This study was a single-center, single-arm, open-label, phase Ⅱ trial in which the key inclusion criteria were aged 18–75 years, histologically confirmed unresectable or metastatic BTC, an Eastern Cooperative Oncology Group performance status of 0–2, an estimated life expectancy of at least 3 months, at least one radiographically measurable lesion, adequate organ function, and ability to understand and sign a written informed consent document. Previous chemotherapy, radiotherapy, or other local ablative therapies must have been completed over 4 weeks before enrollment and patients must have shown radiologically confirmed disease progression. The key exclusion criteria included active, known or suspected autoimmune disease, known brain metastasis or active central nervous system disease, being treated with either corticosteroids (>10 mg daily prednisone equivalent) or other immunosuppressive medications within 14 days of enrollment, and previous treatment with anti-PD-1/PD-L1 antibodies. Details of the inclusion and exclusion criteria are presented in the study protocol (online supplementary file 1). Eligible patients were assigned to cohort A (resistant to gemcitabine-based or cisplatin-based chemotherapy) and cohort B (chemotherapy-naive) based on their previous systemic therapies.

Written informed consent based on Declaration of Helsinki principles was provided by patients or their representatives before study entry.

All enrolled patients in both cohort A and cohort B were administered the combination therapy consisting of gemcitabine 1000 mg/m² on day 1 and day 5, cisplatin 75 mg/m² on day 1, and nivolumab 3 mg/kg on day 3 infused intravenously every 3 weeks for up to six cycles. Afterwards, patients with responsive or stable disease (SD) switched to maintenance therapy in which nivolumab and gemcitabine were administered every 6–8 weeks until disease progression, intolerable toxicity, death, withdrawal of consent, or any other reasons. Dose reductions were permitted according to the protocol. Adverse events were graded in accordance with National Cancer Institute Common Terminology Criteria for Adverse Events version 5.0, and the causal association with study drugs was determined by investigators. Tumor responses were assessed every two cycles by site investigators according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1.16 Positron emission tomography-CT was mandated to confirm response evaluation if targeted tumors were assessed by CT scans with contrast or MRI as showing a complete response (CR). Patients achieving PR or progressive disease (PD) were advised to undergo on-therapy site-matched tumor biopsy. Tumor cell PD-L1 expression was assessed on either archival or fresh pretreatment biopsy samples by immunohistochemistry using the Dako 22C3 pharmDx assay (Dako North America, Carpinteria, California, USA). Positive tumor PD-L1 expression was defined as at least 1% of tumor cells being membrane stained at any intensity in a section that contained at least 100 evaluable tumor cells.

Genomic DNA was isolated from tumor biopsies and matched peripheral blood mononuclear cell samples using the GeneRead DNA FFPE Kit. All sample capture libraries were prepared using the Agilent SureSelect Human All Exon V6 Kit (Agilent Technologies, Santa Clara, California, USA) according to the manufacturer’s instructions. Libraries were sequenced on an Illumina HiSeq 6000 platform. Primary sequence data were processed by filtering adaptor sequences and removing low-quality reads, which were defined as those with a >10% N rate and/or with >10% bases with a quality score of <20 using SOAPnuke (V.1.5.6). The clean reads were mapped to hg19 using BWA-mem (V.0.7.12). Single nucleotide variants and small insertions and deletions (indels) were detected using VarScan (V.2.4.1). The mutations were further filtered using inhouse software to remove false positive mutations. Tumor mutational burden (TMB) was determined by analyzing non-silent somatic mutations, including coding base substitution and indels per megabase. PyClone was employed to detect subclones and calculate the cancer cell fraction. The ratio of these subclones to all mutations was interpreted as intratumor heterogeneity. Microsatellite instability (MSI) detection was performed by interrogating 344 available genomic microsatellites using an MSI sensor. The percentage of unstable sites was reported as the MSI sensor score. Human lymphocyte antigen-I (HLA-I) typing of tumors and adjacent normal samples was performed using Polysolver (V.1.0). All non-silent mutations were translated into 21-mer peptide sequences. Then, 9-mer to 11-mer peptide sequences were extracted using a sliding window. NetMHCpan (V.3.0) was used to predict the major histocompatibility complex (MHC) class I binding affinity of peptides with the patient-specific HLA alleles. The predicted peptides were selected and ranked by inhouse software. Peptides with scores higher than 0 were selected. Tumor neoantigen burden (TNB) was measured as the number of such peptides per megabase. In the neoantigen fitness model, we calculated the neoantigen recognition potential for each neoantigen using a recently developed method.17

Peripheral blood samples were collected every cycle prior to the infusion of study drugs to test the concentration level of cytokines, including interleukin (IL)-1β, IL-2, IL-4, IL-6, IL-8 (CXCL8), IL-10, IL-12p70, IL-17A, IL-18, IL-23, IL-33, interferon (IFN)-α2, IFN-γ, tumor necrosis factor (TNF)-α, soluble Fas, soluble FasL (sFasL), granzyme A, granzyme B, perforin, granulysin, and monocyte chemotactic protein-1 (MCP-1) C-C motif chemokine ligand 2 (CCL2), using BioLegend LEGENDplex bead-based immunoassays, the LEGENDplex Human Inflammation Panel (Cat: 740118), and the Human CD8/NK Panel (Cat: 740267).

