Patient characteristics

Between September 2020 and April 2023, a total of 51 patients were enrolled, comprising 27 patients in the BTC cohort and 24 patients in the PDAC cohort, all of whom received the combination treatment of SHR-1701 and famitinib (Fig. 1a). At the data cutoff date of May 1, 2024, 45 patients (25 in the BTC cohort and 20 in the PDAC cohort) were assessable for response evaluation, while all 51 patients were included in the safety analysis (Fig. 1b). 2 patients in the BTC cohort and 4 patients in the PDAC cohort were excluded from the response assessment population due to early discontinuation of treatment before the first scheduled post-baseline imaging assessment. Treatment discontinuation occurred in 46 patients (90.2%), primarily due to disease progression. Advanced events (AE) led to treatment discontinuation in four patients (7.8%). Five patients were still receiving active treatment, with two patients continuing on both study drugs and three patients receiving famitinib monotherapy (Supplementary Fig. S1a and S1b).

Fig. 1

Study design and clinical outcomes of SHR-1701 plus famitinib in advanced BTC and PDAC. a Upper: Schematic representation of the bifunctional fusion protein SHR-1701. Lower: Treatment schedule of (1) SHR-1701 and (2) famitinib in the study. Q3w, every three weeks; i.v., intravenously; p.o., orally. b CONSORT (Consolidated Standards of Reporting Trials) diagram of patient disposition in the biliary tract cancer (BTC) and pancreatic ductal adenocarcinoma (PDAC) cohorts. Six patients were not evaluable for tumor response due to an early death (n = 3) or withdrawal of consent (n = 3) before the first scheduled post-baseline imaging assessment. c Waterfall plot of the best percent change from baseline in target lesion size and the best response for evaluable patients in the BTC and PDAC cohorts. *, tumor growth < 5 mm, still classified as stable disease per RECIST criteria. d Kaplan-Meier curve of progression-free survival (PFS) for the BTC cohort. Median PFS: 5.1 months (95% CI, 2.6–7.6). CI, confidence interval. e Kaplan-Meier curve of PFS for the PDAC cohort. Median PFS: 2.1 months (95% CI, 0.7-3.5). f Kaplan-Meier curve of overall survival (OS) for the BTC cohort. Median OS: 16.0 months (95% CI, 6.1-NE). NE, not estimable. g Kaplan-Meier curve of OS for the PDAC cohort. Median OS: 5.3 months (95% CI, 4.0–6.5)

Baseline characteristics were summarized in Table 1. In the overall population, the median age was 61 years, 52.9% of patients were male, and all patients had an Eastern Cooperative Oncology Group (ECOG) performance status of 1. In the BTC cohort, most patients had metastatic cancer in the liver (22/27, 81.5%), lungs (10/27, 37%), peritoneum (8/27, 29.6%), and distant lymph nodes (8/27, 29.6%). And the median number of prior systemic treatments at the start of therapy was 1. Fourteen patients (51.9%) had intrahepatic cholangiocarcinoma, 5 (18.5%) had extrahepatic cholangiocarcinoma, and 8 (29.6%) had gallbladder cancer. In the PDAC cohort, 75% (18/24) of patients experienced liver metastasis, 62.5% (15/24) experienced lung metastasis, and 54.2% (13/24) experienced peritoneal metastasis. And participants had received a median of two prior systemic treatments.

Table 1 Baseline demographic and clinical characteristicsEfficacy

In the BTC cohort, 25 patients were evaluable for response, and 7 patients (28.0%; 95% confidence interval [CI]: 12.1-49.4) achieved an objective response, including 2 complete responses (CR) and 5 partial responses (PR). In the PDAC cohort, 20 patients were evaluable for response, and 3 patients (15.0%; 95% CI: 3.2-37.9) achieved an objective response, including 2 CRs and one PR (Table 2). The disease control rate (DCR) was 80.0% (95% CI: 59.3-93.2) and 60.0% (95% CI: 36.1-80.9) for the BTC and PDAC cohorts, respectively. Figure 1c illustrated the best percent of change from baseline in the target lesion size and the best response for evaluable patients in the two cohorts.

