Efficacy, Safety, and Biomarker Analysis of Combined PD-L1 (Atezolizumab) and VEGF (Bevacizumab) Blockade in Advanced Malignant Peritoneal Mesothelioma

Malignant peritoneal mesothelioma (MPeM) is a rare and lethal cancer with annual incidence of 0.11/100,000 and a five-year survival lower than 20% (1, 2). MPeM arises from mesothelial cells that line the serosal layer of the peritoneum and typically presents with abdominal discomfort, distension, and ascites. In contrast to its more familiar analogue, malignant pleural mesothelioma (MPM), MPeM is far less frequent (estimated 275 vs. 2,458 new cases per year in the United States) and understudied (Supplementary Fig. S1A and S1B; ref. 2). MPeM has a weaker association with asbestos exposure (attributable risk: 50% vs. 88%), affects women more frequently (44% vs. 19%), occurs at younger age (median: 63 vs. 71 years), and is diagnosed more often with advanced disease (73% vs. 65%) compared with MPM (1–3). MPeM and MPM also appear to be molecularly dissimilar with copy-number gains and BAP1 mutations more common in MPeM (4, 5).

Treatment strategies for MPeM can vary by patient and disease factors (6). Although optimal cytoreductive surgery [CRS; completeness of cytoreduction score (CCS) 0/1: residual disease <2.5 mm] and hyperthermic intraoperative peritoneal perfusion with chemotherapy (HIPEC) or early postoperative intraperitoneal chemotherapy (EPIC) result in good outcomes for select patients, a substantial proportion need systemic therapy and have limited survival (1, 6). Despite these recognized clinicomolecular and epidemiologic differences, systemic therapy for MPeM is largely based on data extrapolated from MPM or scant retrospective/prospective evidence; hence, consensus regarding optimal treatment is lacking (7, 8). The National Comprehensive Cancer Network (NCCN) recommends first-line platinum–pemetrexed chemotherapy for both mesotheliomas, but after failure of this first-line therapy, there is no recommended standard or FDA-approved therapy for advanced MPeM, and a critical unmet need of novel therapies for this orphan disease exists (9, 10).

MPeM harbors a complex immune milieu and a proinflammatory microenvironment with 50% to 60% of cases expressing PD-L1 (11–13). Although immune-checkpoint inhibition (ICI) has shown efficacy in MPM, data in patients with MPeM are limited and efficacy is low (Supplementary Table S1; ref. 14). Key studies of ICI in mesothelioma, such as Checkmate-743 and PROMISE-meso trials, were exclusively designed for MPM and excluded patients with MPeM (15, 16). Consequently, the body of evidence and approval for ICI is restricted to patients with MPM. The VEGF pathway is functional in MPeM, and VEGF inhibition results in decreased proliferation and metastasis in vivo (17, 18). An active VEGF axis also facilitates immune evasion (19). VEGF inhibition, by converting an immunosuppressive tumor microenvironment to an immunopermissive one through increased infiltration of immune effector cells and better antigen presentation, can augment responses to ICI (20, 21). We hypothesized that combining ICI and antiangiogenic therapy can have synergistic activity in MPeM.

Atezolizumab is a humanized monoclonal antibody (mAb) that targets PD-L1, blocks its interactions with PD-1 and B7-1 (CD80) receptors, and reverses T-cell suppression. Bevacizumab is a mAb against VEGFA that inhibits angiogenesis and tumor growth. Atezolizumab combined with bevacizumab (AtezoBev) has shown robust activity in advanced hepatocellular carcinoma and is approved by the FDA for this indication (21). We conducted this phase II trial to assess the safety and efficacy of AtezoBev in patients with advanced previously treated MPeM and to identify biomarkers of treatment response.

Between March 30, 2017, and February 12, 2019, 20 patients were enrolled and treated with AtezoBev. Baseline characteristics are shown in Table 1. Then median age was 63 years (range, 33–87). Most patients were women (60%) and self-reported no prior asbestos exposure (75%). Biphasic histology was seen in two (10%) patients, whereas the remaining were epithelioid. Twelve (60%) of these patients had prior CRS and HIPEC in addition to systemic chemotherapy. All patients had received prior platinum–pemetrexed therapy (only one patient had prior bevacizumab), and eight (40%) patients had received ≥2 lines of therapy pre-enrollment. The median time from last systemic therapy to trial enrollment was 1.5 months. Seventeen (85%) patients had documented disease progression, and three (15%) were intolerant to prior platinum–pemetrexed therapy.

