Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a predictive biomarker for immune checkpoint inhibitors in advanced biliary tract cancer

Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a predictive biomarker for immune checkpoint inhibitors in advanced biliary tract cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a predictive biomarker for immune checkpoint inhibitors in advanced biliary tract cancer Changhoon Yoo, Yeong Hak Bang, Choong-kun Lee, Kyunghye Bang, and 18 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3839367/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The combination of anti-PD-1/L1 with gemcitabine and cisplatin (GemCis) has recently shown significant survival benefits in randomized phase 3 trials for advanced biliary tract cancer (BTC). However, no biomarker predictive of benefit has been established for anti-PD-1/L1 in BTC. Here, we evaluated tumor-infiltrating lymphocytes (TILs) using artificial intelligence-powered immune phenotype (AI-IP) analysis in advanced BTC treated with anti-PD-1. Data and images of BTC cohort from The Cancer Genome Atlas (TCGA) were initially analyzed to evaluate the transcriptomic and mutational characteristics of various AI-IPs in BTC. The inflamed IP showed increased cytolytic activity scores and an interferon-gamma signature compared to the non-inflamed IP. Next, pre-treatment H&E-stained whole-slide images from 339 advanced BTC patients who received anti-PD-1 monotherapy as second-line treatment or beyond, were retrospectively utilized for AI-IP analysis. Overall, AI-IPs were classified as inflamed (high intratumoral TIL [iTIL]) in 40 patients (11.8%), immune-excluded (low iTIL and high stromal TIL) in 167 (49.3%), and immune-deserted (low TIL overall) in 132 (38.9%). The inflamed IP group showed a significantly higher overall response rate compared to the non-inflamed IP groups (27.5% vs. 7.7%, P < 0.001). Median overall survival (OS) and progression-free survival (PFS) were significantly longer in the inflamed IP group than in the non-inflamed IP group (OS: 12.6 vs. 5.1 months, P = 0.002; PFS: 4.5 vs. 1.9 months, P < 0.001). IP classified by AI-powered spatial TIL analysis was effective in predicting the efficacy outcomes of advanced BTC patients treated with anti-PD-1 therapy. Further validation is necessary in the context of anti-PD-1/L1 plus GemCis. Health sciences/Biomarkers/Predictive markers Biological sciences/Cancer/Gastrointestinal cancer/Biliary tract cancer Biliary tract cancer artificial intelligence tumor-infiltrating lymphocyte immune-checkpoint inhibitor Figures Figure 1 Figure 2 Figure 3 MAIN TEXT Biliary tract cancer (BTC) is a malignancy that arises from the epithelium of the intrahepatic and extrahepatic bile ducts as well as the gallbladder. Approximately 60–70% of patients with BTC present with unresectable or metastatic stage, and the prognosis for these patients is poor, with a median overall survival (OS) of 1 year 1 . Although several targeted agents against FGFR2 gene fusion, IDH1 mutations, BRAF V600E mutation, and ERBB2 amplification have shown promising efficacy outcomes, only certain groups of patients could benefit from these agents 2–8 . Immune checkpoint inhibitors (ICIs) such as programmed cell death-1 (PD-1) and programmed cell death ligand-1 (PD-L1) antibodies have led to significant advancements in treating various types of cancer and have been widely investigated for patients with advanced BTC. In BTC, anti-PD-1 monotherapy demonstrated only modest efficacy, with an objective response rate (ORR) of 5–13% and a median progression-free survival (PFS) of 1.4–3.7 months in previously treated patients with unresectable or metastatic BTC 9–11 . However, two recent pivotal phase 3 trials (Keynote-966 and TOPAZ-1 trials) have demonstrated a statistically significant benefit in OS with the addition of pembrolizumab (anti-PD-1) or durvalumab (anti-PD-L1) in the first-line setting compared to gemcitabine plus cisplatin (GemCis) alone 12,13 . Despite first-in-decade breakthroughs in first-line treatment for advanced BTC and promising long-term survival, there are concerns about the relatively small OS benefits, with a median OS improvement of 1.3 to 1.8 months and a hazard ratio (HR) of 0.80 to 0.83 12,13 . This is particularly concerning given the high cost of ICIs. This underscores the urgent need for reliable biomarkers for the response of patients with advanced BTC to ICIs. Tumor-infiltrating lymphocytes (TILs) play a major role in the anti-tumor activity of ICIs, and the spatial heterogeneity of TILs has been suggested to be associated with the efficacy outcomes of ICIs 14,15 . However, quantifying relevant TILs has been challenging due to the labor-intensive nature of the method and limitations imposed by spatial distribution in whole-slide images (WSI) as well as interobserver heterogeneity 16–18 . To overcome this limitation, an artificial intelligence (AI)-powered spatial TILs analyzer (Lunit SCOPE IO®, Lunit Inc., Seoul, Republic of Korea) using hematoxylin and eosin (H&E) WSI was developed to classify the immune phenotype (IP) of the tumor microenvironment (TME) into one of three distinct categories: inflamed (TILs distributed intratumorally), immune-excluded (TILs excluded from the cancer stroma), and immune-desert (scant TILs in TME). This AI-powered spatial TILs analysis has demonstrated its effectiveness in stratifying patients with non-small cell lung cancer for outcomes with ICIs 19,20 . Given the spatial heterogeneity of TILs and their potential correlation with the efficacy outcomes of anti-PD-1 therapy in BTC 21,22 , we hypothesized that biomarkers identified via AI-powered IPs (AI-IPs) may be effective in identifying patient subgroups with a favorable tumor response to anti-PD-1 treatment in BTC. Here, we conducted a multicenter retrospective study to assess the predictive efficacy of AI-IP based on AI-powered spatial TILs analysis in patients with unresectable or metastatic BTC who were treated with anti-PD-1. RESULTS Artificial Intelligence-powered Immune Phenotype The Lunit SCOPE IO model is an AI-IP analyzer that classifies the IP of the TME based on spatial distribution and density of TILs. The model was updated from a previously published version 19 To perform spatial analysis of TIL distribution within WSI of various sizes, each WSI was divided into 0.25 mm 2 grids, with the IP of each grid classified based on the following criteria 23,24 (Fig. 1 a): grid-level inflamed IP, grid-level immune-excluded IP, and grid-level immune-desert IP. The overall WSI-level inflamed score, immune-excluded score, and immune-desert score were calculated by dividing the number of grids with the respective phenotype by the total number of grids analyzed. The representative IP for each WSI was as follows: inflamed when the inflamed score was ≥ 33.3%, immune-excluded when the immune-excluded score was ≥ 33.3% and the inflamed score was < 33.3%, and immune-desert otherwise. Genetic and Transcriptomic Landscape of Immune Phenotype To evaluate the transcriptomic and mutational characteristics of various AI-IPs in BTC, data and images from The Cancer Genome Atlas (TCGA) were included. In 36 H&E-stained WSIs of the TCGA BTC cohort, 5 (13.9%), 14 (38.9%), and 17 (47.2%) were classified as inflamed, immune-excluded, and immune-desert IP, respectively. Gene Set Enrichment Analysis (GSEA) was performed to compare inflamed IP and non-inflamed IP groups. Transcriptomic signatures were determined by measuring cell abundance using CIBERSORT with the LM22 matrix 25 . Pathway enrichment analysis revealed high enrichment of interferon-gamma (IFNγ) and IL6/JAK/STAT3 in inflamed IP, while non-inflamed IP showed enrichment of epithelial mesenchymal transition, TGF-beta, and WNT beta-catenin signaling (Fig. 1 b). Gene Set Enrichment Analysis (GSEA) revealed that T cell proliferation pathways are enriched in inflamed IP, while keratinization pathways are enriched in non-inflamed IP (Fig. 1 c). The inflamed IP exhibited a significantly increased interferon-gamma signature 26 , cytolytic score 27 and immunologic constant of rejection (ICR) score 28 , indicating a more immune-active microenvironment compared to the non-inflamed IP (Fig. 1 d). When IP was further classified using previously defined transcriptomics-based TME subtypes 29 , most of the inflamed IP were immune-enriched subtypes (4 out of 5, 80.0%), while more than half of the patients with immune-excluded IP and immune-desert IP were classified as fibrotic or immune-enriched/fibrotic subtypes (8 out of 14, 57.1%) and as depleted subtype (11 out of 17, 64.7%), respectively ( Extended Data Fig. 1 ). The CIBERSORT algorithm indicated an enrichment of CD8-positive T cells and M1 macrophages in the inflamed IP ( Extended Data Fig. 2 ). In the non-inflamed IP, oncogenic mutations in KRAS , ERBB2 , and PIK3CA were observed at a frequency of 6.7% each, while they were not observed in inflamed IP. In addition, the polybromo-1 ( PRBM1 ) gene, which has been reported as a negative predictive biomarker for ICIs in non-small cell lung cancer 30 , was only observed in non-inflamed IP (26.7%) (Fig. 1 e). Clinical Cohort for Correlative Analysis For the correlational study of IPs in BTC patients treated with anti-PD-1 therapy, patient data and pathological slides were retrospectively collected from two tertiary referral cancer institutions (N = 339; Asan Medical Center [AMC] and Yonsei Cancer Center [YCC], Seoul, Republic of Korea). Key eligibility criteria included histologically confirmed unresectable or metastatic BTC (intrahepatic and extrahepatic cholangiocarcinoma, and gallbladder cancer), radiological progression on first-line GemCis, anti-PD-1 monotherapy as second- or subsequent-line therapy, and no prior ICIs before anti-PD-1 therapy. In the sub-cohort for multicolor flow cytometry analysis using peripheral blood mononuclear cells (PBMCs), clinical data and blood were prospectively collected. Between December 2017 and November 2022, a total of 445 patients were treated with second-line or subsequent anti-PD-1 monotherapy (pembrolizumab or nivolumab) for unresectable or metastatic BTC after progression on first-line GemCis at AMC and YCC in Seoul, Korea. Among them, 339 (76.2%) patients were eligible for current analysis after excluding 89 and 17 patients due to the lack of clinical information or archival tissue, and being ineligible for meeting the tissue quality criteria for AI-IP analysis, respectively (Fig. 2 a). Baseline patient characteristics are summarized in Table 1 . Intrahepatic cholangiocarcinoma was the most prevalent type of BTC (N = 155, 45.7%), and 55.8% of patients (N = 189) received anti-PD-1 as second-line treatment. Pembrolizumab was administered to 227 (67.0%) patients, while nivolumab was given to 112 (33.0%) patients. Additionally, 43.2% of the patients (137 out of 317) had PD-L1 CPS ≥ 1. According to the IP classification, 40 patients (11.8%) were classified as inflamed, 167 patients (49.3%) as immune-excluded, and 132 patients (38.9%) as immune-desert IP. There was a trend showing an increased proportion of inflamed IP in patients with higher PD-L1 CPS (6.1%, 14.0%, and 21.6% in CPS < 1, 1–9, and ≥ 10, respectively; P = 0.020) (Fig. 2 b-c). Table 1 Baseline characteristics according to immune phenotypes Total (N = 339) Inflamed (N = 40) Immune-excluded (N = 167) Immune-desert (N = 132) P-value Age ≥ 65 years 172 (50.7%) 24 (60.0%) 85 (50.9%) 63 (47.7%) 0.396 Sex 0.095 Male 144 (42.5%) 12 (30.0%) 68 (40.7%) 64 (48.5%) Female 195 (57.5%) 28 (70.0%) 99 (59.3%) 68 (51.5%) ECOG 0.745 0 29 (8.6%) 4 (10.0%) 17 (10.2%) 8 (6.1%) 1 250 (73.7%) 28 (70.0%) 122 (73.1%) 100 (75.8%) 2 60 (17.7%) 8 (20.0%) 28 (16.8%) 24 (18.2%) Site of origin < 0.001 Extrahepatic 100 (29.5%) 7 (17.5%) 72 (43.1%) 21 (15.9%) Gallbladder 84 (24.8%) 10 (25.0%) 35 (21.0%) 39 (29.5%) Intrahepatic 155 (45.7%) 23 (57.5%) 60 (35.9%) 72 (54.5%) Tissue harvest method < 0.001 Biopsy 190 (56.0%) 16 (40.0%) 67 (40.1%) 107 (81.1%) Surgery 149 (44.0%) 24 (60.0%) 100 (59.9%) 25 (18.9%) Tissue site < 0.001 Primary site 271 (79.9%) 27 (67.5%) 144 (86.2%) 100 (75.8%) Metastatic site 43 (12.7%) 4 (10.0%) 17 (10.2%) 22 (16.7%) Lymph node 25 (7.4%) 9 (22.5%) 6 (3.6%) 10 (7.6%) Pathology 0.155 Adenocarcinoma 311 (91.7%) 35 (87.5%) 158 (94.6%) 118 (89.4%) Others 28 (8.3%) 5 (12.5%) 9 (5.4%) 14 (10.6%) PD-L1 CPS* 0.020 < 1 180 (56.8%) 11 (32.4%) 96 (61.1%) 73 (60.3%) 1–9 86 (27.1%) 12 (35.3%) 41 (25.3%) 33 (27.3%) ≥ 10 51 (16.1%) 11 (32.4%) 25 (15.4%) 15 (12.4%) Time from tissue harvest to ICI < 0.001 < 12 months 211 (62.2%) 22 (55.0%) 87 (52.1%) 102 (77.3%) ≥ 12 months 128 (37.8%) 18 (45.0%) 80 (47.9%) 30 (22.7%) ICI agent 0.521 Pembrolizumab 227 (67.0%) 25 (62.5%) 109 (65.3%) 93 (70.5%) Nivolumab 112 (33.0%) 15 (37.5%) 58 (34.7%) 39 (29.5%) Treatment line 0.207 Second-line 189 (55.8%) 24 (60.0%) 85 (50.9%) 80 (60.6%) Third-line or later 150 (44.2%) 16 (40.0%) 82 (49.1%) 52 (39.4%) Data are presented as no. (%) * CPS: combined positive score, PD-L1 status was not assessed in 22 patients. Correlation between Immune Phenotype and Efficacy of Anti-PD-1 Therapy At a median follow-up of 18.9 months (95% confidence interval [CI], 15.4–25.9), the median OS, PFS, and overall response rate (ORR) per RECIST v1.1 were 5.7 months (95% CI, 4.7–6.6), 2.0 months (95% CI, 1.8–2.4), and 10.0%, respectively, in the study patients as a whole. The median OS in the inflamed IP group was 12.6 months (95% CI, 9.1–not reached), which was significantly longer than that in the immune-excluded IP group (5.9 months [95% CI, 4.6–7.7]) and the immune-desert IP group (4.5 months [95% CI, 3.9–5.6]) (inflamed IP vs. non-inflamed IP, 12.6 vs. 5.1 months, hazard ratio [HR] = 0.46 [0.29–0.73], P < 0.001; Fig. 3 a). The median PFS in the inflamed IP group was significantly longer than those in the immune-excluded and immune-desert IP groups (4.5 [95% CI, 1.9–17.6], 2.0 [95% CI, 1.8–2.6], and 1.9 [95% CI, 1.6–2.3] months, respectively) (inflamed IP vs. non-inflamed IP, 4.5 vs. 1.9 months, HR = 0.49 [95% CI, 0.33–0.73], P < 0.001; Fig. 3 b). The ORR was significantly higher in the inflamed IP group compared to the immune-excluded and immune-desert IP groups (27.5% vs. 6.6% vs. 9.1%, P < 0.001; inflamed IP vs. non-inflamed IP, 27.5% vs. 7.7%, P < 0.001). The association between IPs and survival outcomes was consistent in the subgroup analysis for each AMC and YCC cohorts ( Extended Data Fig. 3 ). The favorable survival outcomes of the inflamed IP group compared to the non-inflamed IP group were consistent across the subgroups (Fig. 3 c, Extended Data Fig. 4 ). In the multivariable analyses including the potential prognostic factors (P < 0.1 in the univariate analyses), inflamed IP was significantly associated with better OS (adjusted HR = 0.44 [95% CI, 0.28–0.69], P < 0.001) and PFS (adjusted HR = 0.50 [95% CI, 0.34–0.75], P < 0.001) (Table 2 ). To determine whether IP classification can serve as a predictive biomarker of anti-PD-1 therapy, we analyzed the PFS with first-line GemCis in the same dataset (N = 334), excluding patients with no PFS data for first-line GemCis; the analysis revealed no significant association between IP and PFS with first-line GemCis (median 2.3 months in the inflamed group, 3.8 months in the immune-excluded group, 2.8 months in the immune-desert group, P = 0.340; Fig. 3 d). In addition, PD-L1 CPS was not associated with OS (CPS < 1 vs. ≥1, median 5.9 vs. 5.1 months, P = 0.301) and PFS (median 1.9 vs. 2.1 months, P = 0.152) ( Extended Data Fig. 5 ). Table 2 Multivariable analyses for progression-free survival and overall survival Variables Overall survival Progression-free survival univariable multivariable univariable multivariable HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value HR (95% CI) P-value