dose for the expansion cohort.

In the expansion cohort, 25 patients were enrolled and randomized to either arm A or arm B. Of the 13 patients randomized to arm A, 10 patients were confirmed evaluable for DLT, and none of these patients experienced any DLTs; one patient experienced an irAE—grade 3 diarrhea. Of the three inevaluable patients, two went off early for disease progression, and a third patient received steroids in the absence of irAEs.

Of the 12 patients randomized to arm B, two went off treatment early (one of whom experienced toxicities prior to receiving any ipilimumab and one who went to hospice) and thus 10 patients on arm B were confirmed evaluable for DLT; two of these 10 experienced DLT including prolonged neutropenia and a grade 3 skin infection. In addition, two patients experienced grade 3+ irAEs: two patients experienced grade 3 diarrhea, and another patient experienced grade 3 hypoxia (Table 2). Hematologic toxicity was modest with four patients (12%) having grade 3 anemia and one patient with febrile neutropenia (comprehensive toxicity assessment, Supplementary Table S2).

Eight of the 33 treated patients (24%) achieved a response, with two achieving a complete response (CR; 6%); in addition, six (18%) patients had stable disease, corresponding to a disease control rate of 42% (Table 3). Eleven (33%) had disease progression as the best response, and eight came off too early for efficacy evaluation. When considering the entire cohort of 33 patients, the median (95% CI) PFS was 2.6 months (1.6–4.6 months; Fig. 1A), with a median follow-up time of 5.5 months among the nine who were censored (range, 0.5–18.5 months). Of the 13 patients with follicular lymphoma, seven responded (54%), two with CR (15%); the median (95% CI) PFS was 5.6 months (1.6–18.4+ months; Fig. 1B). Considering the entire cohort, 27 of 33 patients (87%) and 12 of the 13 patients (92%) with follicular lymphoma were considered refractory to anti-CD20 therapy.

Blood for correlative analysis was drawn before treatment and on days 1, 8, 15, 60, and 90. This permitted the measurement of plasma cytokine levels (17 cytokines) by the Multiplex Bead-based Luminex platform and lymphocyte subsets (T cells, B cells), as well as T-cell subset populations by flow cytometry. For these analyses, blood was available for 18 patients—five responders and 13 nonresponders.

Randomization during the expansion cohort allowed for testing the hypothesis that the magnitude of BCD would be increased when I and R were administered simultaneously. B-cell levels were compared in patients randomized to initial treatment with the combination of R + I (group A) versus R and delayed I (group B; Fig. 2A). Although group A had fewer B cells on D8 and D15 when compared with group B (Fig. 2A), this reduction did not reach statistical significance [P value = 0.15 (linear mixed-effects model) and P value = 0.08 (D8 Student t test)]. However, the increase in B-cell reduction in group A (R + I) is consistent with enhanced R-mediated ADCC in group A (Fig. 2A). Also of interest is that the initial effect of simultaneous R+I persisted through weeks 10 and 12 after patients on both arms were receiving ipilimumab, although this difference was attenuated. This ultimately will need to be explored further in future studies given that this trial was not sufficiently powered to detect this difference.

Exploratory and descriptive analyses were also planned to identify patterns which would explain the treatment effects, or would merit further study with the goal of identifying a potential predictive biomarker.

For the 17 plasma cytokines measured, mean values (±SD) were graphed against time according to response and based on the arm with the expansion phase (Supplementary Fig. S2). Visually, the plasma concentrations of IL2 and TNF stood out as being consistently higher in nonresponders, when compared with responders (Supplementary Fig. S2A). This difference was most pronounced on day 70 (P = 0.044 and 0.050, TNF and IL2, respectively; Supplementary Fig. S2B). In these analyses, none of the cytokine associations (Supplementary Fig. S2; Supplementary Table S3) remained significant after adjustment for multiple testing using the Bonferroni method.