At data cutoff, the median follow-up was 23.5 months (range, 10.1–36.1). Patients received a median of 15 (range, 4–38) cycles, and six (30%) patients continued treatment at time of data cutoff. Reasons for treatment discontinuation were disease progression [10 (50%)], toxicity [2 (10%)], death [1 (5%)], and withdrawal of consent [1 (5%)]. Among 20 evaluable patients, the primary endpoint of confirmed objective response rate (ORR) per Response Evaluation Criteria in Solid Tumors-version1.1 (RECISTv1.1) by independent radiology review (IRR) was 40% (95% CI, 19.1–64.0; 8/20 patients; Fig. 1A). Responses were ongoing in six (75%) of these eight patients at data cutoff. The median duration of response (DoR) was 12.8 months (Fig. 1B) in responders. In a post hoc analysis, similar ORR was observed across key clinical subgroups reflecting patient, disease, and prestudy treatment factors (Supplementary Fig. S2). Disease control at 12 and 18 weeks per RECISTv1.1 was seen in 19 (95%) and 17 (85%) patients, respectively. Supplementary Fig. S3A–S3D illustrates key responses in patients on study. No pseudoprogression was seen in our cohort. At data cutoff, 10 (50%) progression events and 7 (35%) deaths were seen. Most patients progressed within the peritoneal cavity, and two (20%) patients had extraperitoneal progression. Median progression-free survival (PFS) was estimated to be 17.6 months [95% CI, 9.1–not reached (NR)], and one-year PFS was 61% (95% CI, 35–80; Fig. 1C). Median overall survival (OS) was not reached at data cutoff. The one-year OS was 85% (95% CI, 60–95; Fig. 1D). Corresponding outcomes per immune-modified RECIST (imRECIST) are shown in Supplementary Table S2.

We performed an exploratory analysis to assess the impact of selection bias pertaining to indolent tumor biology in confounding outcomes by evaluating patient and population dynamics of the study cohort (all 20 patients) prior to enrollment (Supplementary Methods S1). We reviewed their treatment course on platinum–pemetrexed chemotherapy prior to study enrollment using electronic medical records. Time to next treatment was defined as the time interval between treatment (platinum–pemetrexed) initiation and commencement of next line of therapy as per treating physician discretion and included time on maintenance pemetrexed, duration of reintroduction, and treatment breaks. Disease regression with chemotherapy was seen in 35% of patients, similar to the reported response rate with platinum–pemetrexed combination in prior studies for this population (9). The median time to next treatment on standard-of-care platinum–pemetrexed treatment prior to enrollment was 8.3 months (95% CI, 6.3–10.3) compared with 17.6 months with AtezoBev on study (Supplementary Figs. S4 and S5). Durable responses to AtezoBev were seen regardless of response characteristics on prior chemotherapy (Supplementary Fig. S5).

All 20 patients were included in safety analyses and received a median of 15 (range, 4–38) cycles of AtezoBev. Mean dose intensity was 99% for atezolizumab and 81% for bevacizumab. Treatment-emergent adverse events (TRAE) of any grade were reported by 17 (85%) patients (Table 2; Supplementary Table S3). Grade 3 TRAEs occurred in 10 (50%) patients; most common were hypertension (40%) and anemia (10%). No grade 4/5 events occurred. Two (10%) patients had grade 3 immune-related adverse events (AE), pancreatitis and thrombocytopenia managed with corticosteroids, that required treatment discontinuation. Proteinuria, the only TRAE that caused dose interruptions (all bevacizumab), occurred in five (25%) patients after a median of six cycles. One patient's death on study was attributed to disease progression. TRAEs that occurred in at least 10% and 20% of patients (or grade 3 events) are listed in Supplementary Table S3 and Table 2, respectively.

