We retrospectively screened patients from the Sun Yat-Sen University Cancer Center and the First Affiliated Hospital of Sun Yat-Sen University in China; these patients were diagnosed with advanced BTC between January 2014 and September 2020. The following inclusion criteria were used during screening: histopathologic diagnosis of ICC, ECC, or GBC; unresectable locally advanced or metastatic BTC; chemotherapy alone or combined with anti-PD1 regimen for more than two cycles; complete radiographic evaluation. The exclusion criteria were as follows: other primary malignant tumors in addition to BTC; radiotherapy before anti-PD1 treatment; uncontrolled intercurrent illness; surgical resection of metastatic lesions. A total of 198 patients were included after screening, and the study protocols were approved by the ethics committees of Sun Yat-Sen University Cancer Center (B2020-190-01) and the First Affiliated Hospital of Sun Yat-Sen University ([2022]069), with the exemption of informed consent. The study was conducted retrospectively without affecting the interests of the patients and adhered to the guidelines of the Declaration of Helsinki.

An IRB-approved protocol was followed to search the electronic medical records of the patients for the treatment options, clinical characteristics, and efficacy evaluations. Two oncologists examined the treatment efficacies in accordance with the Response Evaluation Criteria in Solid Tumors (RESIST) criteria (1.1) (Armato and Nowak, 2018). Any inconsistencies in the results were solved by consensus. The effects were evaluated by measuring the changes in the target lesions through imaging. The ORR, DCR, PFS, and OS values were used to assess the treatment efficacy. PFS refers to the time between receiving immunotherapy or chemotherapy and the earliest response between progressive disease or mortality; OS is defined as the time range from registration into the study to the date of death. A standard follow-up was conducted for all participants, and the total follow-up period was more than 24 months unless death occurred. The last follow-up visit for the participants was October 2022 or death.

The beginning and end dates of treatment, initial dose, and dose modifications were systematically collected. The anti-PD1 drugs included pembrolizumab, sintilimab, toripalimab, camrelizumab, and tirelizumab. The immunotherapy treatment regimen included administration of anti-PD1 in combination with an antiangiogenic or chemotherapy or transarterial chemoembolization or hepatic arterial infusion chemotherapy.

The ICC (GSE32225) and ECC (GSE132305) datasets were obtained from the Gene Expression Omnibus (GEO) database before being log normalized (log2x+1) and merged. The “limma” R package was used for differential expression analysis, where the log2Foldchange parameter was set to 0.5 and adjusted p-value was 0.05. The intersection of the differentially expressed genes and immune-related genes in the ImmPort dataset (https://www.immport.org/shared/genelists) was selected, and a volcano map was drawn to show the differentially expressed immune-related genes.

The stromal, immune, and ESTIMATE scores of the ICC and ECC samples in the two datasets were calculated using the ESTIMATE algorithm in the “estimate” R package (Yoshihara et al., 2013). The tumor immune dysfunction and exclusion (TIDE) (http://tide.dfci.harvard.edu/) algorithm was used to predict the potential ICB therapy responses (Jiang et al., 2018). TISIDB (http://cis.hku.hk/TISIDB) was used to compare the differences in the immune-related molecules, including immunostimulators, immunoinhibitors, major histocompatibility complex (MHC) molecules, chemokines, and chemokine receptors (Ru et al., 2019).

We used five algorithms to calculate the differences in the tumor-infiltrating immune cells between the GSE32225 and GSE132305 datasets. Xcell (http://xcell.ucsf.edu/) (Aran et al., 2017) and MCP-counter (http://github.com/ebecht/MCPcounter/) (Becht et al., 2016) were used as the quantification methods for the marker-gene-based tumor-infiltrating immune cells, while CIBERSORT (https://cibersort.stanford.edu/) (Newman et al., 2015), TIMER (https://cistrome.shinyapps.io/timer/) (Li et al., 2017), and quanTIseq (http://icbi.i-med.ac.at/software/quantiseq/doc/index.html) (Finotello et al., 2019) were used as the quantification methods for the tumor-infiltrating immune cells based on expression feature deconvolution of the cell mixture.

The cutoff date (October 2022) for our report was used to generate the summary of baseline characteristics and efficacy assessments. The Cox proportional hazards model and Kaplan–Meier method were used to estimate the PFS, OS, 6-month and 12-month OS (PFS), and hazard ratios (HRs); the log-rank test was then used for comparative analyses. The Chi-square and Fisher’s exact tests were used to compare the categorical variables, while the Student’s t-test was used for the continuous variables. A value of p < 0.05 was set as the statistically significant threshold. SPSS software (SPSS Statistics 25) and R4.2.2 were used for the statistical analyses.