Age (≥ 65 years vs. < 65 years) 0.97 (0.76–1.24) 0.799 0.93 (0.74–1.17) 0.538 Sex (Male vs. Female) 0.91 (0.71–1.17) 0.460 0.85 (0.67–1.06) 0.149 ECOG (2 vs. 0–1) 2.97 (2.16–4.09) < 0.001 3.38 (2.44–4.70) < 0.001 2.69 (1.99–3.63) <0.001 2.63 (1.95–3.55) < 0.001 Location Extrahepatic Reference Reference Gallbladder 1.15 (0.83–1.60) 0.410 1.17 (0.86–1.59) 0.323 Intrahepatic 0.94 (0.71–1.26) 0.689 1.05 (0.800–1.37) 0.745 Immune phenotype 0.46 (0.29–0.73) < 0.001 0.44 (0.28–0.69) < 0.001 0.49 (0.33–0.73) < 0.001 0.50 (0.34–0.75) < 0.001 (Inflamed vs. Non-inflamed) PD-L1 CPS (≥ 1vs. < 1) 0.87 (0.67–1.13) 0.301 0.84 (0.66–1.07) 0.152 Treatment line (≥ 3 vs. 2) 1.31 (1.02–1.67) 0.032 1.51 (1.18–1.94) 0.001 1.19 (0.95–1.49) 0.140 HR, hazard ratio; CI, confidence interval; CPS, combined positive score; ICI, immune checkpoint inhibitor Immune Cell Subpopulation in PBMCs According to Immune Phenotype Multi-color flow cytometry of PBMCs was performed on a subgroup of 29 patients enrolled in the prospective cohort study. Peripheral blood samples were collected at baseline (i.e., cycle 1 day 1 of anti-PD-1 therapy). In this subgroup, 6.9% (N = 2), 48.3% (N = 14), and 44.8% (N = 13) of patients were classified into the inflamed, immune-excluded, and immune-desert IP, respectively. PBMCs from the inflamed IP group showed a higher proportion of CD69 + CD8 + T cells, which are a marker of resident memory T cells 31,32 , and CD8 + effector memory T cells, compared to the non-inflamed IP group ( Extended Data Fig. 6 ). DISCUSSION In this study, we showed that AI-IPs classified using AI-powered spatial analysis of TILs on H&E-stained tumor slides were effective in stratifying patients with unresectable or metastatic BTC for the efficacy of anti-PD-1 monotherapy. In the analysis using the TCGA dataset, the inflamed IP showed increased cytolytic activity scores and an interferon-gamma signature, and in the prospective sub-cohort for PBMC analysis, inflamed IP was associated with higher proportions of resident memory T cells and CD8 + effector memory T cells than non-inflamed IPs; provided biological background. Among the 339 patients included in the clinical cohort for the correlative analysis of AI-IP for its relationship with the efficacy of anti-PD-1 therapy, 11.8% of them were classified into the inflamed IP group, which was significantly associated with better ORR, PFS, and OS with anti-PD-1 monotherapy compared to those in the non-inflamed IP groups (i.e., immune-excluded and immune-desert IPs). As there was no relationship between IPs and PFS with prior first-line GemCis, our findings suggest that this AI-IP model may serve as a predictive biomarker rather than a prognostic marker. To the best of our knowledge, this is one of the first studies to predict the effectiveness of ICI treatment in patients with BTC. The inflamed IP group showed significantly higher ORR (27.5% vs. 7.7%) and better PFS (4.5 vs. 1.9 months) and OS (12.6 vs. 5.1 months) compared to the non-inflamed IP groups. The results were consistent in the subgroup analysis for each group of participating institutions. Its association with PFS and OS remained significant in the multivariable analyses, which included key clinicopathological factors. The AI-IP model used in the present analysis also indicated that inflamed IP could predict the efficacy of ICIs in addition to PD-L1 immunohistochemistry (IHC) for patients with advanced non-small cell lung cancer (NSCLC) 19 . In the prospective randomized phase 3 ATTLAS trial comparing atezolizumab-bevacizumab plus chemotherapy with chemotherapy alone, the PFS benefit of adding atezolizumab-bevacizumab was significant in the subgroup with a high inflamed score as assessed by AI-IP 20 . Despite these findings in NSCLC that support our findings, further investigations are required to validate the role of current AI-IP in predicting the efficacy of ICI in BTC, especially using prospective cohorts. In the context of BTC, there is no established biomarker for predicting the efficacy of ICIs. PD-L1 expression, assessed using IHC, has been the most extensively investigated biomarker for ICIs 33 ; however, it was not predictive of the efficacy of anti-PD-1 monotherapy 9 or GemCis plus pembrolizumab or durvalumab in previous trials 12,13 . Our analysis also showed that PD-L1 CPS was not associated with PFS or OS with anti-PD-1 monotherapy in advanced BTC. Although high microsatellite instability/mismatch repair deficiency (MSI-H/dMMR) is a tissue-agnostic biomarker for anti-PD-1 therapy 34 , its incidence in BTC is extremely rare, with an incidence rate of less than 1% in patients with BTC 9,35 . Furthermore, no genetic alterations were associated with the advantage of administering additional durvalumab in the exploratory analysis of the TOPAZ-1 trial 36 . Our results showing a correlation between AI-IP and efficacy outcomes with anti-PD-1 monotherapy, may have clinical significance given that anti-PD-1/L1 has now become the standard of care for the treatment of unresectable or metastatic BTC. However, approvals for anti-PD-1/L1 agents in BTC are currently limited to combination therapy with GemCis in the first-line setting, and our study did not include these patient populations. Future research is needed to evaluate the impact of AI-IP analysis in patients undergoing treatment with pembrolizumab or durvalumab in combination with GemCis. This study has limitations, the most prominent of which is its retrospective design, which introduces unintentional biases. As mentioned earlier, the combination of anti-PD-1/L1 with GemCis is now the standard of care, while we focused on patients treated with anti-PD-1 monotherapy. Furthermore, we could not delineate the relationship between MSI-H/dMMR and AI-IP as it was not assessed in this study. These may limit the application of our findings in daily practice. In addition, while we attempted to analyze the patterns of immune cell subsets in PBMC based on the AI-IP, we were unable to derive meaningful results due to the small sample size in the prospective cohort. In conclusion, AI-IP analysis using H&E-stained slides could effectively stratify the IPs in patients with advanced BTC. We found that inflamed IP was predictive of the efficacy outcomes with anti-PD-1 monotherapy. AI-IP could serve as a novel biomarker for patients with advanced BTC treated with ICIs. DECLARATIONS Data availability All requests for raw and analyzed data and materials should be directed to C.Y. and C.-k.L., and will be promptly reviewed by the Asan Medical Center and Yonsei Cancer Center to determine whether the request is subject to any intellectual property or confidentiality obligations. Patient-related data not included in this paper were generated as part of the clinical trial, and may be subject to patient confidentiality. Any data and materials that can be shared will be released via a material transfer agreement. Code availability All analyses were conducted using publicly available software, as detailed in the Methods section. The raw scripts used to generate the RNA-seq analysis figures presented in this paper are available at https://github.com/SGI-pan/BTC-IP Acknowledgments This study was supported in part by the Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea [grant number 2020IL0018], and Lunit, Inc. The authors appreciate the patients and their families who generously donated their tissues to TCGA, as well as the members of TCGA who collected and disclosed the valuable data. The authors thank Joon Seo Lim, PhD, from the Scientific Publications Team at Asan Medical Center (Seoul, South Korea) for providing editorial assistance. Author Contributions Conceptualization: YHB, CYO, CY; Methodology: YHB, CKL, HDK, CYO, JS, CY; Formal analysis: YHB, CKL, CYO, JS, CY; Investigation: YHB, CKL, HDK, CYO, JS, CY; Resources: CKL, KB, HDK, KPK, JHJ, IP, BYR, DKL, HJC, CT, SHJ, ECS, CO, SK, YL, GP, CHA, CYO; Original draft: YHB, CKL, CYO, CY; Writing-review and editing: All authors; Visualization: YHB; Supervision: JS, CY Competing Interests Statement CY received honoraria from Servier, Bayer, AstraZeneca, Merck Sharp & Dohme, Eisai, Celgene, Bristol Myers Squibb, Ipsen, Novartis, Boryung Pharmaceuticals, Mundipharma, and Roche; and received research grants from Servier, Bayer, AstraZeneca, Ono Pharmaceuticals, Ipsen, Boryung Pharmaceuticals, and Lunit Inc. C-kL received honoraria from AstraZeneca, Servier, Dong-A ST, Boryung Pharmaceuticals, Mundipharma, and Roche; consulting fees from Roche, and Daiichi Sankyo; and received research grants or supports from Ono Pharmaceuticals, Celltrion, Boryung Pharmaceuticals, GC Biopharma and Lunit Inc. HDK received honoraria from AstraZeneca, Bristol Myers Squibb, Ono Pharmaceuticals, Boryung Pharmaceuticals, and Boostimmune, received research grants from AstraZeneca and served as a consultant for Mustbio. RSF reports grants or contracts to their institution from Adaptimmune, Bayer, Bristol Myers Squibb, Eisai, Eli Lilly, Pfizer, Roche, and Genentech; consulting fees to themself from Merck, AstraZeneca, Bayer, Bristol Myers Squibb, Exelixis, Cstone, Hengrui, Eisai, Eli Lilly, MSD, Pfizer, Roche, and Genentech; payment or honoraria to themself from Genentech; and participation on a data safety monitoring or advisory board from AstraZeneca, and Hengrui.CO. HJC reports an advisory role at AstraZeneca and Roche. CO, SK, YL, and GP are employees, and CHA and CYO are employees and stock holder of Lunit Inc. REFERENCES Valle JW, Kelley RK, Nervi B, et al: Biliary tract cancer. Lancet 397:428-444, 2021 Goyal L, Meric-Bernstam F, Hollebecque A, et al: Futibatinib for FGFR2-Rearranged Intrahepatic Cholangiocarcinoma. N Engl J Med 388:228-239, 2023 Abou-Alfa GK, Sahai V, Hollebecque A, et al: Pemigatinib for previously treated, locally advanced or metastatic cholangiocarcinoma: a multicentre, open-label, phase 2 study. Lancet Oncol 21:671-684, 2020 Abou-Alfa GK, Macarulla T, Javle MM, et al: Ivosidenib in IDH1-mutant, chemotherapy-refractory cholangiocarcinoma (ClarIDHy): a multicentre, randomised, double-blind, placebo-controlled, phase 3 study. Lancet Oncol 21:796-807, 2020 Javle M, Borad MJ, Azad NS, et al: Pertuzumab and trastuzumab for HER2-positive, metastatic biliary tract cancer (MyPathway): a multicentre, open-label, phase 2a, multiple basket study. Lancet Oncol 22:1290-1300, 2021 Subbiah V, Lassen U, Élez E, et al: Dabrafenib plus trametinib in patients with BRAF V600E -mutated biliary tract cancer (ROAR): a phase 2, open-label, single-arm, multicentre basket trial. Lancet Oncol 21:1234-1243, 2020 Yoo C, Hyung J, Chan SL: Recent Advances in Systemic Therapy for Advanced Intrahepatic Cholangiocarcinoma. Liver Cancer:1-17, 2023 Lee CK, Chon HJ, Cheon J, et al: Trastuzumab plus FOLFOX for HER2-positive biliary tract cancer refractory to gemcitabine and cisplatin: a multi-institutional phase 2 trial of the Korean Cancer Study Group (KCSG-HB19-14). Lancet Gastroenterol Hepatol 8:56-65, 2023 Piha-Paul SA, Oh D-Y, Ueno M, et al: Efficacy and safety of pembrolizumab for the treatment of advanced biliary cancer: Results from the KEYNOTE-158 and KEYNOTE-028 studies. Int J Cancer 147:2190-2198, 2020 Kim RD, Chung V, Alese OB, et al: A Phase 2 Multi-institutional Study of Nivolumab for Patients With Advanced Refractory Biliary Tract Cancer. JAMA Oncol 6:888-894, 2020 Ueno M, Ikeda M, Morizane C, et al: Nivolumab alone or in combination with cisplatin plus gemcitabine in Japanese patients with unresectable or recurrent biliary tract cancer: a non-randomised, multicentre, open-label, phase 1 study. Lancet Gastroenterol Hepatol 4:611-621, 2019 Oh D-Y, Ruth He A, Qin S, et al: Durvalumab plus Gemcitabine and Cisplatin in Advanced Biliary Tract Cancer. NEJM Evidence 1:EVIDoa2200015, 2022 Kelley RK, Ueno M, Yoo C, et al: Pembrolizumab in combination with gemcitabine and cisplatin compared with gemcitabine and cisplatin alone for patients with advanced biliary tract cancer (KEYNOTE-966): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet 401:1853-1865, 2023 Lopez de Rodas M, Nagineni V, Ravi A, et al: Role of tumor infiltrating lymphocytes and spatial immune heterogeneity in sensitivity to PD-1 axis blockers in non-small cell lung cancer. J Immunother Cancer 10, 2022 Liu D, Heij LR, Czigany Z, et al: The role of tumor-infiltrating lymphocytes in cholangiocarcinoma. J Exp Clin Cancer Res 41:127, 2022 Van Bockstal MR, François A, Altinay S, et al: Interobserver variability in the assessment of stromal tumor-infiltrating lymphocytes (sTILs) in triple-negative invasive breast carcinoma influences the association with pathological complete response: the IVITA study. Mod Pathol 34:2130-2140, 2021 Khoury T, Peng X, Yan L, et al: Tumor-Infiltrating Lymphocytes in Breast Cancer: Evaluating Interobserver Variability, Heterogeneity, and Fidelity of Scoring Core Biopsies. Am J Clin Pathol 150:441-450, 2018 Swisher SK, Wu Y, Castaneda CA, et al: Interobserver Agreement Between Pathologists Assessing Tumor-Infiltrating Lymphocytes (TILs) in Breast Cancer Using Methodology Proposed by the International TILs Working Group. Ann Surg Oncol 23:2242-8, 2016 Park S, Ock C-Y, Kim H, et al: Artificial Intelligence–Powered Spatial Analysis of Tumor-Infiltrating Lymphocytes as Complementary Biomarker for Immune Checkpoint Inhibition in Non–Small-Cell Lung Cancer. J Clin Oncol 40:1916-1928, 2022 Park S, Kim TM, Han J-Y, et al: A Phase 3, Randomized study of atezolizumab plus bevacizumab and chemotherapy in patients with EGFR or ALK mutated in non-small cell lung cancer (ATTLAS, KCSG-LU19-04). Journal of Clinical Oncology 0:10.1200/JCO.23.01891 Yoon JG, Kim MH, Jang M, et al: Molecular Characterization of Biliary Tract Cancer Predicts Chemotherapy and Programmed Death 1/Programmed Death-Ligand 1 Blockade Responses. Hepatology 74:1914-1931, 2021 Kim HD, Kim JH, Ryu YM, et al: Spatial Distribution and Prognostic Implications of Tumor-Infiltrating FoxP3- CD4+ T Cells in Biliary Tract Cancer. Cancer Res Treat 53:162-171, 2021 Amgad M, Stovgaard ES, Balslev E, et al: Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group. NPJ Breast Cancer 6:16, 2020 Chen DS, Mellman I: Elements of cancer immunity and the cancer–immune set point. Nature 541:321-330, 2017 Newman AM, Steen CB, Liu CL, et al: Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol 37:773-782, 2019 Ayers M, Lunceford J, Nebozhyn M, et al: IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 127:2930-2940, 2017 Rooney MS, Shukla SA, Wu CJ, et al: Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell 160:48-61, 2015 Bertucci F, Finetti P, Simeone I, et al: The immunologic constant of rejection classification refines the prognostic value of conventional prognostic signatures in breast cancer. Br J Cancer 119:1383-1391, 2018 Bagaev A, Kotlov N, Nomie K, et al: Conserved pan-cancer microenvironment subtypes predict response to immunotherapy. Cancer Cell 39:845-865.e7, 2021 Zhou H, Liu J, Zhang Y, et al: PBRM1 mutation and preliminary response to immune checkpoint blockade treatment in non-small cell lung cancer. NPJ Precis Oncol 4:6, 2020 Kumar BV, Ma W, Miron M, et al: Human Tissue-Resident Memory T Cells Are Defined by Core Transcriptional and Functional Signatures in Lymphoid and Mucosal Sites. Cell Rep 20:2921-2934, 2017 Luoma AM, Suo S, Wang Y, et al: Tissue-resident memory and circulating T cells are early responders to pre-surgical cancer immunotherapy. Cell 185:2918-2935.e29, 2022 Davis AA, Patel VG: The role of PD-L1 expression as a predictive biomarker: an analysis of all US Food and Drug Administration (FDA) approvals of immune checkpoint inhibitors. J Immunother Cancer 7:278, 2019 Le DT, Uram JN, Wang H, et al: PD-1 Blockade in Tumors with Mismatch-Repair Deficiency. N Engl J Med 372:2509-2520, 2015 Lowery MA, Ptashkin R, Jordan E, et al: Comprehensive Molecular Profiling of Intrahepatic and Extrahepatic Cholangiocarcinomas: Potential Targets for Intervention. Clin Cancer Res 24:4154-4161, 2018 Valle JW, Qin S, Antonuzzo L, et al: 68O Impact of mutation status on efficacy outcomes in TOPAZ-1: A phase III study of durvalumab (D) or placebo (PBO) plus gemcitabine and cisplatin (+GC) in advanced biliary tract cancer (BTC). Ann Oncol 33:S1457, 2022 METHODS Development of an artificial intelligence-powered immune phenotype (AI-IP) analyzer The current version of AI-IP (Lunit SCOPE IO ® ) used in this study contains updated versions of the cell detection AI model and the tissue segmentation AI model, compared to previous versions 19 . This section describes the differences compared to the previous models. Cell Detection Model The cell detection model identifies the location of lymphocytes and tumor cells. This model is based on a convolutional neural network (CNN). In the previous model, a Faster R-CNN 37 architecture was used. In contrast, the current version utilizes a DeepLabV3+ 38 architecture with a Resnet-34 39 backbone as the feature extractor. With this model, we framed the detection task as a dense pixel prediction problem. To train the network, a circle with a radius of 0.95 μm was drawn around the point annotation of each cell. The class value associated with the cell at that location (lymphocyte or tumor cell) is then assigned. The training patches cover an area of 151,414 μm 2 (linearly resized to an image of 2048 x 2048 pixels). In each training step, a random portion of images sized 1024 x 1024 pixels is cropped, and this cropped image undergoes on-the-fly data augmentation. The model generates probability maps of size 256 x 256, which are then linearly interpolated to match the original input dimensions (1024 x 1024 pixels), ensuring a 1-to-1 pixel correspondence with the pixel annotations. Since the model predicts the likelihood of the cells being at each pixel, a post-processing stage is required to extract the cell locations. In this stage, the likelihood maps undergo Gaussian filtering (σ = 3), followed by local maximum detection with a radius of 0.57 μm. The model was optimized using the Adam optimizer 40 with a learning rate of 0.002. The learning rate was decayed by a multiplicative factor of 0.2 when the validation loss did not decrease for a period of 5 epochs. Mini-batches of 24 samples were used. The model's performance in detecting lymphocytes and tumor cells, as measured by the F1-score, was 0.69 and 0.74, respectively. Tissue Segmentation Model The AI model was similar to the previous version, except that the backbone for feature extraction was ResNet-34 39 Similarly, it determines whether a pixel belongs to the cancer area, cancer stroma, or background regions. The model was trained using patches that covered an area of 605,657 μm 2 , which were then linearly resized to images sized 1024 x 1024 pixels. The model generates probability maps of size 256 x 256, which are then linearly interpolated to match the original input dimensions (1024 x 1024 pixels), ensuring a 1-to-1 pixel correspondence with the pixel annotations. In this case, there was an unbalanced sampling of patches from different types of organs. The segmentation model was optimized using the Adam optimizer 40 with a learning rate of 0.0001. The learning rate was decayed by a multiplicative factor of 0.5 when the validation loss did not decrease for a period of 5 epochs. Mini-batches of 32 samples were used. The model's performance in segmenting the cancer area (CA) and cancer stroma (CS) was evaluated using the Intersection-over-Union metric, yielding scores of 0.78 and 0.64, respectively. Datasets for developing the AI models The WSIs that were used to develop the previously described AI models included more than 16 different primary tumor types and origins. The Cell Detection Model was developed using patches extracted from 3,334 WSIs (N=2,485 for training and N=849 for tuning). From these WSIs, 5,698 and 1,925 patches (151,414 μm 2 per patch) were extracted for training and tuning, respectively. The Tissue Segmentation Model was developed using patches extracted from 15,830 WSIs (N=15,004 for training, and N=826 for tuning). From these WSIs, 56,545 and 2,971 patches (605,657 μm 2 per patch) were extracted for training and validation, respectively. Tissue Quality Criteria Qualified H&E-stained slides were included in the current analysis. Criteria for tissue quality included: (1) total cancer area in WSI ≥ 0.125 mm 2 and (2) total number of analyzed grids (containing cancer area or cancer stroma ≥ 5% in a 0.25 mm 2 -sized grid) in WSI ≥ 10. Classification of Artificial Intelligence-powered Immune Phenotype The Lunit SCOPE IO model is an AI-IP analyzer that classifies the IP of the TME based on spatial distribution and density of TILs. The model was updated from a previously published version 19 through additional training and optimization (tuning) using 14.6×10 10 μm 2 H&E-stained tissue regions and 6.2 x 10 5 TIL from 18,679 H&E stained WSI of 17 solid tumor types, including BTC, which were not utilized in the development of the original AI models. To perform spatial analysis of TIL distribution within WSI of various sizes, each WSI was divided into 0.25 mm 2 grids, with the IP of each grid classified based on the following criteria 23,24 : grid-level inflamed IP, if the TIL density within the total cancer area (CA) in the grid is ≥ 130/mm 2 (where CA refers to cancer epithelium or, in the case of non-epithelial tumors, the non-stromal tumor cells); grid-level immune-excluded IP, if the TIL density within the total CA is 260/mm 2 ; grid-level immune-desert IP, if the TIL densities are below threshold in both the CA and CS within the grid. The overall WSI-level inflamed score, immune-excluded score, and immune-desert score were calculated by dividing the number of grids with the respective phenotype by the total number of grids analyzed. The representative IP for each WSI was as follows: inflamed when the inflamed score was ≥ 33.3%, immune-excluded when the immune-excluded score was ≥ 33.3% and the inflamed score was < 33.3%, and immune-desert otherwise. Investigation of Transcriptomics and Mutational Characteristics of Immuno Phenotype To investigate the transcriptomics and mutational characteristics of IPs, we used publicly available H&E slides images, clinical, mutation data, and transcriptomic data for 36 patients with BTC from TCGA. Clinical information, mRNA expression data, and gene-level somatic mutation data were downloaded from the cBioPortal (https://www.cbioportal.org) and the GDC portal (https://portal.gdc.cancer.gov), respectively. Pathway enrichment scores were calculated using the R function "gsva". Hallmark gene sets from the Molecular Signature Database (MSigDB) were used. Gene Set Enrichment Analysis (GSEA) was performed to compare inflamed IP and non-inflamed IP groups using gseGO in the clusterProfiler package (version 4.4.4). The gseGO was carried out with the following options: Type="SYMBOL", pvalue Cutoff=0.05, OrgDb=org.Hs.eg.db, and pAdjustMethod="fdr". Transcriptomic signatures were determined by measuring cell abundance using CIBERSORT with the LM22 matrix 25 . Interferon-gamma (IFNG) signaling, cytolytic scores, and the immunologic constant of rejection (ICR) signature were calculated using the methods described by Ayers et al. 26 , Rooney et al. 27 , and Bertucci et al. 28 , respectively. TME subtype was classified based on the method described by Bagaev et al. 29 . The somatic mutation data were filtered based on a variant allele frequency (VAF) of ≥ 3%, a total tumor depth of ≥ 10, and an alternative tumor depth of ≥ 2. Non-synonymous coding mutations were used to calculate tumor mutation load (TMB). Patients cohort For the correlational study of IPs in BTC patients treated with anti-PD-1 therapy, patient data and pathological slides were retrospectively collected from two tertiary referral cancer institutions (N = 339; Asan Medical Center [AMC] and Yonsei Cancer Center [YCC], Seoul, Republic of Korea). Key eligibility criteria included histologically confirmed unresectable or metastatic BTC (intrahepatic and extrahepatic cholangiocarcinoma, and gallbladder cancer), radiological progression on first-line GemCis, anti-PD-1 monotherapy as second- or subsequent-line therapy, and no prior ICIs before anti-PD-1 therapy. If there is no clinical data available for tumor response following anti-PD-1 therapy, or if there is no archival tissue or slides, patients were excluded. This study was approved by the Institutional Review Board (IRB) of each participating center (AMC; S2023-0521-0001, and YCC; 4-2020-1378), and conducted in accordance with the Declaration of Helsinki. Informed consent was waived due to the retrospective nature of this study and the inclusion of the overall patient population. In the sub-cohort for multicolor flow cytometry analysis using peripheral blood mononuclear cells (PBMCs), clinical data and blood were prospectively collected, and written informed consent was provided (ClinicalTrials.gov identifier, NCT03695952, and IRB approval number 2019-1231). PD-L1 Immunohistochemistry PD-L1 IHC was performed using the PD-L1 IHC 22C3 PharmDx assay and the Dako Autostainer Link 48 (Agilent Technologies, CA, USA) following the manufacturer's instructions. PD-L1 expression was assessed using the combined positive score (CPS), which represents the number of PD-L1 staining cells (e.g., tumor cells, lymphocytes, and macrophages) relative to all viable tumor cells. The interpretation was conducted by an academic pathologist (JS) who received regular training to interpret the 22C3 PharmDx assay. Multi-color Flow Cytometry of PBMCs Multi-color flow cytometry of PBMCs was performed on a subgroup of 29 patients enrolled in the prospective cohort study. Peripheral blood samples were collected at baseline (i.e., cycle 1 day 1). Plasma was obtained by centrifuging whole blood. PBMCs were isolated using standard Ficoll-Paque (GE Healthcare) density gradient centrifugation. Plasma samples were stored at -70°C. PBMCs were resuspended in freezing medium (RPMI 1640; Corning) supplemented with 20% fetal bovine serum (FBS) and 10% DMSO, and then stored in liquid nitrogen until use. After thawing, PBMCs were stained with the LIVE/DEAD Fixable Red Dead Cell Stain Kit (Invitrogen, Waltham, MA, USA) to exclude dead cells. The cells were washed once, stained with fluorochrome-conjugated antibodies against surface markers for 30 minutes at 4°C, and then washed again. For intracellular staining, surface-stained cells were permeabilized using a Foxp3 staining buffer kit (eBioscience, San Diego, CA, USA), and then further stained for intracellular proteins. For dextramer staining, the cells were pretreated with 50 nmol/L dasatinib (Axon Medchem, Groningen, Netherlands) at 37°C for 30 minutes 41 . Then these cells were stained with dextramers for 20 minutes at room temperature, washed twice, and then stained using the described protocols. Flow cytometry was performed using an LSR II instrument and FACSDiva software (BD Biosciences, San Jose, CA, USA). The data were analyzed using FlowJo software (Treestar, San Francisco, CA, USA). The proportions of various immune cell populations in PBMCs at baseline were analyzed per IP. Statistical Analysis Categorical variables were compared using the chi-squared test or Fisher's exact test as appropriate, and the differences for continuous variables between groups were assessed by the Kruskal-Wallis test. OS was defined as the duration from the start of anti-PD-1 therapy to the date of death from any cause. PFS was defined as the duration from the start of anti-PD-1 therapy to the date of disease progression determined by the RECIST v1.1. PFS with GemCis was defined as the duration from the initiation of first-line GemCis to the date of disease progression according to the RECIST v1.1. The Kaplan–Meier method was utilized to estimate survival outcomes, which were compared between IPs using the log‐rank test. Univariate and multivariable analyses of PFS and OS were performed using Cox proportional hazards models. Interactions between the IPs and key baseline clinicopathological characteristics were analyzed using a stratified Cox proportional hazards model. A P‐value of < 0.05 was considered statistically significant. All statistical analyses were conducted using R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria). Methods-only References 37. Ren S, He K, Girshick R, et al: Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 39:1137-1149, 2017 38. Chen L-C, Zhu Y, Papandreou G, et al: Encoder-decoder with atrous separable convolution for semantic image segmentation, Proceedings of the European conference on computer vision (ECCV), 2018, pp 801-818 39. He K, Zhang X, Ren S, et al: Deep residual learning for image recognition, Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp 770-778 40. Kinga D, Adam JB: A method for stochastic optimization In: International Conference on Learning Representations (ICLR). arXiv preprint arXiv:1412.6980, 2014 41. Dolton G, Tungatt K, Lloyd A, et al: More tricks with tetramers: a practical guide to staining T cells with peptide-MHC multimers. Immunology 146:11-22, 2015 Additional Declarations Yes there is potential Competing Interest. CY received honoraria from Servier, Bayer, AstraZeneca, Merck Sharp & Dohme, Eisai, Celgene, Bristol Myers Squibb, Ipsen, Novartis, Boryung Pharmaceuticals, Mundipharma, and Roche; and received research grants from Servier, Bayer, AstraZeneca, Ono Pharmaceuticals, Ipsen, Boryung Pharmaceuticals, and Lunit Inc. C-kL received honoraria from AstraZeneca, Servier, Dong-A ST, Boryung Pharmaceuticals, Mundipharma, and Roche; consulting fees from Roche, and Daiichi Sankyo; and received research grants or supports from Ono Pharmaceuticals, Celltrion, Boryung Pharmaceuticals, GC Biopharma and Lunit Inc. HDK received honoraria from AstraZeneca, Bristol Myers Squibb, Ono Pharmaceuticals, Boryung Pharmaceuticals, and Boostimmune, received research grants from AstraZeneca and served as a consultant for Mustbio. RSF reports grants or contracts to their institution from Adaptimmune, Bayer, Bristol Myers Squibb, Eisai, Eli Lilly, Pfizer, Roche, and Genentech; consulting fees to themself from Merck, AstraZeneca, Bayer, Bristol Myers Squibb, Exelixis, Cstone, Hengrui, Eisai, Eli Lilly, MSD, Pfizer, Roche, and Genentech; payment or honoraria to themself from Genentech; and participation on a data safety monitoring or advisory board from AstraZeneca, and Hengrui.CO. HJC reports an advisory role at AstraZeneca and Roche. CO, SK, YL, and GP are employees, and CHA and CYO are employees and stock holder of Lunit Inc. Supplementary Files ExtendedFigure231225.pdf Extended Figure 1-6 Supplementarymaterial231225.docx Extended Figure Lgend Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3839367","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":266129753,"identity":"f122a7e5-edd6-47ef-92a1-8ad6ecbfdc47","order_by":0,"name":"Changhoon Yoo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvElEQVRIiWNgGAWjYLCCD3BWAYMBUToYZ8CZBkRqYeYhSYvB8eaDn21qtskz8B8++JjHgMHYvIGQljPHkqVzjt02bJBISzYGajGTOUBIy40cM+YcttsJDBI8ZtJALTYSBB0G0mLxD6iF//w3ErQwtgG1MOSwgbSYEdQiCfSLZG/fbcM2iTRjwzkGEsYEtfABQ+zDj2+35fn5Dz988KbCxnAGIS0KB6AMNghF0A4GBvkGwmpGwSgYBaNgpAMA+u01sCfJHkwAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-1451-8455","institution":"Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":true,"prefix":"","firstName":"Changhoon","middleName":"","lastName":"Yoo","suffix":""},{"id":266129754,"identity":"f724b848-a280-4d9b-a716-99df63a4c77e","order_by":1,"name":"Yeong Hak