Nivolumab and Relatlimab for the treatment of patients with unresectable or metastatic mismatch repair proficient colorectal cancer

Abstract

Purpose:

Combined inhibition of lymphocyte-activation gene 3 (LAG-3) and programmed cell death protein 1 (PD-1) improves outcomes in patients with melanoma. Increased LAG-3 expression in colorectal cancer (CRC) correlates with reduced survival. Higher mucin and PD-L1 expression in the mismatch repair proficient (pMMR)

CRC tumor microenvironment (TME) was associated with increased LAG-3 and retrospectively with prolonged progression-free survival upon PD-1 blockade. This led to the hypothesis that LAG-3/PD-1 inhibition would improve clinical outcomes in this pMMR CRC subset.

Methods:

NCT03642067 was a phase 2 study evaluating combining relatlimab (LAG-3 inhibitor) and nivolumab (PD-1 inhibitor) in patients with previously treated metastatic pMMR CRC. Patients were enrolled onto one of three cohorts, A: mucin/PD-L1 high, B: mucin/PD-L1 low, or C: mucin/PD-L1 unselected. The primary endpoint for each cohort was overall response rate.

Results:

We enrolled 59 evaluable patients; best treatment responses were partial response: 3, stable disease: 6, progressive disease: 50. Response rates did not differ significantly between cohorts. Subgroup analyses demonstrated 2 of 5 patients with lung-only metastases had a partial response. Comparison of liver and lung metastases identified higher baseline dendritic cell densities in lung lesions. Nivolumab/relatlimab resulted in increased intratumoral cytotoxic T cells. Lower baseline intratumoral Tregs and ADAM10⁺ cancer cells correlated with clinical response.

Conclusions:

This investigation did not reach its primary endpoint for any of the three treatment cohorts, but does provide critical insight into the effects of combining nivolumab/relatlimab on the CRC TME and identifies subgroups that may derive greater benefit from this combination.

Statement of Translational Relevance

Lymphocyte-activation gene 3 (LAG-3) is an emerging cancer immunotherapy target with studies in melanoma demonstrating promising clinical activity in combination with programmed cell death protein 1 (PD-1). We tested whether combining LAG-3 (relatlimab) and PD-1 (nivolumab) inhibition improved outcomes in patients with metastatic mismatch repair proficient colorectal cancer. Based on previous investigations, we hypothesized that elevated mucin and PD-L1 expression in tumors would predict treatment response to checkpoint blockade and included a high mucin/PD-L1 cohort. Combining nivolumab and relatlimab failed to provide clinical benefit in the overall study population or high mucin/PD-L1 cohort. Subgroup and correlative analyses identified patients with baseline lung-only metastases, low densities of Tregs and ADAM10 expression were all enriched in patients that responded. While this study did not meet its primary endpoint, it provides important biologic insight into the effects of LAG-3/PD-1 inhibition, identifies subgroups for further investigation of LAG-3/PD-1, and informs future combinatorial approaches.

Introduction

Colorectal cancer (CRC) represents the second leading cause of cancer related death in the United States.^1,2 While immune checkpoint inhibitors (ICIs) have demonstrated significant success in mismatch repair deficient (dMMR) colorectal cancer. Some but too few patients with mismatch repair proficient (pMMR) colorectal cancer (CRC) have benefited from this therapeutic revolution.^3,4 The limited clinical response is partially driven by an immunosuppressive tumor microenvironment (TME) enriched for immune checkpoint expression, regulatory T cells (Tregs) and immunosuppressive myeloid cell populations that circumvent the development of a productive anti-tumor immune response through inducing CD8⁺ lymphocyte apoptosis and anergy.⁵

Lymphocyte-activation gene 3 (LAG-3) is a potent mediator of immunosuppression found on immune cells within the CRC TME. LAG-3 binds to major histocompatibility complex-II (MHC II) molecules on the surface of antigen presenting cells (APCs) with a higher affinity than CD4 receptors producing inhibitory signaling and directly hindering CD4⁺ T cell activation.^6–9 Pre-clinical studies suggest inhibition of LAG-3 enhances effector T cell (Teff) activation and proliferation to promote a more productive immune response.¹⁰ These observations led to development of the FDA approved LAG-3 inhibitor, relatlimab. The addition of relatlimab to nivolumab leads to significant clinical benefit for patients with melanoma and is currently approved as first line treatment for patients with metastatic disease.¹¹

In patients with CRC, prior correlative studies have linked increased LAG-3 levels in the TME or draining lymph nodes with reduced overall survival albeit overall LAG-3 expression levels are lower than in melanoma.^8,9,12 These observations prompted optimism that inhibiting this immune checkpoint in combination with programmed cell death protein 1 (PD-1)/programmed cell death protein ligand 1 (PD-L1) blockade might lead to anti-tumor responses in patients with metastatic pMMR CRC through the simultaneous inhibition of these two immunosuppressive signals (PD-1 and LAG-3).

Previous investigations at our institution in pMMR and dMMR CRC identified increased tumor associated extracellular mucin and PD-L1 expression were more common in patients that had favorable responses to immune checkpoint blockade.¹³ Of note, the majority of patients in this analysis with PD-1 responsiveness had dMMR disease. Additionally, an exploratory analysis from the phase I/II trial of pembrolizumab (aPD1)/favezelimab (aLAG3) identified a numerically higher objective response rate (ORR) and overall survival (OS) in patient whose tumors had a PD-L1 CPS ≥ 1.¹⁴ These data suggested these features might help select patients likely to benefit from a relatlimab/nivolumab approach.

To evaluate whether the addition of a LAG-3 inhibitory antibody to anti-PD-1 would improve outcomes in selected or unselected patients, we initiated a clinical trial NCT03642067 to evaluate the clinical activity of relatlimab and nivolumab in patients with metastatic pMMR CRC.