The primary objective of this study was to assess the objective response rate (ORR) for nivolumab plus gemcitabine and cisplatin combination therapy. The secondary objectives included determining the frequency and severity of adverse events occurring up to 120 days after the last dose of study drugs, disease control rate (DCR), progression-free survival (PFS), PFS at 6 months, overall survival (OS), and OS at 12 months. ORR was defined as the proportion of all treated patients with either a confirmed CR or PR per RECIST version 1.1. PFS was defined as the time from the first dose to the first documented disease progression or to death from any cause. OS was defined as the time from the first dose to death from any cause. The exploratory objective was to assess pathological, immunological or clinical predictive biomarkers for response and prognosis.

The A’Hern single-stage phase II study design was used to determine the sample size of this clinical trial of nivolumab plus gemcitabine and cisplatin chemotherapy. According to previously reported data, the ORR of gemcitabine and cisplatin chemotherapy for patients with advanced BTCs is up to 26%. Based on this, we set the null hypothesis to be that 26% or fewer patients would have an objective response versus the alternative hypothesis that it was 55% or higher. At least 25 patients would need to be enrolled with a two-sided significance level of 0.05% and 90% power. Safety analysis was performed in patients who received at least one dose of nivolumab in combination with gemcitabine and cisplatin, and efficacy analysis was performed in patients who underwent one or more post-treatment scans. The proportions of patients with an objective response and adverse events were summarized by descriptive statistics with Wilson 95% CIs. Response differences among clinical subgroups were assessed with Fisher’s exact test. For PFS, patients without disease progression were censored at the time of last radiological imaging. For OS, patients still surviving were censored at the time of data cut-off. Survival was analyzed using the Kaplan-Meier method and compared using the log-rank test. Immune biomarker changes were detected by paired t-tests between pretreatment and post-treatment, and differences among groups were evaluated by t-test or the Mann-Whitney U test. All statistical analyses were completed using Stata/SE V.15.1.

Between November 16, 2017 and December 31, 2018, 32 eligible patients with advanced unresectable or metastatic BTCs were enrolled, of whom 7 patients were resistant to gemcitabine-based or cisplatin-based chemotherapy and 25 patients were chemotherapy-naive (figure 1). All enrolled patients, including 1 (3%) patient with regional unresectable disease, 6 (19%) with metastatic disease, and 25 (78%) with recurrent disease (defined as patients who had regionally relapsed disease or distant metastases after complete resection or locoregional or systemic therapies), were administered at least one cycle of nivolumab plus gemcitabine and cisplatin combination therapy (table 1). Patients who did not meet the inclusion criteria or were participating in other trials were excluded (n=9). At the time of data cut-off (September 31, 2019), all patients in cohort A and cohort B were eligible for safety analyses, of whom 6 in cohort A and 21 in cohort B were qualified for clinical activity analyses; 5 patients discontinued treatment within the first cycle due to rapidly deteriorating tumor-related complications (1 from cohort A and 3 from cohort B) or adverse events unrelated to study drugs (1 from cohort B). The detailed baseline demographics and characteristics of all enrolled patients are summarized in table 1. The median age was 60 years (range 27–69). Fourteen (44%) patients had target lesions larger than or equal to 5 cm. Liver metastases were detected in 28 (88%) patients, while abdominal lymphatic metastases were detected in 21 (66%). PD-L1 status was evaluable in 26 tumor samples (81%), of which 12 (37%) were positive for PD-L1 expression and 14 (44%) were negative.

Safety data from cohort A and cohort B were summarized and analyzed together. All 32 enrolled patients experienced at least one treatment-related adverse event. The most frequent adverse events were nausea (29 patients; 91%), neutropenia (26 patients; 81%), fatigue (21 patients; 66%), thrombocytopenia (20 patients; 62%), and anemia (19 patients; 59%) (table 2). The most common grade 3 or higher treatment-related adverse events were thrombocytopenia, reported in 18 (56%) patients, and neutropenia, reported in 7 (22%) patients. Other severe adverse events included elevated alanine aminotransferase (1 patient; 3%, grade 3), elevated aspartate aminotransferase (1 patient; 3%, grade 4), elevated lipase (1 patient; 3%, grade 3), hyponatremia (1 patient; 3%, grade 3), and hypertension (2 patients; 6%, grade 3). One (3%) patient had immune-related adverse events (rash, grade 1). There were no treatment-related deaths at the time of analysis.

