Pilot Clinical Trial of Perioperative Durvalumab and Tremelimumab in the Treatment of Resectable Colorectal Cancer Liver Metastases

This article is featured in Highlights of This Issue, p. 2959

Colorectal cancer is the second most common cancer in the United States with an estimated 147,950 new cases and 53,200 deaths in 2020 (1). Approximately one third of these patients will develop liver metastases within 3 years of initial diagnosis (2). Surgery is potentially curative in the 15%–20% of patients who meet criteria for resection (2–5). For surgical candidates, complete surgical resection is associated with a 20%–50% overall survival (OS) rate at 5 years (6–8). Unfortunately, the majority of resected patients ultimately recur, and data demonstrating the OS benefit from perioperative systemic chemotherapy are limited (9, 10). To date, the role of immunotherapy in the perioperative setting in metastatic colorectal cancer has not been investigated.

The tumor microenvironment represents a complex collection of cell types that modulate tumor development. As a result, even highly immunogenic tumors can have a suppressed immune environment depending on the cells that make up the microenvironment (11, 12). Tumor-infiltrating lymphocytes (TIL) have the capacity to control the growth of many types of cancers and are emerging as an important biomarker in predicting the efficacy and outcome of treatment (13). Blockade of immune-checkpoint inhibitors such as cytotoxic T-lymphocyte–associated antigen 4 (CTLA-4), programmed cell death 1 (PD-1), and programmed cell death ligand 1 (PD-L1) can exhibit clinical activity in a wide range of tumor types. PD-1 based therapy has become the standard of care for patients with mismatch-repair–deficient (dMMR) metastatic colorectal cancer (14, 15). However, patients with mismatch-repair proficient (pMMR) metastatic colorectal cancer have shown minimal responses to immune-checkpoint monotherapy. The combination of CTLA-4 inhibition (tremelimumab) with anti–PD-L1 inhibition (durvalumab) compared with best supportive care was recently examined in the Canadian Cancer Trials Group CO.26 Study, which demonstrated an improvement in OS in refractory metastatic colorectal cancer, although no difference in response rate or progression-free survival was found (16). In a separate study, durvalumab and tremelimumab were demonstrated to be safe when combined with liver directed yttrium-90 resin microsphere-based radioembolization (17).

To better understand the mechanisms of immune resistance in pMMR colorectal cancer, we initiated a pilot clinical trial combining CTLA-4 inhibition (tremelimumab) with anti–PD-L1 inhibition (durvalumab) in the perioperative resection setting where the suppressive impact of the tumor microenvironment would be addressed with surgical resection and the acquisition of tumor tissue for immune characterization would be possible. In addition, adjuvant durvalumab was allowed postoperatively to evaluate the additional impact of PD-L1–based therapy, given the high risk of residual microscopic disease following colorectal cancer liver resection.

This study was an open-label, single-center pilot trial assessing the safety and feasibility of adding neoadjuvant tremelimumab 75 mg i.v. flat dose and durvalumab 1,500 mg i.v. flat dose given preoperatively for 1 cycle prior to colorectal cancer liver metastasis resection. Postoperative therapy was at the discretion of the treating physician, and patients were eligible to receive durvalumab 1,500 mg i.v. every 4 weeks for 4 cycles. Liver resection was scheduled approximately 4 to 8 weeks after durvalumab/tremelimumab infusion.

The primary endpoints were feasibility and safety assessed by the rate of on-trial surgical resection of liver metastases, postoperative toxicity graded by the Clavien-Dindo classification (18), and treatment-related toxicity graded by CTCAE v5. The combination was defined as feasible if at least 80% of patients could undergo resection or if between 60% and 80% could undergo resection with a positive toxicity and efficacy profile. Secondary endpoints included the translational evaluation of various immune-relevant factors, preoperative response rate by RECIST v1.1, OS, and relapse-free survival (RFS).

The trial was conducted in accordance with the Declaration of Helsinki. The protocol (ClinicalTrials.gov identifier: NCT02754856) was approved by the Institutional Review Board at University of Texas MD Anderson Cancer Center (Houston, TX), and written informed consent was obtained for all patients before performing study-related procedures.