Methods

Study design

NCT03642067 was an open-label, single center phase II investigation at the Sidney Kimmel Comprehensive Cancer Center (SKCCC) at Johns Hopkins University (JHU) evaluating the efficacy of combining relatlimab and nivolumab in the treatment of metastatic pMMR CRC which had progressed on at least 1 line of prior therapy. Study results are reported per the CONSORT trial guidelines. There was no limit on the maximum number of prior lines of treatment. This study enrolled patients onto one of three cohorts with the initial two using a previously described composite PD-L1/Mucin (CPM) score to place patients into CPM positive: Cohort A (sum of % mucin + % PD-L1 staining greater than or equal to 15%) and CPM negative: Cohort B (sum of % mucin + % PD-L1 staining less than 15%) cohorts.¹³ Patients enrolled on Cohorts A and B received relatlimab 160mg and nivolumab 480mg intravenously every 4 weeks. Based on prior work we hypothesized that patients in the CPM positive group would likely have better treatment responses.¹³ An additional cohort was later added (Cohort C) enrolling pMMR CRC patients without biomarker selection with a higher planned dose of relatlimab to determine if this would lead to better anti-tumor activity. This decision was based on data suggesting improved pharmacokinetics and LAG-3 receptor occupancy on CD8⁺ effector memory T cells with higher relatlimab doses. Patients in Cohort C were planned for dosing at 960mg relatlimab/480mg nivolumab, however, following two cases of fatal myocarditis/myositis in patients treated with 960mg relatlimab/480mg nivolumab on a hepatocellular carcinoma trial, all patients on active treatment were initially switched to 160mg relatlimab (3 patients total). Further safety analysis was performed and the decision was made to enroll all subsequent patients at 480mg of relatlimab. After completion of enrollment onto cohort C, in consultation with the study sponsor, it was decided against further enrollment onto Cohorts A/B.

Each cohort was designed to enroll up to 32 evaluable patients with a Simon two-stage minimax design for each cohort. Initially, 18 patients per cohort were enrolled with assessment for response. If at least one of those 18 demonstrated a response, then an additional 14 patients were enrolled in Stage 2 for a total of 32 patients. Tumor biopsies were obtained at the time of study enrollment and prior to cycle 2 day 1. Biopsies were required only as long as they were considered safe for the patient. Peripheral blood mononuclear cells (PBMC), serum, and plasma were obtained at baseline, 4 weeks, and every 16 weeks thereafter, and computed tomography (CT) imaging was obtained at baseline, 12 weeks, and every 8 weeks thereafter during treatment for clinical assessment and correlative analyses.

Patients were treated until radiographic or clinical disease progression (allowing patients to continue treatment beyond radiographic progression once if suspicion for pseudoprogression and clinically stable), unacceptable adverse events, or withdrawal of consent. The protocol was approved by the institutional review board (IRB) at Johns Hopkins University and US Food and Drug Administration (FDA), and complied with the International Ethical Guidelines for Biomedical Research Involving Human Subjects and the Declaration of Helsinki. All patients provided written informed consent for this study.

Patients

Patients were eligible for this trial if they were 18 years or older with histologically confirmed metastatic CRC. Patients were required to have Eastern Cooperative Oncology Group (ECOG) performance status 0 or 1 and adequate organ function as defined by absolute neutrophil count ≥ 1500/µL, platelets ≥ 100 000/µL, hemoglobin ≥ 8.5g/dL, serum creatinine ≤ 1.5 x ULN, total bilirubin ≤ 1.5 x ULN, AST and ALT ≤ 3 x ULN, and left ventricular ejection fraction (LVEF) ≥ 50%. Patients were excluded if they had prior treatment with anti-PD-1, anti-PD-L1, anti-PDL2, anti-CTLA4, or anti-Lag-3 antibodies. The patient’s tumor had to be deemed microsatellite stable and/or pMMR by a Clinical Laboratory Improvement Amendment (CLIA) certified laboratory using either polymerase chain reaction or next generation-based microsatellite testing or immunohistochemistry for MMR proteins. For cohorts A and B, CPM was used to determine which cohort they would be enrolled in with patients with a CPM ≥ 15 (designated as CPM⁺) enrolling onto Cohort A and CPM < 15 (designated as CPM⁻) enrolling onto Cohort B. Patients were required to have measurable disease according to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 criteria.

Assessment of Clinical Endpoints

The primary endpoint for this study was objective response rate (ORR). ORR was estimated as the proportion of subjects whose best overall response was either a CR or PR by RECIST 1.1. Patients were considered evaluable for the primary endpoint if they received at least one dose of study drug combination. Each cohort was analyzed separately for ORR with an ORR of 5% or less leading to the regimen being considered inactive and not an improvement over standard of care treatment whereas an ORR of 20% or greater would suggest further investigation is warranted.^15,16

Safety was assessed throughout trial participation with adverse events recorded and graded using common terminology criteria for adverse events (CTCAE) version 5.0. A safety run-in was used for the first six patients for Cohorts A and B only. If more than 1 of the first 6 patients experienced an unacceptable toxicity within the first cycle, then enrollment would be halted and the overall risk-benefit ratio of the study reconsidered. At any time thereafter, if more than 33% of patients were observed to experience unacceptable toxicity within the first cycle, enrollment would be halted and the safety of the combination reevaluated.

Exploratory end points included progression-free survival (PFS), overall survival (OS), disease control rate (DCR), and duration of response (DOR). PFS and OS were defined as time from the first day of study treatment to the date of event of interest. DOR was defined as the time between the date of initial complete or partial response to the date of the first documented tumor progression or death due to any cause.

Prior clinical investigations have demonstrated CRC with liver metastases is less responsive to immunotherapy than CRC with other sites of metastasis.^17,18 Therefore as an exploratory analysis we assessed the influence of metastatic location on response looking at three different patient subgroups by disease location agnostic of treatment cohort (with liver metastases, without liver metastases, and lung-only metastases).