Bang","email":"","orcid":"","institution":"Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":false,"prefix":"","firstName":"Yeong","middleName":"Hak","lastName":"Bang","suffix":""},{"id":266129755,"identity":"f3bbf950-b54c-47be-ab6c-eb17eef5e341","order_by":2,"name":"Choong-kun Lee","email":"","orcid":"https://orcid.org/0000-0001-5151-5096","institution":"Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Choong-kun","middleName":"","lastName":"Lee","suffix":""},{"id":266129756,"identity":"2854c4ce-b64a-4e39-961c-aafcbdbbe41e","order_by":3,"name":"Kyunghye Bang","email":"","orcid":"","institution":"Department of Oncology, Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":false,"prefix":"","firstName":"Kyunghye","middleName":"","lastName":"Bang","suffix":""},{"id":266129757,"identity":"c7b219a2-639e-4114-95ff-e808a3637d92","order_by":4,"name":"Hyung-Don Kim","email":"","orcid":"","institution":"Department of Oncology, Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":false,"prefix":"","firstName":"Hyung-Don","middleName":"","lastName":"Kim","suffix":""},{"id":266129758,"identity":"4e923e28-74d6-4df6-bbf1-fc31bd88a7b5","order_by":5,"name":"Baek-Yeol Ryoo","email":"","orcid":"https://orcid.org/0000-0002-9052-833X","institution":"Asan Medical Center, University of Ulsan College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Baek-Yeol","middleName":"","lastName":"Ryoo","suffix":""},{"id":266129759,"identity":"2a9826ef-7ceb-4940-aaca-dfb8ca2764d3","order_by":6,"name":"Kyu-pyo Kim","email":"","orcid":"","institution":"Department of Oncology, Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":false,"prefix":"","firstName":"Kyu-pyo","middleName":"","lastName":"Kim","suffix":""},{"id":266129760,"identity":"8ee046a7-4b8f-415d-bbf6-62ffe0afeefe","order_by":7,"name":"Jae Ho Jeong","email":"","orcid":"","institution":"Asan Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Jae","middleName":"Ho","lastName":"Jeong","suffix":""},{"id":266129761,"identity":"9d73040c-0c7b-455b-b32d-ffe2673f73e5","order_by":8,"name":"Inkeun Park","email":"","orcid":"","institution":"Department of Oncology, Asan Medical Center, University of Ulsan College of Medicine, 88, Olympic-ro 43-gil, Songpa-gu, Seoul 05505, Republic of Korea.","correspondingAuthor":false,"prefix":"","firstName":"Inkeun","middleName":"","lastName":"Park","suffix":""},{"id":266129762,"identity":"d8add282-a67d-4c2b-bcd1-3bf7b4cd58d8","order_by":9,"name":"Dong Ki Lee","email":"","orcid":"","institution":"Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"Ki","lastName":"Lee","suffix":""},{"id":266129763,"identity":"28c8c4cf-dc95-4f27-aaae-2bfe538e1196","order_by":10,"name":"Hye Jin Choi","email":"","orcid":"","institution":"Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hye","middleName":"Jin","lastName":"Choi","suffix":""},{"id":266129764,"identity":"fb7dbfbf-6b2b-4523-bef5-7048937a2010","order_by":11,"name":"Taek Chung","email":"","orcid":"","institution":"Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Taek","middleName":"","lastName":"Chung","suffix":""},{"id":266129765,"identity":"18868a38-37af-42ab-a971-79a5a736a95e","order_by":12,"name":"Seung Hyuck Jeon","email":"","orcid":"","institution":"Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology (KAIST)","correspondingAuthor":false,"prefix":"","firstName":"Seung","middleName":"Hyuck","lastName":"Jeon","suffix":""},{"id":266129766,"identity":"c042677f-4f4f-4056-a464-1bd3e0af55f5","order_by":13,"name":"Eui-Cheol Shin","email":"","orcid":"","institution":"National Institutes of Health, DHHS","correspondingAuthor":false,"prefix":"","firstName":"Eui-Cheol","middleName":"","lastName":"Shin","suffix":""},{"id":266129767,"identity":"9f7f2d50-726f-44d7-bb5a-ad0ae4230822","order_by":14,"name":"Chiyoon Oum","email":"","orcid":"","institution":"Lunit","correspondingAuthor":false,"prefix":"","firstName":"Chiyoon","middleName":"","lastName":"Oum","suffix":""},{"id":266129768,"identity":"e416409f-d14c-44d3-b2d2-fd8f009b326e","order_by":15,"name":"Seulki Kim","email":"","orcid":"https://orcid.org/0000-0001-5505-1022","institution":"Lunit","correspondingAuthor":false,"prefix":"","firstName":"Seulki","middleName":"","lastName":"Kim","suffix":""},{"id":266129769,"identity":"d2b7373a-89c6-4cc9-a9b4-8de38d08072d","order_by":16,"name":"Yoojoo Lim","email":"","orcid":"","institution":"Lunit","correspondingAuthor":false,"prefix":"","firstName":"Yoojoo","middleName":"","lastName":"Lim","suffix":""},{"id":266129770,"identity":"3fe07f02-cd17-4fc0-98f8-7de9381da0e6","order_by":17,"name":"Gahee Park","email":"","orcid":"","institution":"Lunit","correspondingAuthor":false,"prefix":"","firstName":"Gahee","middleName":"","lastName":"Park","suffix":""},{"id":266129771,"identity":"0f9d8bef-175c-4634-b7ec-7cfdf939e927","order_by":18,"name":"Changho Ahn","email":"","orcid":"","institution":"Lunit Inc.","correspondingAuthor":false,"prefix":"","firstName":"Changho","middleName":"","lastName":"Ahn","suffix":""},{"id":266129772,"identity":"33d9df3d-ce23-4ba3-a91c-391b16e31dab","order_by":19,"name":"Richard Finn","email":"","orcid":"","institution":"David Geffen School of Medicine at UCLA","correspondingAuthor":false,"prefix":"","firstName":"Richard","middleName":"","lastName":"Finn","suffix":""},{"id":266129773,"identity":"91a98d62-774c-40bf-9cca-2de673a73325","order_by":20,"name":"Chan-Young Ock","email":"","orcid":"","institution":"Lunit","correspondingAuthor":false,"prefix":"","firstName":"Chan-Young","middleName":"","lastName":"Ock","suffix":""},{"id":266129774,"identity":"57f7b091-d278-49b0-9dd5-d30fe0a78e09","order_by":21,"name":"Jinho Shin","email":"","orcid":"","institution":"Department of Pathology, Asan Medical Center, University of Ulsan College of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jinho","middleName":"","lastName":"Shin","suffix":""}],"badges":[],"createdAt":"2024-01-06 09:25:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3839367/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3839367/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49434420,"identity":"daa702e5-5ba0-4dba-9d21-fe4736a6f784","added_by":"auto","created_at":"2024-01-10 19:31:18","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1088462,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTranscriptomics and mutational characteristics of immune phenotype assessed by AI-powered\u003c/strong\u003e \u003cstrong\u003espatial analysis of TIL\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea. \u003c/strong\u003eClassification of the grid-level immune phenotype (IP) and scoring of the whole slide image (WSI)-level immune phenotype scores. Each WSI was divided into 0.25 mm\u003csup\u003e2\u003c/sup\u003e grids, with the IP of each grid classified based on the following criteria: grid-level IIP, if the TIL density within the total cancer area (CA) in the grid is ≥ 130/mm\u003csup\u003e2\u003c/sup\u003e; grid-level Immune-excluded IP, if the TIL density within the total CA is \u0026lt;130mm2 and that within the total cancer stroma (CS) is \u0026gt;260/mm2; grid-level immune-desert IP, if the TIL densities are below the threshold in both total CA and CS within the grids. The overall WSI-level inflamed score, immune-excluded score, and immune-desert score were calculated by dividing the number of grids with that respective phenotype by the total number of grids analyzed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb. \u003c/strong\u003eGene set variation analysis across immune phenotypes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ec. \u003c/strong\u003eGene ontology enrichment analysis between inflamed and non-inflamed immune phenotypes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ed.\u003c/strong\u003e Comparison of cytolytic, interferon-gamma (IFN-γ) signature and immunologic constant of rejection (ICR) scores and\u003cem\u003e \u003c/em\u003etumor mutation burden (TMB)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ee.\u003c/strong\u003e Mutational landscape of biliary tract cancers in the TCGA dataset. The mutation plot shows the most frequent and most clinically meaningful genetic mutations according to the immune phenotypes.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/308d33f7d76de59e30a9462b.jpg"},{"id":49434077,"identity":"df66889b-0028-448f-b739-cc3f59e8bf19","added_by":"auto","created_at":"2024-01-10 19:23:17","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1196974,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverall immune phenotype landscape assessed by AI-powered spatial analysis of TILs\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea.\u003c/strong\u003e Study flow diagram.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb.\u003c/strong\u003e Proportion of inflamed phenotype (red), immune-excluded phenotype (green), and immune-desert phenotype (blue) according to PD-L1 CPS.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ec.\u003c/strong\u003e Clinicopathological landscape of AI-powered immune phenotypes. Proportion of inflamed score (red), immune-excluded score (green), and immune-desert score (blue, first row), representative IP (second row). Histology (adenocarcinoma: yellow, others: orange), tumor location (light purple: extrahepatic, light pink: gallbladder, pink: intrahepatic), PD-L1 CPS (brown: \u0026gt;10, apricot: 1-9, ocher: \u0026lt;1 and light teal: Not available for PD-L1 staining), and best overall response by RECIST v1.1 (purple: complete response or partial response, and blank: stable disease or progressive disease)\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/4f4f180bc731cace0616aed9.jpg"},{"id":49434079,"identity":"5a059188-2b84-47a5-a4e6-e8e4b5b4fc2f","added_by":"auto","created_at":"2024-01-10 19:23:17","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1166336,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSurvival outcomes according to the immune phenotypes assessed by AI-powered spatial analysis of TIL\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea.\u003c/strong\u003e Overall survival and\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eb. \u003c/strong\u003eprogression-free survival with anti-PD-1 therapy according to the immune phenotypes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ec.\u003c/strong\u003e Forest plot of HR (dot) and 95% CI (arrow) for progression-free survival according to each clinicopathologic subgroup.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ed.\u003c/strong\u003e Progression-free survival with first-line gemcitabine plus cisplatin (GemCis) according to the immune phenotypes.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/7c1bff3ce1961c2fa857e8ee.jpg"},{"id":50804956,"identity":"09196746-213d-48be-aa74-8292f33b3ab0","added_by":"auto","created_at":"2024-02-07 15:10:23","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1205307,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/50555328-741d-4a4f-b2c1-2256a61cf313.pdf"},{"id":49434080,"identity":"f4772bf2-0ae1-44be-aeee-7a5aacd95c9b","added_by":"auto","created_at":"2024-01-10 19:23:18","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":873912,"visible":true,"origin":"","legend":"\u003cp\u003eExtended Figure 1-6\u003c/p\u003e","description":"","filename":"ExtendedFigure231225.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/51570787e6f28e1b359c0275.pdf"},{"id":49434078,"identity":"c837e8cd-a8b8-473b-86f3-47e808b9dd5f","added_by":"auto","created_at":"2024-01-10 19:23:17","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":15116,"visible":true,"origin":"","legend":"\u003cp\u003eExtended Figure Lgend\u003c/p\u003e","description":"","filename":"Supplementarymaterial231225.docx","url":"https://assets-eu.researchsquare.com/files/rs-3839367/v1/57d888a00b80b401e4eead60.docx"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nCY received honoraria from Servier, Bayer, AstraZeneca, Merck Sharp \u0026 Dohme, Eisai, Celgene, Bristol Myers Squibb, Ipsen, Novartis, Boryung Pharmaceuticals, Mundipharma, and Roche; and received research grants from Servier, Bayer, AstraZeneca, Ono Pharmaceuticals, Ipsen, Boryung Pharmaceuticals, and Lunit Inc. C-kL received honoraria from AstraZeneca, Servier, Dong-A ST, Boryung Pharmaceuticals, Mundipharma, and Roche; consulting fees from Roche, and Daiichi Sankyo; and received research grants or supports from Ono Pharmaceuticals, Celltrion, Boryung Pharmaceuticals, GC Biopharma and Lunit Inc. HDK received honoraria from AstraZeneca, Bristol Myers Squibb, Ono Pharmaceuticals, Boryung Pharmaceuticals, and Boostimmune, received research grants from AstraZeneca and served as a consultant for Mustbio. RSF reports grants or contracts to their institution from Adaptimmune, Bayer, Bristol Myers Squibb, Eisai, Eli Lilly, Pfizer, Roche, and Genentech; consulting fees to themself from Merck, AstraZeneca, Bayer, Bristol Myers Squibb, Exelixis, Cstone, Hengrui, Eisai, Eli Lilly, MSD, Pfizer, Roche, and Genentech; payment or honoraria to themself from Genentech; and participation on a data safety monitoring or advisory board from AstraZeneca, and Hengrui.CO. HJC reports an advisory role at AstraZeneca and Roche. CO, SK, YL, and GP are employees, and CHA and CYO are employees and stock holder of Lunit Inc.","formattedTitle":"Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a predictive biomarker for immune checkpoint inhibitors in advanced biliary tract cancer","fulltext":[{"header":"MAIN TEXT","content":"\u003cp\u003eBiliary tract cancer (BTC) is a malignancy that arises from the epithelium of the intrahepatic and extrahepatic bile ducts as well as the gallbladder. Approximately 60\u0026ndash;70% of patients with BTC present with unresectable or metastatic stage, and the prognosis for these patients is poor, with a median overall survival (OS) of 1 year\u003csup\u003e1\u003c/sup\u003e. Although several targeted agents against \u003cem\u003eFGFR2\u003c/em\u003e gene fusion, \u003cem\u003eIDH1\u003c/em\u003e mutations, \u003cem\u003eBRAF V600E\u003c/em\u003e mutation, and \u003cem\u003eERBB2\u003c/em\u003e amplification have shown promising efficacy outcomes, only certain groups of patients could benefit from these agents\u003csup\u003e2\u0026ndash;8\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eImmune checkpoint inhibitors (ICIs) such as programmed cell death-1 (PD-1) and programmed cell death ligand-1 (PD-L1) antibodies have led to significant advancements in treating various types of cancer and have been widely investigated for patients with advanced BTC. In BTC, anti-PD-1 monotherapy demonstrated only modest efficacy, with an objective response rate (ORR) of 5\u0026ndash;13% and a median progression-free survival (PFS) of 1.4\u0026ndash;3.7 months in previously treated patients with unresectable or metastatic BTC\u003csup\u003e9\u0026ndash;11\u003c/sup\u003e. However, two recent pivotal phase 3 trials (Keynote-966 and TOPAZ-1 trials) have demonstrated a statistically significant benefit in OS with the addition of pembrolizumab (anti-PD-1) or durvalumab (anti-PD-L1) in the first-line setting compared to gemcitabine plus cisplatin (GemCis) alone\u003csup\u003e12,13\u003c/sup\u003e. Despite first-in-decade breakthroughs in first-line treatment for advanced BTC and promising long-term survival, there are concerns about the relatively small OS benefits, with a median OS improvement of 1.3 to 1.8 months and a hazard ratio (HR) of 0.80 to 0.83\u003csup\u003e12,13\u003c/sup\u003e. This is particularly concerning given the high cost of ICIs. This underscores the urgent need for reliable biomarkers for the response of patients with advanced BTC to ICIs.