Table 2 Efficacy outcomes

For the full analysis population, the median follow-up duration was 26.4 months (95% CI 19.6–33.2). Among BTC patients, the median duration of response (mDoR) was 7.6 months (95% CI: 2.2–13.0 months; Table 2 and Supplementary Fig. S1a) and the median PFS (mPFS) was 5.1 months (95% CI: 2.6–7.6 months; Fig. 1d). The median OS (mOS) was 16.0 months (95% CI: 6.1-not estimable [NE]; Fig. 1f), with a 1-year OS rate of 54.4% (95% CI: 38.3%–77.3%; Table 2) and a 2-year OS rate of 48.1% (95% CI: 31.5%–73.5%). Notably, seven patients (26.0%) in the BTC cohort received subsequent systemic treatment post-progression. For the PDAC group, among the three patients who achieved a response, one achieved PR with a DoR of 6.3 months, while the other two achieved CR with DoRs of 9.8 and 14 months, respectively (Supplementary Fig. S1b). However, it should be noted that the patient with a DoR of 14 months died due to an accidental fall from a building while working, unrelated to disease progression. The mPFS was 2.1 months (95% CI: 0.7-3.5 months; Fig. 1e), and the mOS was 5.3 months (95% CI: 4.0-6.5 months; Fig. 1g), with a 1-year OS rate of 39.6% (95% CI: 23.9%-65.6%; Table 2). 4 patients (16.7%) in the PDAC cohort received subsequent systematic treatment post-progression.

The CR rates of 8% in BTC and 10% in PDAC observed in our study were notable, considering the rarity of CR cases reported in previous trials of metastatic cancer.24,25 A comprehensive characterization of these exceptional responders was summarized in Supplementary Data S1. Representative magnetic resonance or computed tomography images of the CR cases were presented in Fig. 2a and Supplementary Fig. S1c, with the longest DoR exceeding 3 years (Supplementary Data S1).

Fig. 2

Safety profile and clinicopathologic parameters analysis of the study. a Representative magnetic resonance (MR) or computed tomography (CT) images of the complete response (CR) cases in our cohorts. Lesions are marked with horizontal lines indicating their longest target diameters. CT and MR images of CR patient 1023 with PDAC (the top row: metastatic lesion 1 in the peritoneum; the middle row: metastatic lesion 2 in the peritoneum; the bottom row: metastatic lesion in the lung). CT images of CR patient 2013 with BTC (metastatic lesion in the peritoneum). Detailed patient background information can be found in Supplementary Data S1. b Overview of treatment-related adverse events (TRAEs) by grade in organ systems with grade 3–4 events. c Forest plot of hazard ratios (HR) and 95% confidence intervals (CI) for overall survival (OS) based on various clinicopathological factors in the overall population. Factors favoring OS are shown on the right side of the plot, while factors against OS are shown on the left side. The category of local treatment includes ablation, radiation and intra-arterial therapy. **, P < 0.01; SOD, sum of diameter; NLR, neutrophil-to-lymphocyte ratio. d Top: comparison of peripheral blood mononuclear cell (PBMC) cluster proportions before treatment (pre-treatment) and 6 weeks post-treatment. N = 37 evaluable patients. Statistical significance was determined by Bonferroni’s multiple comparison test. *P < 0.05; **P < 0.01; ***P < 0.001. Bottom: comparison of CD4+CD25+CD127low, CD3-CD16+CD56+, CD3+HLA-DR+ and CD3+HLA-DR- cell cluster proportion pre- and post-treatment split by clinical response. N = 8 clinical responders (R), N = 29 clinical non-responders (NR). Statistical significance was determined using the Mann-Whitney U test

Safety

At the time of data cutoff, the median duration of exposure was 2.4 months (range, 0.1–41.0 months) for the study treatment (SHR-1701: median, 1.5 months; range, 0.03–26.4 months; famitinib: median, 2.1 months; range, 0.1–41.0 months). The majority of patients (88.2%) experienced at least one treatment-related adverse event (TRAE). The most common any-grade TRAEs were proteinuria (47.1%), anemia (39.2%), and positive urinary occult blood (31.4%; Supplementary Table S1). Grade 3 or 4 TRAEs occurred in 15 patients (29.4%; Fig. 2b), with the most common being anemia (13.7%) and hypertension (7.8%). In addition, 16 cases (31.4%) experienced potential immune-related adverse events, with the most common being rash (23.5%) and hypothyroidism (13.7%; Supplementary Table S2). Among them, 2 cases (3.9%) were grade 3-4. A total of 8 subjects received immunosuppressive therapy, with 4 cases of topical corticosteroids for skin toxicity. No grade 5 TRAEs were reported.