Assessment of Correlative Endpoints

Correlative studies were performed to identify baseline biomarkers that might prospectively identify subjects likely (or unlikely) to respond to treatment and establish circulating and intratumoral changes with nivolumab/relatlimab treatment. Circulating immune cells were profiled using mass cytometry on PBMCs collected pre-treatment and prior to cycle 2 day 1. Tumor microenvironment assessments were made on biopsy specimens obtained pre-treatment and prior to cycle 2 day 1 using immunohistochemistry and imaging mass cytometry. Immunohistochemistry staining for PD-L1 (clone 28–8) and LAG-3 (clone 17B4) testing was performed by Labcorp (Burlington, NC) with % positive staining confirmed by a second pathologist at JHU (R. Anders) while mass cytometry and imaging mass cytometry were performed in the Johns Hopkins Cancer Convergence Institute Mass Cytometry Facility.

CyTOF staining

Viably frozen aliquots of PBMCs were recovered with 250 µL of Benzonase media. Cells were washed with 2 mM EDTA diluted in Maxpar PBS. Viability staining was carried out using 5 μM cis-platinum (Standard Biotools) in Maxpar PBS for 2.5 minutes at room temperature (RT), followed by media quenching. After two washes in Maxpar Cell Staining Buffer (CSB), samples were stained with CD45-based barcodes for 25 minutes at RT using a 7-choose-3 scheme to allow up to 35-plexing. After three washes in Maxpar CSB, samples were batched, Fc blocked (Invitrogen), and surface stained using an antibody cocktail for chemokine receptors at 37°C for 10 minutes and then rest of the markers at RT for 30 minutes (Supplementary Table 1). Cells were subsequently washed and fixed-permeabilized using the Fix/Perm Kit (eBioscience) to be stained with intracellular antibody cocktail (Supplementary Table 1). Fully stained cells were fixed with 1.6% PFA (Thermofisher) for 15 minutes at RT and then labeled with 1:500 rhodium (Standard Biotools) for 30 minutes at RT. Data was acquired using Helios at the Johns Hopkins Mass Cytometry Facility. Research resource identifiers (RRIDs) for key reagents are listed in (Supplementary Table 1).

CyTOF preprocessing and analysis

Data were randomized, normalized using beads, and cleared of beads using CyTOF software (v6.7, Standard Biotools) (RRID:SCR_021055). Using FlowJo (v10.5, BD Biosciences) (RRID:SCR_008520) cells were gated based on rhodium and event lengths, platinum signal to gate out dead events, and then debarcoded into individual samples based on hierarchical gating of 7 CD45 barcode channels. Processed FCS files were loaded onto R v4.2.2 (RRID:SCR_001905) for further analysis. The FlowSOM (RRID:SCR_016899) was used to cluster the data.¹⁹

CPM scoring (mucin and PD-L1)

Formalin-fixed paraffin embedded (FFPE) tissue sections from resected colon tumors obtained at diagnosis were stained with a hematoxylin and eosin (H&E) combination. Extracellular mucin pools were defined as the collection of mucin not associated with malignant epithelial cells and quantified as the percentage of the tumor surface area replaced by mucin.¹³ Digital quantification of tumor and mucin areas was performed utilizing the HALO™ image analysis platform (RRID:SCR_018350) from Indica labs. FFPE colon tumor tissue sections were also stained for PD-L1 (clone SP263) as previously reported. PD-L1 was scored at the invasive front, which is the region where the tumor tissue juxtaposes the normal colonic tissue.¹³ Composite PD-L1/Mucin (CPM) score was calculated as the sum of % mucin + % PD-L1 staining.

IMC staining and image acquisition

Among the 59 patients enrolled, only 38 patients had paired samples for baseline and at cycle 2. Metastatic sites of biopsy ranged from liver, lung, abdominal wall, peritoneal soft tissue, lymph node, and bone. Tissue sites absent in either responders or non-responders were considered prone to biasing the correlative analysis; samples from liver, lung, and abdominal wall were prioritized. Review of tissue quality based on H&E, ensuring the presence of viable tumor, led to the final selection of 24 patient’s samples, including 15 paired and 9 unpaired for IMC. Slides were baked for 2 hours in preheated oven at 60°C, deparaffinized with xylene for 20 minutes with light agitation, and rehydrated using an alcohol gradient (100%, 95%, 80%, 70% EtOH for 5 minutes each). The slides were washed in Maxpar water (Standard Biotools) and then placed in target retrieval solution (Antigen Retrieval Agent, pH 9, 1:10 in Maxpar water) equilibrated at sub-boiling temperature (90~95°C) for one hour followed by Maxpar PBS wash for 10 minutes with gentle agitation. Sections were blocked using 3% BSA for 45 minutes at RT followed by incubation in the IMC antibody cocktail solution (Supplementary Table 2) overnight at 4°C. For nuclear counterstaining, sections were stained with Cell-ID Intercalator-Ir (Standard Biotools) diluted at 1:400 in Maxpar PBS for 30 minutes at RT. Tissues were counterstained with ruthenium tetroxide 0.5% (Electron Microscopy Sciences PN 20700–05) at 1:2000 for 3 minutes at RT. After washing in Maxpar water, slides were air-dried and loaded in the Hyperion Imaging Plus System (Standard Biotools) (RRID:SCR_023195) for image acquisition at the Johns Hopkins Mass Cytometry Facility. Regions of interest with representative viable tumor tissue from each core needle biopsy specimen were selected for ablation by an expert pathologist (RAA) based on H&E. RRIDs for key reagents are listed in (Supplementary Table 2).

IMC data preprocessing and analysis

Representative images were created and exported using MCD Viewer 1.0560.2 (Standard Biotools) (RRID:SCR_023007). A software pipeline based on DeepCell (RRID:SCR_022197) and HistoCAT 1.76 (RRID:SCR_026499) were used for image segmentation. Single-cell datasets were imported into R 4.2.2 for further analysis.^20,21 Clusters were generated with PhenoGraph (RRID:SCR_016919), a single-cell clustering software that automatically determines and generates the optimal number of metaclusters in an unsupervised manner.²² The data was first clustered globally, generating 57 metaclusters that were then annotated into 11 broad cell types based on the expression profiles of canonical markers. To enable granularity of cell phenotypes, epithelial and T cell clusters were subsequently sub-clustered separately into 36 metaclusters annotated into 5 epithelial subtypes and 16 metaclusters annotated into 7 T cell subtypes, respectively. Upon annotation, the dataset constituted a total of 22 clusters, as shown in the Results section. X and Y coordinates of each cell were used for calculating distance relationships using spatstat. Plots were generated using ggplot and igraph.