First, we evaluated whether established biomarkers of response to ICI used clinically [microsatellite instability (MSI), PD-L1, and tumor mutational burden (TMB) status] in other tumors predicted response in our patients with MPeM treated with AtezoBev on study. PD-L1 and TMB status was determined in 13 (65%) and 14 (70%), respectively. PD-L1 expression of 0%, 1% to 50%, and ≥50% was seen in four (31%), six (46%), and three (23%) patients, respectively. Responses were seen in both PD-L1–positive and PD-L1–negative cases [44.5% vs. 25.0%; odds ratio (OR) 2.4; 95% CI, 0.2–38.0; P > 0.99] and median PD-L1 expression did not differ between responders and nonresponders (Fig. 2A). Median TMB for all patients was 0.8 mutations/megabase (range, 0.4–20.7), and one patient had high-TMB (≥10 mutations/megabase; Supplementary Fig. S6). Median TMB was similar among responders and nonresponders (0.8 vs. 0.8; P = 0.83; Fig. 2A). Using whole-exome sequencing (WES), we did not find any association between response and specific gene alterations (mutations or copy-number variations; Fig. 2A; Supplementary Fig. S7).

We then explored plausible biology underlying efficacy of combined PD-L1 and VEGF blockade using transcriptomic profiling in 14 (70%) patients with evaluable pretreatment tumors. Gene-expression scores were calculated for each patient using normalized RNA expression fitted to established signatures based on associations with following biology: immune sensitivity, angiogenesis, and epithelial–mesenchymal transition (EMT; Supplementary Methods S2; refs. 22–24). Heat map of genes delimiting immune sensitivity and angiogenesis failed to show any distinct subgroups, and median gene signature scores were similar between responders and nonresponders (Fig. 2B; Supplementary Fig. S8; refs. 22, 23). Conversely, the EMT gene signature scores were lower (favoring an epithelial over mesenchymal phenotype) in responders (median: −0.4 vs. 0.8) compared with nonresponders and correlated with the magnitude of response (spearman r: 0.8; 95% CI, 0.5–0.9; P = 0.0010; Fig. 2B; ref. 24). High EMT gene expression was associated with poorer ORR (14% vs. 86%; OR, 0.03; 95% CI, 0.0–0.6; P = 0.029), but no differences were seen with other scores (Supplementary Fig. S9). To identify modulations following treatment, we compared change in gene signature scores between baseline and on-treatment samples and found no significant association with response (Supplementary Fig. S10).

To delineate a specific immune milieu predictive of response to AtezoBev, we examined pretreatment immune cell subsets across tumor and stroma using multiplex immunofluorescence (mIF) in 15 (75%) available patient samples. Density of key immune effector cells (number of cells/mm²), such as total lymphocytes (CD3⁺), cytotoxic T cells (CD3⁺CD8⁺), and regulatory T cells (FOXP3⁺) and macrophages (CD68⁺), and ratio of effector/suppressor cells did not differ significantly between responder and nonresponders (Fig. 2C; Supplementary Fig. S11).

MPeM is a life-threatening malignancy with limited treatment options beyond first-line platinum–pemetrexed-based chemotherapy. Despite this unmet need, the rarity of MPeM has hindered attentive research. Dedicated trials are missing as efforts have focused on MPM, creating evidence gaps in how we treat these patients. Our study designed specifically for MPeM, acknowledging differences between MPeM and MPM, is a singular but necessary venture in this rare tumor (25). In this study, AtezoBev demonstrated a confirmed ORR of 40% in platinum–pemetrexed-treated MPeM. These responses were durable with median DoR in excess of 12 months. Notably, we observed meaningful PFS and OS of 61% and 85% at one year, respectively. Therapy was very well tolerated with a safety profile consistent with prior reports (21). Most AEs were grade 1/2, and no grade 4/5 toxicity occurred. Grade 3 AEs were readily manageable and did not require treatment discontinuation barring immune-mediated AEs in two patients.

