Cabozantinib and atezolizumab for recurrent or metastatic esophageal squamous cell carcinoma after platinum-based chemotherapy failure: a single-arm phase II study and biomarker analysis Authors: Hung-Yang Kuo, Ta-Chen Huang, Chien-Huai Chuang, Tsung-Che Wu, Yen-Lin Huang, Chia-Chi Lin, Jang-Ming Lee, Jhe-Cyuan Guo, Chih-Hung Hsu Published: 2026 ABSTRACT AbstractPurposeAnti-PD-1 immune-checkpoint inhibitors (ICIs) are approved for treating patients with recurrent or metastatic (R/M) esophageal squamous cell carcinoma (ESCC) who have failed platinum-based chemotherapy. However, their efficacy as monotherapy remains limited. We hypothesized that adding cabozantinib, a multikinase inhibitor with antiangiogenic and immunomodulatory properties, could enhance the therapeutic efficacy of atezolizumab, an anti-PD-L1 ICI, in patients with R/M ESCC.Patients and methodsA single-arm, single-institution phase II trial was conducted (NCT05007613) that included patients with ESCC who experienced disease progression following a platinum-based chemotherapy for R/M ESCC or within 6 months after chemo(radio)therapy for locoregional disease. Participants received cabozantinib 40 mg daily and atezolizumab 1200 mg every 3 weeks until disease progression or intolerable toxicity was reached. The primary endpoint was investigator-assessed objective response rate (ORR), and the secondary endpoints were progression-free survival (PFS), overall survival (OS), and toxicity profile.ResultsBetween June 2021 and March 2024, 24 patients were enrolled in this study. With a median follow-up of 7.0 months (range, 0.5–38.1), 7 patients achieved partial responses and 7 had stable disease. The ORR was 29.1% (95% CI = 11.7%-60.1%), and the disease control rate (DCR) was 58.3% (95% CI = 31.9–97.9%). The median duration of response was 6.5 months (95% CI = 0.9–12.1), the median PFS was 2.2 months (95% CI = 0.64–3.76), and the median OS was 7 months (95% CI = 2.36–11.64). Of the 24 patients, 12 (50%) experienced ≥ grade 3 treatment-related adverse events. Patients with high PD-L1 expression and increased intratumoral or stromal CD8+ tumor infiltrating lymphocytes had numerically higher ORR and improved OS.ConclusionThe combination of cabozantinib and atezolizumab has an acceptable safety profile and moderate antitumor activity as a second-line treatment for patients with R/M ESCC.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00262-026-04408-w.Keywords: ESCC, Cabozantinib, Atezolizumab, PD-L1, Immune checkpoint inhibitor, Antiangiogenesis Introduction Esophageal cancer is one of the most lethal malignancies worldwide, with estimations indicating 511 054 new cases and 445 391 deaths occurred in 2022 [1]. Esophageal squamous cell carcinoma (ESCC), the predominant histological subtype of such cancer, accounts for approximately 85% of new cases [2]. Most patients with ESCC have locally advanced or metastatic disease at diagnosis, and their prognosis is dismal. Prior to 2020, platinum-based combination chemotherapy was the standard first-line treatment for patients with recurrent or metastatic ESCC. Following disease progression, the median overall survival (OS) in such patients was approximately 6–8 months [3, 4]. Since 2019, immune-checkpoint inhibitors (ICIs) have been the standard second-line treatment for platinum-refractory metastatic or recurrent ESCC, after several randomized phase III clinical trials demonstrated them to have superior efficacy to that of chemotherapy [3–6]. However, in these trials, anti-PD-1 monotherapy yielded a modest objective response rate (ORR) of 16.0% to 20.4% and limited OS benefit, with the median OS ranging from 8.2 to 10.9 months in the ICI group and from 6.2 to 8.4 months in the chemotherapy group (HRs, 0.70–0.78). Thereafter, clinical trials combining an ICI with a partner drug to improve the ICI’s therapeutic efficacy are warranted. Cabozantinib is an orally bioavailable small-molecule multikinase inhibitor (MKI) which potently targets VEGFR and several other tyrosine kinases, including MET, AXL, and RET. The combination of antiangiogenic agents, such as anti-VEGF antibodies and