Statistical Methods

Survival data for each cohort was plotted using a Kaplan-Meier curve and compared using Log-Rank testing. Hazard ratio (HR) and 95% confidence intervals (CIs) were assessed by Cox regression models. For comparisons between cohorts, cohort C was used as the reference population for hazard ratios. Differences in clinical benefit between disease locations was determined using Fisher exact testing. R 4.2.2 and GraphPad Prism 9 (RRID:SCR_002798) were used for paired t-tests for timepoint comparisons and unpaired t-tests for response comparisons of cytometric correlatives. All tests were 2-sided and p <0.05 was considered statistically significant.

Results

Patient characteristics

From 2/2019 to 5/2023, a total of 59 patients received study treatment on this trial. The demographic characteristics of these patients are presented in Table 1 and Supplementary Table 3. Patients had received 1 to 7 lines of prior treatment. Supplementary Table 4 is included to outline the representativeness of these study participants to the general population with CRC.

Table 1:

Patient Characteristics for NCT03642067: Nivolumab and Relatlimab in pMMR Colorectal Cancer

	Cohort A (N=12)	Cohort B (N=15)	Cohort C (N=32)	Overall (N=59)
Age (years)  Mean (SD)  Median [Min, Max]	54.9 (11.0) 52.5 [40.0, 83.0]	56.3 (8.12) 58.0 [44.0, 71.0]	52.4 (11.6) 53.5 [27.0, 74.0]	53.9 (10.6) 54.0 [27.0, 83.0]
Gender  Female  Male	4 (33.3%) 8 (66.7%)	9 (60.0%) 6 (40.0%)	17 (53.1%) 15 (46.9%)	30 (50.8%) 29 (49.2%)
Race  Asian  Black  White  Other	1 (8.3%) 3 (25%) 8 (66.7%) 0 (0%)	1 (6.7%) 4 (26.7%) 9 (60.0%) 1 (6.7%)	1 (3.1%) 9 (28.1%) 21 (65.6%) 1 (3.1%)	3 (5.1%) 16 (27.1%) 38 (64.4%) 2 (3.4%)
Ethnicity  Hispanic or Latino  Not Hispanic or Latino	1 (8.3%) 11 (91.7%)	2 (13.3%) 13 (86.7%)	0 (0%) 32 (100%)	3 (5.1%) 56 (94.9%)
Cancer Site  Left Sided Colon  Rectum  Right Sided Colon	8 (66.7%) 1 (8.3%) 3 (25%)	9 (60.0%) 4 (26.7%) 2 (13.3%)	16 (50.0%) 5 (15.6%) 11 (34.4%)	33 (55.9%) 10 (16.9%) 16 (27.1%)
Differentiation  Moderately differentiated  Poorly differentiated  Unknown	9 (75%) 3 (25%) 0 (0%)	12 (80.0%) 2 (13.3%) 1 (6.7%)	29 (90.6%) 3 (9.4%) 0 (0%)	50 (84.7%) 8 (13.6%) 1 (1.7%)
Prior Surgery  Yes	12 (100%)	15 (100%)	32 (100%)	59 (100%)
Prior Radiation Therapy  No  Yes	11 (91.7%) 1 (8.3%)	9 (60.0%) 6 (40.0%)	21 (65.6%) 11 (34.4%)	41 (69.5%) 18 (30.5%)
Number of Prior Chemotherapy Regimens  Mean (SD)  Median [Min, Max]	2.92 (1.31) 2.50 [1.00, 5.00]	3.07 (1.49) 3.00 [1.00, 7.00]	2.97 (1.36) 3.00 [1.00, 6.00]	2.98 (1.36) 3.00 [1.00, 7.00]
RAS-RAF status  Wild type  Altered  Unknown	3 (25%) 8 (66.7%) 1 (8.3%)	9 (60.0%) 6 (40.0%) 0 (0.0%)	10 (31.3%) 21 (65.6%) 1 (3.1%)	22 (37.3%) 35 (59.3%) 2 (3.4%)
Liver Metastasis (at baseline)  No  Yes	4 (33.3%) 8 (66.7%)	4 (26.7%) 11 (74.4%)	7 (21.9%) 25 (78.1%)	15 (25.4%) 44 (74.6%)
Lung Metastasis  No  Yes	3 (25%) 9 (75%)	2 (13.3%) 13 (86.7%)	10 (31.3%) 22 (68.8%)	15 (25.4%) 44 (74.6%)
Lung Metastasis Only  No  Yes	12 (100%) 0 (0%)	14 (93.3%) 1 (6.7%)	28 (87.5%) 4 (12.5%)	54 (91.5%) 5 (8.5%)

Safety

A total of 59 patients were treated on study including 27 patients (12 in Cohort A and 15 in Cohort B) who received relatlimab 160mg. Of the patients in Cohort C, 11 patients were treated with 960mg of relatlimab for cycle 1 day 1. In three instances, these patients received subsequent doses at 160mg due to safety concerns raised by another trial. Twenty one patients were treated at 480mg of relatlimab. All patients received nivolumab 480mg IV every 4 weeks. Hematological and nonhematological toxicities are shown in Supplementary Tables 5/6/7. Toxicity was evaluable for all patients that started treatment. Of note, there were no grade 4 or 5 treatment related AEs. Grade 3 and 4 toxicities occurred in 78% of cohorts A/B and 47% of patients on Cohort C, the majority of which were disease related. The most frequent grade 3/4 toxicities were abdominal pain: 5 (8%), anemia: 6 (10%), acute kidney injury: 4 (7%), hypertension: 4 (7%), and small bowel obstruction: 5 (8%). Overall, 3 of the 59 evaluable patients had a grade 3 treatment related AE (adrenal insufficiency, arthritis, and diabetic ketoacidosis/diabetes). No patients discontinued study treatment due to drug-related AEs although one patient developed unacceptable toxicity – arthritis at the time of disease progression.