Eligible patients were required to have histologically confirmed colorectal adenocarcinoma with liver-only metastases deemed resectable by a liver surgeon (resectability may involve the use of ablative techniques to some but not all liver metastases), measurable disease per RECIST v1.1, Eastern Cooperative Oncology Group (ECOG) Performance Status ≤1, normal organ and marrow function, any number of prior lines of systemic chemotherapy, and known MMR status. In addition, patients with an intact primary tumor that was planned for surgical resection were eligible.

When feasible, tumors were subdivided for fresh flow cytometry analysis. Cells were first stained for surface antigens and a live/dead dye followed by fixation and permeabilized for intracellular staining as previously described (19). The antibody resources and dilutions used are listed in Supplementary Table S2. Gating was determined by using the fluorescence minus one approach. Samples were acquired using a BD Fortessa X20 and analyzed using FlowJo v 10.0.7. The gating strategy for T-cell and myeloid cell phenotyping is depicted in Supplementary Fig. S4D and S4E.

We performed RNA sequencing on RNA extracted from fresh-frozen tumor tissue. The sequencing reads were aligned to the UCSC GRCh37 assembly of the human genome with TopHat2 (20), and the read counts were resolved using HTSeq (21). The average coverage per base is ∼118. We next applied DESeq2 to calculate differential expression between tissue groups of interest and transform count data for downstream analysis with variance stabilizing transformations (VST; ref. 22). Genes with an adjusted P value of < 0.05 were determined as differentially expressed genes (DEG). We plotted heat maps for DEGs and performed hierarchical clustering utilizing pheatmap, an R package (23). To further analyze the biological differences at a pathway level, we performed gene set enrichment analysis (GSEA) using the default settings of the GSEA software (24, 25). Multiple precollected gene sets from the Molecular Signatures Database (MSigDB) were also selected, and included Hallmark gene sets, C2 curated gene sets, and C7 immunologic gene sets (25).

The mIF analysis was conducted by a pathologist in five intratumoral areas using 660 μm × 500 μm (0.33 mm²) region of interest (ROI) at ×20 magnification to cover a total intratumoral area of 1.65 mm². In cases where five ROIs did not cover 1.65 mm² of intratumoral area, additional ROIs were included in the analysis. The final results were expressed as the average cell densities in any given area by mm² (cells/mm²) (26). The mIF panels utilized are present in Supplementary Table S2. IHC staining and scoring of PD-L1, CD20, and CD73 are included in the Supporting Information.

Pretreatment fecal baseline samples were collected from colorectal cancer patients. In brief, genomic DNA was isolated using the QIAamp DNA stool mini kit (Qiagen), according to the manufacturer's protocol, modified to include an intensive bead-beating lysis step. The V4 region of 16S rRNA gene was amplified by PCR from 10 ng of each of extracted and purified genomic DNA using 515 forward and 806 reverse primer pairs (27). The amplicon pool was purified with QIAquick gel extraction kit (Qiagen) and sequenced on the Illumina Miseq sequencer platform using 2 × 250 bp paired-end protocol. After sequencing, paired-end reads were demultiplexed by QIIME and then merged and dereplicated for chimeras using VSEARCH. UNOISE 3 command algorithm was used to perform denoising of reads (28). Operational taxonomic units (OUT) were classified using Mothur method with the Silva database version 138. For differential taxa-based univariate analysis, abundant microbiome taxa at species, genus, family, class, and order levels were analyzed using the Mann–Whitney U test after logit transformation. The detailed computational pipeline of analysis has been previously described (29). For exploratory analyses, P values have not been adjusted for multiple comparisons. Three patients analyzed for microbiome had antibiotics in the preceding six months. We did not observe any obvious or dramatic derangements in the fecal microbiome composition of patients treated with antibiotics previously, though numbers were small.

The planned study sample size was 25 patients, enabling a Bayesian 95% credible interval (ci) of surgery received to be (0.62–0.92), assuming that the proportion of patients successfully getting to surgery is 80%. This utilized a beta-binomial model with a noninformative prior distribution of Beta (1.2, 0.8). Patients underwent interim analyses for futility, regimen toxicity, and postsurgical complications based on a Bayesian sequential monitoring design (30, 31). The trial was able to continue by these rules, but was closed prior to completion of full enrollment due to programmatic reasons. Estimates of successful surgery rates are provided from the posterior distribution due to the multiple Bayesian interim analyses throughout the trial. OS and RFS were calculated and plotted by Kaplan–Meier methods (32). Relationships between baseline characteristics, immune markers, and outcomes (response, OS, or RFS) were explored with logistic or Cox models as previously described (33). Individual immune markers between pre and posttreatment samples were compared with unpaired Student t test. For the analysis across flow cytometry, IHC, and RNA sequencing with efficacy, patients were stratified by RFS into long RFS of >1.5 years or short RFS ≤1.5 years.