TRAEs led to treatment interruption and treatment discontinuation of SHR-1701 in 7 (13.7%) and 8 (15.7%) patients, respectively. Regarding famitinib, TRAEs resulted in 17 (33.3%) treatment interruption, 11 (21.6%) dose reduction, and 5 (9.8%) treatment discontinuation cases. Overall, 4 patients (7.8%) discontinued all treatment due to TRAEs.

Comprehensive clinicopathologic features associated with treatment benefit

Multivariate analysis demonstrated that in the overall population, patients who had previously undergone primary tumor resection exhibited longer OS (HR = 0.11, 95% CI = 0.02–0.45; Fig. 2c and Supplementary Fig. S1d), while a higher number of lines of previous systemic treatment was identified as a risk factor for OS (HR = 2.4, 95% CI = 1.29–4.48; Fig. 2c). Additionally, distant lymph node metastases and neutrophil-to-lymphocyte ratio (NLR) were identified as risk factors for PFS (Supplementary Fig. S1e and S1f).

The presence or absence of common metastases, including liver, lung, peritoneum, and distant lymph nodes, did not show a significant correlation with treatment response (Supplementary Fig. S1g). Similarly, PD-L1 expression (Supplementary Fig. S1h) and microsatellite stability (MSS) status (Supplementary Fig. S1i) did not correlate with treatment response in our cohort. However, baseline carcinoembryonic antigen (CEA) levels were significantly higher in BTC non-responders (Supplementary Fig. S1j), and baseline carbohydrate antigen 19-9 (CA 19-9) levels were higher in PDAC non-responders (Supplementary Fig. S1k).

In the post hoc analysis, we investigated the association between peripheral blood immunophenotype and clinical outcome by collecting peripheral blood mononuclear cells (PBMCs) from participants before treatment and 6 weeks post-treatment. Our analysis identified revealed notable changes in several cell subtypes, implying that the combination of SHR-1701 and famitinib treatment markedly improved the tumor microenvironment (Fig. 2d). The decrease in CD4+CD25+CD127low regulatory T cells (Tregs) signified a decrease in immunosuppressive function,26 while the increase in CD3-CD16+CD56+ NK cells suggested an increase in cytotoxic activity against tumors.27 Additionally, the increase in CD3+HLA-DR+ activated T cells and the decrease in CD3+HLA-DR- cells (Fig. 2d) suggested a shift towards a more activated T cell subset.28 Interestingly, these consistent alterations were predominantly observed in responders, implying a more favorable immune microenvironment for anti-tumor responses, whereas non-responders exhibited varied changes. These findings suggested that changes in peripheral blood immunology within 6 weeks after treatment may be indicative of treatment efficacy and could potentially serve as a predictive biomarker in the context of SHR-1701 plus famitinib therapy, warranting further validation in larger clinical trials.

Distinct immune and metabolic profiles associated with treatment response

RNA-seq data of baseline primary tumor samples revealed distinct gene expression patterns between BTC and PDAC patients (Supplementary Fig. S2a). Despite cancer type-specific mechanisms, differentially expressed genes (Supplementary Fig. S2b) and pathway enrichment analysis identified common features related to treatment efficacy in both responders and non-responders (Fig. 3a; Supplementary Data S2). Responders exhibited upregulation of immune-related pathways, including leukocyte adhesion, migration, and activation, as well as T cell differentiation and activation. In contrast, non-responders showed enrichment of metabolic pathways, such as amino acid metabolism and catabolism, implying the potential role of heightened metabolic activity in treatment resistance.

Fig. 3

Immune and metabolic profiles associated with treatment response. a Pathway enrichment analysis of differentially expressed genes between responders and non-responders in BTC (left) and PDAC (right) cohorts. Red for upregulated pathways in responders and blue for downregulated ones. b Comparison of the estimated abundance of immune cell types in the tumor microenvironment (TME) between responders (R) and non-responders (NR) using xCell deconvolution analysis. Statistical significance was determined using the Mann-Whitney U test. c Heatmap displaying the enrichment scores of 12 gene set signatures in responders (R) and nonresponders (NR) across BTC and PDAC cohorts. The signatures were categorized into three main groups: immune cell-related signatures, antigen presentation-related signatures, and cytokine-related signatures. Statistical significance was determined using the Mann-Whitney U test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease. d Bubble plot showing upregulated metabolic pathways in non-responders (NR) compared to responders (R) in BTC (red) and PDAC (blue) cohorts. Bubble size represented the normalized enrichment score (NES). N.S. (gray) denoted non-significant (P value ≥ 0.05) pathways. e Correlation matrix showing the interplay between NR-high metabolic pathways and immune signatures. Color scale represented Spearman correlation coefficient. Statistical significance was determined using the Mann-Whitney U test. *, P < 0.05, **, P < 0.01, ***, P < 0.001. f Scatter plot illustrating the negative correlation between the peroxisomal lipid metabolism pathway score and the CD8+ T cell effector signature score in non-responders (NR, blue) and responders (R, red). Spearman correlation coefficient (r) = -0.735, P = 7.7e-04. g Scatter plot demonstrating the negative correlation between the insulin secretion pathway score and the MHC class I signature score in non-responders (NR, blue) and responders (R, red). Spearman correlation coefficient (r) = -0.696, P = 1.9e-03

To elucidate differences in immune cell recruitment and activation, we estimated the abundance of diverse immune cell types in the TME using deconvolution analysis and scored 12 gene set signatures previously reported to be associated with immunotherapy efficacy for each patient (Supplementary Data S3). Our analysis revealed a higher level of overall immune cell infiltration in the responder group (P = 0.015), particularly in the proportions of CD8+ T cells (P = 0.03), activated dendritic cells (P = 0.006), and class-switched memory B cells (P = 0.015; Fig. 3b and Supplementary Fig. S2c). Moreover, responders harbored significantly elevated scores in most signatures related to immune cell activation, antigen presentation, and cytokine activity (Fig. 3c).

Counterintuitively, we observed elevated levels of TGFB1 expression and TGF-β signaling in responders of our cohort (Supplementary Fig. S2d), despite their usual association with poor responses as reported. Moreover, genes and pathways related to angiogenesis and hypoxia were also upregulated in responders (Supplementary Fig. S2e-f), further supporting the notion that SHR-1701 plus famitinib might help overcome these potentially immunosuppressive factors, thereby enhancing immunotherapy efficacy.29,30

Metabolic reprogramming in treatment resistance and its interplay with the immune landscape

Considering the complexity of tumor metabolic networks, we employed gene set enrichment analysis (GSEA) to comprehensively profile the upregulated metabolic pathways in patients with poor response (Supplementary Data S4). Certain pathways consistently elevated in non-responders across both cancer types were involved in the metabolism of amino acids (glycine and arginine, etc.), lipids (such as butanoate, propanoate, and steroid biosynthesis), and vitamins, reflecting a broad enhancement of metabolic processes in treatment-resistant patients (Fig. 3d and Supplementary Fig. S2g-h). Additionally, we also observed cancer type-specific preferences in upregulated metabolic modules. In BTC non-responders, the predominantly upregulated pathways were related to drug metabolism, fatty acid degradation, and bile acid metabolism, suggesting a focus on detoxification and lipid handling; in contrast, PDAC-specific active processes were associated with carbohydrate metabolism and pancreatic functions, underlining the unique metabolic demands of pancreatic tumors.

To uncover the interplay between metabolic reprogramming and the immune landscape, we calculated scores for the top non-responder-enriched (NR-high) metabolic pathways and analyzed their correlation with immune signatures. Remarkably, several commonly upregulated metabolic pathways, such as peroxisomal lipid metabolism, propanoate metabolism, and the metabolism of vitamins and cofactors, exhibited strong negative correlations with CD8+ T cell effector and MHC class I signatures, suggesting that their upregulation might shape an immunosuppressive microenvironment, thereby hindering patients’ responses to immune-based therapy like SHR-1701 (Fig. 3e, f). Interestingly, as an integral part related to cellular metabolism, insulin secretion pathways also displayed negative correlations with immune signatures (Fig. 3g).