Treatment efficacy

Among the 59 patients enrolled with response data, there were 3 patients who achieved a PR, 6 with stable disease, and 50 with progressive disease (PD) as their best response. By cohort, the best responses were Cohort A - SD: 1, PD: 11, Cohort B - PR: 1, SD: 3, PD: 11, and Cohort C - PR: 2, SD: 2, PD: 28. The ORR for the overall trial was 5.1% with response rates for Cohort A: 0%, Cohort B: 6.7%, and Cohort C: 6.3%. Median progression-free survival for the overall study was 2.56 months. Progression-free survival was 2.17 months, 2.76 months, and 2.56 months for Cohorts A, B, and C respectively. This identified a worse PFS for patients treated on cohort A (CPM-) (HR: 1.98 (1.00–3.92), p-value=0.050) (Fig. 1A). Median overall survival for the study was 12.43 months. Median overall survival was 4.34 months, 14.17 months, and 12.46 months for Cohorts A, B, and C respectively (Fig. 1B). This identified worse OS for patients treated on cohort A (CPM-) (HR: 2.40 (1.19–4.84), p-value=0.015). Of note, no patient within Cohorts A or B who had mucin present in their primary tumor resection specimen had a radiographic response.

Figure 1. Kaplan-Meier curves for progression-free and overall survival in microsatellite stable colorectal cancer patients treated with relatlimab and nivolumab.

Following administration of relatlimab and nivolumab per protocol, patients were followed for survival outcomes measured as time from cycle 1 day 1 of treatment to (A) disease progression on radiographic imaging or (B) death respectively. At the time of data censoring, no patients were still on active treatment and 6 were still alive.

Response rates based on the location of a patient’s metastasis were also assessed looking specifically at patients with liver metastases, without radiographic visible liver metastases at the time of study enrollment, and lung-only metastases. Patients in the without liver metastases group could have had previously treated liver metastases and/or metastases to lymph nodes, peritoneal, or other sites. For these three populations the response rate was 2.3%, 13.3%, and 40% respectively (Supplementary Table 3). The PFS and OS for the patients with liver metastases was 2.48 months and 8.68 months. The PFS and OS for patients without liver metastases was 2.73 months and 17.49 months. Patients with lung-only metastases had a median PFS of 5.0 months and OS of 32.55 months. Patients without liver metastases had significantly improved progression-free survival (HR:0.49 (0.26–0.90) p-value=0.021) (Fig. 2A) and overall survival (HR:0.36 (0.17–0.75) p-value=0.006). (Fig. 2B)

Figure 2. Evaluation of metastatic site-specific responses to nivolumab and relatlimab.

To understand the influence that location of metastases had on outcomes for patients with pMMR CRC treated with nivolumab and relatlimab, survival outcomes as a function of whether the patient had active liver metastases at the time of study enrollment was evaluated. We took the group without liver metastases and also looked individually at the patients with lung-only metastases. Progression-free survival (A) and overall survival (B) were compared between patients with these different sites of metastasis. A “site specific” RECIST was calculated for liver and the other sites independently for patients with liver and lung (C), liver and peritoneum (D), and liver and other sites (E) to assess whether tumors in different locations responded differently to nivolumab/relatlimab within the same patient.

Evaluation of Site-Specific Treatment Responses

Having observed differences in survival outcomes and responses based on sites of metastases raised the question of whether these differential responses related to intrinsic tumor biology that influenced all sites of metastasis and outcomes or whether factors related to the metastatic niche established within the liver specifically suppressed anti-tumor immune responses in that location. As an exploratory analysis, we looked specifically at patients with liver metastases and at least one other site of RECIST measured disease which we categorized as lung, peritoneal, or other sites. We then looked at the RECIST response in each site of disease at their initial on-treatment scan (approximately 12 weeks after cycle 1 day 1) to establish a “site specific response”. For patients with liver and lung RECIST measured disease, liver lesions showed progression in 13 (62%) patients, stable disease in 7 (33%), and partial response in 1 (5%). Lung lesions in contrast showed PD in 6 (29%), SD in 14 (67%), and PR in 1 (5%) (Fig. 2C). RECIST responses in the peritoneum and liver demonstrated similar outcomes with both sites of disease (liver: 5 PD, 3 SD, peritoneum: 5 PD, 3 SD) (Fig. 2D). In patients with RECIST-measured liver lesions and other sites of disease which represented predominantly adenopathy, the liver disease showed PD in 10 patients and SD in 8 whereas the other sites of disease showed PD in 2 patients, SD in 15 patients, and PR in 1 patient (Fig. 2E). This demonstrated a trend towards disease control in lung (p-value=0.062) and statistically significantly better disease control in other sites of disease (p-value=0.011) even in the presence of concomitant liver metastases.

Evaluation of baseline differences in the tumor microenvironment of liver and lung metastases in colorectal cancer.

We explored what baseline differences in the TME of CRC liver and lung metastases might account for these differential clinical outcomes to relatlimab/nivolumab treatment. Comparing the baseline liver (n=13) and lung (n=7) biopsy samples we identified that lung metastases had statistically higher levels of dendritic cells (p-value=0.0012) and a trend towards higher B cells (p-value=0.078) compared to liver metastases (Supplementary Figure 1A). Representative images are shown in Supplementary Figure 1B.

Despite the generally more favorable outcomes for patients without baseline liver metastases, we had one patient with liver metastases who had a partial response to relatlimab/nivolumab. We compared the TME of this patient’s liver metastases to those of other patients with biopsies of their liver metastases. This identified that this patient’s tumor had one of the highest densities of both cytotoxic T cells and memory T helper cells along with a low density of baseline Tregs. This was in contrast to the other liver metastases evaluated where high levels of cytotoxic T cells were accompanied by higher Tregs (Supplementary Figure 2A/B/C).

Assessment of PD-L1 and LAG-3 expression as a biomarker of treatment response to relatlimab and nivolumab.

