ZEBRA: A Multicenter Phase II Study of Pembrolizumab in Patients with Advanced Small-Bowel Adenocarcinoma

SBA is a rare malignancy affecting approximately 3,600 to 4,400 new patients in the United States with a similar incidence in Europe annually (1). Because of difficulty visualizing the epithelial surface of the small intestine except the most proximal and distal tumors, SBA is often diagnosed in its late stages. As a rare cancer, it remains historically understudied and has been traditionally managed as an extension of colorectal cancer. Within the past decade, several small prospective phase II clinical trials of oxaliplatin-based chemotherapy in combination with fluoropyrimidines have been explored in the first-line treatment setting of advanced SBA. These studies suggest that FOLFOX (2–4), CAPOX (5), and CAPIRINOX (6) with or without bevacizumab are active and safe and have emerged as the standard of care for first-line therapy, as affirmed by recently published National Comprehensive Cancer Network guidelines (7).

With the lack of prospective trial data, there is no clearly established therapy beyond the first line for SBA. FOLFIRI efficacy is supported by retrospective data only (8). In the only successful prospective study of second- and later-line patients, nab-paclitaxel was shown to have promising activity, with an overall response rate of 20% [50% disease control rate (DCR)] in 10 patients with refractory SBA (9). This contrasts starkly to the lack of taxane activity in colorectal cancer (9, 10).

In addition to the above suggestion that there are unique cytotoxic mechanisms between the two cancers, recent comparative genomic analyses have found key molecular differences between SBA and colorectal cancer (11, 12). Unfortunately, the most frequent SBA gene mutations are not currently clinically relevant targets, including KRAS, TP53, APC, and SMAD4. Although HER2 (ERBB2), BRAF (often non-V600), and PI3KCA mutations are seen (11, 13, 14), the low frequency (<16%) in a rare cancer makes it difficult to study drug efficacy.

Recent successes using immunotherapy to durably treat a number of advanced cancers have spurred interest in investigating the SBA immune tumor microenvironment. While approximately 4% of stage IV colorectal cancers are MSI-H, which is predictive of response to immune checkpoint inhibitors, including pembrolizumab, twice that frequency is seen in SBA (11). In addition, per our earlier assessment, programmed death ligand 1 (PD-L1) is expressed in up to 50% of SBA tumor or stromal cells (15, 16), a similar frequency to gastroesophageal cancers (GEC), although this prevalence varies by histologic subtype (17), MSI/MMR status (17), and underlying small-bowel comorbidities (18). Because of the improved understanding and discrimination of SBA tumor signaling and microenvironment composition, we hypothesized that PD-1 immune checkpoint inhibition with pembrolizumab would induce meaningful tumor response in patients with refractory metastatic SBA, unselected by MSI status.

The study was conceived and designed through the International Rare Cancers Initiative (IRCI) Small Bowel Adenocarcinoma Working Group. From May 2017 to September 2018, 40 patients with advanced SBA were enrolled at six institutions via the ACCRU network to a single-arm phase II study to receive pembrolizumab, 200 mg i.v. once every 3 weeks until treatment intolerance, disease progression, or completion of 35 cycles. To qualify, each subject had biopsy-proven adenocarcinoma of the small intestine at any site (duodenum, jejunum, ileum). Ampullary and appendiceal cancers were excluded. Inclusion criteria included age ≥18 years old, ≥1 prior line of systemic chemotherapy for unresectable or metastatic disease, measurable disease per RECIST v1.1, ECOG performance status 0 to 1, and adequate baseline organ function. Prospective participants were excluded if they had nonadenocarcinoma histology, prior immune checkpoint inhibitor exposure; immunodeficiency including HIV; chronic steroid or immunosuppressive therapy within 1 week prior to registration; active CNS metastases; active autoimmune disease; past pneumonitis; active infections including hepatitis B and C; prior tuberculosis; or confounding comorbidities.

Ethics board review and human research protections monitoring were performed at each institution. All patients signed written, informed consent. The trial was conducted according to ICH GCP, and in accordance with the Declaration of Helsinki. Trial identifiers were RU021502I (ACCRU), MISP No. 52192 (Merck), NCT02949219 (clinicaltrials.gov).

