Atezolizumab following definitive chemoradiotherapy in patients with unresectable locally advanced esophageal squamous cell carcinoma – a multicenter phase 2 trial (EPOC1802)
Main
Esophageal cancer ranks as the seventh most prevalent cancer and the sixth leading cause of cancer-related deaths worldwide, accounting for more than half a million cancer-related mortality annually1. According to the Comprehensive Registry of Esophageal Cancer in Japan, ESCC invading adjacent structures, such as the aorta, vertebral bodies or trachea (T4b), accounts for approximately 6.7% of all cases of esophageal cancer (approximately 1,500 patients per year in Japan)2. As T4b ESCC is unresectable, the standard treatment globally is dCRT with curative intent using platinum-based regimen plus 50–60 Gy of radiation3,4,5; however, clinical trial results using this regimen have shown that cCR rates are low (11–25 %), resulting in a median overall survival (OS) of 9–10 months6,7,8. Recent real-world data from 175 patients with unresectable locally advanced ESCC from 2013 to 2020 indicated that the cCR rate was 24% (ref. 9), as treatment regimens have not changed since the 1990s (ref. 10). Therefore, new treatment regimens with improved response and survival rates are urgently required.
Among such treatment regimens, immunotherapy with immune-checkpoint inhibitors (ICIs) including anti-cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), anti-PD-1 and anti-PD-L1 antibodies has revolutionized the treatment of advanced cancers, including esophageal cancer. Currently, 5-FU/cisplatin + pembrolizumab, 5-FU/cisplatin + nivolumab, and nivolumab + ipilimumab are the standard first-line therapies for patients with advanced metastatic ESCC3,11,12. In addition, in patients with resectable locally advanced esophageal adenocarcinoma and ESCC, neoadjuvant CRT followed by surgery and adjuvant 1-year nivolumab has been the standard therapy3,13,14; however, the efficacy and safety of ICIs in patients with unresectable locally advanced ESCC treated with dCRT remain unclear.
Basic research suggests that the use of ICIs combined with ionizing radiation is a promising approach, owing to the synergistic efficacy of these therapies. The mechanisms by which radiation facilitates ICI activity include increased tumor antigen release, activated innate immune system, increased T cell infiltration, increased antigen presentation and modulated immunosuppressive cells15,16. Chemotherapy and radiotherapy may mediate the release of interferon (IFN)-γ produced by CD8+ T cells, resulting in PD-L1 upregulation in various tumor cells16,17. As lymphocytes are radiation sensitive, we hypothesized that sequential combinations of anti-PD-L1 agents immediately after CRT completion would enhance the therapeutic efficacy of the anti-PD-L1 agent. Indeed, among patients with unresectable locally advanced cell lung cancer, 12 months of treatment with the anti-PD-L1 antibody durvalumab after platinum-based CRT significantly improved both progression-free survival (PFS) and OS, irrespective of PD-L1 expression before CRT18,19.
Based on this information, we planned a phase II proof-of-concept clinical trial (EPOC1802, TENERGY) to evaluate the safety and efficacy of the anti-PD-L1 antibody atezolizumab following definitive CRT in patients with unresectable, locally advanced ESCC.
Results
Patient characteristics and treatment protocol
In this phase II study, patients with unresectable locally advanced ESCC treated with two cycles of cisplatin/5-FU plus 60 Gy of radiation were enrolled. The treatment consisted of atezolizumab monotherapy every 3 weeks for a maximum of 12 months (Fig. 1a). Of the 51 candidate patients with unresectable locally advanced ESCC who were treated with dCRT, 2 patients with lung metastasis after CRT, 3 patients with prolonged CRT-related adverse events, 2 patients who refused to participate, 1 patient with tracheal fistula after CRT, 1 patient whose general condition deteriorated and 2 patients treated with ineligible radiation were not included in the study. Thus, 40 patients were enrolled in the primary cohort between December 2018 and March 2021. In addition, ten patients with locoregionally recurrent ESCC after radical surgery were enrolled in the exploratory cohort because additional surgical resection is difficult and they are generally treated with dCRT. Sex was not considered in the study design and the sex of the participants was determined based on self-reports. In the primary cohort, the median age was 64 years and 82.5% (33 of 40) of the patients were male (Table 1). A total of 95.0% (38 of 40) of patients had clinical T4 disease and 92.5% (37 of 40) had lymph node metastasis. Furthermore, 25.0% (10 of 40) of the patients had extraregional M1 lymph node metastasis. Accordingly, 97.5% of the patients were diagnosed with clinical stage IVA or IVB ESCC (Table 1).