To determine if PD-L1 and/or LAG-3 expression levels correlated with treatment response, we performed immunochemistry on pre-treatment biopsy specimens for PD-L1 and LAG-3. PD-L1 tumor proportion score (TPS) and combined proportion score (CPS) were available on 40 patients with the remaining patients having insufficient biopsies for assessment. LAG-3 staining was only available on 27 patients as this was only performed if the patient’s sample had sufficient tumor infiltrating immune cells to allow LAG-3 evaluation. LAG-3 expression was reported as the percentage of immune cells staining positive for this marker. Of our three patients who achieved a partial response, 2 of them were negative for PD-L1 expression by both TPS and CPS while the third had insufficient viable tumor. None of the three were evaluable by LAG-3, one because of a lack tumor infiltrating immune samples and the other 2 for insufficient sample. Of the patients that had stable disease as best treatment response the highest LAG-3 expression was <1% (n=2) with none having positive PD-L1 expression by TPS and one having 1% expression by CPS (n=4). A summary table of PD-L1 and LAG-3 expression results is available in Supplementary Table 8.

Relatlimab and nivolumab treatment enhances T cell activation

To compare systemic immune responses, we analyzed PBMCs at pre-treatment and week 4 timepoints. We selected a subset of patients for correlative analysis that had a spectrum of treatment responses, adequate peripheral blood samples, and biopsy specimens of metastatic lesions. With 15 paired patient blood samples, we performed 36-marker mass cytometry analysis (Cytometry by Time-of-Flight; CyTOF) to distinguish lymphoid and myeloid cell subtypes and their functional states, clustering the dataset into 20 annotated cell types (Supplementary Fig. 3A). While none of the three pre-specified cohorts met the primary endpoint for overall response rate to treatment, we sought to determine if there were any immunologic correlates of clinical benefit. Patients were categorized as responders or non-responders based on whether they had a partial response (responder; R) to treatment versus stable or progressive disease (non-responder; NR) as their best treatment response. Compared to non-responders, responders displayed significantly lower proportions of naïve CCR7⁺CD45RA⁺ helper (p=0.044) and cytotoxic (p<0.0001) T cell populations in circulation (Fig. 3A). Responders also exhibited lower proportions of GZMB⁺ NK cells (p=0.044) and higher proportions of CCR6⁺ B cells (p=0.038) (Fig. 3B).

Figure 3. Circulating immune cell changes following relatlimab (anti-LAG-3) and nivolumab (anti-PD-1) treatment in pMMR CRC.

To determine the circulating immune cell changes with relatlimab and nivolumab treatment we performed mass cytometry consisting of a panel of markers to identify immune cell populations, their subsets, and the expression of different activation and exhaustion markers on peripheral blood mononuclear cells obtained prior to cycle 1 day 1 (PreTx) and after 4 weeks of treatment (OnTx). Boxplots showing proportions of (A) naïve T cell subsets and (B) NK and B cells in responders (R) and nonresponders (NR; stable and progressive disease). (C) Line plots show proportions of effector memory T cells and regulatory T cells at PreTx and OnTx timepoints for each patient. Expression of (D) costimulatory/activation markers, (E) CD25, and (F) coinhibitory/exhaustion markers by helper T cell subsets are also shown as line plots. Non-responders (NR) and Responders (R) are indicated by red solid and dotted blue lines, respectively. Each shape represents a unique patient. All data are representative of 15 paired patient samples. P-values are calculated by paired t-tests. Abbreviations: Acv, activated; CM, central memory; EM, effector memory; Ex, exhausted; Tc, cytotoxic T cells; Th, helper T cells.

We also analyzed the effects of the treatment on the T cell subsets. In terms of proportional abundances, activated effector memory helper T cells (CD45RA⁻CD45RO⁺ 41BB⁺Ki67⁺; ThEM_Acv) and regulatory T cells (CD25^hiCD127⁻CTLA4^hi; T_reg) significantly increased within the bloodstream after treatment, suggesting an increased overall inflammatory state systemically (Fig. 3C). When comparing the expression of the therapeutic targets pre- vs. on-treatment, LAG3 appeared to significantly increase in most T cell subtypes (Supplementary Fig. 3B). PD1 significantly decreased with treatment. This reduction in PD1 is an expected result of occupancy effect related to the EG12.2H7 antibody clone being blocked by nivolumab, validating our workflow (Supplementary Fig. 3C). To determine whether T cells were becoming more activated as a result of relatlimab/nivolumab treatment, we next assessed the presence of activation/co-stimulatory markers and exhaustion/coinhibitory markers on CD4⁺ T cells and CD8⁺ T cells. In the CD4⁺ T cell population, there was a significant rise in the expression of OX40, HLADR, 41BB, Ki67, and CD28 (p<0.05, Fig. 3D), whereas CD25 dropped (p<0.005, Fig. 3E). The increase in activation markers was accompanied by an increase in exhaustion/coinhibitory markers, with an increase in CD39, TOX, CTLA4, TIGIT, and CD95 (p<0.005) on effector memory T cells (Fig. 3F). In contrast, effector memory CD8⁺ T cells did not show a significant rise in GZMB expression with treatment but did show higher levels of several exhaustion markers, CD39, CTLA4, and CD95 (p<0.005, Supplementary Fig. 3D). Treg cells also had a significant increase in exhaustion markers (Supplementary Fig. 3E). This is consistent with LAG-3 inhibition having a more direct global effect on CD4⁺ T cell populations.

TME analysis identifies lower baseline levels of Tregs and ADAM10⁺ cancer cells as correlates of treatment response.