VEGFR inhibitors, was reported to enhance the efficacy of ICIs across various cancer types [7]. The proposed immunomodulatory mechanisms of these combinations include enhancing infiltration and activation of effector T cells, along with reducing populations of regulatory T cells and myeloid-derived suppressor cells [8]. Additionally, research indicates that cabozantinib may exert immunostimulatory effects by inhibiting AXL and MERTK signaling, thereby reversing the polarization of immunosuppressive M2 macrophages [9]. Phase II and III clinical trials have demonstrated the efficacy of combining antiangiogenic MKIs with ICIs. For example, the cabozantinib plus nivolumab regimen has been approved by the U.S. Food and Drug Administration as the first-line treatment for metastatic renal cell carcinoma [10]. In addition, in a phase II study, the combination of cabozantinib and pembrolizumab demonstrated promising efficacy in recurrent and metastatic head and neck squamous cell carcinoma, with an ORR of 52% and a median progression-free survival (PFS) of 14.6 months [11]. In a phase II trial focused on ESCC, camrelizumab (an anti-PD-1 ICI) plus axitinib (a VEGFR tyrosine kinase inhibitor [TKI]) revealed encouraging results in patients with cisplatin-refractory metastatic or recurrent disease (ORR = 44%; median PFS = 6.8 months) [12]. These findings suggest that a combination of anti-PD-1/PD-L1 ICIs with VEGFR MKIs may be more effective than ICI monotherapy is. On the basis of this, we conducted a phase II trial to evaluate the efficacy and safety of cabozantinib plus atezolizumab, an anti-PD-L1 antibody, in patients with cisplatin-refractory, locally advanced unresectable, and recurrent or metastatic ESCC. Patients and methods Study design and patient eligibility This was a single-arm, single-center, open-label phase II investigator-initiated trial (ClinicalTrials.gov Identifier: NCT05007613) conducted in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice international guidelines. The study protocol was approved by the Research Ethics Committee of National Taiwan University Hospital (No. 202101041MIPB). All patients provided written informed consent prior to participation in the study. The clinical data were collected by the investigators and research staff. To be eligible, patients were required to be aged ≥ 20 years and have histologically confirmed unresectable, locally advanced, recurrent, or metastatic ESCC that had progressed following first-line platinum-based chemotherapy or within 6 months after neoadjuvant, definitive, or adjuvant chemoradiotherapy (for locally advanced ESCC). Additional inclusion criteria were an Eastern Cooperative Oncology Group performance status of 0–1; adequate organ function; at least 1 measurable lesion according to the Response Evaluation Criteria in Solid Tumours (RECIST), version 1.1. In addition, tumor tissue and plasma samples were required for biomarker analysis. The exclusion criteria were prior exposure to anti-PD-1 or anti-PD-L1 ICIs or small-molecule VEGFR kinase inhibitors, any other active malignancy diagnosed within 2 years, active autoimmune disease, or any condition requiring chronic systemic corticosteroids or immunosuppressants. Patients were also excluded if they were receiving anticoagulants, had untreated brain metastases, or had a disease associated with bleeding or fistula risk. Patients who had received radiotherapy exceeding a total dose of 70 Gy (curative or palliative intent) to a single site (including esophageal or metastatic tumors) were not included. A minimum of 4 months were required to have elapsed since the end of the patient’s concurrent chemoradiotherapy, or at least 8 weeks was required if radiotherapy was administered alone. Additional exclusion criteria were an inability to swallow tablets, known hypersensitivity to monoclonal antibodies, and having undergone major surgery (eg, gastrointestinal surgery or brain metastasis resection/biopsy) within 8 weeks prior to treatment initiation. The eligibility criteria are listed in greater detail in the protocol (supplementary file). Treatment The participants received