PD-L1 expression was centrally assessed (QualTek) by IHC assays using the Merck 22C3 antibody (Agilent/Dako). Samples required ≥50 viable neoplastic cells. Modified proportion score (MPS) was based on a calculation of the tumor proportion score of quantified PD-L1 expression in reactive tumor cells and other PD-L1 expressing mononuclear inflammatory cells located in tumor nests (19). PD-L1 positivity within surrounding stroma was not quantified, but presence/absence was reported qualitatively. An MPS ≥1 with qualitative presence of PD-L1 at the stromal interface is analogous to CPS ≥1.

Next-generation sequencing (NGS) to assess somatic mutations and TMB levels was performed on 27 sufficient tissue specimens using a commercially available panel (Tempus xT) with methods as described previously (20). DNA mismatch repair (MMR)/MSI status was assessed locally via IHC or NGS/PCR. Patients without local MMR/MSI assessment were performed centrally via NGS (Tempus xT) or IHC with submitted tissue depending upon the amount of tissue available or via NGS using circulating tumor DNA (ctDNA) collected at screening (Guardant) when tissue was not available. ctDNA was used for MMR and individual genetic mutation status (noted as * in Table 4), but not for TMB assessment. Local clinical genomic testing, where performed and available, was also collected for analysis.

Primary endpoint was confirmed ORR per RECIST v1.1 criteria with a second scan at least 4 weeks after the initial scan showing response. All patients who met eligibility criteria, signed consent, and began treatment were included in the primary analysis. The null hypothesis was 10% response rate, and an alternative hypothesis of 30% was used to confirm pembrolizumab activity in SBA. This design has a significance level of 0.06 when the confirmed response rate is 10%, and 94% power to detect a confirmed response rate of 30% in 35 eligible patients, where ≥7 confirmed responses (20%) would signify adequate activity to pursue further study. A 12-week scanning interval was selected due to unknowns of immunotherapy-induced pseudoprogression that existed at time of study design. Treatment was allowed to continue past progression if the investigator deemed ongoing clinical benefit.

Secondary endpoints included toxicity/safety, PFS, and OS, with preplanned stratification of survival outcomes by primary tumor site, PD-L1 expression, and MSI/MMR status. Adverse events (AEs) were assessed at each encounter and graded by CTCAE v4.03 criteria. PFS was defined as the time from study registration to the first of either PD or death. OS was defined as the time from study registration to death from any cause. Time-to-event distributions were estimated with the Kaplan–Meier method (21) and log-rank tests were used to compare groups of interest. Univariate and multivariate analyses of clinical and genetic factors with overall survival were performed. None reached statistical significance (Supplementary Tables S2A and S2B).

Planned correlative analyses included response and survival associations with multiple biomarkers, including TMB, tumor-infiltrating lymphocytes (TILs), MPS, and TP53, KRAS, BRAF, or APC mutations. Given the small number of confirmed responses, we associated the biomarker data with grouped response data [stable disease (SD) + PR vs. PD] or singular group associations (PR vs. SD vs. PD). These associations were assessed via Wilcoxon rank-sum tests, Fisher exact tests, Kruskal–Wallis tests, and log-rank tests. SAS version 9.4 (SAS Institute Inc.) was used for statistical analysis. Two-sided P values <0.05 were considered statistically significant.

From May 2017 to September 2018, 40 eligible patients with advanced SBA were enrolled and treated with pembrolizumab (Supplementary Fig. S1). The majority had primary duodenal tumors (24, 60%). Median age was 63 years (29–85) with male predominance (24, 60%; Table 1). Twenty-five percent were non-white, and 67.5% had at least two prior lines of chemotherapy.

Of the 40 evaluable patients, 36 (90%) had tissue or blood available to characterize MMR/MSI status (Table 1). Thirty-two patients (80%) were found to have pMMR/MSS/MSI-low tumors, and four patients (10%) were categorized as having MSI-H tumors (although one of these had heterogenous dMMR and was MSS by Tempus), concordant with expected rates seen in the general advanced SBA population. Ten of these 36 subjects (28%) required MSI status determination by ctDNA due to lack of evaluable tissue, and dMMR/MSI-H was not detected in any of these 10 patients.

Within the entire study population, we observed three confirmed PRs for an 8% confirmed ORR [95% confidence interval (CI), 2%–20%; Fig. 1A]. Two of four patients with MSI-H achieved confirmed PR (50%). One (3%) of 32 patients with MSS/MSI-L status confirmed PR and another had unconfirmed PR at week 12, but progressed at the week-24 scan (Fig. 1A and B). One patient with apparent target lesion PR had progression at nontarget and new lesions [Fig. 1A, (*) notation]. None achieved CR. The DCR (PR + SD) was 38% regardless of MSI status (three confirmed PR + one unconfirmed PR + 11 SD; Fig. 1A). Four patients were treated past progression due to investigator-assessed clinical benefit (Fig. 1B).Three patients did not have post-baseline scans and were thus inevaluable for response, two died within 1 month of starting therapy, and one other died after four cycles before imaging. They have been marked NE (not evaluable) in Table 4.

The median duration of response (DOR) for the four patients with confirmed or unconfirmed PR as best response was 28.5 months (range, 3.0–32.1). Eleven patients with SD had a median DOR of 2.8 months (range, 1.4–16.7; Fig. 1B). Among responders, the average DOR was 28.5 months (range, 26.5–30.5) in patients with MSI-H status and 17.5 months (range, 3.0–32.1) for patients with MSS/MSI-L. Among 10 patients with SD and MSS/MSI-L, the median DOR was 2.6 months (range, 1.3–16.7). Figure 1C shows DOR by response category, suggesting responses are durable.

All 40 patients were evaluable for secondary survival endpoints. Median PFS was 2.8 months overall (95% CI, 2.7–4.2 months; Fig. 2A) with no significant survival difference (P = 0.93) when stratified by primary tumor location (duodenum, jejunum, ileum; Supplementary Fig. S2A). When stratified by MSI/MMR status, the median PFS was not reached for patients with confirmed MSI-H (Supplementary Fig. S2B) with an HR of 0.29 compared with patients with MSS/MSI-L (95% CI, 0.07–1.22; P = 0.07), which trended toward statistical significance.

As of the censor date (October 14, 2020), seven patients were living. In this refractory SBA population, median OS was 7.1 months (95% CI, 5.1–17.1; Fig. 2B). Again, no statistically significant survival difference was seen among groups stratified by either tumor site or MSI/MMR status (Supplementary Fig. S3A and S3B), possibly secondary to small cohort sizes.

Of the 40 patients who received pembrolizumab, seven (18%) experienced grade 3 or higher AEs attributed as at least possibly related to treatment (Table 2), and 25 (63%) experienced grade 3 or higher adverse events, regardless of the relationship to the study treatment (Supplementary Table S1). The most common AEs attributed to pembrolizumab include hepatobiliary enzyme abnormalities and thyroid dysfunction, whereas severe (grade 3–5) AEs were rare and affected only one patient (4%) in each reported disorder (Table 2).

Although eight patients (20%) experienced grade 5 AEs during the study and reporting period (Supplementary Table S1), all except two events (one possibly related sepsis, one possibly related respiratory failure) were deemed not related or unlikely related by the treating investigator (Table 2).

Exploratory biomarker analysis using archived tissue was performed for 27 patients with sufficient quality specimens. PD-L1 was positive in 19 of 27 patients (70%), although only 26% had MPS > 5 (Tables 3 and 4). TILs penetrated tumor nests to varying degrees in all 27 tumors, with 30% of tumors having CD8⁺ lymphocytes present to a high degree (3+; Table 4). Analysis of other cellular constituents of the tumor immune microenvironment is underway.

Acknowledging small cohort sizes, we found that TMB was significantly associated with response (P = 0.01, Table 3), where the two patients with ≥10 mt/MB of 13 patients with known TMB status and evaluable for response were more likely to achieve PR compared with patients with lower TMB (100% vs. 0%). The degree of tumor infiltration by lymphocytes was not a statistically significant response predictor (P = 0.13, Table 3). Among commonly mutated gene drivers of SBA carcinogenesis and metastasis, including KRAS, BRAF, APC, and TP53, only TP53 mutation status was statistically significantly associated with tumor response to pembrolizumab (P = 0.04, Table 3), although the small PR cohort and the bimodal distribution seen with TP53 mutation in the PD cohort suggests this is possibly due to chance. The associations of response with TILs, MPS, and TMB are further depicted in Supplementary Fig. S4A–S4C, respectively. MSI-H patients generally had higher TIL colocalization and were also more likely to respond with a PR (Supplementary Fig. S4A). There were no significant associations with survival stratified by genetic and immunologic phenotypes (data not shown) though a trend toward significantly worse PFS was seen in patients with the BRAF mutation compared with wild-type (P = 0.0717). Again, cohort sizes were limited in this exploratory analysis.

This trial represents the largest study conducted to date for SBA and is the first study of immune checkpoint inhibition in this disease. Although open at only six sites and with a low predicted annual incidence in the United States across all stages, this trial accrued rapidly and suggests that future SBA clinical research with larger randomized cohorts are feasible.