Since relatlimab/nivolumab successfully enhances T cell activation in non-responders and responders alike, we sought to determine what features within the TME correlate with successful conversion of immune responses to clinical responses. Based on single-stain immunohistochemistry, tumor positivity of LAG3 or PD1 expression did not correlate with response (Supplementary Table 8). To undertake a more detailed analysis of the TME, we selected a subcohort of 24 patients with core needle biopsies available for analysis, of which 13 were from the liver and 7 from lung metastatic sites (and the rest from primary and abdominal tumors). 15 of the 24 were available as pairs of pre- and on-treatment tumor biospecimens. Using an antibody panel of 45 markers (Supplementary Table 2), we analyzed the TME in a total of 96 regions of interest (ROIs) guided by pathologist (R. Anders) review using imaging mass cytometry (IMC) (Fig. 4A, Supplementary Fig. 4). Due to poor tissue/staining quality, 18 of 96 ROIs were excluded from analysis. The final set of images represented 92.5mm² tumor area with an average of 3.85mm² per patient. We segmented all of the images into single cells (Supplementary Fig. 4A) and clustered the data using PhenoGraph into 5 epithelial, 1 stromal, 1 parenchymal, and 14 immune cell types based on their expression profiles (Supplementary Fig. 4B), the sum of which totaled 198,931 annotated cells. T cell subsets were defined by CD3 along with CD4 and CD8 (Supplementary Fig. 4C). The CD8⁺ cytotoxic T cells (Tc) were then further subtyped into CD45RO⁺ memory Tc (TcM) and CXCR3⁺ Type-I Tc (Tc1). Similarly, CD4⁺ T cells (Th) were further subtyped into CD45RO⁺ memory Th (ThM), CXCR3⁺ Type-I Th (Th1), and FOXP3⁺ regulatory T cells (Treg). Other major immune cell types were identified by CD20 for B cells, CD68 for macrophages, CD15 for neutrophils, and DCLAMP for dendritic cells. Macrophages were also subtyped by a marker for immunosuppressive states (CD163); stromal fibroblast clusters were defined by varying combinations of vimentin (VIM) and α-smooth muscle actin (SMA); epithelial/cancer cell clusters were distinguished by higher levels of CDX2 or pan-cytokeratin (CK) (Supplementary Fig. 4B). Based on prior studies demonstrating the role of metalloproteases ADAM10 and ADAM17 in mediating LAG-3 turnover with low levels corresponding to anti-PD-1 resistance and fibrinogen-like protein 1 (FGL1) as another ligand of LAG3, epithelial/cancer cell clusters were further subtyped into 5 subclusters: “Epi_I” (CDX2+HLADR+ADAM10^lo), “Epi_II” (CDX2⁺ADAM10^hi), “Epi_III” (CDX2⁻CK⁺ADAM17^hiFLG1⁺), “Epi_pSTAT3” (CDX2⁺FGL1⁺pSTAT3^hi), and “Epi_KI67” (CDX2⁺FGL1⁺KI67^hi) (Fig. 4B).^23–25 Representative images of ADAM10+CDX2+ cells are shown in Supplementary Fig. 4D.

Figure 4. TME correlates of relatlimab/nivolumab.

(A) Representative IMC results for two multicolored sets of phenotyping markers are shown in six distinct regions. Scale bar: 200µm. (B) Detailed expression profiles for epithelial/cancer cell clusters are shown as scaled heatmap. (C) Proportions of cell types are shown as boxplots stratified by pre- (PreTx, n=57) vs. on-treatment (OnTx, n=38) timepoints. (D) Proportions of cell types at only baseline timepoint are compared between representative regions analyzed by IMC from non-responders (NR; stable and progressive disease, n=49) and responders (R; partial response, n=8). P-values are calculated by unpaired t-tests.

Based on these annotated clusters, we observed that tumor-infiltrating Tc significantly increased with relatlimab/nivolumab (average of 0.81% to 2.23%, p=0.0052) along with CD68⁺CD163^lo macrophages (Mac_I, 5.45% to 14.13%, p=0.048) (Fig. 4C), confirming the pro-inflammatory effects of the treatment. Accordingly, treatment also led to a significant decrease in Epi_I cells (p=0.034). We then compared the TME compositions between responders and non-responders. This identified Treg cells (p-value=0.0027) and B cells (p-value=0.015) density to be significantly higher in non-responder patients than those that had a partial response (Fig. 4D). We next assessed whether intratumoral Tc or Treg density correlated with best radiographic response to treatment. We took the subset of patients that had undergone IMC testing and stratified them by whether their pre-treatment Tc or Treg density was above or below the median for that cell type (Supplementary Fig. 5A and 5B for Tc and Treg respectively). This demonstrated that higher Tcs and lower Tregs tended to be enriched in patients with better radiographic responses to relatlimab/nivolumab. Comparing the epithelial cell subtypes between responders and non-responders indicated that higher presence of ADAM10^hi cancer cells was associated with absence of response to relatlimab/nivolumab (p-value=0.017). These findings suggest that a ADAM10-high microenvironment, along with the enriched presence of other immunosuppressive barriers such as Tregs may restrict the benefit of anti-LAG3 therapy.

Evaluating immune cell spatial relationships reveals Treg-mediated resistance to relatlimab/nivolumab.

Using the spatial data, we then determined how the different cell types were spatially coordinated by calculating the average shortest distances between every cell type and visualized them as networks based on response. This revealed that distance relationships among Tregs and memory T cell subtypes were particularly different between non-responders and responders (Fig. 5A). When looking specifically at memory CD45RO⁺CD8⁺ T cells (TcM), memory CD45RO⁺CD4⁺ T cells (ThM), and helper T cell (Th) subsets, their distances to Tregs were significantly greater in responders compared to non-responders (Fig. 5B).

Figure 5. Distance relationships among cell types in the TME of non-responders and responders.

(A) Average shortest distances between every cell type were calculated and plotted as networks. The size of the node indicates the abundance of the cluster and the thickness of the line along with the relative proximity correlates with distance. Dashed regions highlight spatial associations among T cell subtypes. (B) Distances from memory CD45RO⁺CD8⁺ T cells (TcM), memory CD45RO⁺CD4⁺ T cells (ThM), and helper T cell (Th) to Tregs for non-responders and responders are shown as violin plots. P-values are calculated by unpaired t-tests.