Although nivolumab and ipilimumab and bevacizumab are available to patients with MPM as standard first-line options based on results of Checkmate-743 and MAPS trials, these are not approved for use in MPeM (16, 26). MPeM and MPM also appear to be dissimilar in their expression of PD-L1 and in their response to ICI (13, 14). PD-L1 expression is seen in nearly 50% of patients with MPeM compared with 30% in MPM (13). In a small study enrolling both MPM and MPeM cases, pembrolizumab showed an ORR of 20% in pleural (N = 56) versus 12.5% in peritoneal (N = 8) mesothelioma (14). The strong therapeutic effect in this study of AtezoBev in patients with MPeM, as measured by a 40% confirmed ORR and one-year OS of 85% compared with the one-year OS of 45% to 56% reported in literature for patients with systemic therapy, is notable (6, 9, 10). This promising response rate and the totality of evidence (with substantial DoR, PFS, and OS) compare very favorably with any therapies available for these patients in clinical practice (Supplementary Table S4), although prospective data and consensus are lacking for these therapies in MPeM. Responses with therapies in second and subsequent lines occur in 10% to 25% of patients and are often short-lived with limited median PFS (2–7 months) and OS (6–18 months). Because historical data regarding MPeM are scarce, we leveraged real-world evidence to further our evidence of benefit from AtezoBev. We showed that responses to AtezoBev occurred regardless of outcomes with prior platinum–pemetrexed, and duration of treatment with AtezoBev was distinctly longer (17.6 vs. 8.3 months) compared with time to next treatment from previous platinum–pemetrexed therapy. The better tolerance of AtezoBev over chemotherapy also allows for this prolonged treatment duration. Notably, our trial population characteristics at baseline and behavior prior to study enrollment were consistent with historical multicenter expanded-access experience in MPeM arguing for a representative cohort (9). Although there are limitations to such intrapatient comparisons, such as inability to compare histology subtype or history of asbestos exposure, not reported in these cohorts, the data presented here have significant clinical implication for this orphan disease. The authors recommend that AtezoBev should be considered a meaningful treatment option for these patients barring clinical trial participation. Although our study allowed prior bevacizumab, we cannot determine the true efficacy of AtezoBev in patients with prior exposure to anti-VEGF drugs, because only one patient had prior bevacizumab. Notably, this patient did achieve a durable response to AtezoBev.

Integration of pretreatment/on-treatment biopsies in this trial demonstrates the feasibility and value of a translationally driven approach in rare cancers. Using this we demonstrated that strong activity of AtezoBev seen in our patients with MPeM did not correlate with clinically established biomarkers (PD-L1 and TMB) of response to ICI in other tumors, although these are not validated in mesothelioma. All tumors were microsatellite-stable as expected because this is a rare occurrence in mesothelioma (27). Responses also occurred in PD-L1–negative patients, although a trend toward a higher response rate was seen with PD-L1 positivity. However, PD-L1 staining and interpretation in patients can vary with the assay (Ventana SP263 clone as used in our study versus Ventana SP142 and Dako 22C3), as sensitivity and specificity are different, although good correlation is seen across the assays (28). Because PD-L1 expression in MPeM may be affected by prior therapy, this is an important consideration in designing future trials in this population. Furthermore, our comprehensive profiling demonstrated EMT phenotype as a resistance mechanism to AtezoBev. EMT gene signature score has been reported to blunt responses to ICI in lung cancer and have a prognostic impact in MPM (24, 29). In our cohort, mesenchymal phenotype (higher EMT gene scores) was associated with poorer PFS on both AtezoBev and prior platinum–pemetrexed chemotherapy (Supplementary Fig. S12). To further the validity of this finding, we explored a sarcomatoid component (S-comp) and EMT gene signature scores derived from metanalysis of published classifications in MPM in our study cohort (30). Although S-comp score overall showed no association with response, we found a strong correlation between the two EMT scores (Spearman r = 0.72; P = 0.005) and a similar trend of responders having lower EMT scores than nonresponders (Supplementary Fig. S13; refs. 24, 30). Even among patients with epithelioid subtype (after excluding biphasic cases), EMT continued to be a predictor of response (median EMT score in responders vs. nonresponders: −0.49 vs. 0.81, P = 0.014). These results indicate that transcriptomic mesenchymal differentiation is a predictor of poor outcomes with AtezoBev in MPeM, including epithelioid MPeM (Supplementary Fig. S13).