While the overall results of the ZEBRA trial do not support universal treatment with pembrolizumab in advanced unselected patients with SBA, some cohorts appear to have promising suggestion of activity. Among these are MSS tumors with TMB ≥10 mt/Mb and dMMR/MSI-H tumors. Clinical trials of PD-1 checkpoint inhibitors in other advanced gastrointestinal malignancies indicate an ORR of 31% to 44% in MSI-H colorectal cancer (22, 23). Recently, data from the dMMR SBA cohort of KEYNOTE-158 found an ORR of 42.1% in 19 patients with three achieving CR and an additional five achieving PR (24). As expected, the patients with MSI-H status in this study had a comparable response rate (50%). Although MSI-H tumors were low frequency in this study of unselected patients with SBA, it did reflect the expected prevalence (10%) among the 36 patients with confirmed MMR status and an additional four patients with unknown MSI status. As durable responses were seen in half of our patients with dMMR/MSI-H with both achieving the maximum number of cycles given, activity is confirmed in the MSI-H/dMMR SBA patient population. This supports the current pembrolizumab recommendation for MSI-H/dMMR SBA. Further exploration is warranted to determine factors that may predict nonresponse in this population.

The activity seen in some patients with MSS tumors with advanced SBA was of great interest. In upper gastrointestinal cancers, immune checkpoint inhibitors have shown response in only 11.6% in later-line therapy (25). The confirmed ORR in our study was disappointing and below that seen in GEJ tumors, but about equal when the unconfirmed responder is considered. While the patient with the unconfirmed response had loss of response at the second imaging timepoint, the confirmed responder exhibited a durable response on that order of that seen in the MSI-H population, and recently had surgical excision of a residual liver lesion that responded but partially persisted 2+ years after starting therapy. The pathologic examination showed only necrotic tissue without viable tumor. Further biomarker analysis of the confirmed and unconfirmed responders’ tumors both showed TMB >10 mt/MB. None of the nonresponders had TMB above this threshold, suggesting possible association with pembrolizumab response in this exploratory analysis. This finding was independently reached prior to the recent tumor-agnostic FDA approval of pembrolizumab for where TMB ≥10 mt/MB, which was approved on the basis of data from non-SBA tumor histologies on KEYNOTE-158 (26). Although we independently identified the 10 mt/MB prediction threshold with the Tempus assay, it is important to note that this is not the FDA-approved companion diagnostic platform (Foundation Medicine) and has variances in mutation calling within the computed TMB compared with the approved assay. This necessitates further validation when applying pembrolizumab approval to TMB calling on other platforms (20, 27). Previously, a retrospective analysis of an NGS database (Foundation Medicine; ref. 11) revealed that, across 317 SBA clinical assays, 5.8% of all patients with MSS tumors had a TMB ≥10 mutations/Mb, confirming the TMB-high incidence in our study population matches that seen across larger cohorts and suggesting a not insignificant population of MSS SBA patients who may benefit from pembrolizumab therapy. Thus, our study supports the approved indication of pembrolizumab for TMB-high MSS SBA, although further studies to confirm these findings are warranted. This also highlights the importance of NGS testing in all patients with advanced SBA.

Interestingly, although the cohort is small, none of the responders had mutated APC, raising the question of whether the APC pathway itself may explain some of the immunotherapy resistance seen in colorectal cancer, and, if so, perhaps this may apply to APC-mutant SBA. APC is mutated in approximately 26.8% of patients with SBA (vs. 75.9% in colorectal cancer; ref. 10). Accordingly, the presence or absence of an APC mutation may provide future biologic rationale for targeting specific subpopulations in SBA and warrants further investigation.

Unfortunately, exploratory analyses of additional more common somatic mutations did not discover correlations predictive of pembrolizumab response. Again, all molecularly altered subgroups were small in absolute number, so we have provided individual patient-level detail for future analyses. While PD-L1 has been a partially predictive biomarker among patients with GEC, our analyses of MPS suggest no association with response at any level of expression and should not be utilized in selecting pembrolizumab for patients with MSS/pMMR SBA at this time outside of clinical trial. Although the patients in this study were not selected for genomic or immunologic response predictors, when one considers the DCR 38% among patients with confirmed MSS/MSI-L and the finding that nine of 32 (28%) of these patients remained on treatment for ≥5 months with some responses as durable as patients with dMMR/MSI-H status, further research is needed to identify candidates who may clinically benefit from immunotherapy.