a, In this phase II study, patients with unresectable locally advanced ESCC treated with two cycles of cisplatin/5-FU plus 60 Gy of radiation were enrolled. The study treatment was 12 months of atezolizumab monotherapy every 3 weeks. For the biomarker study, serial biopsies from the primary site and blood collection were performed at three time points (before treatment, after definitive chemoradiotherapy and 4 weeks after the first atezolizumab dose). b,c, Kaplan–Meier plots of PFS (n = 40 patients) and OS (n = 40 patients) in the primary cohort for primary unresectable locally advanced ESCC are shown. Vertical lines denote patients who were censored. d, Swimmer plots of dCRT, the treatment protocol, first CR and PD in the primary cohort (n = 40 patients) are shown. e,f, Kaplan–Meier plots of PFS (cCR group, n = 16 patients, non-cCR group, n = 24 patients) and OS (cCR group, n = 16 patients; non-cCR group, n = 24 patients) in patients with cCR (red) and without cCR (blue) in the primary cohort are shown. Vertical lines denote patients who were censored.
Source data
In the primary cohort, the median relative dose intensity (RDI) for atezolizumab was 97.3%. Most patients discontinued atezolizumab due to disease progression (57.5%, 23 of 40), and 27.5% (11 of 40) completed atezolizumab monotherapy (Supplementary Table 1). The median RDI of atezolizumab was 95.6% and 60.0% (6 of 10) of the patients in the exploratory cohort discontinued owing to disease progression.
Efficacy and adverse events
Because long-term survival with dCRT is expected only in patients with cCR, we decided that the cCR rate determined by an investigator’s assessment in the primary cohort was the primary end point in this study. The cCR rate of the first consecutive 38 patients in the primary cohort was 42.1% (90% confidence interval (CI) 28.5–56.7%). As the lower limit of the 90% CI for the cCR rate was >20%, the primary end point was met. Thus, this treatment merits further development. The cCR rates in the primary and exploratory cohorts were 40.0% and 50.0%, respectively (Table 2). In addition, of the 51 candidate patients treated with dCRT in the primary cohort, the cCR rate was 35.3% (18 of 51), as 2 of the 11 (18.2%) patients who were not included in the trial exhibited cCR. In the primary cohort, the median follow-up period was 12.5 (range 0.6–31.4) months. The median PFS and 12-month PFS rate were 3.2 months and 29.6%, respectively (Fig. 1b). Of the 25 patients with disease progression, 16 had local regrowth, 6 had metastatic disease, 2 had both and 1 had clinical disease progression. After the study treatment, radical resection and additional dCRT were performed in seven (one before and six after PFS events) and two patients, respectively. Palliative radiotherapy, another ICI and chemotherapy were administered to 5, 3 and 19 patients, respectively (Supplementary Table 2). At a median follow-up duration of 12.5 (range 0.6–31.4) months, the median OS and 12-month OS rate were 31.0 months and 65.8%, respectively (Fig. 1c,d). We determined the median PFS and OS from the starting date of dCRT, which is the common starting date, at 5.5 and 33.0 months, respectively. In patients with cCR, the median PFS was not reached, and the 12-month PFS rate was 67.5%; only one patient died (Fig. 1e,f). Of the 35 patients with evaluable PD-L1 tumor proportion score (TPS), the cCR rates with PD-L1 TPS < 1% and ≥1% were not significantly different at 44.4% and 41.2%, respectively (Supplementary Table 3). The peripheral blood lymphocyte counts decreased during dCRT and recovered during the study period. The lowest peripheral blood lymphocyte counts were not significantly different between patients with and without cCR (Supplementary Table 4).
Although 5.0% and 10.0% of patients in the primary and exploratory cohorts, respectively, showed grade 3 or higher pneumonitis, only one (2.5%) case of grade 2 adrenal insufficiency, one (2.5%) with grade 1 hyperthyroidism and three (7.5%) with grades 1–2 hypothyroidism were observed in the primary cohort. No treatment-related deaths were observed (Supplementary Table 5). Atezolizumab monotherapy following dCRT increases long-term survival without compromising safety.