\u003c/p\u003e \u003cp\u003eTumor-infiltrating lymphocytes (TILs) play a major role in the anti-tumor activity of ICIs, and the spatial heterogeneity of TILs has been suggested to be associated with the efficacy outcomes of ICIs\u003csup\u003e14,15\u003c/sup\u003e. However, quantifying relevant TILs has been challenging due to the labor-intensive nature of the method and limitations imposed by spatial distribution in whole-slide images (WSI) as well as interobserver heterogeneity\u003csup\u003e16\u0026ndash;18\u003c/sup\u003e. To overcome this limitation, an artificial intelligence (AI)-powered spatial TILs analyzer (Lunit SCOPE IO\u0026reg;, Lunit Inc., Seoul, Republic of Korea) using hematoxylin and eosin (H\u0026amp;E) WSI was developed to classify the immune phenotype (IP) of the tumor microenvironment (TME) into one of three distinct categories: inflamed (TILs distributed intratumorally), immune-excluded (TILs excluded from the cancer stroma), and immune-desert (scant TILs in TME). This AI-powered spatial TILs analysis has demonstrated its effectiveness in stratifying patients with non-small cell lung cancer for outcomes with ICIs\u003csup\u003e19,20\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eGiven the spatial heterogeneity of TILs and their potential correlation with the efficacy outcomes of anti-PD-1 therapy in BTC\u003csup\u003e21,22\u003c/sup\u003e, we hypothesized that biomarkers identified via AI-powered IPs (AI-IPs) may be effective in identifying patient subgroups with a favorable tumor response to anti-PD-1 treatment in BTC. Here, we conducted a multicenter retrospective study to assess the predictive efficacy of AI-IP based on AI-powered spatial TILs analysis in patients with unresectable or metastatic BTC who were treated with anti-PD-1.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n\u003ch2\u003eArtificial Intelligence-powered Immune Phenotype\u003c/h2\u003e\n\u003cp\u003eThe Lunit SCOPE IO model is an AI-IP analyzer that classifies the IP of the TME based on spatial distribution and density of TILs. The model was updated from a previously published version\u003csup\u003e19\u003c/sup\u003e To perform spatial analysis of TIL distribution within WSI of various sizes, each WSI was divided into 0.25 mm\u003csup\u003e2\u003c/sup\u003e grids, with the IP of each grid classified based on the following criteria\u003csup\u003e23,24\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ea): grid-level inflamed IP, grid-level immune-excluded IP, and grid-level immune-desert IP. The overall WSI-level inflamed score, immune-excluded score, and immune-desert score were calculated by dividing the number of grids with the respective phenotype by the total number of grids analyzed. The representative IP for each WSI was as follows: inflamed when the inflamed score was \u0026ge;\u0026thinsp;33.3%, immune-excluded when the immune-excluded score was \u0026ge;\u0026thinsp;33.3% and the inflamed score was \u0026lt;\u0026thinsp;33.3%, and immune-desert otherwise.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n\u003ch2\u003eGenetic and Transcriptomic Landscape of Immune Phenotype\u003c/h2\u003e\n\u003cp\u003eTo evaluate the transcriptomic and mutational characteristics of various AI-IPs in BTC, data and images from The Cancer Genome Atlas (TCGA) were included. In 36 H\u0026amp;E-stained WSIs of the TCGA BTC cohort, 5 (13.9%), 14 (38.9%), and 17 (47.2%) were classified as inflamed, immune-excluded, and immune-desert IP, respectively. Gene Set Enrichment Analysis (GSEA) was performed to compare inflamed IP and non-inflamed IP groups. Transcriptomic signatures were determined by measuring cell abundance using CIBERSORT with the LM22 matrix\u003csup\u003e25\u003c/sup\u003e. Pathway enrichment analysis revealed high enrichment of interferon-gamma (IFN\u0026gamma;) and IL6/JAK/STAT3 in inflamed IP, while non-inflamed IP showed enrichment of epithelial mesenchymal transition, TGF-beta, and WNT beta-catenin signaling (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eb). Gene Set Enrichment Analysis (GSEA) revealed that T cell proliferation pathways are enriched in inflamed IP, while keratinization pathways are enriched in non-inflamed IP (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ec). The inflamed IP exhibited a significantly increased interferon-gamma signature\u003csup\u003e26\u003c/sup\u003e, cytolytic score\u003csup\u003e27\u003c/sup\u003e and immunologic constant of rejection (ICR) score\u003csup\u003e28\u003c/sup\u003e, indicating a more immune-active microenvironment compared to the non-inflamed IP (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ed). When IP was further classified using previously defined transcriptomics-based TME subtypes\u003csup\u003e29\u003c/sup\u003e, most of the inflamed IP were immune-enriched subtypes (4 out of 5, 80.0%), while more than half of the patients with immune-excluded IP and immune-desert IP were classified as fibrotic or immune-enriched/fibrotic subtypes (8 out of 14, 57.1%) and as depleted subtype (11 out of 17, 64.7%), respectively (\u003cstrong\u003eExtended Data\u003c/strong\u003e Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). The CIBERSORT algorithm indicated an enrichment of CD8-positive T cells and M1 macrophages in the inflamed IP (\u003cstrong\u003eExtended Data\u003c/strong\u003e Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). In the non-inflamed IP, oncogenic mutations in \u003cem\u003eKRAS\u003c/em\u003e, \u003cem\u003eERBB2\u003c/em\u003e, and \u003cem\u003ePIK3CA\u003c/em\u003e were observed at a frequency of 6.7% each, while they were not observed in inflamed IP. In addition, the polybromo-1 (\u003cem\u003ePRBM1\u003c/em\u003e) gene, which has been reported as a negative predictive biomarker for ICIs in non-small cell lung cancer\u003csup\u003e30\u003c/sup\u003e, was only observed in non-inflamed IP (26.7%) (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ee).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n\u003ch2\u003eClinical Cohort for Correlative Analysis\u003c/h2\u003e\n\u003cp\u003eFor the correlational study of IPs in BTC patients treated with anti-PD-1 therapy, patient data and pathological slides were retrospectively collected from two tertiary referral cancer institutions (N\u0026thinsp;=\u0026thinsp;339; Asan Medical Center [AMC] and Yonsei Cancer Center [YCC], Seoul, Republic of Korea). Key eligibility criteria included histologically confirmed unresectable or metastatic BTC (intrahepatic and extrahepatic cholangiocarcinoma, and gallbladder cancer), radiological progression on first-line GemCis, anti-PD-1 monotherapy as second- or subsequent-line therapy, and no prior ICIs before anti-PD-1 therapy. In the sub-cohort for multicolor flow cytometry analysis using peripheral blood mononuclear cells (PBMCs), clinical data and blood were prospectively collected.\u003c/p\u003e\n\u003cp\u003eBetween December 2017 and November 2022, a total of 445 patients were treated with second-line or subsequent anti-PD-1 monotherapy (pembrolizumab or nivolumab) for unresectable or metastatic BTC after progression on first-line GemCis at AMC and YCC in Seoul, Korea. Among them, 339 (76.2%) patients were eligible for current analysis after excluding 89 and 17 patients due to the lack of clinical information or archival tissue, and being ineligible for meeting the tissue quality criteria for AI-IP analysis, respectively (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ea). Baseline patient characteristics are summarized in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e. Intrahepatic cholangiocarcinoma was the most prevalent type of BTC (N\u0026thinsp;=\u0026thinsp;155, 45.7%), and 55.8% of patients (N\u0026thinsp;=\u0026thinsp;189) received anti-PD-1 as second-line treatment. Pembrolizumab was administered to 227 (67.0%) patients, while nivolumab was given to 112 (33.0%) patients. Additionally, 43.2% of the patients (137 out of 317) had PD-L1 CPS\u0026thinsp;\u0026ge;\u0026thinsp;1. According to the IP classification, 40 patients (11.8%) were classified as inflamed, 167 patients (49.3%) as immune-excluded, and 132 patients (38.9%) as immune-desert IP. There was a trend showing an increased proportion of inflamed IP in patients with higher PD-L1 CPS (6.1%, 14.0%, and 21.6% in CPS\u0026thinsp;\u0026lt;\u0026thinsp;1, 1\u0026ndash;9, and \u0026ge;\u0026thinsp;10, respectively; P\u0026thinsp;=\u0026thinsp;0.020) (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb-c).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eBaseline characteristics according to immune phenotypes\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTotal\u003c/p\u003e\n\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;339)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eInflamed\u003c/p\u003e\n\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;40)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eImmune-excluded\u003c/p\u003e\n\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;167)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eImmune-desert\u003c/p\u003e\n\u003cp\u003e(N\u0026thinsp;=\u0026thinsp;132)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eP-value\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e \u0026ge; 65 years\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e172 (50.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e24 (60.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e85 (50.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e63 (47.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.396\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eSex\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.095\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMale\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e144 (42.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e12 (30.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e68 (40.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e64 (48.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eFemale\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e195 (57.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e28 (70.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e99 (59.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e68 (51.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eECOG\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.745\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e29 (8.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e4 (10.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e17 (10.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e8 (6.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e250 (73.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e28 (70.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e122 (73.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e100 (75.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e60 (17.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e8 (20.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e28 (16.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e24 (18.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eSite of origin\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eExtrahepatic\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e100 (29.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e7 (17.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e72 (43.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e21 (15.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGallbladder\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e84 (24.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e10 (25.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e35 (21.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e39 (29.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eIntrahepatic\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e155 (45.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e23 (57.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e60 (35.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e72 (54.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eTissue harvest method\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBiopsy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e190 (56.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e16 (40.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e67 (40.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e107 (81.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSurgery\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e149 (44.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e24 (60.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e100 (59.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e25 (18.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eTissue site\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePrimary site\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e271 (79.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e27 (67.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e144 (86.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e100 (75.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMetastatic site\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e43 (12.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e4 (10.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e17 (10.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e22 (16.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLymph node\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e25 (7.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e9 (22.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e6 (3.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e10 (7.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003ePathology\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.155\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAdenocarcinoma\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e311 (91.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e35 (87.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e158 (94.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e118 (89.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOthers\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e28 (8.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e5 (12.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e9 (5.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e14 (10.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003ePD-L1 CPS*\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.020\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026lt; 1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e180 (56.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e11 (32.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e96 (61.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e73 (60.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u0026ndash;9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e86 (27.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e12 (35.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e41 (25.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e33 (27.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026ge; 10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e51 (16.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e11 (32.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e25 (15.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e15 (12.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eTime from tissue harvest to ICI\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;12 months\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e211 (62.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e22 (55.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e87 (52.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e102 (77.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026ge;\u0026thinsp;12 months\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e128 (37.