After a median follow-up of 12.8 months (95% CI 10.8 to 14.8), 27 response-evaluable patients in cohort A and cohort B received a median of 4 cycles of nivolumab plus gemcitabine and cisplatin combination therapy (IQR, 3–6). Fifteen (55.6%) patients achieved a confirmed objective response, including 5 (18.6%) CRs and 10 (37%) PRs (table 3, figure 2A). Disease control was achieved in 25 (92.6%) patients, including 10 (37%) patients who had SD as their best response. The radiological changes of each response-evaluable patient are summarized in online supplementary file 2. In cohort A, six of seven patients who were resistant to gemcitabine-based or cisplatin-based regimens were response-evaluable, of whom one patient achieved CR and one patient obtained PR; the ORR and DCR were 33.3% and 83.3%, respectively. In cohort B, 13 of 21 (61.9%) chemotherapy-naive patients achieved CR or PR, and the proportion of patients with disease control was 95.2%. Responses were ongoing at the time of data cut-off in two patients with CR and one patient with PR (figure 2B,C). Analysis of 27 response-evaluable patients found that the PD-L1 expression level could not be used as a biomarker for predicting clinical response (p=0.395; online supplementary figure S1A). Whole-exome sequencing was performed on patients’ biopsied tumor samples and paired peripheral blood mononuclear cells, which were allocated to a responding group (CR+PR) and non-responding group (SD+PD) according to the clinical response. TMB and TNB were generally low in this study (online supplementary figure S2). However, the median value of TMB, TNB, and fitness was higher in the responding group than in the non-responding group, while the median value of heterogeneity was lower in the responding group; of these differences, fitness had statistical difference (p=0.041; figure 3A). Mutations of RYR2, MUC4, and APOB were detected only in samples from the responding group (figure 3B). We performed exploratory analysis to study the association between the activation of peripheral T cells and clinical antitumor activity. The evaluation of T cells in peripheral blood showed that the baseline percentage of CD3+ cells in responders was higher than that in non-responders (p=0.046; online supplementary figure S3A). The proportion of HLA-DR+ CD3+ cells in patients’ peripheral blood increased after the start of the combination therapy, especially in patients with an objective response (p=0.009). However, a statistical difference was not observed between responders and non-responders (online supplementary figure S3B,C). The association between changes in peripheral serum cytokines and chemokines at baseline (C1D0) and the day before the first dose of the third cycle (C3D0) and clinical response was also assessed. The concentrations of serum sFasL and granzyme A were higher in non-responders than in responders after two cycles of combination treatment (p=0.042 and p=0.048, respectively), while the concentrations of IL-2, IL-18, sFasL and CCL2 dropped significantly in responders compared with non-responders (p=0.036, p=0.047, p=0.012 and p=0.042, respectively; online supplementary figure S4A,B).

The median PFS in this study was 6.1 months (95% CI 3.4 to 8.2), and the proportions of patients who were progression-free at 6 months and 12 months were 51.9% (95% CI 31.9 to 68.6) and 18.5% (95% CI 6.8 to 34.8), respectively (figure 2D). A comparison between cohort A and cohort B showed that chemotherapy-naive patients could obtain longer median PFS than those who had received chemotherapy; however, there was no statistical difference. Further analysis found that patients who were administered more than four cycles of combination treatment had longer PFS than those who received four cycles or less (p=0.024), and PD-L1 expression status could not be established as a biomarker for predicting PFS (p=0.125; online supplementary figure S1B). We also analyzed the impact of TMB, TNB, and fitness on PFS in this study. However, there was no correlation between the above biomarkers and PFS (figure 3C, online supplementary figure S5). Analysis of peripheral serum cytokines found that patients whose concentration of IFN-γ decreased following combination therapy could obtain longer PFS than those whose concentration of IFN-γ increased or remained the same after treatment (p=0.033; figure 3D), and a similar association was found between the decrease in MCP-1 and PFS (p=0.019; figure 3D).

The median OS was 8.5 months (95% CI 5.0 to 12.5), and the 12-month OS rate and 18-month OS rate were 33.3% (95% CI 16.8 to 50.9) and 24.7% (95% CI 10.2 to 42.4), respectively (figure 2D). There was no statistical difference between the median OS of cohort A and that of cohort B. Four cycles or more of combination therapy was a parameter that could be correlated with longer OS (HR 0.595 (95% CI 0.398 to 0.89), p=0.012; online supplementary figure S1C), while a correlation between PD-L1 expression and OS was not established (p=0.499; online supplementary figure S1C). Whole-exome sequencing results showed that a 1.37 neoantigens/Mb cut-off value, as used for the TNB in this study, had prognostic value, and patients with TNB greater than 1.37 neoantigens/Mb had significantly longer OS than those with TNB of 1.37 neoantigens/Mb or fewer (p=0.048; figure 3C). Analysis of serum cytokines detected that the concentration of sFasL and IFN-γ dropped significantly in patients with longer OS (p=0.00076, p=0.032; figure 3E). The changes in granulysin, MCP-1, IL-17a, IL-23, TNF-α and granzyme B in serum following combination therapy had no statistical influence on OS (figure 3E; online supplementary figure S6).