Clinically, ICI and VEGF blockade individually have demonstrated limited activity with modest response rates as single agents in mesothelioma (Supplementary Table S1). Clinical trials have shown response rates between 7% to 20% and 0% to 6% for ICI and VEGF inhibition, with corresponding median PFS of 2.5–6.2 months and 2.2–4.1 months, respectively. Both response rate and PFS with AtezoBev in our study appear to be much better than would be expected with single-agent activity. Our gene-expression analyses showing lack of any predictive impact of angiogenic and immune signatures further the proof-of-component with AtezoBev and argue that the efficacy of AtezoBev in our cohort is conceivably a result of complex synergistic interactions of dual PD-L1 and VEGF blockade rather than each drug alone. To investigate this, we explored prognostic transcriptomic profiles described in MPM (hot/immune-checkpoint⁺/angiogenic⁺, VEGF2⁺/VISTA⁺, and cold/angiogenic) in our cohort, but found no clear predictive signatures for AtezoBev in our MPeM cohort (Supplementary Fig. S14; ref. 31). However, because these analyses are limited by size and hypothesis generating in nature, further investigations are required to elucidate the mechanism of this synergy. Efforts using comprehensive integrated multiomic analyses aimed at understanding the molecular continuum and heterogeneity in these patients can help uncover this biology. Additionally, although our study offers a much-needed treatment option, a subset of patients fail to derive any benefit with AtezoBev. Trials investigating the addition of AtezoBev to chemotherapy and use of novel immune targets and combinatorial immunomodulator strategies are needed. Future collaborative efforts should be considered high priority if we are to benefit more patients.

In conclusion, AtezoBev was well tolerated and led to robust and durable responses in patients with MPeM who had progressed on or were intolerant to prior platinum–pemetrexed chemotherapy with meaningful prolongation of survival. This study establishes a promising treatment option for our patients who suffer from this morbid cancer and represents an unprecedented effort to bridge the gap of dedicated research in this orphan disease.

Eligible patients were ≥18 years old, had histologically confirmed advanced MPeM not amenable to definitive CRS (according to peritoneal-multidisciplinary tumor board) and had progressed on or were intolerant to at least one line of systemic chemotherapy involving platinum–pemetrexed doublet. Prior bevacizumab was allowed (Supplementary Methods S2). Patients were required to have measurable disease according to RECISTv1.1, an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 or 1 and normal organ/bone marrow function. Extraperitoneal metastases including pleural and lung metastases were allowed. Key exclusion criteria were any prior immunotherapy, diagnosis of active autoimmune disease or immunodeficiency, concurrent malignancy, any known history of active/untreated central nervous system metastases, and ongoing systemic immunosuppressive therapy at the time of enrollment. Full eligibility criteria are provided in the study protocol (Supplementary Protocol).

This phase II single-center study was designed as an open-label, single-stage, multicohort basket trial for evaluation of AtezoBev in a variety of advanced rare cancers, including MPeM. Each cohort of 20 patients had an individual analysis planned, and this article reports on the results of the MPeM cohort. Atezolizumab was administered at a fixed dose of 1,200 mg in combination with bevacizumab at a dose of 15 mg/kg intravenously every 21 days until disease progression or unacceptable toxicity. Dose modifications were not permitted. Dose interruptions were allowed for bevacizumab if a patient had related AEs and patients were permitted to continue atezolizumab alone. Patients with initial progression could continue therapy if they were clinically well and assessed for pseudoprogression per defined criteria. Details are provided in the study protocol (Supplementary Protocol). The objective of this study was to determine the clinical efficacy and safety of AtezoBev for patients with advanced MPeM who have failed prior systemic treatment with platinum–pemetrexed chemotherapy. The primary endpoint was confirmed ORR per RECISTv1.1 by IRR. Key prespecified secondary endpoints were safety, disease control rate (DCR; percentage of patients who achieved confirmed response or confirmed stable disease), DoR (defined as time interval between date of first confirmed response to progression), PFS, and OS. ORR, DCR, DoR, and PFS were also assessed per imRECIST criteria. Exploratory objective was to examine tissue correlates for clinical activity.

The protocol and all amendments were approved by the University of Texas MD Anderson Institutional Review Board. The study was conducted in accordance with Declaration of Helsinki and International Conference on Harmonization Good Clinical Practice guidelines. Patients provided written informed consent before study enrollment. The full protocol is provided (Supplementary Protocol; NCT03074513).

Tumor assessments were done using either CT or MRI to include chest, abdomen, and pelvis at baseline and every nine weeks until disease progression. Consistent imaging modality was used for tumor assessment for each patient. Tumor measurements by IRR using RECISTv1.1 were performed by Institutional Quantitative Imaging Analysis Core. Responses, partial (PR) or complete (CR), were confirmed on subsequent scans in nine weeks (at least four weeks as per protocol) after the initial response. TRAEs were assessed from the date of initiation of protocol therapy until ≥28 days after last dose. AEs were classified and graded according to the Common Terminology Criteria for Adverse Events-version4.0 (CTCAEv4.0). Baseline (within seven days prior to treatment initiation) and on-treatment (cycle 2 day 1 ± 7 days) tumor specimens were obtained from all patients. Blood was also collected for correlative analyses. Details are provided in the study protocol (Supplementary Protocol).