A total of 24 metastatic colorectal cancer patients were enrolled from November 2016 to September 2020, and 23 patients received trial treatment and are evaluable (Fig. 1A). One patient withdrew from the study prior to study treatment and is not included in the study results. The median study follow-up is 2.3 years. Table 1 lists the baseline characteristics of the evaluable patients. Eighteen patients (78%) received preoperative chemotherapy with a median of 3.8 weeks from last chemotherapy dose to durvalumab and tremelimumab treatment. Of the 23 patients, 20 underwent surgical exploration (87%; 95% CI: 67%–96%) and 17 underwent surgical resection (74%; 95% CI: 53%–88%; Table 2). The Bayesian posterior distribution estimate for the proportion of patients undergoing surgery with 95% ci is 85% (69%–96%), while for patients undergoing complete surgical resection was 73% (54%–88%). Immune-related toxicity did not prevent surgical intervention in any patients. The reasons for not undergoing surgical exploration were progression of previously noted subcentimeter lung nodules in all three patients. Of the three patients who underwent surgical exploration but not resection, one patient had chemotherapy induced liver toxicity (this patient had received previous FOLFOXIRI chemotherapy) and two patients demonstrated more extensive liver metastatic disease than appreciated on imaging. Two of these patients subsequently went on to definitive liver treatment after additional interval therapy (liver resection in one patient and combined radiation/microwave ablation in one patient).

A major hepatectomy, defined as ≥4 liver segments, was performed in 6 of the 17 resected patients, 30%. The median time from durvalumab/tremelimumab to surgical resection was 30 days (range, 17–69 days). Sixteen of the 17 resected patients received adjuvant durvalumab in the postoperative setting, while one patient received adjuvant FOLFOX due to physician decision.

A total of 6 postsurgical complications occurred in 8 of 20 patients (40%), with 2 complications rated grade 3 or 4. A grade 3 anastomotic leak complication occurred in a patient who underwent concurrent ultra-low anterior resection of his rectal primary and a grade 4 colonic perforation occurred at another patient's primary tumor location. In the eight patients with ≤28 days between durvalumab/tremelimumab and surgical resection, two patients (25%) experienced a postoperative complication.

Durvalumab and/or tremelimumab-related toxicities are presented in Supplementary Table S1. Five grade 3 or 4 treatment-related adverse events occurred: fatigue, AST elevation, lipase elevation, oral mucositis, and thromboembolic event.

Presurgical radiographic response per RECIST v1.1 was stable disease in 15 (65%) patients, partial response in 3 (13%) patients and progressive disease in 5 (22%) patients (Fig. 1B, waterfall plot). Percent tumor cellularity in the 17 resected patients is shown in Fig. 1C. Four patients, 2 of whom were dMMR, demonstrated a complete pathologic response with the presence of residual mucin in all cases. Of the two pMMR cases with complete pathologic response, both had POLE P286R mutations with a tumor mutation burden of 61 per Foundation One assay in one patient and 42 mutations on the 138 gene Oncomine panel. OS and RFS are shown in Fig. 1D and E and Table 2. As 5 of 17 patients did not receive prestudy systemic chemotherapy, we evaluated the various efficacy outcomes, percent tumor cellularity, RFS, and OS between these two groups, and no significant differences were seen (Supplementary Fig. S1A–S1C). Imaging and markers of the pMMR patients with POLE mutation are shown in Supplementary Fig. S2.

Pretreatment mandatory tumor biopsy and posttreatment surgery samples were collected along with pretreatment microbiome samples (Fig. 1A). A total of 21 pretreatment biopsies were collected but only 10 biopsies demonstrated >10% malignant cells. Reasons for inadequate biopsies were: normal liver in 4, necrotic tumor only in 2, and minimal tumor amount in 5. Posttreatment surgical resection samples were collected in 13 patients with 11 demonstrating >10% malignant cells and 2 demonstrating pathologic complete response, no evidence of tumor cells in the samples. In total, 6 paired (pre- and post-surgery) samples were obtained, but one post-surgery sample was a pathologic complete response with acellular mucin (pMMR patient) and thus 5 pairs with tumor tissue were available for analyses. Due to the known differences between pMMR and dMMR colorectal cancer, the two dMMR were excluded from all analyses except for when directly compared with pMMR samples.