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e18 (45.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e80 (47.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e30 (22.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eICI agent\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.521\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePembrolizumab\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e227 (67.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e25 (62.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e109 (65.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e93 (70.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNivolumab\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e112 (33.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e15 (37.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e58 (34.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e39 (29.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eTreatment line\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.207\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSecond-line\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e189 (55.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e24 (60.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e85 (50.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e80 (60.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eThird-line or later\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e150 (44.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e16 (40.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e82 (49.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e52 (39.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003ctfoot\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"6\"\u003eData are presented as no. (%)\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"6\"\u003e* CPS: combined positive score, PD-L1 status was not assessed in 22 patients.\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tfoot\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n\u003ch2\u003eCorrelation between Immune Phenotype and Efficacy of Anti-PD-1 Therapy\u003c/h2\u003e\n\u003cp\u003eAt a median follow-up of 18.9 months (95% confidence interval [CI], 15.4\u0026ndash;25.9), the median OS, PFS, and overall response rate (ORR) per RECIST v1.1 were 5.7 months (95% CI, 4.7\u0026ndash;6.6), 2.0 months (95% CI, 1.8\u0026ndash;2.4), and 10.0%, respectively, in the study patients as a whole. The median OS in the inflamed IP group was 12.6 months (95% CI, 9.1\u0026ndash;not reached), which was significantly longer than that in the immune-excluded IP group (5.9 months [95% CI, 4.6\u0026ndash;7.7]) and the immune-desert IP group (4.5 months [95% CI, 3.9\u0026ndash;5.6]) (inflamed IP vs. non-inflamed IP, 12.6 vs. 5.1 months, hazard ratio [HR]\u0026thinsp;=\u0026thinsp;0.46 [0.29\u0026ndash;0.73], P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ea). The median PFS in the inflamed IP group was significantly longer than those in the immune-excluded and immune-desert IP groups (4.5 [95% CI, 1.9\u0026ndash;17.6], 2.0 [95% CI, 1.8\u0026ndash;2.6], and 1.9 [95% CI, 1.6\u0026ndash;2.3] months, respectively) (inflamed IP vs. non-inflamed IP, 4.5 vs. 1.9 months, HR\u0026thinsp;=\u0026thinsp;0.49 [95% CI, 0.33\u0026ndash;0.73], P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eb).\u003c/p\u003e\n\u003cp\u003eThe ORR was significantly higher in the inflamed IP group compared to the immune-excluded and immune-desert IP groups (27.5% vs. 6.6% vs. 9.1%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001; inflamed IP vs. non-inflamed IP, 27.5% vs. 7.7%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The association between IPs and survival outcomes was consistent in the subgroup analysis for each AMC and YCC cohorts (\u003cstrong\u003eExtended Data\u003c/strong\u003e Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The favorable survival outcomes of the inflamed IP group compared to the non-inflamed IP group were consistent across the subgroups (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ec, \u003cstrong\u003eExtended Data Fig.\u0026nbsp;4\u003c/strong\u003e). In the multivariable analyses including the potential prognostic factors (P\u0026thinsp;\u0026lt;\u0026thinsp;0.1 in the univariate analyses), inflamed IP was significantly associated with better OS (adjusted HR\u0026thinsp;=\u0026thinsp;0.44 [95% CI, 0.28\u0026ndash;0.69], P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and PFS (adjusted HR\u0026thinsp;=\u0026thinsp;0.50 [95% CI, 0.34\u0026ndash;0.75], P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). To determine whether IP classification can serve as a predictive biomarker of anti-PD-1 therapy, we analyzed the PFS with first-line GemCis in the same dataset (N\u0026thinsp;=\u0026thinsp;334), excluding patients with no PFS data for first-line GemCis; the analysis revealed no significant association between IP and PFS with first-line GemCis (median 2.3 months in the inflamed group, 3.8 months in the immune-excluded group, 2.8 months in the immune-desert group, P\u0026thinsp;=\u0026thinsp;0.340; Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003ed). In addition, PD-L1 CPS was not associated with OS (CPS\u0026thinsp;\u0026lt;\u0026thinsp;1 vs. \u0026ge;1, median 5.9 vs. 5.1 months, P\u0026thinsp;=\u0026thinsp;0.301) and PFS (median 1.9 vs. 2.1 months, P\u0026thinsp;=\u0026thinsp;0.152) (\u003cstrong\u003eExtended Data Fig.\u0026nbsp;5\u003c/strong\u003e).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eMultivariable analyses for progression-free survival and overall survival\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth rowspan=\"3\" align=\"left\"\u003e\n\u003cp\u003eVariables\u003c/p\u003e\n\u003c/th\u003e\n\u003cth colspan=\"4\" align=\"left\"\u003e\n\u003cp\u003eOverall survival\u003c/p\u003e\n\u003c/th\u003e\n\u003cth colspan=\"4\" align=\"left\"\u003e\n\u003cp\u003eProgression-free survival\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003cth colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eunivariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003emultivariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eunivariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003emultivariable\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eHR (95% CI)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eP-value\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eHR (95% CI)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eP-value\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eHR (95% CI)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eP-value\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eHR (95% CI)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eP-value\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eAge\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(\u0026ge;\u0026thinsp;65 years vs. \u0026lt; 65 years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.97\u003c/p\u003e\n\u003cp\u003e(0.76\u0026ndash;1.24)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.799\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.93\u003c/p\u003e\n\u003cp\u003e(0.74\u0026ndash;1.17)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.538\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eSex\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(Male vs. Female)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003cp\u003e(0.71\u0026ndash;1.17)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.460\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.85\u003c/p\u003e\n\u003cp\u003e(0.67\u0026ndash;1.06)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.149\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eECOG\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(2 vs. 0\u0026ndash;1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.97\u003c/p\u003e\n\u003cp\u003e(2.16\u0026ndash;4.09)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.38\u003c/p\u003e\n\u003cp\u003e(2.44\u0026ndash;4.70)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.69\u003c/p\u003e\n\u003cp\u003e(1.99\u0026ndash;3.63)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.63\u003c/p\u003e\n\u003cp\u003e(1.95\u0026ndash;3.55)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eLocation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExtrahepatic\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eReference\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eReference\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGallbladder\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.15\u003c/p\u003e\n\u003cp\u003e(0.83\u0026ndash;1.60)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.410\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.17\u003c/p\u003e\n\u003cp\u003e(0.86\u0026ndash;1.59)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.323\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eIntrahepatic\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.94\u003c/p\u003e\n\u003cp\u003e(0.71\u0026ndash;1.26)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.689\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.05\u003c/p\u003e\n\u003cp\u003e(0.800\u0026ndash;1.37)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.745\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eImmune phenotype\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.46\u003c/p\u003e\n\u003cp\u003e(0.29\u0026ndash;0.73)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.44\u003c/p\u003e\n\u003cp\u003e(0.28\u0026ndash;0.69)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.49\u003c/p\u003e\n\u003cp\u003e(0.33\u0026ndash;0.73)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003cp\u003e(0.34\u0026ndash;0.75)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e(Inflamed vs. Non-inflamed)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003ePD-L1 CPS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(\u0026ge; 1vs. \u0026lt; 1)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.87\u003c/p\u003e\n\u003cp\u003e(0.67\u0026ndash;1.13)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.301\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.84\u003c/p\u003e\n\u003cp\u003e(0.66\u0026ndash;1.07)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.152\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eTreatment line\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(\u0026ge;\u0026thinsp;3 vs. 2)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.31\u003c/p\u003e\n\u003cp\u003e(1.02\u0026ndash;1.67)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.032\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.51\u003c/p\u003e\n\u003cp\u003e(1.18\u0026ndash;1.94)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.19\u003c/p\u003e\n\u003cp\u003e(0.95\u0026ndash;1.49)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.140\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003ctfoot\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"9\"\u003eHR, hazard ratio; CI, confidence interval; CPS, combined positive score; ICI, immune checkpoint inhibitor\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tfoot\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n\u003ch2\u003eImmune Cell Subpopulation in PBMCs According to Immune Phenotype\u003c/h2\u003e\n\u003cp\u003eMulti-color flow cytometry of PBMCs was performed on a subgroup of 29 patients enrolled in the prospective cohort study. Peripheral blood samples were collected at baseline (i.e., cycle 1 day 1 of anti-PD-1 therapy). In this subgroup, 6.9% (N\u0026thinsp;=\u0026thinsp;2), 48.3% (N\u0026thinsp;=\u0026thinsp;14), and 44.8% (N\u0026thinsp;=\u0026thinsp;13) of patients were classified into the inflamed, immune-excluded, and immune-desert IP, respectively. PBMCs from the inflamed IP group showed a higher proportion of CD69\u003csup\u003e+\u003c/sup\u003eCD8\u003csup\u003e+\u003c/sup\u003eT cells, which are a marker of resident memory T cells\u003csup\u003e31,32\u003c/sup\u003e, and CD8\u003csup\u003e+\u003c/sup\u003e effector memory T cells, compared to the non-inflamed IP group (\u003cstrong\u003eExtended Data Fig.\u0026nbsp;6\u003c/strong\u003e).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this study, we showed that AI-IPs classified using AI-powered spatial analysis of TILs on H\u0026amp;E-stained tumor slides were effective in stratifying patients with unresectable or metastatic BTC for the efficacy of anti-PD-1 monotherapy. In the analysis using the TCGA dataset, the inflamed IP showed increased cytolytic activity scores and an interferon-gamma signature, and in the prospective sub-cohort for PBMC analysis, inflamed IP was associated with higher proportions of resident memory T cells and CD8\u003csup\u003e+\u003c/sup\u003e effector memory T cells than non-inflamed IPs; provided biological background. Among the 339 patients included in the clinical cohort for the correlative analysis of AI-IP for its relationship with the efficacy of anti-PD-1 therapy, 11.8% of them were classified into the inflamed IP group, which was significantly associated with better ORR, PFS, and OS with anti-PD-1 monotherapy compared to those in the non-inflamed IP groups (i.e., immune-excluded and immune-desert IPs). As there was no relationship between IPs and PFS with prior first-line GemCis, our findings suggest that this AI-IP model may serve as a predictive biomarker rather than a prognostic marker. To the best of our knowledge, this is one of the first studies to predict the effectiveness of ICI treatment in patients with BTC.\u003c/p\u003e \u003cp\u003eThe inflamed IP group showed significantly higher ORR (27.5% vs. 7.7%) and better PFS (4.5 vs. 1.9 months) and OS (12.6 vs. 5.1 months) compared to the non-inflamed IP groups. The results were consistent in the subgroup analysis for each group of participating institutions. Its association with PFS and OS remained significant in the multivariable analyses, which included key clinicopathological factors. The AI-IP model used in the present analysis also indicated that inflamed IP could predict the efficacy of ICIs in addition to PD-L1 immunohistochemistry (IHC) for patients with advanced non-small cell lung cancer (NSCLC)\u003csup\u003e19\u003c/sup\u003e. In the prospective randomized phase 3 ATTLAS trial comparing atezolizumab-bevacizumab plus chemotherapy with chemotherapy alone, the PFS benefit of adding atezolizumab-bevacizumab was significant in the subgroup with a high inflamed score as assessed by AI-IP\u003csup\u003e20\u003c/sup\u003e. Despite these findings in NSCLC that support our findings, further investigations are required to validate the role of current AI-IP in predicting the efficacy of ICI in BTC, especially using prospective cohorts.\u003c/p\u003e \u003cp\u003eIn the context of BTC, there is no established biomarker for predicting the efficacy of ICIs. PD-L1 expression, assessed using IHC, has been the most extensively investigated biomarker for ICIs\u003csup\u003e33\u003c/sup\u003e; however, it was not predictive of the efficacy of anti-PD-1 monotherapy\u003csup\u003e9\u003c/sup\u003e or GemCis plus pembrolizumab or durvalumab in previous trials\u003csup\u003e12,13\u003c/sup\u003e. Our analysis also showed that PD-L1 CPS was not associated with PFS or OS with anti-PD-1 monotherapy in advanced BTC. Although high microsatellite instability/mismatch repair deficiency (MSI-H/dMMR) is a tissue-agnostic biomarker for anti-PD-1 therapy\u003csup\u003e34\u003c/sup\u003e, its incidence in BTC is extremely rare, with an incidence rate of less than 1% in patients with BTC\u003csup\u003e9,35\u003c/sup\u003e. Furthermore, no genetic alterations were associated with the advantage of administering additional durvalumab in the exploratory analysis of the TOPAZ-1 trial\u003csup\u003e36\u003c/sup\u003e. Our results showing a correlation between AI-IP and efficacy outcomes with anti-PD-1 monotherapy, may have clinical significance given that anti-PD-1/L1 has now become the standard of care for the treatment of unresectable or metastatic BTC. However, approvals for anti-PD-1/L1 agents in BTC are currently limited to combination therapy with GemCis in the first-line setting, and our study did not include these patient populations. Future research is needed to evaluate the impact of AI-IP analysis in patients undergoing treatment with pembrolizumab or durvalumab in combination with GemCis.