For each of subset populations of T cells, the log-transformed counts (±SD) were graphed against time for responders and nonresponders (Fig. 2B and C; gating strategy shown in Fig. 2B and Supplementary Fig. S3). This revealed that increased percentages of naïve CD8⁺T cells and IFNγ-secreting CD8⁺T cells were associated with response to therapy at the first two time points (Fig. 2C–E). In contrast, effector CD4⁺T cells or effector memory CD8⁺T cells were increased in nonresponders (Fig. 2C and E). NK T cells, invariant NK T cells, and NK cells did not appear to be associated with response to therapy (Supplementary Fig. S4).

The linear mixed-effect model revealed that B cells and T cells (both quantified as log percent of peripheral blood mononuclear cells [PBMCs]) were changed over time (P = 0.005 and 0.015, respectively). B cells were calculated as the percentage of live lymphocytes that were CD19 positive and HLADR positive. The gating strategy used to identify live lymphocytes is presented in Supplementary Fig. S3, and the gating strategies to identify T cells and B cells are presented in Figs. 2B and 3B, respectively. In terms of response to therapy, the percentage of B cells (P ≤ 0.001) and CD4⁺effector cells (P = 0.047) were associated with treatment outcome and naïve CD8⁺T cells and IFNγ-secreting cytotoxic T cells trended toward significance (P = 0.064 and 0.094, respectively; Supplementary Table S4). In terms of actual lymphocyte percentages, pretreatment B-cell percentages were 21.5 (SD 16.1) and 9.0 (SD 6.7) for nonresponders and responders, respectively. These levels fell to a posttherapy low of 11.19 (SD 11.5) and 1.4 (SD 0.5), respectively. Of note, responders could be successfully classified from nonresponders by their percentages of B cells, even before initiation of therapy (Fig. 3D).

In a follow-up analysis, we examined whether the ratio of suppressive Tregs (CTLA-4⁺CD4⁺CD25⁺Foxp3⁺cells) or the ratio of CD45RA⁻Tregs (CD4⁺CD25⁺Foxp3⁺CD45RA⁻) to total Tregs could better separate responders from nonresponders (Fig. 4A–C). While when assessed independently no Treg subpopulation (Treg, CD45RA⁻Treg, suppressive Treg) was strongly associated with response to therapy (Fig. 4A), the ratio of CD45RA⁻Treg to Treg was significantly elevated in responders when compared with nonresponders at all time points (Fig. 4A and C) including at baseline (Fig. 4D). When normalized to total Treg, no other cellular subset was significantly associated with response to therapy (Supplementary Fig. S5), and no other possible paired combinations among all cell populations were clearly associated with response. To assess for the potential role of I-mediated Treg depletion, we assessed the predictive potential of the ratio in group A versus group B, and no predictive difference was observed (data not shown).

Although the majority of patients with NHL initially respond to immunochemotherapy, most eventually relapse. Resistance and intolerance to chemotherapy increase over time, making the development of nonchemotherapeutic strategies for NHL of clinical relevance. The use of CD20-targted approaches remains the standard of care both as initial therapy and in relapsed disease. Many attempts to improve the effectiveness of antibody-based CD20-targeted therapeutics have focused on strategies to enhance host immune effector mechanisms particularly in patients that are considered refractory to anti–CD20-based therapy.

In the planned dose-escalation phase of this trial, one of six patients evaluable for toxicity at the ipilimumab dose of 3 mg/kg experienced DLT. Due to non-DLT AEs in this trial as well as AEs experienced at higher doses in other trials, in collaboration with Cancer Therapy Evaluation Program (CTEP), a phase II dose of 3 mg/kg was recommended without establishing an MTD. An additional 25 patients were enrolled onto the randomized expansion cohort at 3 mg/kg of ipilimumab. The combination had moderate but manageable toxicity. Eight patients came off study due to treatment-related toxicities, the most common being grade 3–4 diarrhea (12%; Table 2). When considering the entire cohort, efficacy was modest with an overall response rate of 24% and a CR rate of 6%, with a median PFS of 2.6 (1.6–4.6) months. However, patients with follicular lymphoma, 92% of whom were refractory to prior anti–CD20-based therapy, had an ORR of 54% and a CR rate of 15% and a median PFS of 5.6 months (1.6–18.4+) months (Fig. 1; Table 3). Although the numbers were small, this compares favorably with a similar group of rituximab-refractory follicular lymphoma patients treated with ofatumumab (16), IL2/rituximab (4), lenalidomide/rituximab (7), bendamustine/obinutuzumab (17), ibrutinib (18), and idelalisib (19).