To assess the impact of the combination of durvalumab/tremelimumab on the tumor immune infiltrates, we performed immunoprofiling using flow cytometry, IHC, mIF, and RNA sequencing of pretreated and posttreated pMMR tumor samples.

Flow cytometry assessment of TIL activation and inhibitory receptors in pretreatment versus posttreatment samples found the intratumoral cytotoxic CD8⁺ T-cell populations exhibited a significant increase in Lag3⁺ expression following treatment (P = 0.017; Fig. 2A). Although limited by sample size and sample heterogeneity, there were trends suggesting potential increases in PD-1⁺ (P = 0.117), ICOS⁺ (P = 0.284), and Tim3⁺ (P = 0.121) expression by CD8⁺ TIL (Fig. 2A). No markers were significantly decreased posttreatment. When comparing posttreatment CD4⁺ T-cell subsets (Fig. 2B), no significant changes were observed within CD4⁺ TIL subsets, except for a trend with increased ICOS⁺ (P = 0.111) levels. Treatment effects on other immune cell populations are shown in Fig. 2C and demonstrate similar levels of various myeloid and dendritic subtypes.

Immune cell distribution and infiltrates were analyzed by mIF to evaluate tumor (Fig. 2D), stroma (Fig. 2E), and total (Fig. 2F) compartments between baseline and posttreatment samples. These findings are represented as cells/mm² and reflect similarities with flow-cytometric analyses. Intratumor distribution illustrated in Fig. 2D demonstrates a trend toward an increase in activated cytotoxic T cells (CD8⁺ GZB⁺, P = 0.094) and no significant decrease in macrophages (P = 0.431). Similar trends in immune changes were also observed in the stromal region (Fig. 2E). The combined tumor and stromal region (Fig. 2F) shows trends toward an increase in activated cytotoxic T cells (CD8⁺ GZB⁺, P = 0.099) and decreases in both T regulatory cells (P = 0.239) and macrophages (P = 0.018) following treatment.

In addition, RNA sequencing analysis identified 188 DEG that were significantly (adjusted P < 0.05) different between baseline and posttreated pMMR samples (Fig. 2G; Supplementary Table S3). GSEA against Hallmark gene sets, C2 curated chemical and genetic perturbations gene sets, and C7 immunologic signatures of treated versus baseline samples was performed. Supplementary Table S3 lists modulation of gene sets involved in the T-cell, myeloid cell, cytokine, and other immune-related pathways between treated and baseline samples. An upregulation in the inflammatory response genes and in signaling pathways such as JAK/STAT3 and STAT5 observed in the Hallmark gene sets along with the downmodulation of FOXP3 targets in CD4 T cells in the C2 gene sets supports the activation of T cells. In addition, a downregulation of proinflammatory monocyte genes in C7 gene and an upregulation of DNA damage response in the C2 gene sets were also observed.

Given the immune changes observed above across the cohort, we next focused on an exploratory analysis of samples that were paired pre and post-surgery from pMMR patients (n = 6). Of the 6 paired (pre- and postsurgery) tumor samples, 5 pairs were analyzed using IHC and immunofluorescence, while 3 pairs passed the quality control metrics to allow flow cytometry subgating of CD8 and CD4 TIL subsets, and myeloid lineage stratifications. Flow analysis demonstrated nonstatistically significant changes, with a decrease in CD8⁺ T-cell percentage, yet trends toward an increase in the percentages of CD8⁺ ICOS⁺, CD8⁺ 41BB⁺, CD8⁺ Tim3⁺, CD8⁺ Lag3⁺, and CD8⁺ PD-1⁺ cells (Fig. 3A–F). Similar results were observed by immunofluorescence staining with nonstatistically significant increases in activated cytotoxic T cells (CD8⁺ GZB⁺), CD8⁺ PD-L1⁺ cells, and effector memory cells in posttreated samples (Fig. 3M–P).

Trends in CD4⁺ T cells were also observed with increases in CD4⁺ Tregs by both flow cytometry (Fig. 3G) and immunofluorescence (Fig. 3Q). Decreases in the myeloid population were seen with nonstatistically significant decreases in both activated monocytes (Fig. 3K) and M2 macrophages (Fig. 3L) by flow cytometry. Interestingly, B-cell densities increased in 3 of 5 samples posttreatment (Fig. 3S).