Stromal characterization of tumors revealed that all patients with evaluable tissue had some degree of T-lymphocyte cells interspersed within the tumor. Surprisingly, there was not a correlation with response to pembrolizumab, although a number of tumors had higher CD8⁺ TIL volume. Our prior studies of PD-L1 expression and tumor microenvironment characterization found that PD-L1 co-localizes in the stroma with macrophages, not lymphocytes (28), suggesting that PD-L1 could prevent TIL-mediated cytotoxicity via microenvironment suppression. As only four of 40 (10%) patients evaluable for response achieved confirmed or unconfirmed PR, in spite of TIL presence and PD-L1 expression suggestive of a microenvironment primed for pembrolizumab-induced tumor cell death, it is perhaps the stroma-based PD-L1 expression that accounts more for ongoing immunosuppression and lack of response. Further correlative studies are underway to investigate this hypothesis and to determine immuno-oncologic combinations to improve outcomes with pembrolizumab in SBA.

Limitations of this study, in addition to those noted above, include the lack of centralized response determination, as well as the lack of prospective randomization to a comparative control arm. In addition, at the time of trial design, the only prospective studies in advanced SBA were all in the first-line setting; thus, there was no standard of care beyond that extrapolated from the treatment of colorectal cancer. As there is only one other prospective clinical trial in a small cohort of patients (n = 10) to report successful outcomes in a refractory SBA population, it is difficult to determine whether survival benefit was enhanced for patients who received pembrolizumab in this study versus historic comparators.

In this first immunotherapy trial for patients with advanced SBA, pembrolizumab is generally safe and is active in patients with MSI-H/dMMR tumors. Although it did not meet the primary response endpoint in all-comers, activity was seen among certain patients with MSS tumors. Elevated TMB appears to be a promising predictive factor for pembrolizumab response in MSS-SBA.

K.S. Pedersen reports grants, nonfinancial support, and other support from ACCRU/Merck during the conduct of the study; grants, personal fees, nonfinancial support, and other support from Array/Pfizer; other support from Beigene; grants and other support from Nouscom; grants from AbbVie, BioLineRx, Boston Biomedical/Sumitomo Dainippon, BMS, Daiichi Sankyo, Ipsen, MedImmune, Novartis, Pierre Fabre, Rafael, Roche/Genentech, Natera; other support from NCCN outside the submitted work; and Medscape (personal reimbursement for CME videos recorded) and UpToDate (royalties for section entitled diagnosis of small intestinal cancers). M.J. Overman reports grants and personal fees from Merck during the conduct of the study. P.M. Boland reports grants and personal fees from Merck and Ipsen, grants from Boerhinger Ingelheim, Taiho, Macrogenics; and personal fees from Bayer, Nouscom outside the submitted work. S.S. Kim reports other support from Merck and other support from GSK outside the submitted work. T. Bekaii-Saab reports other support from Merck outside the submitted work. R.R. McWilliams reports grants from Merck during the conduct of the study, personal fees from Zeno Pharmaceuticals, and personal fees from NewLink Genetics outside the submitted work. No disclosures were reported by the other authors.

K.S. Pedersen: Data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. N.R. Foster: Data curation, Formal analysis, writing–original draft, writing–review and editing. M.J. Overman: Conceptualization, data curation, investigation, writing–original draft, writing–review and editing. P.M. Boland: Investigation, writing–review and editing. S.S. Kim: Investigation, writing–review and editing. K.A. Arrambide: Investigation, writing–review and editing. B.L. Jaszewski: Data curation, formal analysis, investigation, writing–review and editing. T. Bekaii-Saab: Investigation, writing–review and editing. R.P. Graham: Formal analysis, investigation, writing–review and editing. J. Welch: Supervision, investigation, writing–review and editing. R.H. Wilson: Conceptualization, writing–review, and editing. R.R. McWilliams: Conceptualization, resources, formal analysis, supervision, investigation, writing–original draft, writing–review and editing.

The authors wish to acknowledge the International Rare Cancers Initiative (IRCI) and Academic and Community Cancer Research United (ACCRU) for successful completion of this trial and Mike Thompson and Wendy Nevala for their assistance with biospecimens. Financial support: Primary study funding and investigational product was provided by the Merck Investigator Studies Program (to R.R. McWilliams), and additional site-specific funding was provided by the Kavanagh Family Foundation (to M.J. Overman) and Kevin T. Doner Memorial Fund (to M.J. Overman).

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.