With regard to our ability to identify predictors of response, flow cytometry proved superior to serum cytokine detection. This result is likely due to the large variation in cytokine levels at baseline among individual patients and the small sample size. Of the cytokines that were elevated in nonresponders, only IL2 and TNF reached significance at several time points, but neither could predict response to therapy or remain significant if adjusted for multiple testing (Supplementary Fig. S2). However, the possibility that IL2 and TNF might be elevated in patients with progressive disease is not surprising. With regard to IL2, malignant B cells can secrete IL2, which in this setting would likely function as a prosurvival mitogen. Furthermore, when used as an immunotherapy agent, IL2 has the potential to expand Tregs, which when expanded predict treatment nonresponsiveness in other tumor types (20). Like IL2, TNF can also be secreted by malignant B cells, and therefore, increased levels of TNF over time might represent continued growth of the B-cell lymphoma. Importantly, in experimental systems, it has been demonstrated that TNF blockade can overcome resistance to anti–PD-1 therapy (21). Thus, the TNF elevation seen in our nonresponders may also be involved in cancer circumvention of immune recognition.

The randomized component of this study was designed to test the hypothesis that ipilimumab will enhance rituximab-mediated ADCC. BCD data for the two groups (R with simultaneous I versus R + delayed I) were consistent with the possibility that the addition of ipilimumab to rituximab augmented BCD (Fig. 2A); however, the difference in BCD was not significant (P = 0.08, D8 Student t test) and thus requires further studies to establish if the effect was real.

As expected, in this trial, B-cell and T-cell levels were dependent on time (P = 0.005 and 0.015, respectively; Supplementary Table S4), which was an a priori prediction given that the patients received rituximab, which depletes B cells, and ipilimumab, which allows for T-cell activation. ADCC is a well-described lymphomacidal mechanism of rituximab, and peripheral blood BCD is a known consequence. Based on this and the dependence on host immune effector function, we hypothesized that peripheral blood BCD could be used as a surrogate marker for the ADCC-mediated antilymphoma response. Indeed, B-cell depletion was not equal between nonresponders and responders [21.5 (SD 16.1) and 9.0 (SD 6.7) to a posttreatment low of 11.19 (SD 11.5) and 1.4 (SD 0.5), respectively]. The linear mixed-effects model also determined that B-cell percentage was significantly altered by responder status (P < 0.0006, and there was no remaining variability in effect size that was unexplained P = 0.72). The difference between B-cell percentages in responders versus nonresponders was also established for individual time points using the Wilcoxon rank-sum test (Fig. 3C). Finally, the ability of peripheral blood BCD to serve as a surrogate for the antilymphoma response was demonstrated by the area under the receiver-operating curves for B-cell percentages at each individual time point. Thus, the percentage of live B cells within the lymphocyte gate can predict response to therapy, that is, identify responders from nonresponders (Fig. 3D; Supplementary Fig. S6). In the future, it will be important to investigate the ability of BCD to predict response in other anti–CD20-based therapies.

Although there was an overall increase in T cells over time, not all T-cell populations behaved similarly. For example, after an initial expansion on day 8, Tregs generally decreased over time (Fig. 4A). This Treg-time association was clearly demonstrated by the linear effects model (P = 0.021, 0.047, and 0.010, for Tregs, CD45RA⁻Tregs, and suppressive Tregs, respectively; Supplementary Table S4). The Treg expansion followed by contraction is likely due in part to the expression of CTLA-4 by the suppressive Tregs. CTLA-4 is the target of ipilimumab which has been demonstrated to deplete Tregs, possibly by ADCC (22, 23). Although our study examined Treg subsets in the peripheral blood, intratumoral depletion of suppressive Tregs has been associated with response in other tumor types.