To identify potential factors that may associate with disease relapse, pMMR tumors were stratified based upon recurrence and assessed for differences in immune populations and gene-expression profiles as well as microbiome profiles at baseline and surgery. At baseline, patients with shorter RFS trended toward lower frequencies of CD8⁺ PD-1⁺ TIL compared with those patients with longer RFS (RFS short vs. long, P = 0.079; Fig. 4A). Other markers of activation or inhibition on CD8⁺ or CD4⁺ T cells were not observed to be different at baseline (Fig. 4A and B). Overall, myeloid populations were not differentially present at baseline between samples with longer RFS and shorter RFS (Fig. 4C).

When comparing surgical cases posttreatment based upon RFS, although CD8⁺ TIL percentage itself did not change (Supplementary Fig. S5A, P = 0.662), we observed a significant difference in the frequency of Tim3 expression on CD8⁺ TIL in longer RFS cases (RFS short vs. long, P = 0.013; Fig. 4E). Expression of other activation and inhibitory receptors on CD4 or CD8 TIL was not different based upon RFS (Fig. 4E and F). Supervised DEG analyses between long and short RFS demonstrated a dramatic B-cell immunoglobulin predominance in the posttreatment samples of the long RFS patients (Fig. 4G; Supplementary Table S4). This B-cell response was characterized by a robust expression of immunoglobulin genes, including IGHG1, IGHG2, IGHG3, and IGHD genes. CD20 staining by IHC demonstrated similar findings with a trend toward an increase in B-cell density in posttreatment samples with longer RFS (P = 0.062; Fig. 4I). Interestingly, in the analysis of pretreatment supervised DEG, there was no evidence for an enrichment of a B-cell response in the long RFS cohort (Supplementary Table S4), and no increase in CD20 density in pretreatment long RFS patients (Fig. 4H; P = 0.308). In addition, higher tumoral CD73 expression (≥10% vs. <10%) posttreatment correlated with improved RFS (S5F, P = 0.002). CD20⁺ lymphoid aggregate densities within the intratumoral compartment were also included in the analysis (Supplementary Fig. S5G).

The genus-level differences in microbiome composition stratified by patients who had short-term RFS and those that had long-term RFS benefit are shown in Supplementary Fig. S5C and Supplementary Table S5. In an exploratory analysis (Fig. 4D), the abundance of certain microbiome species was correlated with disease RFS with an enrichment of Murimonas intestine (P = 0. 038) and a diminution of species Blautia caecimuris (P = 0037), Blautia hominis (P = 0.019), Enterococcus malodoratus (P = 0.006), Anaerostipes caccae (P = 0.023), Fusobacterium simiae (P = 0.008), Lactobacillus salivarius (P = 0.039), Scardovia wiggsiae (P = 0.039), Maihella massiliensis (P = 0.037), Actinomyces viscosus (P = 0.039), Dialister propionicifaciens (P = 0.039), and genus Lachnoclostridium (P = 0.013), Scardovia (P = 0.039), Desulfovibrionaceae unclass (P = 0.037) in patients with shorter RFS.

To assess the baseline immune characteristics in pMMR and dMMR tumors, we evaluated tumor immune markers using flow cytometry and compared 2 dMMR samples with pMMR samples in Fig. 5. The expression pattern of activation and inhibitory receptors within the CD8⁺ (Fig. 5A) and CD4⁺ non-Tregs (Fig. 5B), and frequencies of TIL (Fig. 5C) and myeloid populations (Fig. 5D) in baseline pMMR versus dMMR samples are also shown. Significantly higher expression of Tim3⁺ (P = 0.013) and surface CTLA4⁺ (P = 0.012) was observed in the CD8⁺ TIL subset in baseline dMMR tumors compared with pMMR tumors, suggesting the presence of a greater number of activated T cells in dMMR patients pretreatment. However, there was no significant difference in the frequency of TIL or myeloid subsets between dMMR and pMMR tumors. Overall, more cases are needed to verify these findings. Given that the 2 dMMR patients demonstrated complete pathologic responses to inhibition of the CTLA-4 and PD-1/PD-L1 axes, it suggests that this combination may be functionally important given the higher expression of CTLA-4 observed in these patient samples.