Tumor MSI and PD-L1 expression status were evaluated using IHC on formalin-fixed, paraffin-embedded (FFPE) tumor slides. All biomarker analyses, except MSI (performed on archived tissue as part of clinical care), were performed on fresh tumor biopsy samples collected pretreatment (within seven days prior to treatment initiation) and on-treatment (1 day prior to cycle 2 day 1) per protocol. Pathologists, blinded to clinical data, with expertise in PD-L1 assessment evaluated staining. PD-L1 was assessed using the SP263 clone and reported as a proportion of tumor cells expressing PD-L1. TMB, determined by WES using standard protocol, was reported as the number of somatic mutations per megabase (Mb) of captured region. Details are provided in Supplementary Methods S2. Fresh-frozen tumor blocks with adequate tumor content were used to extract DNA and RNA. Samples meeting prespecified quantity/quality criteria underwent WES and RNA sequencing on Illumina Hi-seq platform. Gene-expression scores were calculated for each patient using normalized RNA expression fitted to established signatures based on associations with following biology: immune-sensitivity, angiogenesis, and EMT (22–24). Patients were divided into high or low groups based on median gene signature score for all patients. mIF on FFPE was performed using an antibody panel to characterize cancer and subsets of tumor-associated immune cells using 10 markers on Opal chemistry and multispectral microscopy Vectra system. Expression of protein markers was examined using infiltrate density scores. Details are provided in Supplementary Methods 2.

Patients who received at least one dose of AtezoBev were included in the primary safety and efficacy analysis. Data are reported as of April 15, 2020. Descriptive statistics were used. The Clopper and Pearson method was used to calculate the exact 95% CIs for primary endpoint analysis and other proportions. Time to event outcomes (PFS and OS) were estimated using the Kaplan–Meier method. The Fisher exact test or χ² test (when appropriate) was used for comparing proportions across groups. An unpaired nonparametric Mann–Whitney (Wilcoxon rank sum) test was used to compare means (and medians) between two distinct groups. Details regarding the statistical plan are provided in the study protocol (Supplementary Protocol). Statistical analysis was performed using R 3.6.3 (http://www.R-project.org), SPSS version 25 (IBM Corp), and GraphPad Prism version 8.0.0 (GraphPad Software).

K. Raghav reports personal fees from AstraZeneca, Bayer, Daiichi-Sankyo, Eisai, and Seattle Genetics and research support from Bayer, Daiichi-Sankyo, Genentech/Roche, Guardant Health, Lilly, and Medimmune outside the submitted work. M.J. Overman reports personal fees from Janssen, Bristol-Myers Squibb, Merck, AbbVie, Medimmune, and Takeda and research support from Merck, Bristol-Myers Squibb, Lilly, Nouscom, Medimmune, and Genentech/Roche outside the submitted work. I.I. Wistuba reports personal fees from Bayer, Bristol-Myers Squibb, Genentech/Roche, AstraZeneca/Medimmune, Pfizer, Merck, HTG Molecular, GlaxoSmithKline, Guardant Health, and MSD and research support from Genentech/Roche, Bayer, Bristol-Myers Squibb, Pfizer, Merck, AstraZeneca/Medimmune, HTG Molecular, Guardant Health, Oncoplex, DepArray, Adaptive, Adaptimmune, EMD Serono, Takeda, Amgen, Karus, Johnson & Johnson, Iovance, 4D, Novartis, Oncocyte, Akoya outside the submitted work. S. Kopetz reports personal fees from Genentech/Roche, Merck, Karyopharm Therapeutics, Amal Therapeutics, Navire Pharma, Symphogen, Holy Stone, Biocartis, Amgen, Novartis, Lilly, Boehringer Ingelheim, Boston Biomedical, AstraZeneca/MedImmune, Bayer, Pierre Fabre, EMD Serono, Redx Pharma, Jacobio, Natera, Repare Therapeutics, Daiichi-Sankyo, Lutris, Pfizer, Ipsen, and HalioDx outside the submitted work. J.C. Yao reports grants and personal fees from Novartis outside the submitted work. D.M. Halperin reports research support from Genentech/Roche during the conduct of the study, personal fees from Advanced Accelerator Applications, Ipsen, Lexicon, ITM, Curium, and AbbVie and research support from Tarveda Therapeutics, ThermoFisher Scientific, Novartis, and Advanced Accelerator Applications outside the submitted work. P. Cotazar, E. McKenna, C. Yun, S. Dervin, K. Schulze, and W.C. Darbonne are full-time employees of and have received stock/stock options from F Hoffmann-La Roche/Genentech. No disclosures were reported by the other authors.