oral cabozantinib 40 mg once daily and intravenous atezolizumab 1200 mg every 3 weeks, with it administered in 21-day cycles, until disease progression, unacceptable toxicity, or withdrawal of consent occurred. The dosing regimen was based on that in the phase I COSMIC-021 study, which evaluated the efficacy of this combination in advanced solid tumors [13, 14]. Cabozantinib dose reductions were permitted in 2 steps (from 40 mg daily to 20 mg daily and subsequently to 20 mg every other day), whereas for atezolizumab, dose modification was not allowed. Dose interruptions were permitted for both agents. Atezolizumab was to be permanently discontinued in cases of severe immune-related adverse effects (eg, grade 3 or 4 myocarditis, pneumonitis, and nephritis), as specified in the study protocol. Detailed criteria for dose interruptions and discontinuations are outlined in the protocol. Endpoints and patient assessment The primary endpoint was the ORR, defined as the proportion of patients achieving complete or partial response, according to RECIST, version 1.1. The secondary endpoints were the median PFS and OS and the toxicity profile. An exploratory endpoint was the correlation of PFS and OS with PD-L1 expression and the abundance of tumor-infiltrating lymphocytes. Adverse events were assessed using the National Cancer Institute Common Terminology Criteria for Adverse Events, version 5.0. All patients underwent baseline physical examination and tumor assessment with contrast-enhanced computed tomography (CT) of the chest, abdomen, and pelvis during the screening phase. During treatment, CT scans were performed every 9 weeks (after every 3 cycles) for the first 49 weeks and every 12 weeks thereafter. Treatment beyond progression was permitted at the investigators’ discretion if doing so was believed to have clinical benefit. After discontinuation of the study treatment, all patients were monitored for at least 30 days for adverse effects. Additionally, OS was tracked until death, withdrawal of consent for study participation, or the end of the study period, whichever occurred first. Biomarker analysis PD-L1 expression Archival tumor specimens were stained with the anti-PD-L1 antibody SP263 on an automatic platform (Ventana Medical Systems, Tucson, AZ, USA). PD-L1 expression was evaluated through immunohistochemical staining and was reported as both the percentage of PD-L1-positive tumor cells (tumor cell score, TC) and the combined positive score (CPS). TC was defined as the proportion of viable tumor cells exhibiting PD-L1 membrane staining, and CPS was calculated by dividing the number of PD-L1-positive cells (including tumor cells, macrophages, and lymphocytes) by the total number of tumor cells and multiplying the resulting value by 100 [15]. PD-L1 scoring was performed by pathologist Dr. Y.L. Huang. CD8+ tumor-infiltrating lymphocytes Pretreatment formalin-fixed, paraffin-embedded (FFPE) archival tumor tissue from biopsy or surgical specimens was stained with hematoxylin and eosin. CD8+ tumor-infiltrating lymphocytes (TILs) were defined as CD8 positive cells by immunohistochemistry staining. The abundance of CD8+TILs within the intratumoral and stromal compartments was scored by pathologist Dr. Y.L. Huang [16]. mRNA expression analysis Total RNA was extracted and purified from pretreatment FFPE archival tumor tissues using a Qiagen RNeasy FFPE Kit (Qiagen, Valencia, CA, USA). Extracted mRNA was analyzed using an nCounter PanCancer Immune Profiling Panel (NanoString, Seattle, WA, USA) which included 770 immune-related genes. Digital data acquisition via the nCounter Digital Analyzer (NanoString, Seattle, WA, USA), was performed by Cold Spring Biotech Corporation (Taipei, Taiwan). Linear normalization of mRNA expression data was generated by nSolver 4.0 (NanoString, Seattle, WA, USA). Statistical analysis The expected ORR for cabozantinib plus atezolizumab was estimated to be approximately 40%, with this based on efficacy data from the COSMIC-021 study regarding solid tumors [13]. By contrast, a study reported single-agent anti-PD-1 therapy to have an ORR of approximately 15% in