Combined dCRT/atezolizumab strengthens antitumor response
To investigate the predictive biomarkers, we performed serial biopsies of the primary tumor and collected blood samples at three time points (before treatment, after dCRT and 4 weeks after the first atezolizumab dose) (Fig. 1a). We used these samples to assess whether dCRT could strengthen antitumor immune responses in the tumor microenvironment (TME). First, we compared the immune-related gene expression in ESCC biopsy samples before and after dCRT. Gene set expression analysis (GSEA) using Gene Ontology Biological Process (GOBP) gene sets showed that both gene signatures related to innate and acquired immunity were significantly upregulated by dCRT (Fig. 2a). Furthermore, consistent with previous reports20,21,22,23, the GSEA results suggest that tumor cells released cytoplasmic DNA and RNA when exposed to dCRT and activated cytosolic DNA- and RNA-sensing pathways (the cGAS–STING pathway20,21 and RIG-I pathway22,23, respectively) (Extended Data Fig. 1a). Furthermore, tumors exhibited type I and II IFN responses16,17,20,21,22,23 (Extended Data Fig. 1b) and HLA-A, HLA-B and HLA-C upregulation24 (n = 28) (Extended Data Fig. 1c). These phenomena promoted TCR stimulation, co-stimulation by CD28 and further PD-1 signaling15,16,17 (Extended Data Fig. 1d). Consistent with these findings, we found that, among the immune-checkpoint molecules, the expression of CD274 was significantly increased by dCRT (n = 28) (Extended Data Fig. 1e).
a,b, RNA extracted from ESCCs during treatment (pretreatment (n = 28 patients) after dCRT (n = 25 patients) and after atezolizumab treatment (n = 25 patients)) was subjected to RNA-seq. Gene expression in ESCCs after dCRT was compared to that in ESCCs before treatment by performing GSEA with GOBP gene sets (a). The top 20 significantly enriched gene sets among GOBP gene sets are shown. ssGSEA scores from the indicated gene sets were calculated from gene expression in ESCCs during treatment (b). Summaries of ssGSEA scores from the indicated gene sets are shown. Gray, pretreatment; red, after dCRT; blue, after atezolizumab (anti-PD-L1 mAb). c–e, TILs extracted from ESCCs during treatment (pretreatment (n = 18 patients), after dCRT (n = 18 patients) and after atezolizumab treatment (n = 18 patients)) were subjected to FCM. Summary of the ratio of intratumor CCR7−CD45RA− (effector memory (EM)) CD8+CD3+CD45+ T cells to FOXP3highCD45RA–CD4+CD3+CD45+ T cells (eTreg cells) during treatment (pretreatment, after dCRT and after atezolizumab treatment) is shown (c). Gray, pretreatment; red, after dCRT; blue, after atezolizumab (anti-PD-L1 mAb). Representative histograms (left) and summary (right) of CD80 (d) and CD86 (e) expression in HLA-DR+CD11c+CD45+ cells (APCs) during treatment courses (pretreatment, after dCRT and after atezolizumab treatment) are shown. Gray, pretreatment; red, after dCRT; blue, after atezolizumab (anti-PD-L1 mAb); light gray, unstained control. P values were calculated by permutation testing in GSEA (a). P values were calculated by one-way analysis of variance (ANOVA) (b–e). For multiple comparisons, Holm‒Sidak’s test was performed. The differences with P ≤ 0.05 were considered significant. Significant differences were found in general lymphocytes (P = 0.002), lineage cytotoxic T (P = 0.08), IFNG28 gene (P = 0.008), cell cycle G1S (P = 0.0000007), cell cycle G2M (P = 0.000003), EM CD8+ T/2Treg cells (P = 0.007), CD80+HLA-DR+CD11c+ (P = 0.03) and CD86+HLA-DR+CD11c+ (P = 0.04) cells. mAb, monoclonal antibody.
Source data
In addition, we calculated single-sample GSEA (ssGSEA) scores25 during the treatment course (n = 28) (Fig. 2b and Extended Data Fig. 1f). The combination of dCRT and sequential atezolizumab robustly increased the ssGSEA scores, reflecting the expression of the gene sets, General_lymphocyte, Lineage_cytotoxicT, IFNG_28-gene, Lineage_CD4, Lineage_CD8, Lineage_B_cell and Lineage_NK_cell, but significantly decreased the ssGSEA scores, reflecting the expression of the gene sets Cell_cycle_G1S, Cell_cycle_G2M and MDSC_ShirleyLiu. These data suggest that dCRT not only directly kills tumor cells, but also enhances antitumor immune responses, possibly increasing the efficacy of sequential atezolizumab treatment.