K. Raghav: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Liu: Data curation, software, formal analysis, validation, methodology, writing–original draft, writing–review and editing. M.J. Overman: Investigation, writing–original draft, writing–review and editing. A.F. Willett: Data curation, investigation, writing–review and editing. M. Knafl: Data curation, investigation, writing–review and editing. S.-C. Fu: Data curation, investigation, writing–review and editing. A. Malpica: Data curation, formal analysis, investigation, writing–review and editing. S. Prasad: Data curation, writing–review and editing. R.E. Royal: Writing–review and editing. C.P. Scally: Writing–review and editing. P.F. Mansfield: Writing–review and editing. I.I. Wistuba: Resources, methodology, writing–review and editing. A.P. Futreal: Resources, writing–review and editing. D.M. Maru: Resources, writing–review and editing. L.M. Solis Soto: Data curation, investigation, writing–review and editing. E.R. Parra Cuentas: Data curation, investigation, writing–review and editing. H. Chen: Data curation, investigation, writing–review and editing. P. Villalobos: Data curation, investigation, writing–review and editing. A. Verma: Data curation, investigation, writing–review and editing. A. Mahvash: Investigation, writing–review and editing. P. Hwu: Writing–review and editing. P. Cortazar: Writing–review and editing. E. McKenna: Writing–review and editing. C. Yun: Writing–review and editing. S. Dervin: Writing–review and editing. K. Schulze: Writing–review and editing. W.C. Darbonne: Writing–review and editing. A.C. Morani: Investigation, writing–review and editing. S. Kopetz: Writing–review and editing. K.F. Fournier: Writing–review and editing. S.E. Woodman: Resources, data curation, formal analysis, methodology, writing–review and editing. J.C. Yao: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. G.R. Varadhachary: Conceptualization, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. D.M. Halperin: Resources, data curation, project administration, writing–review and editing.

This research was funded by F Hoffmann-La Roche/Genentech and the NIH CCSG Award (P30 CA016672). We would like to first and foremost thank our patients, their families, and caregivers for participating in this trial. We thank all site personnel [including Gastrointestinal Medical Oncology (GIMO) research and regulatory team (Alma Delagarza, Christina J. Williams, Kimberly D. Ross, Laurel Deaton, Marily Elopre, Mari Gray, Michelle Escano, Shaelynn Riley, Shanequa Manuel, Tracy Trevino), clinical pharmacists, and other clinical staff] for clinical trial support; William Betsy for coordinating trial logistics and regulatory provisions; Arvind Dasari, MD (MD Anderson Cancer Center, Houston, TX), and Jonathan Loree, MD (BC Cancer, University of British Columbia, Vancouver, Canada) for critical review of the manuscript and input, and the Mesothelioma Applied Research Foundation whose patient travel grant program assisted some patients with travel and housing costs. This study was supported by the NIH CCSG Award [CA016672; Institutional Tissue Bank (ITB) and Research Histology Core Laboratory (RHCL)], Adaptive Patient-Oriented Longitudinal Learning and Optimization (APOLLO) Moonshot Program, Strategic Alliances, and the Translational Molecular Pathology-Immunoprofiling lab (TMP-IL) at the Department Translational Molecular Pathology, The University of Texas MD Anderson Cancer Center. We would like to acknowledge the support staff Grace Mathew and Wenhua Lang (APOLLO); Celia Garcia-Prieto, Liren Zhang, and Julia Mendoza-Perez (Strategic alliance); Wei Lu and Jianling Zhou (IHC lab technician); Mei Jiang and Auriole Tamegnon (Multiplex lab technician); Renganayaki Krishna Pandurengan and Shanyu Zhang (Data consolidation), and Beatriz Sanchez-Espiridion and Sandesh Subramanya (TMP-IL) for their support to the translational analyses.

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.