Other studies have suggested that the ratio of effector cells to Tregs may be increased in patients who respond to immunotherapy (24), but this phenomenon is likely treatment-specific as the ratio was not informative in our study. However, over the past decade Tregs (historically CD4⁺CD25⁺Foxp3⁺T cells) have been further subdivided into more defined subpopulations depending on their surface and functional phenotypes. Of these, the memory Treg subpopulation has the capacity to secrete proinflammatory cytokines and may even play a pathogenic role in the setting of autoimmunity (25). The ability of specific Treg subpopulation ratios to predict response to cancer immunotherapy, specifically the CD45RA⁻Treg to total Treg ratio, has not been previously studied. We hypothesized here that the potential inflammatory nature of some Tregs makes them more desirable in patients receiving immunotherapy. Indeed, the ratio of CD45RA⁻Tregs to conventional Tregs differed between responders and nonresponders at every time point in our study including baseline (Fig. 4A, C, and D; Supplementary Fig. S7), which highlights the reproducibility of this finding. The exact mechanism that explains the strength of the association of this ratio to response, compared with the degree of association of the absolute levels of each component, is unclear and needs further study but may suggest that there is a complex interplay within the Treg compartment. Moreover, the observation that the ratio is predictive at baseline as well as the lack of predictive potential in group A versus group B suggests that ipilimumab-induced suppressive Treg depletion (in the peripheral blood) alone is not solely responsible for enhancing immune-mediated tumor responses.

Given the relatively small sample size and the heterogeneity of the NHL subtypes, future studies are needed to determine the reproducibility of these findings. Nevertheless, the identification of a biomarker that can predict response is potentially clinically significant in patients receiving immunotherapy for lymphoma as well as other malignancies. Our study did not include validation sample sets, and thus this finding would be considered hypothesis generating, and further evaluation of the CD45RA⁻/Treg ratio to predict response to therapy is warranted.

In summary, this clinical study in relapsed and refractory B-cell lymphoma demonstrated that the combination of ipilimumab and rituximab was moderately tolerated at the dose studied. Although efficacy was modest in the entire cohort, encouraging efficacy was observed in patients with mostly anti-CD20 refractory follicular lymphoma. Moreover, the results of this trial suggest that CD45RA⁻Treg to Treg ratio has the potential to identify patients who are likely to respond to this regimen, even prior to initiation of therapy. The efficacy of this combination and the predictive potential of this biomarker need validation in larger studies of patients with follicular lymphoma.

J.M. Tuscano recieved research funding from Celgene, Novartis, Spectrum, and Takada. J.M. Tuscano received honoraria from Celgene, Amgen, and Seattle Genetics. E. Maverakis received research support from Roche, Pfizer, Amgen, and Abbvie. The family of A. Beaven has equity in GSK. M.A. Schroeder has consultancy with Amgen, Incyte, Sanofi, Flatron, and from Merk and Takada. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.M. Tuscano, E. Maverakis, S. Groshen, M. Kirschbaum, E.M. Newman

Development of methodology: J.M. Tuscano, E. Maverakis, G. Luxardi, M. Kirschbaum, E.M. Newman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Tuscano, E. Maverakis, G. Luxardi, A. Beaven, L. Popplewell, R. Chen, M. Kirschbaum, M.A. Schroeder, E.M. Newman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.M. Tuscano, E. Maverakis, S. Groshen, D. Tsao-Wei, G. Luxardi, A.A. Merleev, J.F. DiPersio, R. Chen, M. Kirschbaum, M.A. Schroeder

Writing, review, and/or revision of the manuscript: J.M. Tuscano, E. Maverakis, S. Groshen, G. Luxardi, A. Beaven, J.F. DiPersio, L. Popplewell, R. Chen, M. Kirschbaum, M.A. Schroeder, E.M. Newman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.M. Tuscano, G. Luxardi, M. Kirschbaum

Study supervision: J.M. Tuscano, E. Maverakis, M. Kirschbaum, E.M. Newman

We would like to thank Stella Khoo for her role in data analysis and editorial assistance.

This work was supported in part by NIH awards U01CA062505, UM1CA186717, P30CA033572, and P30CA093373, and the Biostatistics Core of P30CA014089. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. E. Maverakis was supported by an early career award from the Burroughs Wellcome Fund and an NIH Director's New Innovator Award (DP2OD008752).

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.