In this 23 patient trial, neoadjuvant use of durvalumab and tremelimumab prior to colorectal cancer liver resection was well tolerated and feasible with 73% of patients undergoing surgical resection. In addition, we show the rapid activity of combined CTLA-4 and PD-L1 therapy in dMMR colorectal cancer with both dMMR patients demonstrating complete pathologic responses after one dose of therapy. Within pMMR colorectal cancer, we show that treatment with durvalumab and tremelimumab was able to produce modest T-cell activation and a posttreatment B-cell signature was found to correlate with long-term RFS benefit.

The Bayesian posterior distribution estimate for the proportion of patients undergoing surgery of 73% in conjuncture with the overall safety demonstrated that perioperative durvalumab and tremelimumab was feasible, meeting the primary endpoint of this study. The perioperative surgical resection setting provides the potential for robust tumor acquisition from surgical resection. However, our results also demonstrate the challenges with pretreatment biopsies in the perioperative setting as over half of our pretreatment biopsies were inadequate for analysis. This may reflect this study's selection of resectable liver metastases, where metastatic tumor size maybe smaller, as reflected by the median size of resected tumors being 2 cm. In this trial, we focused upon quantifying T- and B-cell percentages, phenotyping T-cell subpopulations, and evaluating the changes in their distribution with combination treatment which was achieved by comparing the results obtained by flow cytometry, transcriptome, and immunofluorescence multiplex analysis. Due to the known unique nature of dMMR colorectal cancer, these two patients were excluded from all analyses, except for the direct comparison between dMMR and pMMR colorectal cancer samples. Overall, we observed a general concordance of results across approaches with an increase in cytotoxic T-cell activation and B-cell enrichment, in pMMR colorectal cancer following treatment with anti–PD-L1 and anti–CTLA-4 combination therapy.

Assessment of immune markers in posttreated specimens showed consistent increases in CD8⁺ T-cell activation, and a decrease in the percentages of macrophages in tumors reflecting a dynamic interplay between the immunosupportive and immunosuppressive cell types in mediating antitumor immune responses. Lag3⁺ in particular was markedly amplified in CD8⁺ T cells following treatment. In other studies, antibody blockade of LAG3 increased proliferation and effector cytokine production of intratumoral T cells isolated from pMMR metastatic colorectal cancer, suggesting a potential role for combination immune-checkpoint inhibition in pMMR (34).

RNA-sequencing analysis showing increases in CD86 (an antigen-presenting cell-specific marker) and CD69 (an early activation marker) gene expression is suggestive of a response to anti–CTLA-4 therapy in these patients that mediates an upregulation in costimulatory signals that are necessary for T-cell activation and survival (Supplementary Table S3). However, the lack of significant increases in the expression of coinhibitory receptors CTLA-4 and PD-1 in patients after 4 weeks of treatment suggests that these T cells have not reached exhaustion and may respond to continuation of therapies that target CTLA-4 and PD-1/PD-L1 (35). In addition, enrichment in interferon gamma signature, JAK/STAT signaling, and inflammatory response genes was observed in posttreated tumors, suggesting a potential immune activation mechanism.

Despite these immunologic changes, the median RFS of 9.7 months and median OS of 24.5 months do not appear improved over the expected outcomes for resected colorectal cancer liver metastases (9, 10). Recently, in the randomized CO.26 clinical trial of durvalumab/tremelimumab versus best supportive care in refractory metastatic colorectal cancer, the response rate of durvalumab/tremelimumab was 1% and median PFS was 1.8 months, which was similar to the best supportive care arm of 1.9 months (P = 0.97; ref. 16). However, an improvement in OS (6.6 vs. 4.1 months, P = 0.07) was seen for durvalumab/tremelimumab. In addition, the NICHE trial investigating nivolumab and ipilimumab prior to the resection of localized primary colon adenocarcinomas demonstrated a 27% pathologic complete response in 20 pMMR patients (36). Whether this signal of activity relates to the earlier nonmetastatic stage of patients is not known. In the context of this study, our findings demonstrating T-cell activation suggest that additional factors potentially related to the immunosuppressive microenvironment of the liver or insufficient T-cell recruitment may be contributors to the limited clinical activity of this combination in pMMR colorectal cancer. A recent study combining the use of durvalumab/tremelimumab following Yttrium-90 resin microsphere-based radioembolization also demonstrated the limited effects of radiation on the infiltration of TILs and tumor microenvironment in pMMR, although there was a transient increase of cyclin-dependent kinase inhibitor 1A (CDKN1A: p21CIP1) and TNF receptor superfamily member 10c (TNFRSF10C: TRAILR3) expression (17).