a platinum-refractory ESCC population, irrespective of PD-L1 status [3]. On the basis of these estimates, the current study calculated the required sample size to be 24 to detect an increase in ORR from 15 to 40% by using an exact binomial test with a power of 82% and a significance level of 5%. Data lock was performed on December 1, 2024, and the analysis was conducted on an intention-to-treat principle. Descriptive statistics was used to analyze baseline characteristics and adverse events. PFS was defined as the time from enrollment to disease progression, death, or the final follow-up (censored). OS was defined as the time from enrollment to death from any cause or the last follow-up (censored). Survival distributions were estimated using Kaplan–Meier curves, and subgroup comparisons were performed using the log-rank test. All statistical analyses were conducted using GraphPad Prism, version 9.01 (GraphPad Software, San Diego, CA, USA). Results Patient characteristics Between June 2021 and March 2024, 24 patients were enrolled. All of the study participants were included in both the efficacy and safety analyses (Fig. 1). As of the data cutoff date on December 1, 2024, all patients had discontinued treatment. Of the 24 participants, 22 (92%) were male, with a median age of 55 years (range, 32–81 years). The majority of the participants had a history of alcohol (20; 83%) or cigarette (20; 83%) use. In addition, 19 patients (79.2%) had distant metastasis, 3 (12.5%) had both locoregional recurrence and distant metastases, and 2 (8.3%) had only locoregional recurrence (Table 1). After treatment discontinuation, 13 (54.2%) patients received subsequent chemotherapy, 1 (4.2%) received targeted therapy in combination with chemotherapy, 1 (4.2%) received immunotherapy, and 9 (37.5%) received supportive care alone. The baseline characteristics of the participants are presented in Table 1. Study CONSORT diagram Table 1Patient characteristics (N = 24)CharacteristicsN (%)Age (years)Median age (range)55 (32–81)< 65 years18 (75%)≥ 65 years6 (25%)SexMale22 (92%)Female2 (8%)ECOG performance status02 (8%)122 (92%)HabitsAlcohol20 (83%)Betel nuts14 (58%)Cigarette20 (83%)Disease statusLoco-regional recurrence1 (4.2%)Distant metastasis20 (83.3%)Both loco-regional and distant metastasis3 (12.5%)Sites of metastasesLung10 (42%)Liver7 (29%)Distant lymph nodes12 (50%)Others (adrenal, soft tissue, peritoneal)5 (21%)Numbers of organs with metastases0–15 (21%)≥ 219 (79%)Prior therapiesSurgery9 (38%)Chemotherapy (all received 1 line)24 (100%)Radiotherapy23 (96%)PD-L1 expressionCPS < 110 (42%)CPS ≧ 114 (58%) Patient characteristics (N = 24) Efficacy At the median follow-up of 7.0 months (range, 0.5–38.1), 7 patients achieved a partial response, and 7 had stable disease as their best response. The ORR was 29.1% (95% CI = 11.7–60.1%), and the disease control rate was 58.3% (95% CI = 31.9–97.9%). The median PFS and OS were 2.2 months (95% CI = 1.8–5.9) and 7.0 months (95% CI = 2.36–11.64), respectively (Fig. 2). Among the 7 responders, the median duration of response was 6.5 months (95% CI = 0.9–12.1). Waterfall plot (A) and Kaplan–Meier survival curves for relapse-free survival (B) and overall survival (C) in the entire cohort Adverse events All patients (100%) experienced at least 1 adverse event (AE) of any grade, and all reported at least 1 treatment-related adverse event (TRAE) of any grade. Grade 3 or 4 TRAEs were observed in 12 (50%) patients. The most common grade 3 or 4 TRAEs were hypertension (4 patients, 16.7%), leucopenia (3 patients, 12.5%), and fistula formation (3 patients, 12.5%). Only 1 patient with a grade 4 TRAE had prolonged hypomagnesemia and hypokalemia and required frequent parenteral electrolyte supplementation. A bleeding event occurred in 1 patient who had grade 3 hemoptysis and grade 2 jejunostomy oozing. Potential immune-related adverse events (irAEs) of any grade were observed in 13 (54.2%) patients, with nearly all being grade 1 or 2. Only 1 patient (4.2%) developed grade 3 pneumonitis. The most common grade 1 or 2 irAEs were skin rash (6 patients, 25%), pyrexia (6 patients, 25%), and hypothyroidism (5 patients, 20.8%). Table 2 presents the incidence of TRAEs and potential irAEs. Table 2Treatment-related adverse eventsTreatment-related adverse eventsAny gradeGrade 1Grade 2Grade 3Grade 4Any adverse events24 (100%)2 (8.3%)9 (37.5%)12 (50%)1 (4.2%)Aspartate or alanine aminotransferase increased12 (50%)9 (37.5%)3 (12.5%)00Diarrhea11 (45.8%)10 (37.5%)1 (4.2%)00Hand foot skin reaction10 (41.7%)1 (4.2%)7 (29.2%)2 (8.3%)0Fatigue9 (37.5%)8 (33.3%)1 (4.2%)00Hypertension8 (33.3%)04 (16.7%)4 (16.7%)0Proteinuria8 (33.3%)3 (12.5%)4 (16.7%)1 (4.2%)0Leucopenia7 (29.2%)2 (8.3%)2 (8.3%)3 (12.5%)0Neutropenia7 (29.2%)05 (20.8%)2 (8.3%)0Hypomagnesemia6 (25%)5 (20.8%)01 (4.2%)0Thrombocytopenia5 (20.8%)1 (4.2%)3 (12.5%)1 (4.2%)0Stomatitis5 (20.8%)1 (4.2%)2 (8.3%)2 (8.3%)0Weight loss5 (20.8%)1 (4.2%)3 (12.5%)1 (4.2%)0Hypothyroidism5 (20.8%)05 (20.8%)00Fistula4 (16.7%)1 (4.2%)03 (12.5%)0Hyponatremia3 (12.5%)1 (4.2%)1 (4.2%)1 (4.2%)0Constipation3 (12.5%)2 (8.3%)1 (4.2%)00Hyperbilirubinemia3 (12.5%)3 (12.5%)000Hypokalemia2 (8.3%)1 (4.2%)01 (4.2%)0Nausea2 (8.3%)1 (4.2%)1 (4.2%)00Anemia2 (8.3%)01 (4.2%)1 (4.2%)0Potential immune-related adverse eventsAny adverse events13 (54.2%)4 (16.7%)8 (33.3%)1 (4.2%)0Skin rash6 (25%)2 (8.3%)4 (16.7%)00Pyrexia6 (25%)4 (16.7%)2 (8.3%)00Hypothyroidism5 (20.8%)05 (20.8%)00Adrenal insufficiency2 (8.3%)02 (8.3%)00Pruritus2 (8.3%)02 (8.3%)00Hepatitis2 (8.3%)01 (4.2%)1 (4.2%)0Pneumonitis1 (4.2%)01 (4.2%)00Cytokine release syndrome1 (4.2%)01 (4.2%)00 Treatment-related adverse events TRAE led to cabozantinib treatment interruptions in 12 patients (50%) and atezolizumab interruptions in 8 patients (33.3%). Dose reduction of cabozantinib from 40 to 20 mg daily because of toxicity was required in 9 patients (37.5%). One (4.2%) patient permanently discontinued cabozantinib because of a tracheoesophageal fistula. Biomarker analysis PD-L1 expression levels Pretreatment ESCC tumor tissues were available for all patients and assessed for PD-L1 expression through immunohistochemistry. Five patients (20.8%) exhibited high PD-L1 expression, defined as CPS ≥ 10. These patients exhibited a numerically longer median OS (16.4 vs 7.0 months, HR = 0.89, P = 0.832) and higher ORR (40% vs 26.3%) compared with those with low PD-L1 expression (CPS < 10). However, no significant difference in PFS was observed between the high and low CPS groups, regardless of PD-L1 cutoff (Fig. 3A and B). Furthermore, 11 patients (45.8%) with high PD-L1 expression, defined by TC ≥ 1, did not exhibit improved OS (6.1 vs 9.1 months, HR = 1.22, P = 0.605) compared with those with low PD-L1 expression (TC < 1). Kaplan–Meier survival curves for progression-free survival and overall survival stratified by PD-L1 expression (combined positive score, CPS ≥ 10 vs < 10) (A and B), intratumoral tumor-infiltrating lymphocytes (TILs) (> 5% vs ≤ 5%) (C and D) and stromal TILs (≥ 10% vs < 10%) (E and F) CD8+ tumor-infiltrating lymphocytes Patients with high intratumoral CD8+ tumor-infiltrating lymphocytes (TILs; ≥ 5%) had a numerically longer median OS compared with that of those with low intratumoral CD8+TILs (< 5%; OS = 13.2 vs 6.1 months, HR = 0.63, P = 0.363; Fig. 3C and D). Additionally, patients with high stromal CD8+TILs (≥ 10%) exhibited a longer median OS compared with that of those with low intratumoral CD8+TILs (< 10%; OS = 10.0 vs 6.1 months, HR = 0.64, P = 0.382; Fig. 3E and F). mRNA expression analysis Total 12 FFPE archival tissues (each sample from each patient) were available with acceptable quality for mRNA expression analysis. The best treatment response recorded for these 12 patients were: partial response in 5 patients, stable disease in 1 patient, and disease progression in 6 patients. They were categorized into two groups: partial response (PR, including one patient with stable disease) and disease progression (PD) groups. By comparing PR and PD groups, 37 genes (MAGEA3, MAGEA1, MAGEA4, MAGEB2, CD70, IL2RA, CXCL3, CCL20, TNFRSF4, IL1RN, CCL3L1, TNFAIP3, CCL3, TNFRSF8, IL10, CD83, IL15RA, PLAUR, etc.) were found to be significantly downregulated in the PR group, whereas 8 genes (LTF, BLNK, TXNIP, MUC1, CTSH, KIT, ITGAE, EP300) were significantly upregulated in the PR group (Fig. 4A and B)(full list and raw data as the Supplementary 1 file). mRNA expression analysis (A) mRNA differential expression comparing the PR group to the PD group. Red and blue dots are representative upregulated and downregulated genes with significant differences (P < 0.05); (B) Gene expression differences between PR and PD groups (all genes with P < 0.05). (C) Fold change and p values of genes with leading statistical significance Discussion Prior to the success of global randomized phase III trials combining ICIs with chemotherapy as first-line treatment for metastatic or recurrent ESCC [17–19], anti-PD-1/PD-L1 monotherapy (eg, pembrolizumab and nivolumab) was a standard second-line treatment option. In consideration of the modest efficacy of anti-PD-1 monotherapy (ORR = approximately 16–20%) [3–5], we conducted this study to assess the combination of atezolizumab and cabozantinib as a second-line regimen in patients with metastatic or recurrent ESCC who had progressed after undergoing at least 1 platinum-based chemotherapy. The combination demonstrated promising efficacy in several patients, although the overall activity was modest in the general cohort. The safety profile was consistent with those reported in previous trials (eg, COSMIC-021) using the same regimen [13, 14], with no new or unexpected safety signals observed. Antiangiogenic TKIs that target the VEGF–VEGFR pathway are among the few agents that have been clinically validated to enhance the efficacy of ICIs. The first antiangiogenic–ICI combination evaluated for treating ESCC was camrelizumab plus apatinib, which yielded a promising ORR of 34.6% (18 of 52 patients) as second-line therapy in the CAP 02 study [12]. In our study, the ORR for cabozantinib plus atezolizumab was 29.1%, which exceeded that for ICI monotherapy (16–20%) but did not meet the predefined efficacy threshold of 40%. Moreover, the median PFS (2.2 months) and OS (7.0 months) in the present study were comparable to those reported for ICI monotherapy in the KEYNOTE-181 (intention-to-treat population: median PFS = 2.1 months, median OS = 7.1 months) and ATTRACTION-3 (median PFS = 1.7 months, median OS = 10.9 months) trials [3, 4]. Although cross-trial comparisons should be made cautiously, several patient-related factors may have contributed to the modest efficacy. Notably, nearly all patients (96%) in our study had received prior radiotherapy, which is a substantially higher proportion than those reported in the ATTRACTION-3 (70.4%) and CAP 02 (38%) trials [4, 12]. Additionally, 47.8% (11 of 23) of the previously irradiated patients received radiotherapy within 6 months prior to receiving cabozantinib and ateozlizumab, and these patients had lower absolute lymphocyte count and higher neutrophil to lymphocyte ratios compared to those who received radiotherapy more than 6 months prior to study treatment [Supplementary 2]. Furthermore, high-volume irradiation of tumor-draining lymph nodes was reported to be associated with reduced activated lymphocyte counts and poorer survival in patients with ESCC undergoing definitive chemoradiation [20]. Our previous study also demonstrated that a high NLR (> 6.4) predicted worse survival in patients with ESCC treated with ICIs [21]. The high prevalence of such immunosuppressed patients in our cohort may have contributed to the modest efficacy and survival outcomes we observed. Another potential reason for differences between our study and those reported in the literature is that the proportion of patients with high PD-L1 expression (CPS ≥ 10) in this study was 20.8% (5 of 24), which is lower than the 35.3% (222 of 628) reported in the KEYNOTE-181 trial. In our study, the ORR in the high PD-L1 subgroup (40%, 2 of 5) of the present study was slightly higher than that in the patients with low PD-L1 expression (26.3%, 5 of 19). Additionally, patients with high levels of intratumoral or stromal TILs in baseline tumor tissues tended to have longer PFS and OS. This finding is consistent with findings from other phase III clinical