Immune/metabolism score as a potential predictive biomarker for immunotherapy response

Next, based on selected immune-metabolic features most strongly associated with treatment response, we stratified patients into two subgroups via clustering algorithms: ‘immune-dominant’ (ID) and ‘metabolic-dominant’ (MD). The ID subgroup showed higher levels of immune cell infiltration, activated immune-related phenotypes, and lower metabolic activity (Fig. 4a), whereas the MD subgroup displayed an enriched metabolic pattern with immune dysfunction. Notably, patients of the ID subtype demonstrated better clinical response rates (Fig. 4b) in our cohort. Applying the same strategy to independent, published immunotherapy cohorts resulted in a similar stratification pattern, with ID subtype patients experiencing greater clinical benefits than MD subtype patients (Supplementary Fig. S3a-c).

Fig. 4

Immune/metabolism score as a potential predictive biomarker for immunotherapy response. a Heatmap showing the ssGSEA scores of immune signatures and metabolic pathways in patients stratified into ‘immune-dominant’ (left, pink) and ‘metabolic-dominant’ (right, purple) subgroups. The color scale represents the normalized expression values. b Proportion of patients in the immune-dominant (ID) and metabolic-dominant (MD) subgroups within the non-responder (NR) and responder (R) groups. c Heatmap displaying the expression levels of the six genes used to calculate the immune/metabolism (I/M) score in responders and non-responders. d Representative mIHC images of tumor samples from patients with different BOR in BTC and PDAC patients. Color-coded for immune cell populations was as illustrated, with the TLS region circled by a dashed line. The scale bar at the bottom left corner of the mIHC images represents 0.1 mm. TLS, tertiary lymphoid structure; SD, stable disease; PD, progressive disease. e Comparison of I/M scores between non-responders (NR) and responders (R; P = 0.0079, Mann-Whitney U test). f Kaplan-Meier curves for overall survival (OS) stratified by I/M score group (high vs. low), with median OS and P value (0.0252, log-rank test). The number at risk table was included. g External validation cohort 1 (n = 289, IMvigor210): I/M scores in R vs. NR (left; P = 2.8e-08, Mann-Whitney U test), and OS stratified by I/M score group (high vs. low) with median OS, P value (2.55e-08, log-rank test), and number at risk table (right). mUC, metastatic urothelial carcinoma. h Scatter plot showing the correlation between I/M score and tumor regression change in non-responders (NR, blue) and responders (R, red). Spearman correlation coefficient (r) = 0.592, with a P value of 1.2e-02. i Receiver operating characteristic (ROC) curve for the response prediction model using the I/M score. The plot showed the performance of the model in the test cohorts (SHR-1701 plus famitinib, AUC = 0.930; mUC-anti-PD-L1, AUC = 0.727) and three external validation cohorts. AUC area under the curve, mUC metastatic urothelial carcinoma, NSCLC non-small cell lung cancer, STAD stomach adenocarcinoma

To enhance clinical practicality, we developed an immune/metabolism (I/M) score to predict immunotherapy efficacy, relying on the pretreatment expression of only six genes (Fig. 4c; Supplementary Fig. S4a). Of particular interest, CXCL13, CXCL9, and IFNG were key genes involved in the tertiary lymphoid structures (TLS) formation, consistent with the increased proportions of dendritic cells and B cells in responders as previously mentioned (Fig. 3b). Multiplex immunofluorescence (mIHC) further confirmed the presence of TLS in responders, while patients with stable diseases (SD) had mainly myeloid cell infiltration, and progression disease (PD) patients displayed an immunologically “cold” tumor state (Fig. 4d and Supplementary Fig. S4b). These findings revealed a contribution of TLS to BTC/PDAC immunotherapy and indicated the biological TME basis of the I/M score.

A markedly higher level of I/M score was observed in responders versus nonresponders (Fig. 4e). Moreover, patients with higher scores had remarkably longer OS (Fig. 4f) and PFS (Supplementary Fig. S4c), which positively correlated with pathological tumor regression change as well (Fig. 4h). Validation of the I/M score in various advanced cancer types treated with PD-(L)1 inhibitors confirmed that higher scores were consistently associated with better treatment response and prolonged OS (Fig. 4g and Supplementary Fig. S4d-f).

Leveraging machine learning (Supplementary Fig. S4g), I/M scores showed promising predictive power in both our cohort (area under the receiver operating characteristic curve [AUC] of 0.930) and multiple external cohorts of pan-cancer (Fig. 4i), supporting its utility as a potential effective immunotherapy biomarker.