The B-cell activation signature seen in posttreated tumor samples with longer RFS suggests a role of B cells in promoting response, leading to important biomarkers for therapy and contributing a novel finding in colorectal cancer. The existing data from literature capture a very limited role of tumor-associated B cells within the tumor microenvironment, with few studies suggesting a positive correlation of B-cell signature with improved outcome. Along with the production of tumor-specific antibodies and cytokines, some of the other functions attributed to B-cell presence within the tumor include presentation of B-cell receptor-cognate antigens to T cells, or their role in enhancing antigen capture and presentation by dendritic cells (37, 38). A recent study linking driver mutations and B-cell response showed that abundance and a high proportion of IgG1 isotype were associated with improved OS for KRAS mutant but not KRAS wild-type lung adenocarcinoma. Here IGH transcript upregulation was reported to be associated with longer survival in melanoma and lung adenocarcinoma (39). A recent phase II trial of neoadjuvant immune-checkpoint blockade in patients with resectable melanoma also reports an increased B-cell infiltration contributing to response to therapy in patients (40). Our results similarly show that patients with longer RFS have increased IGH levels (Supplementary Table S4).

Another contributing factor in modulating the host inflammatory response and influencing the outcome of cancer therapy is the gut microbiome and dysbiosis in the colon (41, 42). Gut microbiota can modulate intestinal immunity, increase inflammation and the risk of colorectal cancer. In addition, gut microbiome has been associated with response to PD-1 and CTLA-4 blockade (43, 44). Analysis of pretreatment microbiome profiles in these patients revealed that the enrichment of species Murimonas intestini correlates with shorter RFS. Similar to reports in the literature that have identified Lachnoclostridium as a marker for noninvasive diagnosis of colorectal cancer, we observed an abundance in this genus in our patient cohort (45, 46). Our results show that the abundance of Blautia species and Anaerostipes caccae is associated with longer RFS. Blautia was previously reported to be associated with pMMR metastatic colorectal cancer (47). Baseline enrichment in Blautia has also been reported to be associated with longer PFS (44). Anaerostipes caccae, a butyrate producer, can convert lactate to butyrate. Butyrate is known to function in the suppression of inflammation and cancer (48).

The main limitations of our study relate to the small number of patients and even smaller patient numbers that were able to be fully analyzed for immune phenotyping. Given these findings and the design of this study for feasibility and safety as the primary endpoint, the efficacy and translational analyses are hypothesis generating and require further confirmation in additional data sets. In addition, this study did not complete enrollment, and the 95% confidence intervals for surgical resection are broad, suggesting that further efforts to confirm the feasibility of immunotherapy window studies prior to colorectal cancer liver metastases resection are needed. This study did not investigate the role of repeated dosing of durvalumab and tremelimumab, as only one dose of each agent was given prior to surgical resection.

In conclusion, the use of anti–CTLA-4 and anti–PD-L1 in metastatic colorectal cancer demonstrated immune activation with regard to both T-cell and B cells, though robust clinical activity was only seen in dMMR patients. However, the finding of a posttreatment B-cell signature and functional POLE mutation suggests the importance of improved understanding of the B-cell context within the immunotherapy treatment space. Finally, this study demonstrates the safety of a perioperative immunotherapy approach in metastatic colorectal cancer and supports further efforts to utilize this disease space in order to make both clinical and translational improvements in colorectal cancer.