A total of 198 patients with advanced BTCs were enrolled in this study and consisted of 142 cases with ICC as well as 56 cases of ECC and GBC (referred to as “Others”). The basic clinical characteristics of the participants are shown in Table 1. At the time of diagnosis, 33.8% (48/142) of the ICC patients had a history of hepatitis B virus (HBV) infection, while this value was 14.3% (8/56) for the Others group. About 61.3% of the ICC patients had distant or intrahepatic metastases while receiving chemotherapy or immunotherapy, and 55.4% of the patients in the Others group had metastases. Moreover, 41.4% (82/198) of the patients received standalone chemotherapy while the remaining 58.6% (116/198) of patients received chemotherapy along with anti-PD1 therapy. The characteristics of the BTC patients receiving anti-PD1 therapy are shown in Table 2. There were no statistically significant differences in the age, gender, pathological grade, clinical stage, metastasis, or HBV infection between patients in the ICC and Others groups. The baseline characteristics of the ICC, ECC, and GBC patients receiving anti-PD1 combination therapy are documented in Supplementary Table S4.

We compared the efficacies of chemotherapy combined with anti-PD1 treatment vs. chemotherapy alone in advanced BTCs and showed that the latter was more effective (mOS: 13.2 vs. 11.2 months, p < 0.001; mPFS: 6.2 vs. 5.1 months, p < 0.001) (Supplementary Figures S1A, B). To compare the responses to immunotherapy from different anatomical sites of the subjects with or without another therapy, we divided the participants into two groups as ICC and Others. In patients with ICC, the OS and PFS when receiving immunotherapy were significantly better than those receiving non-immunotherapy treatments (mOS: 15.1 vs. 10.6 months, p < 0.001; mPFS: 9.5 vs. 4.9 months, p < 0.001), and HRs for OS and PFS were 2.06 (95% confidence interval (CI): 1.42–2.99) and 2.76 (95% CI: 1.90–4.00), respectively (Figures 1A, B). In the Others group, the patients receiving immunotherapy fared as well as those receiving chemotherapy alone (mOS: 10.7 vs. 12.1 months; mPFS: 6.2 vs. 5.7 months, p > 0.05 for both OS and PFS) (Figures 1C, D). To this end, we further studied the differences in anti-PD1 effects between the ICC and Others groups. Interestingly, among all the patients receiving anti-PD1, the ICC group achieved a mOS of 15.1 months, while the Others group achieved a mOS of 10.7 months (HR: 1.85, 95% CI: 1.11–3.07, p = 0.02); the ICC group achieved a mPFS of 9.5 months, while the Others group achieved a mPFS of 6.2 months (HR: 1.77, 95% CI: 1.11–2.81, p = 0.02) (Figures 2A, B; Supplementary Table S1). In the cohort receiving chemotherapy alone, the clinical outcomes of the two groups were compared; there were no significant differences in the mOS and mPFS values between the ICC and Others groups (mOS: 10.6 vs. 12.1 months, p = 0.20; mPFS: 4.9 vs. 5.7 months, p = 0.83) (Supplementary Figure S1C, D). In patients receiving anti-PD1 treatment, the ORR was 31.1% (95% CI: 21.8%–41.7%) and DCR was 85.6% (95% CI: 76.6%–92.1%) for the ICC group, with 2 complete response (CR), 26 partial response (PR), 49 stable disease (SD), and 13 progressive disease (PD) instances. Meanwhile, for the Others group, the ORR was 19.2% (95% CI: 6.6%–39.4%) and DCR was 80.8% (95% CI: 60.6%–93.4%), with 5 PR, 16 SD, and 5 PD cases. There were no significant differences in the ORR and DCR between the two groups (p = 0.35 and p = 0.77, respectively) (Supplementary Table S1).