Next, we performed flow cytometry (FCM) to assess the immunological phenotypes of tumor-infiltrating lymphocytes (TILs) during the course of treatment (n = 18) (Fig. 2c–e and Extended Data Fig. 2a). The ratio of CCR7−CD45RA− (effector memory (EM)) CD8+CD3+ T cells to effector regulatory T (eTreg) cells, defined as FOXP3highCD45RA−CD4+CD3+ T cells26 and the proportions of Ki-67+ and/or EOMES+T-bet+ cells, which were regarded as effector T cells27 in CD8+ T cells were significantly elevated by dCRT, whereas PD-1 expression in CD8+ T cells was slightly but not significantly elevated (n = 18) (Fig. 2c and Extended Data Fig. 2a). The proportion of eTreg cells among the CD45+ cells decreased during treatment (Extended Data Fig. 2b). Consistent with these results, multiplexed immunohistochemistry (mIHC) revealed a mild, although nonsignificant, increase in intratumor CD8+CD3+ T cells and PD-1+CD8+CD3+ T cells, likely because of the small sample size, and that intratumor FOXP3+CD8−CD3+ T cells, which are indicative of Treg cells, were decreased (n = 40) (Extended Data Fig. 2c,d). Furthermore, CTLA-4 expression on the surface of eTreg cells was increased by dCRT, although PD-1 expression in eTreg cells did not increase (n = 18) (Extended Data Fig. 2b). dCRT increased not only the frequency of antigen-presenting cells (APCs), defined as HLA-DR+CD11c+CD45+ cells, among CD45+ cells, but also the expression of CD80, CD86 and PD-L1 in APCs (n = 18) (Figs. 2d,e and Extended Data Fig. 2e). APC-induced PD-L1 expression increased after CRT and decreased after atezolizumab treatment. This result should be interpreted with caution because atezolizumab occupies the binding site of the anti-PD-L1 antibody used for FCM analysis (Extended Data Fig. 2f). Nonetheless, the FCM and mIHC data support the notion that dCRT recruits and activates APCs and effector T cells in the TME.
We performed FCM to evaluate the phenotypes of peripheral blood mononuclear cells (PBMCs) during the treatment course involving the combination of dCRT and sequential atezolizumab. This treatment combination increased the proportion of CD8+CD3+ T cells among CD45+ cells and Ki-67 expression in CD8+ T cells (n = 17) (Extended Data Fig. 3a). The proportion of eTreg cells among CD45+ cells, expression of CTLA-4 and PD-1 on eTreg cells, ratio of CCR7−CD45RA− (EM) CD8+CD3+ T cells to eTreg cells, and expression of CD80, CD86 and PD-L1 on APCs were not significantly altered during treatment (n = 17) (Extended Data Fig. 3b,c). These PBMC data suggest that combining dCRT and sequential atezolizumab treatment could strengthen the systemic antitumor immune response, especially that caused by effector CD8+ T cells, in addition to the antitumor immune response in the TME.
Collectively, these data from genomic and immunological analyses suggest that dCRT enhances IFN responses in tumor cells via cytoplasmic DNA- and RNA-sensing pathways and recruits and activates APCs and effector T cells in the TME. Consequently, augmented PD-L1 expression in APCs strengthens PD-1 signaling in effector T cells and hampers efficient antitumor immune responses. These findings support our hypothesis that sequential administration of anti-PD-L1 agents immediately after completing dCRT could vigorously enhance antitumor immune responses and optimize clinical efficacy.
Genomic and immunological ESCCs analysis before treatment
To identify possible predictive biomarkers, we analyzed the genomic status of 27 ESCCs in the primary cohort before treatment. Whole-exome sequencing (WES) showed MYC amplification (n = 2), EGFR mutation (n = 1), PIK3CA mutation (n = 1), PTEN mutation (n = 1) and KEAP1 mutation (n = 1), which have been reported to be associated with ICI resistance mechanisms28,29,30,31,32,33, in 17 patients with non-cCR (n = 27) (Fig. 3a). Tumor mutational burden (TMB) was not significantly associated with patient outcomes (n = 27) (Fig. 3b). GSEA of HALLMARK gene sets demonstrated that gene expression signatures reflective of epithelial–mesenchymal transition (EMT) and TGF-β signaling were significantly enriched in patients with non-cCR before dCRT (Fig. 3c), which is consistent with previous reports suggesting that cancers with these phenotypes can be resistant to cancer immunotherapy34,35. Furthermore, differentially expressed genes (DEGs) in ESCCs in the primary cohort were significantly associated with patient outcomes (Fig. 3d). Although neither the gene expression of human leukocyte antigen (HLA) and immune-checkpoint molecules nor the ssGSEA scores for immune-related gene sets before treatment were significantly associated with patient outcomes (n = 28) (Extended Data Fig. 4a–c), MMP1 and MMP13 were significantly expressed in non-cCR patient tumors (Fig. 3d,e). Matrix metalloproteinase (MMP) family members, including MMP1 and MMP13, are reportedly associated with EMT36,37, which probably causes resistance to various treatment strategies, including radiotherapy38 and immunotherapy35.
a,b, DNA extracted from 27 pretreatment ESCC cases was subjected to WES. Representative driver gene alterations (filled sections) and the TMB (columns) are shown (a). