Randomized phase 2 study to assess the role of single-agent nivolumab to maintain remission in acute myeloid leukemia Open Access

For patients with acute myeloid leukemia (AML) who are not candidates for allogeneic stem cell transplant (SCT) after induction/consolidation chemotherapy, the standard of care until the recent approval of ONUREG¹ was observation alone. However, with observation alone, >50% of patients will experience a disease relapse.² The outcome of older patients with AML (aged ≥60 years) is even more dismal, with an estimated 2-year progression-free survival (PFS) of 20% without SCT.³ Thus, new postremission strategies are needed to improve clinical outcomes for patients with AML. Recently, ONUREG, an oral formulation of azacitidine, was approved for adults aged ≥55 years, for the maintenance of first complete remission (CR) after induction therapy for AML, based on the QUAZAR AML-001 trial.¹ 

With the advent of the era of immunotherapy, administration of antibodies targeting immune checkpoints has led to dramatic clinical benefit in a subset of patients with advanced solid tumors^4-7; however, patients with AML have not broadly benefited from these drugs, and these failures remain poorly understood. Multiple tumor escape mechanisms are activated in patients with active AML, which are capable of potent suppression of the antitumor immune response and help mediate disease recurrence.⁸ These include expansion of regulatory T cells⁹ and myeloid-derived suppressor cells,¹⁰ and induction of exhaustion in antigen-specific T cells, by engagement of multiple immune checkpoint axes, including the programmed cell death protein 1 (PD-1)/programmed death-ligand 1 (PD-L1) pathway.^11-13 These observations may explain why the only randomized clinical trial adding immune checkpoint inhibition to standard-of-care therapy in patients with newly diagnosed AML failed to show benefit.¹⁴ It is possible that immunotherapy for AML might have more impact in the measurable residual disease (MRD) setting, in which AML-related immune suppression is likely to be diminished. In support of this hypothesis, allogeneic SCT, for which curative potential relies largely on the antitumor immune response, is optimally performed remission to demonstrate benefit.^15-17 

Based on these data, we hypothesized that enhancing the host antileukemia immune response by targeting PD-1 might represent a powerful strategy to eliminate MRD in AML and prevent relapse. In this study, initiated and accrued before the approval of ONUREG, we conducted a multicenter randomized phase 2 trial to assess the efficacy of nivolumab (Nivo) as maintenance therapy for patients with AML who were not candidates for SCT. Patients were followed up for disease relapse, adverse events (AEs), nonrelapse mortality, and survival. The primary end point was PFS at 2 years. Secondary end points were overall survival (OS) and evaluation of AEs after Nivo administration. Further exploratory end points/objectives included eradication of MRD during treatment and elucidation of broader molecular predictors of response to immunotherapy (ClinicalTrials.gov identifier: NCT02275533).

Adult, nonpregnant patients with a history of AML were required to have histologically confirmed CR or CR with incomplete count recovery (CRi) after induction (with or without any number of cycles of consolidation) with agents per the discretion of the treating physician. Patients aged <60 years in the European LeukemiaNet favorable risk group were excluded from the study because they have an excellent PFS and are unlikely to benefit from maintenance therapy.³ Enrolled patients were not candidates for SCT because of either advanced age or comorbidities; the enrollee did not have a donor available; the enrollee declined SCT due to personal belief; or SCT was not the standard of care based on the risk category of disease (favorable risk group in patients aged ≥60 years). Patients were required to have a life expectancy of >6 months, Karnofsky performance status of at least 70%, and normal cardiac function. The following clinical laboratory values were required: serum creatinine of ≤1.5× the institutional upper limit of normal (ULN); or calculated creatinine clearance of ≥5.0 mL/min; bilirubin of 1.5× the institutional ULN; aspartate aminotransferase and alkaline phosphatase of <2.5× the institutional ULN; amylase and lipase of 1.5× the institutional ULN; white blood cells of >1.5 × 10⁹/L (1500/μL); absolute neutrophil count (ANC) of >1 × 10⁹/L (1000/μL); and platelet count of >50 × 10⁹/L (100 000/μL). Patients were excluded if they had prior treatment with an anti–PD-1, anti–PD-L1, anti–PD-L2, anti–CTLA-4 antibody, or any other antibody or drug specifically targeting T-cell costimulation or immune checkpoint pathways; known active central nervous system involvement; history of severe hypersensitivity reaction to any monoclonal antibody; uncontrolled intercurrent illness; HIV or hepatitis B or C, unless the viral load by polymerase chain reaction (PCR) was undetectable with/without active treatment and absolute lymphocyte count of ≥0.35 × 10³/μL; history of severe autoimmune disease; or if they had a condition requiring systemic treatment with corticosteroids at >10 mg daily prednisone equivalents. All patients voluntarily gave their written informed consent before therapy. The protocol was approved by National Cancer Institute Central Institutional Review Board and acknowledged by the local institutional review board of the participating centers. The trial was conducted through the Experimental Therapeutics Clinical Trials Network.

The REMAIN trial was designed as a prospective randomized, phase 2 study to compare PFS between observation (Obs) and Nivo maintenance treatment; and was activated through the Experimental Therapeutics Clinical Trials Network at the end of May 2015. Between October 2015 and November 2019, 80 patients with AML in first CR or CRi after chemotherapy were randomized onto an Obs arm or to Nivo treatment. Nivo was provided by Bristol Myers Squibb. Patients who were assigned to the Nivo arm were planned to receive Nivo at 3 mg/kg IV every 2 weeks for 2 years (46 doses). Patients on both arms were to be monitored for relapse, AEs, and survival for a minimum of 2 years after randomization. Upon disease relapse in the Obs arm, patients had the option to receive Nivo at the dose and schedule of the treatment arm (Figure 1).

Morphologic CR was defined by International Working Group criteria¹⁸ as bone marrow blasts of <5%; absence of extramedullary disease, ANC of >1.0 × 10⁹/L (1000/μL); platelet count of >100 × 10⁹/L (100 000/μL); and independence of red cell transfusions. Morphologic CRi: definition as per CR but ANC may be <1000/μL and/or platelet count of <100 000/μL with a cellular marrow (≥20%). Patients on study were monitored for disease relapse (as defined by morphologic relapse) and for cumulative incidence of relapse as defined in “Statistical Methods.”

DNA or RNA was extracted from bone marrow samples at the time of randomization using the AllPrep DNA/RNA Mini kit (QIAGEN). DNA was quantified using the Quant-iT PicoGreen double-stranded DNA assay kit and RNA using the Qubit RNA High Sensitivity assay kit (ThermoFisher Scientific).

For the 55 patients with DNA available, targeted next-generation sequencing was performed using an error-corrected duplex sequencing assay (early access version of the DuplexSeq AML MRD Library Preparation Kit for V2 Chemistry, part number 06-1021-02, TwinStrand Biosciences, Inc) targeting 36 genes frequently mutated in AML (supplemental Table 1), as previously described.¹⁹ The resulting libraries were subjected to paired-end 150–base pair sequencing using unique dual indices on a NovaSeq 6000 (Illumina) and raw FASTQ files were analyzed using the TwinStrand DuplexSeq AML-XP application on the DNAnexus platform. Resulting variant calls were filtered to remove likely germ line variants (variant allele frequency [VAF] of >30% in known predisposition gene) and identify residual variants predicted to have a deleterious variant consequence based on prior conditions deemed prognostic in the de novo setting,^20,21 which includes the removal of variants in DNMT3A, TET2, ASXL1, and TP53 (DTAT) and a VAF of ≥ 0.1% in non-DTAT genes or a VAF of ≥0.01% in NPM1 or FLT3 internal tandem duplication. Remaining variants underwent manual curation to confirm pathogenicity (supplemental Table 2). Of the patients with only RNA available, 2 patients were known to have AML harboring mutations trackable by targeted RNA sequencing (NPM1 and CBFB-MYH11). For these patients, error-corrected targeted RNA sequencing was performed on 250 ng of RNA, as previously described.²² 

The primary end point was PFS, defined as the time from randomization to disease relapse or death from any cause. Analysis was by intent-to-treat, except for the exclusion of 1 patient randomized to the Nivo arm who never received any treatment (discussed hereafter). Patients alive and progression free were censored at the date of the last negative bone marrow examination. The design of the study called for patients to be accrued over a 2-year period with follow-up continuing for 2 years after the last patient was enrolled. PFS was estimated by the Kaplan-Meier (1958) method and compared between the 2 treatment arms using a log-rank test.²³ In addition, Cox (1972) proportional hazards models were fit to provide estimates of the hazard ratio (HR), along with its associated 95% confidence interval (CI).

The sample size was determined assuming a 2-year PFS rate in the control (Obs) arm of 25% based on the literature.³ Nivo was expected to increase the 2-year PFS rate from 25% to 45% (HR, 0.58). To detect an effect of this magnitude with 80% power, a total of 80 patients were required (40 per treatment arm), based on a 1-sided test at the 0.10 α level (a more liberal α level was specified, given the phase 2 nature of the trial). A total of 62 PFS events were projected to occur over the course of the trial. An interim futility analysis was to be conducted after half the number of expected events (ie, 31/62) were observed. If, at this point, the HR did not favor the experimental arm, the study would be stopped for futility. This futility rule is associated with a minimal (<2%) reduction in power.²⁴ 

OS was analyzed in a manner similar to PFS. AEs were summarized by type and severity (grade). AE rates were compared between groups using the Fisher exact test. Patients with enrollment bone marrow samples assayed as MRD positive (MRD⁺) or MRD negative (MRD⁻) were further interrogated for PFS and OS. The cumulative incidence of relapse (treating nonrelapse mortality as a competing risk) was compared between MRD⁺ or MRD⁻ patients using the Fine-Gray test.²⁵ 

In addition to the intent-to-treat analysis, we performed a sensitivity analysis to examine the effects of treatment discontinuation in the Nivo arm and of crossover to Nivo after disease relapse in the Obs arm. In this analysis, patients were censored at the time of treatment discontinuation/crossover. It is known, however, that such analyses are subject to bias.²⁶ It must be assumed that patients who discontinue/switch their treatment at time “t” (and are thereby censored) do not differ from those who continue on treatment at time t, an assumption that cannot be verified.

We used the Delta Gene assay panel from the Advanta Immuno-Oncology Expression Assays, Panel A (96 gene identities) and B (79 gene identities) from Standard Biotools (former Fluidigm). This panel was custom converted from TaqMan to DeltaGene assay format for this study. Data were collected using the BioMark HD system (Standard Biotools). RNA samples were converted to complementary DNA and underwent target specific amplification preamplification, which was performed with a mix of all assays (separately for pool A and B) per manufacturer’s protocols. All samples were preamplified with 14 PCR cycles. Preamplified complementary DNA samples were diluted 1:5 and used for setting up reverse-transcription PCR reactions with individual assays using the 96.96 integrated fluidic circuit format. Fluidigm Real-Time PCR Analyses software was used to collect the data, and the following analysis parameters were applied: quality threshold at 0.5, baseline correction linear, and Ct threshold method on Auto.

A total of 82 patients were enrolled in the trial between October 2015 and November 2019 (Figure 2). However, the first 2 patients inadvertently received open-label Nivo. One patient randomized to the Nivo arm never received any drug and was declared nonevaluable, leaving n = 79 for analysis. In terms of completeness of follow-up, 1 patient in the Obs arm withdrew, and 1 was lost to follow-up before 2 years (2 additional patients withdrew, and 1 was lost to follow-up after disease progression). Two patients in the Nivo arm withdrew and 1 was lost to follow-up before 2 years (1 additional patient withdrew and 1 was lost to follow-up after disease progression). In the Nivo arm, 17 patients progressed or died on treatment and 8 patients completed the protocol-stipulated 1 year of therapy. Thirteen patients discontinued Nivo before 2 year: 8 because of AEs, 2 refused further treatment, 1 withdrew, 1 stopped because of intercurrent illness, and 1 discontinued for other reasons. Four patients in the Obs arm elected to cross over to Nivo upon disease progression. The clinical characteristics of the patients are listed in Table 1. A total of 55 events (disease relapse or death) were observed.

Regarding the primary end point, PFS, a total of 29 patients on the Obs arm (70.7%) relapsed or died (1 died without prior relapse) compared with 26 patients (68.4%) on the Nivo arm (3 died without prior relapse). The PFS curves were not significantly different (log-rank P = .38, 1-sided; Figure 3.) Median PFS was 13.2 months in the Nivo arm (95% CI, 8.5-21.8) and 10.9 months in the Obs arm (95% CI, 5.4-14.9). The 2-year PFS rate was 30.3% in the Nivo arm (95% CI, 16.2-45.8) and 30.0% in the Obs arm (95% CI, 16.8-44.4). The Cox regression model–generated HR for Nivo/Obs was 0.92, with a 95% CI ranging from 0.54 to 1.56. In the 4 patients on the Obs arm who crossed over to receive Nivo treatment at the time of disease progression the duration of survival after progression was 13 months, 4 months, and 7 months with 1 patient lost to follow-up, respectively.

The secondary end points were OS and AEs after Nivo administration. In terms of OS, 20 patients on the Obs arm (48.8%) died compared with 16 patients on the Nivo arm (42.1%). The OS curves were not significantly different (log-rank P = .23, 1-sided; Figure 4.). The median OS was 53.9 months in the Nivo arm (95% CI, 23.4 to not estimable) and 30.9 months in the Obs arm (95% CI, 14.4 to not estimable). The 2-year OS was 60.0% in the Nivo arm (95% CI, 41.9-74.1) and 52.8% in the Obs arm (95% CI, 35.9-67.2). The Cox regression model–generated HR for Nivo/Obs was 0.78, with a 95% CI ranging from 0.40 to 1.51.

Sensitivity analyses were performed to examine the effects of compliance to Nivo treatment and crossovers on the results. As shown in the CONSORT diagram, 13 patients discontinued Nivo before the prescribed 2 years of treatment, and 4 patients in the Obs arm crossed over to Nivo upon disease relapse. If these observations are censored at the time of treatment discontinuation/crossover, the HR for PFS changes from 0.92 to 0.81 (95% CI, 0.45-1.45; 1-sided P = .24), that is, a little stronger but still not statistically significant. The HR for OS changes from 0.78 to 0.83 (95% CI, 0.39-1.77; P = .31).

There were more AEs of any type (regardless of attribution) on the Nivo arm; 5 patients (12%) on the Obs arm had a grade ≥3 AE compared with 27 patients (71%) on the Nivo arm (P < .001; Table 2.)

Eight patients on the Nivo arm (21%) had a serious AE (SAE) compared with 3 patients (7%) on the Obs arm (P = .11). The AEs on the Nivo arm were all expected and manageable; the most common nonhematologic AEs on the Nivo arm were fatigue (42%); hypertension and aspartate transferase elevation (37%); diarrhea (34%); and hyperglycemia (32%; Table 3).

Further predefined exploratory end points included measurement of MRD at the time of randomization, to attempt to identify biomarkers of benefit with Nivo. DNA or RNA at the time of randomization before treatment was available for 57 patients (72%) and MRD was assessed by ultrasensitive duplex DNA sequencing (n = 55) or targeted RNA sequencing (n = 2). A total of 126 potentially deleterious residual variants were detected by duplex DNA sequencing across 26 genes with a median VAF of 0.42% (range, 0.01%-41.1%) and 2 residual variants (NPM1 and CBFB-MYH11) by targeted RNA sequencing (Figure 5A; supplemental Table 2). The most frequent residually mutated gene was PPM1D. Of the 57 patients analyzed, 43 (75%) had residual variants detected before randomization. Patients with MRD (MRD⁺) at the time of randomization had significantly increased rates of relapse (MRD⁺ vs MRD⁻ at 2 years, 74% vs 37%; P = .046) and decreased PFS (MRD⁺ vs MRD⁻ at 2 years, 21% vs 63%; P = .032) compared with patients without MRD (MRD⁻; Figure 5B). However, subgroup analysis of MRD⁺ patients did not demonstrate any significant impact of Nivo on relapse (Nivo vs Obs at 2 years, 81% vs 67%; P = .36), PFS (Nivo vs Obs at 2 years, 14% vs 29%; P = .43), or OS (Nivo vs Obs at 2 years, 51% vs 52%; P = .92; supplemental Figure 1).

Differential expression analysis of bone marrow sample RNA expression, using the Fluidigm platform, did not demonstrate an effect of Nivo. (supplemental Figure 2). There were no genes significantly different between enrollment bone marrow samples assessed as positive or negative for MRD (supplemental Figure 3).

The results of this randomized phase 2 trial did not demonstrate evidence that maintenance Nivo prolonged PFS or OS in patients with AML who were not candidates for SCT. Treatment with Nivo was associated with a greater number of AEs, the most common of which were fatigue, anemia, hypertension, aspartate transferase elevation, diarrhea, and hyperglycemia, all of which were all manageable. Genomic analysis of pretreatment bone marrow samples revealed evidence of MRD in a subset of assayed samples, and these patients subsequently had significantly increased rates of relapse and decreased PFS, in keeping with previous studies.²⁷ Subgroup analysis in the cohort of MRD⁺ patients did not demonstrate any impact of Nivo on clinical outcomes, but these comparisons were underpowered. Commensurate with these results, longitudinal RNA profiling of bone marrow samples did not demonstrate an effect of Nivo, nor a distinct transcriptome in MRD⁺ bone marrow.

This study represents, to our knowledge, the first randomized trial of PD-1 blockade in AML and did not show benefit of PD-1 in the setting of CR after chemotherapy. Checkpoint blockade is thought to induce antitumor activity in an antigen-dependent manner. As such, it is possible that checkpoint inhibition may only show benefit in an MRD⁺ setting. Indeed, in a previously published small single-arm phase 2 trial of the anti–PD-1 antibody Nivo in 15 high-risk patients in CR who were not being evaluated for SCT, 2 of 9 patients with MRD at the beginning of the study cleared this disease while on Nivo and had durable remissions.²⁸ Subgroup analysis of the effect of Nivo in the cohort of MRD⁺ patients in our study did not show benefit with treatment, but the trial was not designed nor powered to answer this question. A recent single-arm study of PD-1 maintenance in the postautologous transplant setting demonstrated the safety and efficacy of PD-1 maintenance in a similar low-disease burden setting,²⁹ but MRD was not assayed before the start of PD-1 therapy, and lack of a comparator arm limits conclusions.

Other immune escape mechanisms, even beyond checkpoint pathways, likely contribute to the persistence of this disease. With negative results using checkpoint inhibitors alone (or with combination with HMA), now in both the MRD setting and in the front line, the utility of targeting the PD1-PDL1 pathway in unselected AML appears limited. It remains possible that further correlatives studies might identify a subset of patients with AML who may responses to immune checkpoint inhibition. A future role for immune checkpoint inhibitors might exist in combination with other agents simultaneously targeting multiple immune pathways. Indeed, the recent demonstration of the benefit of ONUREG in this group of patients, regardless of MRD status, is pertinent.

In summary, Nivo maintenance after AML chemotherapy failed to improve PFS and OS in this randomized phase 2 study. There were increased AEs and SAEs with Nivo maintenance, but these AEs and SAEs were expected and manageable. Further studies should be done to evaluate the potential use of immune checkpoint blockade in the context of MRD.

This work was supported, in part, by the Intramural Research Program of the National Heart, Lung, and Blood Institute. Additional support was provided by a University of Chicago Comprehensive Cancer Center pilot grant, the Ullman Scholar Award, and the Elsa U. Pardee Foundation (H.L.). T.F.G. was, in part, supported by the P30 CA014599-41S4 grant (titled “Administrative Supplements to Support Biomarker Studies Associated with National Cancer Institute [NCI]-Supported Clinical Trials of Immunotherapy”), serving as Principal Investigator; and the Cancer Center Support grant-Experimental Therapeutics Clinical Trials Network (ETCTN) Administrative Supplement (Leadership), titled “P30 Cancer Center Support Grant Administrative Supplements to NCI Approved Clinical Trial Proposals from NCI-Designated Cancer Centers not Affiliated with the NCI ETCTN for Investigator-Initiated Trials Utilizing CTEP IND agents in the ETCTN,” was awarded for the years 2016 to 2022 and was led by Michelle Le Beau.

Contribution: A.R.P. analyzed the data and wrote the manuscript; L.W.D., T.G.K., Y.Z., N.F., G.G., and G.A. analyzed the data; E.S., H.S., K.S., G.Y., J.J.L., B.A.J., A.D.S., S.G., A.M.Z., G.C.H., C.H.L., R.J.M., N.P., A.S., D.T., M.R.B., T.L.L., P.P., H.C., W.M.S., O.O., R.A.L., and T.F.G. enrolled patients on the study; C.S.H. and W.S. designed the research and enrolled patients on the study; and H.L. wrote the protocol, designed the research, and enrolled patients in the study.

Conflict-of-interest disclosure: H.L. served on an advisory board meeting for Rigel and Incyte, and received consultation fees from AbbVie in the past 24 months. B.A.J. served in a consultancy/advisory role for AbbVie, Bristol Myers Squibb (BMS), Daiichi Sankyo, Gilead, Kura, Rigel, Schrodinger, Syndax, and Treadwell; served on the data monitoring committee for Gilead; received travel reimbursement/support from Rigel; and received research funding to institution from AbbVie, Amgen, Aptose, AROG, Biomea, BMS, Celgene, Forma, Forty Seven, Genentech/Roche, Gilead, GlycoMimetics, Hanmi, Immune-Onc Therapeutics, Jazz, Kymera, Loxo, Pfizer, Pharmacyclics, and Treadwell. A.M.Z. participated on advisory boards, had a consultancy with, and received honoraria from, AbbVie, Pfizer, Celgene/BMS, Jazz, Incyte, Agios, Servier, Boehringer Ingelheim, Novartis, Astellas, Daiichi Sankyo, Geron, Taiho, Seattle Genetics, Otsuka, BeyondSpring, Takeda, Ionis, Amgen, Janssen, Genentech, Epizyme, Syndax, Gilead, Kura, Chiesi, ALX Oncology, BioCryst, Notable, Orum, Mendus, Zentalis, Schrodinger, Regeneron, Syros, and Tyme; and served on clinical trial committees for Novartis, AbbVie, Gilead, Syros, BioCryst, AbbVie, ALX Oncology, Kura, Geron, and Celgene/BMS. T.L.L. served in a consulting or advisory role at Servier, Jazz Pharmaceuticals, Daiichi Sankyo, Syndax, and received research payments to the institution from BioPath Holdings Inc, Astellas Pharma, Celyad, Aptevo Therapeutics, Cleave Biosciences, CicloMed, Jazz Pharmaceuticals, Cardiff Oncology, and Kura Oncology. A.D.S. has received research funding from Takeda Pharmaceuticals, BMS, and Medivir AB; received consulting fees/honorarium from Takeda, Novartis, Jazz, and Otsuka Pharmaceuticals; is named on a patent application for the use of double-negative T cells cells to treat acute myeloid leukemia; and is a member of the medical and scientific advisory board of the Leukemia and Lymphoma Society of Canada. R.A.L. has acted as a consultant or adviser to Ariad/Takeda, CVS/Caremark, Daiichi Sankyo, Epizyme/Ipsen, and Novartis; has received clinical research support to his institution from Astellas, Biomea Fusion, Cellectis, Daiichi Sankyo, and Novartis; and has received royalties from UpToDate. P.P. is an employee of Servier Pharmaceuticals since April 2022. A.R.P. is an employee of AbbVie Pharmaceuticals since January 2024. E.S. reports consulting with D.E. Shaw Research and served on the advisory board at Mallinckrodt Pharmaceuticals. A.S. received research funding from Rigel, BMS, and Orcabio; served on the speakers bureau with Sanofi; and served on the advisory boards of Servier and Incyte. N.P. reports consulting with BMS, Servier, and Rigel. O.O. received research funding paid to the institution from Astex, AbbVie, Kartos Therapeutics, Lox Therapeutics, Immune-Onc Therapeutics, AstraZeneca, and Shattuck Labs; participated on scientific advisory boards for AbbVie, Rigel, Servier, and Incyte; and served on the data/safety monitoring board at Threadwell Therapeutics. The remaining authors declare no competing financial interests.

Correspondence: Hongtao Liu, University of Wisconsin-Madison, Medicine, Wisconsin Institutes for Medical Research, 1111 Highland Ave, Madison, WI 53705-2275; email: [email protected].