C. Haymaker reports other from AstraZeneca during the conduct of the study, as well as other from Briacell Therapeutics and Mesothelioma Applied Research Foundation outside the submitted work. I.I. Wistuba reports grants and personal fees from Genentech/Roche, Bayer, Astra Zeneca/MedImmune, Pfizer, HTG Molecular, Merck, and Guardant Health; personal fees from Bristol Myers Squibb, GlaxoSmithKline, Oncocyte, MSD, and Platform Health; and grants from Adaptive, Adaptimmune, EMD Serono, Takeda, 4D, Novartis, Amgen, Karus, Iovance, Akoya, and Johnson & Johnson outside the submitted work. R.S.S. Tidwell reports grants from NIH during the conduct of the study. B. Kee reports grants from MedImmune during the conduct of the study and ownership of Medtronic stock. K. Raghav reports personal fees from Bayer, Daiichi, and AstraZeneca outside the submitted work. V. Morris reports grants from Pfizer, Bristol Myers Squibb, and EMD Serono outside the submitted work. R.R. Jenq reports personal fees from Maat Pharma, Seres Therapeutics, Prolacta Bioscience, and LISCure Biosciences and grants and personal fees from Kaleido Bioscience outside the submitted work. A. Tam reports grants from Guerbet LLC, grants and personal fees from Boston Scientific, and personal fees from Cello Therapeutics and Endocare outside the submitted work. C. Bernatchez reports grants from Astra Zeneca during the conduct of the study, as well as other from Myst Therapeutics and grants and personal fees from Iovance Biotherapeutics outside the submitted work. S. Kopetz reports personal fees from Roche, Genentech, Merck, Karyopharm Therapeutics, Amal Therapeutics, Navire Pharma, Symphogen, Holy Stone, Biocartis, Amgen, Novartis, Lilly, Boehringer Ingelheim, Boston Biomedical, AstraZeneca/MedImmune, Bayer Health, Pierre Fabre, EMD Serono, Redx Pharma, Jacobio, Natera, Repare Therapeutics, Daiichi Sankyo, Lutris, Pfizer, Ipsen, and HalioDx outside the submitted work. M.J. Overman reports grants and personal fees from MedImmune during the conduct of the study. No disclosures were reported by the other authors.

P. Kanikarla Marie: Formal analysis, writing–original draft, writing–review and editing. C. Haymaker: Conceptualization, formal analysis, supervision, funding acquisition, validation, visualization, methodology, writing–original draft, writing–review and editing. E.R. Parra: Software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Y.U. Kim: Software, formal analysis, validation, investigation, visualization, and methodology. R. Lazcano: Formal analysis, validation, investigation, visualization, and methodology. S. Gite: Formal analysis, supervision, validation, investigation, and methodology. D. Lorenzini: Formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. I.I. Wistuba: Software, formal analysis, supervision, investigation, and methodology. R.S.S. Tidwell: Formal analysis, validation, methodology, writing–original draft, writing–review and editing. X. Song: Formal analysis, supervision, investigation, visualization, methodology, writing–review and editing. W.C. Foo: Supervision, investigation, visualization, and methodology. D.M. Maru: Supervision, investigation, and visualization. Y.S. Chun: Conceptualization, supervision, investigation, and visualization. A. Futreal: Conceptualization, supervision, and investigation. B. Kee: Supervision, funding acquisition, investigation, writing–review and editing. D. Menter: Funding acquisition, investigation, visualization, writing–original draft, project administration, writing–review and editing. L. Solis: Software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. C.-W. Tzeng: Formal analysis, validation, investigation, methodology, writing–review and editing. C. Parseghian: Supervision, investigation, writing–review and editing. K. Raghav: Conceptualization, supervision, investigation, writing–review and editing. V. Morris: Conceptualization, supervision, and investigation. C.-C. Chang: Software, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. R. Jenq: Conceptualization, supervision, investigation, methodology, writing–review and editing. A. Tam: Conceptualization, supervision, investigation, writing–review and editing. C. Bernatchez: Conceptualization, supervision, investigation, visualization, and methodology. S. Kopetz: Conceptualization, formal analysis, supervision, investigation, methodology, writing–review and editing. J.-N. Vauthey: Conceptualization, resources, formal analysis, supervision, investigation, methodology, writing–review and editing. M.J. Overman: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported through a Medimmune/AstraZeneca MD Anderson Cancer Center alliance, by the NCI through both the Cancer Center Support Grant P30CA16672 [Institutional Tissue Bank (ITB) and Research Histology Core Laboratory (RHCL)], and SPORE Grant P50CA221707, and supported by the generous philanthropic contributions to The University of Texas MD Anderson Cancer Center Moon Shots Program. Adaptive Patient-Oriented Longitudinal Learning and Optimization (APOLLO) Moonshot Program, Strategic Alliances and the Translational Molecular Pathology-Immunoprofiling lab (TMP-IL) at the Department of Translational Molecular Pathology, the University of Texas MD Anderson Cancer Center, and the NCI Cooperative Agreement U24CA224285 (to the MD Anderson Cancer Center CIMAC).

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.