\u003c/p\u003e \u003cp\u003eThis study has limitations, the most prominent of which is its retrospective design, which introduces unintentional biases. As mentioned earlier, the combination of anti-PD-1/L1 with GemCis is now the standard of care, while we focused on patients treated with anti-PD-1 monotherapy. Furthermore, we could not delineate the relationship between MSI-H/dMMR and AI-IP as it was not assessed in this study. These may limit the application of our findings in daily practice. In addition, while we attempted to analyze the patterns of immune cell subsets in PBMC based on the AI-IP, we were unable to derive meaningful results due to the small sample size in the prospective cohort.\u003c/p\u003e \u003cp\u003eIn conclusion, AI-IP analysis using H\u0026amp;E-stained slides could effectively stratify the IPs in patients with advanced BTC. We found that inflamed IP was predictive of the efficacy outcomes with anti-PD-1 monotherapy. AI-IP could serve as a novel biomarker for patients with advanced BTC treated with ICIs.\u003c/p\u003e"},{"header":"DECLARATIONS","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll requests for raw and analyzed data and materials should be directed to C.Y. and C.-k.L., and will be promptly reviewed by the Asan Medical Center and Yonsei Cancer Center to determine whether the request is subject to any intellectual property or confidentiality obligations. Patient-related data not included in this paper were generated as part of the clinical trial, and may be subject to patient confidentiality. Any data and materials that can be shared will be released via a material transfer agreement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll analyses were conducted using publicly available software, as detailed in the Methods section. The raw scripts used to generate the RNA-seq analysis figures presented in this paper are available at \u003ca href=\"https://github.com/SGI-pan/BTC-IP\"\u003ehttps://github.com/SGI-pan/BTC-IP\u003c/a\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported in part by the Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea [grant number 2020IL0018], and Lunit, Inc. The authors appreciate the patients and their families who generously donated their tissues to TCGA, as well as the members of TCGA who collected and disclosed the valuable data. The authors thank Joon Seo Lim, PhD, from the Scientific Publications Team at Asan Medical Center (Seoul, South Korea) for providing editorial assistance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization: YHB, CYO, CY; Methodology: YHB, CKL, HDK, CYO, JS, CY; Formal analysis: YHB, CKL, CYO, JS, CY; Investigation: YHB, CKL, HDK, CYO, JS, CY; Resources: CKL, KB, HDK, KPK, JHJ, IP, BYR, DKL, HJC, CT, SHJ, ECS, CO, SK, YL, GP, CHA, CYO; Original draft: YHB, CKL, CYO, CY; Writing-review and editing: All authors; Visualization: YHB; Supervision: JS, CY\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCY received honoraria from Servier, Bayer, AstraZeneca, Merck Sharp \u0026amp; Dohme, Eisai, Celgene, Bristol Myers Squibb, Ipsen, Novartis, Boryung Pharmaceuticals, Mundipharma, and Roche; and received research grants from Servier, Bayer, AstraZeneca, Ono Pharmaceuticals, Ipsen, Boryung Pharmaceuticals, and Lunit Inc. C-kL received honoraria from AstraZeneca, Servier, Dong-A ST, Boryung Pharmaceuticals, Mundipharma, and Roche; consulting fees from Roche, and Daiichi Sankyo; and received research grants or supports from Ono Pharmaceuticals, Celltrion, Boryung Pharmaceuticals, GC Biopharma and Lunit Inc. HDK received honoraria from AstraZeneca, Bristol Myers Squibb, Ono Pharmaceuticals, Boryung Pharmaceuticals, and Boostimmune, received research grants from AstraZeneca and served as a consultant for Mustbio. RSF reports grants or contracts to their institution from Adaptimmune, Bayer, Bristol Myers Squibb, Eisai, Eli Lilly, Pfizer, Roche, and Genentech; consulting fees to themself from Merck, AstraZeneca, Bayer, Bristol Myers Squibb, Exelixis, Cstone, Hengrui, Eisai, Eli Lilly, MSD, Pfizer, Roche, and Genentech; payment or honoraria to themself from Genentech; and participation on a data safety monitoring or advisory board from AstraZeneca, and Hengrui.CO. HJC reports an advisory role at AstraZeneca and Roche. CO, SK, YL, and GP are employees, and CHA and CYO are employees and stock holder of Lunit Inc.\u003c/p\u003e"},{"header":"REFERENCES","content":"\u003col\u003e\n\u003cli\u003eValle JW, Kelley RK, Nervi B, et al: Biliary tract cancer. Lancet 397:428-444, 2021\u003c/li\u003e\n\u003cli\u003eGoyal L, Meric-Bernstam F, Hollebecque A, et al: Futibatinib for FGFR2-Rearranged Intrahepatic Cholangiocarcinoma. N Engl J Med 388:228-239, 2023\u003c/li\u003e\n\u003cli\u003eAbou-Alfa GK, Sahai V, Hollebecque A, et al: Pemigatinib for previously treated, locally advanced or metastatic cholangiocarcinoma: a multicentre, open-label, phase 2 study. Lancet Oncol 21:671-684, 2020\u003c/li\u003e\n\u003cli\u003eAbou-Alfa GK, Macarulla T, Javle MM, et al: Ivosidenib in IDH1-mutant, chemotherapy-refractory cholangiocarcinoma (ClarIDHy): a multicentre, randomised, double-blind, placebo-controlled, phase 3 study. Lancet Oncol 21:796-807, 2020\u003c/li\u003e\n\u003cli\u003eJavle M, Borad MJ, Azad NS, et al: Pertuzumab and trastuzumab for HER2-positive, metastatic biliary tract cancer (MyPathway): a multicentre, open-label, phase 2a, multiple basket study. Lancet Oncol 22:1290-1300, 2021\u003c/li\u003e\n\u003cli\u003eSubbiah V, Lassen U, \u0026Eacute;lez E, et al: Dabrafenib plus trametinib in patients with BRAF\u003csup\u003eV600E\u003c/sup\u003e-mutated biliary tract cancer (ROAR): a phase 2, open-label, single-arm, multicentre basket trial. Lancet Oncol 21:1234-1243, 2020\u003c/li\u003e\n\u003cli\u003eYoo C, Hyung J, Chan SL: Recent Advances in Systemic Therapy for Advanced Intrahepatic Cholangiocarcinoma. Liver Cancer:1-17, 2023\u003c/li\u003e\n\u003cli\u003eLee CK, Chon HJ, Cheon J, et al: Trastuzumab plus FOLFOX for HER2-positive biliary tract cancer refractory to gemcitabine and cisplatin: a multi-institutional phase 2 trial of the Korean Cancer Study Group (KCSG-HB19-14). Lancet Gastroenterol Hepatol 8:56-65, 2023\u003c/li\u003e\n\u003cli\u003ePiha-Paul SA, Oh D-Y, Ueno M, et al: Efficacy and safety of pembrolizumab for the treatment of advanced biliary cancer: Results from the KEYNOTE-158 and KEYNOTE-028 studies. Int J Cancer 147:2190-2198, 2020\u003c/li\u003e\n\u003cli\u003eKim RD, Chung V, Alese OB, et al: A Phase 2 Multi-institutional Study of Nivolumab for Patients With Advanced Refractory Biliary Tract Cancer. JAMA Oncol 6:888-894, 2020\u003c/li\u003e\n\u003cli\u003eUeno M, Ikeda M, Morizane C, et al: Nivolumab alone or in combination with cisplatin plus gemcitabine in Japanese patients with unresectable or recurrent biliary tract cancer: a non-randomised, multicentre, open-label, phase 1 study. Lancet Gastroenterol Hepatol 4:611-621, 2019\u003c/li\u003e\n\u003cli\u003eOh D-Y, Ruth He A, Qin S, et al: Durvalumab plus Gemcitabine and Cisplatin in Advanced Biliary Tract Cancer. NEJM Evidence 1:EVIDoa2200015, 2022\u003c/li\u003e\n\u003cli\u003eKelley RK, Ueno M, Yoo C, et al: Pembrolizumab in combination with gemcitabine and cisplatin compared with gemcitabine and cisplatin alone for patients with advanced biliary tract cancer (KEYNOTE-966): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet 401:1853-1865, 2023\u003c/li\u003e\n\u003cli\u003eLopez de Rodas M, Nagineni V, Ravi A, et al: Role of tumor infiltrating lymphocytes and spatial immune heterogeneity in sensitivity to PD-1 axis blockers in non-small cell lung cancer. J Immunother Cancer 10, 2022\u003c/li\u003e\n\u003cli\u003eLiu D, Heij LR, Czigany Z, et al: The role of tumor-infiltrating lymphocytes in cholangiocarcinoma. J Exp Clin Cancer Res 41:127, 2022\u003c/li\u003e\n\u003cli\u003eVan Bockstal MR, Fran\u0026ccedil;ois A, Altinay S, et al: Interobserver variability in the assessment of stromal tumor-infiltrating lymphocytes (sTILs) in triple-negative invasive breast carcinoma influences the association with pathological complete response: the IVITA study. Mod Pathol 34:2130-2140, 2021\u003c/li\u003e\n\u003cli\u003eKhoury T, Peng X, Yan L, et al: Tumor-Infiltrating Lymphocytes in Breast Cancer: Evaluating Interobserver Variability, Heterogeneity, and Fidelity of Scoring Core Biopsies. Am J Clin Pathol 150:441-450, 2018\u003c/li\u003e\n\u003cli\u003eSwisher SK, Wu Y, Castaneda CA, et al: Interobserver Agreement Between Pathologists Assessing Tumor-Infiltrating Lymphocytes (TILs) in Breast Cancer Using Methodology Proposed by the International TILs Working Group. Ann Surg Oncol 23:2242-8, 2016\u003c/li\u003e\n\u003cli\u003ePark S, Ock C-Y, Kim H, et al: Artificial Intelligence\u0026ndash;Powered Spatial Analysis of Tumor-Infiltrating Lymphocytes as Complementary Biomarker for Immune Checkpoint Inhibition in Non\u0026ndash;Small-Cell Lung Cancer. J Clin Oncol 40:1916-1928, 2022\u003c/li\u003e\n\u003cli\u003ePark S, Kim TM, Han J-Y, et al: A Phase 3, Randomized study of atezolizumab plus bevacizumab and chemotherapy in patients with EGFR or ALK mutated in non-small cell lung cancer (ATTLAS, KCSG-LU19-04). Journal of Clinical Oncology 0:10.1200/JCO.23.01891\u003c/li\u003e\n\u003cli\u003eYoon JG, Kim MH, Jang M, et al: Molecular Characterization of Biliary Tract Cancer Predicts Chemotherapy and Programmed Death 1/Programmed Death-Ligand 1 Blockade Responses. Hepatology 74:1914-1931, 2021\u003c/li\u003e\n\u003cli\u003eKim HD, Kim JH, Ryu YM, et al: Spatial Distribution and Prognostic Implications of Tumor-Infiltrating FoxP3- CD4+ T Cells in Biliary Tract Cancer. Cancer Res Treat 53:162-171, 2021\u003c/li\u003e\n\u003cli\u003eAmgad M, Stovgaard ES, Balslev E, et al: Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group. NPJ Breast Cancer 6:16, 2020\u003c/li\u003e\n\u003cli\u003eChen DS, Mellman I: Elements of cancer immunity and the cancer\u0026ndash;immune set point. Nature 541:321-330, 2017\u003c/li\u003e\n\u003cli\u003eNewman AM, Steen CB, Liu CL, et al: Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol 37:773-782, 2019\u003c/li\u003e\n\u003cli\u003eAyers M, Lunceford J, Nebozhyn M, et al: IFN-\u0026gamma;-related mRNA profile predicts clinical response to PD-1 blockade. J Clin Invest 127:2930-2940, 2017\u003c/li\u003e\n\u003cli\u003eRooney MS, Shukla SA, Wu CJ, et al: Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell 160:48-61, 2015\u003c/li\u003e\n\u003cli\u003eBertucci F, Finetti P, Simeone I, et al: The immunologic constant of rejection classification refines the prognostic value of conventional prognostic signatures in breast cancer. Br J Cancer 119:1383-1391, 2018\u003c/li\u003e\n\u003cli\u003eBagaev A, Kotlov N, Nomie K, et al: Conserved pan-cancer microenvironment subtypes predict response to immunotherapy. Cancer Cell 39:845-865.e7, 2021\u003c/li\u003e\n\u003cli\u003eZhou H, Liu J, Zhang Y, et al: PBRM1 mutation and preliminary response to immune checkpoint blockade treatment in non-small cell lung cancer. NPJ Precis Oncol 4:6, 2020\u003c/li\u003e\n\u003cli\u003eKumar BV, Ma W, Miron M, et al: Human Tissue-Resident Memory T Cells Are Defined by Core Transcriptional and Functional Signatures in Lymphoid and Mucosal Sites. Cell Rep 20:2921-2934, 2017\u003c/li\u003e\n\u003cli\u003eLuoma AM, Suo S, Wang Y, et al: Tissue-resident memory and circulating T cells are early responders to pre-surgical cancer immunotherapy. Cell 185:2918-2935.e29, 2022\u003c/li\u003e\n\u003cli\u003eDavis AA, Patel VG: The role of PD-L1 expression as a predictive biomarker: an analysis of all US Food and Drug Administration (FDA) approvals of immune checkpoint inhibitors. J Immunother Cancer 7:278, 2019\u003c/li\u003e\n\u003cli\u003eLe DT, Uram JN, Wang H, et al: PD-1 Blockade in Tumors with Mismatch-Repair Deficiency. N Engl J Med 372:2509-2520, 2015\u003c/li\u003e\n\u003cli\u003eLowery MA, Ptashkin R, Jordan E, et al: Comprehensive Molecular Profiling of Intrahepatic and Extrahepatic Cholangiocarcinomas: Potential Targets for Intervention. Clin Cancer Res 24:4154-4161, 2018\u003c/li\u003e\n\u003cli\u003eValle JW, Qin S, Antonuzzo L, et al: 68O Impact of mutation status on efficacy outcomes in TOPAZ-1: A phase III study of durvalumab (D) or placebo (PBO) plus gemcitabine and cisplatin (+GC) in advanced biliary tract cancer (BTC). Ann Oncol 33:S1457, 2022\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"METHODS","content":"\u003cp\u003e\u003cstrong\u003eDevelopment of an artificial intelligence-powered immune phenotype (AI-IP) analyzer\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe current version of AI-IP (Lunit SCOPE IO\u003csup\u003e\u0026reg;\u003c/sup\u003e) used in this study contains updated versions of the cell detection AI model and the tissue segmentation AI model, compared to previous versions\u003csup\u003e19\u003c/sup\u003e. This section describes the differences compared to the previous models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eCell Detection Model\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe cell detection model identifies the location of lymphocytes and tumor cells. This model is based on a convolutional neural network (CNN). In the previous model, a Faster R-CNN\u003csup\u003e37\u003c/sup\u003e architecture was used. In contrast, the current version utilizes a DeepLabV3+\u003csup\u003e38\u003c/sup\u003e architecture with a Resnet-34\u003csup\u003e39\u003c/sup\u003e backbone as the feature extractor. With this model, we framed the detection task as a dense pixel prediction problem. To train the network, a circle with a radius of 0.95 \u0026mu;m was drawn around the point annotation of each cell. The class value associated with the cell at that location (lymphocyte or tumor cell) is then assigned. The training patches cover an area of 151,414 \u0026mu;m\u003csup\u003e2\u003c/sup\u003e (linearly resized to an image of 2048 x 2048 pixels). In each training step, a random portion of images sized 1024 x 1024 pixels is cropped, and this cropped image undergoes on-the-fly data augmentation. The model generates probability maps of size 256 x 256, which are then linearly interpolated to match the original input dimensions (1024 x 1024 pixels), ensuring a 1-to-1 pixel correspondence with the pixel annotations. Since the model predicts the likelihood of the cells being at each pixel, a post-processing stage is required to extract the cell locations. In this stage, the likelihood maps undergo Gaussian filtering (\u0026sigma; = 3), followed by local maximum detection with a radius of 0.57 \u0026mu;m. The model was optimized using the Adam optimizer\u003csup\u003e40\u003c/sup\u003e with a learning rate of 0.002. The learning rate was decayed by a multiplicative factor of 0.2 when the validation loss did not decrease for a period of 5 epochs. Mini-batches of 24 samples were used. The model\u0026apos;s performance in detecting lymphocytes and tumor cells, as measured by the F1-score, was 0.69 and 0.74, respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTissue Segmentation Model\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe AI model was