To compare the efficacies with different lines of anti-PD1 therapy between the ICC and Others groups, we divided the anti-PD1 therapy cohort into first and second lines. Based on survival analysis of the first-line anti-PD1 therapy, the OS of the ICC group was better than that of the Others group, with mOS of 16.0 months (95% CI: 13.5–19.3) vs. 11.8 months (95% CI: 8.2–14.5) and p = 0.04; the mPFS was longer in the ICC cohort, while no statistically significant difference was found between the two cohorts at 7.3 months (95% CI: 6.1–8.8) vs. 5.3 months (95% CI: 3.0–10.7) and p = 0.10. The 6-, 12-, and 18-month survival rates of the ICC group were 60.0%, 15.7%, and 5.7%, while those of the Others group were 42.1%, 10.5%, and 5.3%, respectively. With regard to tumor shrinkage evaluation, the ORR was 30.0% (95% CI: 19.6%–42.1%) and DCR was 85.7% (95% CI: 75.3%–92.9%) for the ICC group, with 1 CR, 20 PR, 39 SD, and 10 PD cases; similarly, for the Others group, the ORR was 15.8% (95% CI: 3.4%–39.6%) and DCR was 78.9% (95% CI: 54.4%–93.9%), with 3 PR, 12 SD, and 4 PD cases (ORR: p = 0.26, DCR: p = 0.49) (Figures 2C, D; Supplementary Table S2).

Based on survival analysis of the second-line anti-PD1 therapy, the mOS values of the ICC and Others groups were 11.9 and 6.1 months, respectively (p = 0.02). Similarly, the mPFS values of the two groups were 6.1 and 2.6 months (p < 0.01). With regard to tumor shrinkage evaluation, the ORR was 30.0% (95% CI: 11.9%–54.3%) and DCR was 85.0% (95% CI: 63.1%–96.8%) for the ICC group, with 1 CR, 5 PR, 11 SD, and 3 PD cases; similarly, for the Others group, the ORR was 28.5% (95% CI: 3.7%–71.0%) and DCR was 85.7% (95% CI: 42.1%–99.6%), with 2 PR, 4 SD, and 1 PD cases (ORR: p = 1.00, DCR: p = 1.00) (Figures 2E, F; Supplementary Table S3).

We further analyzed the efficacies of anti-PD1 therapy in patients with ICC, ECC, and GBC. The results indicated that the OS of ICC patients was superior to those of the ECC and GBC patients (p = 0.049). However, the PFS was not statistically significant (p = 0.066), likely owing to the small sample size (Supplementary Figure S2).

By analyzing clinical cases of BTCs, we found that patients diagnosed with ICC had better responses to immunotherapy than those diagnosed with other types of BTCs. To further understand the reasons for the high response rate of ICC to ICB therapy, two datasets from the GEO database, namely GSE32225 (for ICC) and GSE132305 (for ECC), were selected for further analyses. The matrices of the two datasets were extracted. ESTIMATE and TIDE were used to predict the responses of both groups to ICB. The results in Figures 3A, B show that the stromal, immune, and ESTIMATE scores for ICC were all lower than those for ECC; furthermore, the tumor purity was higher in ICC than ECC. Conversely, the results in Figures 3C, D show that the TIDE and dysfunction scores were lower in ICC than ECC, while the exclusion scores had no statistically significant difference. Considering that the TIDE prediction scores account for both T-cell dysfunction and T-cell rejection whereas most other biomarkers address only one of these factors, we believe that the TIDE prediction score might be superior to the other scores. Therefore, ICC was considered to have better immune effects than ECC, which is consistent with the results observed in clinical patients.

TISIDB was used to calculate the differences in the immune-related molecules between ICC and ECC. The results shown in Figure 3E indicate that most chemokines, chemokine receptors, MHC molecules, immunostimulators, and immunoinhibitors were stronger in ICC than ECC. Among these, CXCL2 was poorly expressed in ICC while the expressions of CCL5, CXCL8, and CXCL9 were high, indicating that there were more immune-related molecules in ICC.

Five algorithms were used to calculate the differences in immune cells between ICC and ECC. The CD8⁺ T cells were higher in ICC than ECC, except when using CIBERSORT. The M1 macrophages were higher in ICC than ECC based on CIBERSORT and XCELL. The proportion of CD4 memory-activated T cells was higher in ICC than ECC (Figure 4).

Differential analyses of the two normalized datasets using the limma package resulted in 560 upregulated genes in ECC and 590 upregulated genes in ICC (Figure 5A). The 1,155 differential genes of ECC and ICC were intersected with the differential genes from the ImmPort database, and 83 immune-related differentially expressed genes were obtained (Figure 5B).

As shown in Figure 5C, these 83 genes were significantly different between ICC and ECC. The expressions of CCL19, TNFRSF19, TNFRSF4, IL-18, and CXCR4 were higher in ICC patients, while expressions of CXCL2, FAS, and IRF1 were higher in ECC patients. The immune-related differentially expressed genes were mainly enriched in terms of antigen processing and presentation as well as cytokine receptors (Figure 5D). Hence, it is suggested that the differential effects of immunotherapy between ICC and ECC may be related to the immune microenvironment.