To further explore how the increased activation of helper CD4⁺ T cells following relatlimab/nivolumab treatment and higher Tregs in non-responders might influence intratumoral CD8⁺ T cell function, we performed a nearest neighbor analysis. Compared to those without, intratumoral CD8⁺ T cells with nearest neighboring relationships, i.e., within 4µm distance, with helper CD4⁺ T cells exhibited higher levels of PD1 (p-value=0.000027), GZMB (p-value=0.0057), HLA-DR (p-value=0.00000021), and CD45RO (p-value=0.000097), suggesting improved activation. In contrast, CD8⁺ T cells with nearest neighboring relationships with Tregs, compared to those that did not, displayed higher levels of PD1 (p-value=7.1e⁻¹¹) and TOX (p-value=1.1e⁻¹³), suggesting greater exhaustion (Supplementary Fig. 6). We assessed the change in intratumoral Treg density in patients with paired pre- and on-treatment biopsies. This identified that in contrast to the increase in Tregs in the peripheral circulation with this combination, there was a trend towards reduced intratumoral Tregs (p-value=0.19).

Discussion

While significant success has been demonstrated with ICI treatment in dMMR CRC and many other tumor types, immunotherapy has been largely unsuccessful in improving outcomes in patients with pMMR disease.^3,26–28 This investigation evaluated whether the addition of relatlimab (anti-LAG3) to nivolumab (aPD1) might overcome the immunosuppressive tumor microenvironment seen with pMMR CRC. In the overall population, this trial failed to meet its primary endpoint of overall response rate.

Based on prior data, we anticipated that responses would be enriched in patients whose primary tumors had increased PD-L1 and mucin expression.^13,14 While there were only a limited number of patients enrolled on Cohorts A (CPM⁺) and B (CPM⁻), the outcomes for the mucin/PD-L1 high group were numerically worse with a 0% ORR and the lowest PFS and OS of the three cohorts suggesting mucin/PD-L1 is not a good biomarker for relatlimab/nivolumab responsiveness in pMMR CRC.^3,29 However correlative investigations identified low densities of Tregs and ADAM10 expression were both enriched in patients that responded to therapy. This opens up the possibility that these biomarkers could be used as selection criteria for PD-1/LAG-3 treatment in pMMR colorectal cancer and in the case of Tregs serve as a target for combinatorial treatment. Observations related to ADAM10 expression are particularly interesting given that prior work demonstrated the importance of its function, as well as that of another similar enzyme ADAM17, in cleaving LAG3.²³ In the setting of anti-PD1 therapy alone, low levels of ADAM10 or engineered expression of noncleavable LAG3 conferred resistance to antitumor immune response, suggesting that the intact presence of LAG3 on the T cell surface is a necessary molecular mechanism for LAG3-mediated immunotherapeutic resistance.²⁴ Thus, our work suggests an important caveat for anti-LAG3 therapy in that ADAM10-high TME may not benefit from LAG3 inhibition as much as ADAM10-low TME. Accordingly, this also suggests that ADAM10-high TME poses immunosuppressive barriers that are independent of LAG3.

Based on emerging pre-clinical and clinical data suggesting liver metastases to be immunosuppressive and predictive of worse outcomes with ICI, we investigated whether patients with liver metastases had worse outcomes than patients with other sites of disease.^17,30–32 This indicated that patients without liver metastases and particularly those with lung-only disease are more likely to clinically benefit from anti-PD1 and anti-LAG-3 treatment. Assessment of TME differences between these metastatic locations identified lung metastases to have higher levels of dendritic cells and a trend towards higher B cells which may speak to be better MHCII based presentation of tumor associated antigens in lung metastases.

While there were a very limited number of patients with lung-only metastases, more half of these patients (2 PR, 1 SD, and 2 PD) had at least stable disease as best response to this treatment approach which was mirrored by lung metastases seen in patients with concomitant liver disease suggesting that there is something intrinsic to lung metastases and their TME that make these lesions better controlled with relatlimab/nivolumab than liver lesions.

We also evaluated pharmacodynamic effects of relatlimab/nivolumab treatment on circulating and intratumoral immune cell populations. This investigation identified that circulating naïve CD4⁺ T cells and naïve CD8⁺ T cells both decreased with LAG-3/PD-1 blockade with an increase in activation as well as exhaustion markers on the surface of CD4⁺ T cells (CTLA-4, TIGIT, TOX, CD39). This was less apparent in CD8⁺ T cells with some evidence of increases in cell surface exhaustion markers (CD39, CTLA-4). We hypothesize that the drop in circulating naïve CD4⁺ T cells and CD8⁺ T cells may represent increased activation and trafficking into the TME. Supporting this hypothesis, there was a marked increase in intratumoral CD8⁺ T cells suggesting enhanced homing of this cell population into the tumor microenvironment although increased intratumoral expansion is an alternative possibility. Concurrent with this increase in intratumoral CD8⁺ T cells, there was a trend towards reduced intratumoral Tregs following relatlimab/nivolumab treatment despite a peripheral increase. These pharmacodynamic effects of relatlimab/nivolumab in the CRC TME suggests important anti-tumor immune changes with the use of this combination but this was insufficient in isolation to promote clinically responses in the majority of patients.

The increased expression of exhaustion markers including CTLA-4 and TIGIT in addition to promising signals of immune activation suggests that blockade of additional immune checkpoints might further enhance immune activation and lead to improved antitumor responses.

Data Availability

The authors declare that the minimal data set for this study cannot be shared publicly due to ethical and legal restrictions on sharing de-identified data that aligns with the consent of research participants. Current JHU compliance policies require data with no direct consent for public open access sharing be under restricted access. We will provide access through Vivli, an established repository for clinical data that provides open access without a fee restricted to approved researchers under a Data Use Agreement. JHU compliance policy for Vivli requires additional anonymization of certain demographics, including use of age ranges and limiters to outlier values for weight, height, and certain rare diseases, while retaining sufficient value for reference and validation of results. Researchers can request more detailed data from the corresponding author shared though an approved collaboration arrangement. Clinical data is available under a Data Use agreement at: https://doi.org/10.25934/PR00011351. Cytometry data is publicly available at: https://doi.org/10.5281/zenodo.14212759