similar to the previous version, except that the backbone for feature extraction was ResNet-34\u003csup\u003e39\u003c/sup\u003e Similarly, it determines whether a pixel belongs to the cancer area, cancer stroma, or background regions. The model was trained using patches that covered an area of 605,657 \u0026mu;m\u003csup\u003e2\u003c/sup\u003e, which were then linearly resized to images sized 1024 x 1024 pixels. The model generates probability maps of size 256 x 256, which are then linearly interpolated to match the original input dimensions (1024 x 1024 pixels), ensuring a 1-to-1 pixel correspondence with the pixel annotations. In this case, there was an unbalanced sampling of patches from different types of organs. The segmentation model was optimized using the Adam optimizer\u003csup\u003e40\u003c/sup\u003e with a learning rate of 0.0001. The learning rate was decayed by a multiplicative factor of 0.5 when the validation loss did not decrease for a period of 5 epochs. Mini-batches of 32 samples were used. The model\u0026apos;s performance in segmenting the cancer area (CA) and cancer stroma (CS) was evaluated using the Intersection-over-Union metric, yielding scores of 0.78 and 0.64, respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eDatasets for developing the AI models\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe WSIs that were used to develop the previously described AI models included more than 16 different primary tumor types and origins. The \u003cem\u003eCell Detection Model\u003c/em\u003e was developed using patches extracted from 3,334 WSIs (N=2,485 for training and N=849 for tuning). From these WSIs, 5,698 and 1,925 patches (151,414 \u0026mu;m\u003csup\u003e2\u003c/sup\u003e per patch) were extracted for training and tuning, respectively. The \u003cem\u003eTissue Segmentation Model\u003c/em\u003e was developed using patches extracted from 15,830 WSIs (N=15,004 for training, and N=826 for tuning). From these WSIs, 56,545 and 2,971 patches (605,657 \u0026mu;m\u003csup\u003e2\u003c/sup\u003e per patch) were extracted for training and validation, respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTissue Quality Criteria\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eQualified H\u0026amp;E-stained slides were included in the current analysis. Criteria for tissue quality included: (1) total cancer area in WSI \u0026ge; 0.125 mm\u003csup\u003e2\u003c/sup\u003e and (2) total number of analyzed grids (containing cancer area or cancer stroma \u0026ge; 5% in a 0.25 mm\u003csup\u003e2\u003c/sup\u003e-sized grid) in WSI \u0026ge; 10.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eClassification of Artificial Intelligence-powered Immune Phenotype\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Lunit SCOPE IO model is an AI-IP analyzer that classifies the IP of the TME based on spatial distribution and density of TILs. The model was updated from a previously published version\u003csup\u003e19\u003c/sup\u003e through additional training and optimization (tuning) using 14.6\u0026times;10\u003csup\u003e10\u003c/sup\u003e \u0026mu;m\u003csup\u003e2\u003c/sup\u003e H\u0026amp;E-stained tissue regions and 6.2 x 10\u003csup\u003e5 \u003c/sup\u003eTIL from 18,679 H\u0026amp;E stained WSI of 17 solid tumor types, including BTC, which were not utilized in the development of the original AI models. To perform spatial analysis of TIL distribution within WSI of various sizes, each WSI was divided into 0.25 mm\u003csup\u003e2\u003c/sup\u003e grids, with the IP of each grid classified based on the following criteria\u003csup\u003e23,24\u003c/sup\u003e: grid-level inflamed IP, if the TIL density within the total cancer area (CA) in the grid is \u0026ge; 130/mm\u003csup\u003e2\u003c/sup\u003e (where CA refers to cancer epithelium or, in the case of non-epithelial tumors, the non-stromal tumor cells); grid-level immune-excluded IP, if the TIL density within the total CA is \u0026lt; 130/mm\u003csup\u003e2\u003c/sup\u003e and that within the total cancer stroma (CS) is \u0026gt; 260/mm\u003csup\u003e2\u003c/sup\u003e; grid-level immune-desert IP, if the TIL densities are below threshold in both the CA and CS within the grid. The overall WSI-level inflamed score, immune-excluded score, and immune-desert score were calculated by dividing the number of grids with the respective phenotype by the total number of grids analyzed. The representative IP for each WSI was as follows: inflamed when the inflamed score was \u0026ge; 33.3%, immune-excluded when the immune-excluded score was \u0026ge; 33.3% and the inflamed score was \u0026lt; 33.3%, and immune-desert otherwise. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInvestigation of Transcriptomics and Mutational Characteristics of Immuno Phenotype \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo investigate the transcriptomics and mutational characteristics of IPs, we used publicly available H\u0026amp;E slides images, clinical, mutation data, and transcriptomic data for 36 patients with BTC from TCGA. Clinical information, mRNA expression data, and gene-level somatic mutation data were downloaded from the cBioPortal (https://www.cbioportal.org) and the GDC portal (https://portal.gdc.cancer.gov), respectively. Pathway enrichment scores were calculated using the R function \u0026quot;gsva\u0026quot;. Hallmark gene sets from the Molecular Signature Database (MSigDB) were used. Gene Set Enrichment Analysis (GSEA) was performed to compare inflamed IP and non-inflamed IP groups using gseGO in the clusterProfiler package (version 4.4.4). The gseGO was carried out with the following options: Type=\u0026quot;SYMBOL\u0026quot;, pvalue Cutoff=0.05, OrgDb=org.Hs.eg.db, and pAdjustMethod=\u0026quot;fdr\u0026quot;. Transcriptomic signatures were determined by measuring cell abundance using CIBERSORT with the LM22 matrix\u003csup\u003e25\u003c/sup\u003e. Interferon-gamma (IFNG) signaling, cytolytic scores, and the immunologic constant of rejection (ICR) signature were calculated using the methods described by Ayers et al.\u003csup\u003e26\u003c/sup\u003e, Rooney et al.\u003csup\u003e27\u003c/sup\u003e, and Bertucci et al.\u003csup\u003e28\u003c/sup\u003e, respectively. TME subtype was classified based on the method described by Bagaev et al.\u003csup\u003e29\u003c/sup\u003e. The somatic mutation data were filtered based on a variant allele frequency (VAF) of \u0026ge; 3%, a total tumor depth of \u0026ge; 10, and an alternative tumor depth of \u0026ge; 2. Non-synonymous coding mutations were used to calculate tumor mutation load (TMB). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePatients cohort\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor the correlational study of IPs in BTC patients treated with anti-PD-1 therapy, patient data and pathological slides were retrospectively collected from two tertiary referral cancer institutions (N = 339; Asan Medical Center [AMC] and Yonsei Cancer Center [YCC], Seoul, Republic of Korea). Key eligibility criteria included histologically confirmed unresectable or metastatic BTC (intrahepatic and extrahepatic cholangiocarcinoma, and gallbladder cancer), radiological progression on first-line GemCis, anti-PD-1 monotherapy as second- or subsequent-line therapy, and no prior ICIs before anti-PD-1 therapy. If there is no clinical data available for tumor response following anti-PD-1 therapy, or if there is no archival tissue or slides, patients were excluded. This study was approved by the Institutional Review Board (IRB) of each participating center (AMC; S2023-0521-0001, and YCC; 4-2020-1378), and conducted in accordance with the Declaration of Helsinki. Informed consent was waived due to the retrospective nature of this study and the inclusion of the overall patient population. In the sub-cohort for multicolor flow cytometry analysis using peripheral blood mononuclear cells (PBMCs), clinical data and blood were prospectively collected, and written informed consent was provided (ClinicalTrials.gov identifier, NCT03695952, and IRB approval number 2019-1231). \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePD-L1 Immunohistochemistry\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePD-L1 IHC was performed using the PD-L1 IHC 22C3 PharmDx assay and the Dako Autostainer Link 48 (Agilent Technologies, CA, USA) following the manufacturer\u0026apos;s instructions. PD-L1 expression was assessed using the combined positive score (CPS), which represents the number of PD-L1 staining cells (e.g., tumor cells, lymphocytes, and macrophages) relative to all viable tumor cells. The interpretation was conducted by an academic pathologist (JS) who received regular training to interpret the 22C3 PharmDx assay.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMulti-color Flow Cytometry of PBMCs\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMulti-color flow cytometry of PBMCs was performed on a subgroup of 29 patients enrolled in the prospective cohort study. Peripheral blood samples were collected at baseline (i.e., cycle 1 day 1). Plasma was obtained by centrifuging whole blood. PBMCs were isolated using standard Ficoll-Paque (GE Healthcare) density gradient centrifugation. Plasma samples were stored at -70\u0026deg;C. PBMCs were resuspended in freezing medium (RPMI 1640; Corning) supplemented with 20% fetal bovine serum (FBS) and 10% DMSO, and then stored in liquid nitrogen until use. \u003c/p\u003e\n\u003cp\u003eAfter thawing, PBMCs were stained with the LIVE/DEAD Fixable Red Dead Cell Stain Kit (Invitrogen, Waltham, MA, USA) to exclude dead cells. The cells were washed once, stained with fluorochrome-conjugated antibodies against surface markers for 30 minutes at 4\u0026deg;C, and then washed again. For intracellular staining, surface-stained cells were permeabilized using a Foxp3 staining buffer kit (eBioscience, San Diego, CA, USA), and then further stained for intracellular proteins. For dextramer staining, the cells were pretreated with 50 nmol/L dasatinib (Axon Medchem, Groningen, Netherlands) at 37\u0026deg;C for 30 minutes\u003csup\u003e41\u003c/sup\u003e. Then these cells were stained with dextramers for 20 minutes at room temperature, washed twice, and then stained using the described protocols. Flow cytometry was performed using an LSR II instrument and FACSDiva software (BD Biosciences, San Jose, CA, USA). The data were analyzed using FlowJo software (Treestar, San Francisco, CA, USA). The proportions of various immune cell populations in PBMCs at baseline were analyzed per IP.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCategorical variables were compared using the chi-squared test or Fisher\u0026apos;s exact test as appropriate, and the differences for continuous variables between groups were assessed by the Kruskal-Wallis test. OS was defined as the duration from the start of anti-PD-1 therapy to the date of death from any cause. PFS was defined as the duration from the start of anti-PD-1 therapy to the date of disease progression determined by the RECIST v1.1. PFS with GemCis was defined as the duration from the initiation of first-line GemCis to the date of disease progression according to the RECIST v1.1. The Kaplan\u0026ndash;Meier method was utilized to estimate survival outcomes, which were compared between IPs using the log‐rank test. Univariate and multivariable analyses of PFS and OS were performed using Cox proportional hazards models. Interactions between the IPs and key baseline clinicopathological characteristics were analyzed using a stratified Cox proportional hazards model. A P‐value of \u0026lt; 0.05 was considered statistically significant. All statistical analyses were conducted using R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods-only References\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e37. Ren S, He K, Girshick R, et al: Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 39:1137-1149, 2017\u003c/p\u003e\n\u003cp\u003e38. Chen L-C, Zhu Y, Papandreou G, et al: Encoder-decoder with atrous separable convolution for semantic image segmentation, Proceedings of the European conference on computer vision (ECCV), 2018, pp 801-818\u003c/p\u003e\n\u003cp\u003e39. He K, Zhang X, Ren S, et al: Deep residual learning for image recognition, Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp 770-778\u003c/p\u003e\n\u003cp\u003e40. Kinga D, Adam JB: A method for stochastic optimization In: International Conference on Learning Representations (ICLR). arXiv preprint arXiv:1412.6980, 2014\u003c/p\u003e\n\u003cp\u003e41. Dolton G, Tungatt K, Lloyd A, et al: More tricks with tetramers: a practical guide to staining T cells with peptide-MHC multimers. Immunology 146:11-22, 2015\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Biliary tract cancer, artificial intelligence, tumor-infiltrating lymphocyte, immune-checkpoint inhibitor","lastPublishedDoi":"10.21203/rs.3.rs-3839367/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3839367/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe combination of anti-PD-1/L1 with gemcitabine and cisplatin (GemCis) has recently shown significant survival benefits in randomized phase 3 trials for advanced biliary tract cancer (BTC). However, no biomarker predictive of benefit has been established for anti-PD-1/L1 in BTC. Here, we evaluated tumor-infiltrating lymphocytes (TILs) using artificial intelligence-powered immune phenotype (AI-IP) analysis in advanced BTC treated with anti-PD-1. Data and images of BTC cohort from The Cancer Genome Atlas (TCGA) were initially analyzed to evaluate the transcriptomic and mutational characteristics of various AI-IPs in BTC. The inflamed IP showed increased cytolytic activity scores and an interferon-gamma signature compared to the non-inflamed IP. Next, pre-treatment H\u0026amp;E-stained whole-slide images from 339 advanced BTC patients who received anti-PD-1 monotherapy as second-line treatment or beyond, were retrospectively utilized for AI-IP analysis. Overall, AI-IPs were classified as inflamed (high intratumoral TIL [iTIL]) in 40 patients (11.8%), immune-excluded (low iTIL and high stromal TIL) in 167 (49.3%), and immune-deserted (low TIL overall) in 132 (38.9%). The inflamed IP group showed a significantly higher overall response rate compared to the non-inflamed IP groups (27.5% vs. 7.7%, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Median overall survival (OS) and progression-free survival (PFS) were significantly longer in the inflamed IP group than in the non-inflamed IP group (OS: 12.6 vs. 5.1 months, P\u0026thinsp;=\u0026thinsp;0.002; PFS: 4.5 vs. 1.9 months, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). IP classified by AI-powered spatial TIL analysis was effective in predicting the efficacy outcomes of advanced BTC patients treated with anti-PD-1 therapy. Further validation is necessary in the context of anti-PD-1/L1 plus GemCis.\u003c/p\u003e","manuscriptTitle":"Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a predictive biomarker for immune checkpoint inhibitors in advanced biliary tract cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-10 19:23:13","doi":"10.21203/rs.3.rs-3839367/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a5907db9-e123-4569-8e63-ffafb1bff17f","owner":[],"postedDate":"January 10th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":28040412,"name":"Health sciences/Biomarkers/Predictive markers"},{"id":28040413,"name":"Biological sciences/Cancer/Gastrointestinal cancer/Biliary tract cancer"}],"tags":[],"updatedAt":"2024-02-07T15:02:16+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-10 19:23:13","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3839367","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3839367","identity":"rs-3839367","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}