trials testing first-line ICIs plus chemotherapy in patients with R/M ESCC, and suggests PD-L1 expression level remains to be one of the predictive markers for the efficacy of ICI. In terms of toxicity, 50% of the patients experienced grade ≥ 3 TRAEs, and the most common of which were hypertension, leucopenia, and fistula formation—typical toxicities associated with antiangiogenic TKIs. Notably, among the 3 patients who developed fistulas during treatment, 1 had enterocutaneous fistula that was suspected to be treatment-related. The other 2 patients developed esophagotracheal and esophagocutaneous fistulas, which were considered to be potentially attributable to both tumor progression and cabozantinib therapy. These AEs were not observed in a phase II trial of cabozantinib combined with pembrolizumab for recurrent or metastatic head and neck squamous cell carcinoma patients. In the CAP 02 study, fistula formation did not occur under treatment with camrelizumab and apatinib. We therefore speculate that the high incidence of prior radiotherapy and the presence of residual esophageal tumors in this cohort may have contributed to the elevated risk of fistula formation observed in the current study. Through mRNA expression analysis, we found that genes contributing to immune cells inactivation (such as TNFAIP3 and IL10), immunosuppressive chemokine production (such as CXCL3, CCL20, CCL3), and cancer antigens (MAGEA1, MAGEA3, MAGEA4, MAGEB2) were downregulated in patients with partial response (PR group) to cabozantinib and atezolizumab. The TNFAIP3/A20-mediated NF-κB suppression contributes to T-cell dysfunction in the tumor microenvironment, and IL-10 has generally been described as immunosuppressive cytokine for T cells [22, 23]. CXCL3 and CCL20 are reported for recruiting immunosuppressive myeloid-derived suppressor cells (MDSC) and regulatory T cells respectively [24, 25]. CCL3 facilitates M2 polarization of macrophage and consequently contributes a more immunosuppressive tumor microenvironment [26]. MAGE (Melanoma-associated antigens) family genes which are responsible of producing cancer–associated antigens have been reported as adverse prognostic markers for several cancer types including ESCC [27]. Taken together, our results suggest that the PR group displays broad downregulation of genes that promote immune suppression and tumor immune evasion. Such a less immunosuppressive molecular profile may create a more favorable microenvironment for effective anti-tumor immune activation during cabozantinib and atezolizumab therapy. This study has several limitations. The major limitation is the single-arm design of this phase II trial could only provide preliminary evidence of the efficacy of the cabozantinib plus atezolizumab regimen. The absence of a control arm makes our study inherently unable to determine the true benefit of the combination therapy over the standard of care, i.e., anti-PD-1 monotherapy. Additionally, the small sample size limited the statistical power of the subgroup survival analyses based on biomarkers such as PD-L1 expression and TILs. Owing to limited available tissues, we could only have 12 patient samples with adequate quality for mRNA expression analysis. Thus, our findings, which although are of translational implication, warrant additional studies for further validation. Moreover, ICIs have been established as part of standard first-line therapy following multiple successful phase III trials [28], and our enrollment of only ICI-naïve patients restricts the applicability of our results to this specific population. Notably, the CAP 02 Re-challenge study, which included patients with ESCC previously treated with ICIs, reported modest efficacy for the combination of camrelizumab and apatinib in that setting [29]. In conclusion, the cabozantinib plus atezolizumab regimen demonstrated promising efficacy in certain patients, but the overall response was modest. Nevertheless, most TRAEs were manageable, and no new safety concern was noted for this combination therapy compared with treatment with each agent alone. Patients with ESCC exhibiting high PD-L1 expression (CPS ≥ 10), greater intratumoral or stromal TILs appeared to derive greater benefit from this combination. Ongoing trials, such as LEAP-014 [30], evaluating ICIs combined with antiangiogenic TKIs as first-line therapy for metastatic ESCC are expected to provide further insights that will help improve outcomes for this patient population. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (CSV 71 KB) Supplementary file2 (JPG 79 KB) Supplementary file3 (PNG 25 KB) Supplementary file4 (PDF 1718 KB) Acknowledgements The authors would like to thank all patients who participated in this clinical trial. We also thank Ms. Shu-Lin Wu for coordinating patient treatments and follow-up and Mr. Hsi-Yu King for patient sample collection for biomarker analysis. Author contributions CHH conceived the study, recruited study resources, conducted the trial as the principle investigator. and revised the manuscript. HYK organized the study conduct, included patients, interpreted data with statistical analyses and wrote the manuscript. JCG assisted the study conduct and data analysis, included patients and revised the manuscript. TCH, CHC, TCW included patients. YLH performed pathology review and IHC analysis. CCL and JML assisted study coordination and resources allocation. Funding This study was granted by Department of Medical Research, National Taiwan University Hospital (111-CTC0027, 112-CTC0026, 113-CTC0018) and the Ministry of Health and Welfare (MOHW113-TDU-B-211-114006 and MOHW114-TDU-B-221-144006), Taiwan. Ipsen provided product support and partial financial support, and Roche provided product support. Data availability All data supporting the findings of this study are available within the paper and its supplementary information. Declarations Conflict of interest J. Guo: Financial Interests, Personal, Invited Speaker, Travel grant: Ipsen; Financial Interests, Personal, Invited Speaker: Roche. C. Lin: Financial Interests, Personal, Advisory Role: AbbVie, Anbogen, Blueprint Medicines, Boehringer-Ingelheim, Bristol Myers Squibb, Daiichi Sankyo, IMPACT, Merck KGaA, Novartis, Pharma Engines, Pfizer; Financial Interests, Personal, Invited Speaker: Boehringer Ingelheim, Eli-Lilly, Novartis, Roche; Financial Interests, Personal, Other, Travel support for attending meetings: BeiGene, Daiichi Sankyo, Eli Lilly, IMPACT. T. Wu: Financial Interests, Personal, Invited Speaker: Roche. C. Hsu: Financial Interests, Institutional, Research Grant: BeiGene, Gilead, Ipsen, Ono Pharmaceutical, and F. Hoffmann-La Roche Ltd; Financial Interests, Personal and Institutional, Principal Investigator, sponsor-initiated clinical trials: AstraZeneca, BeiGene, Bristol Myers Squibb, Merck Sharp & Dohme, Ono Pharmaceutical, F. Hoffmann-La Roche Ltd, NGM Biopharmaceuticals, Pfizer, and Surface Oncology; Financial Interests, Personal, Advisory Role: AstraZeneca, BeiGene, Ono Pharmaceutical, and F. Hoffmann-La Roche Ltd; Financial Interests, Personal, Invited Speaker: BMS, Daiichi Sankyo, Eisai, Merck Sharp & Dohme, Ono Pharmaceutical, and F. Hoffmann-La Roche Ltd; Financial Interests, Personal, Other, support for attending meetings and/or travel: F. Hoffmann-La Roche Ltd; Financial Interests, Personal, Advisory Board: Daiichi Sankyo and F. Hoffmann-La Roche Ltd. All other authors have declared no conflicts of interest. Ethical approval The study protocol was approved by the research ethics committee of our institute (approval number: 202101041MIPB). The study was conducted in accordance with the Declaration of Helsinki and ICH guidelines for Good Clinical Practice. Consent to participate Written informed consent was obtained from all patients prior to inclusion. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Jhe-Cyuan Guo, Email: jhecyuanguo@gmail.com. Chih-Hung Hsu, Email: chihhunghsu@ntu.edu.tw. References Associated Data Supplementary Materials Supplementary file1 (CSV 71 KB) Supplementary file2 (JPG 79 KB) Supplementary file3 (PNG 25 KB) Supplementary file4 (PDF 1718 KB) Data Availability Statement All data supporting the findings of this study are available within the paper and its supplementary information.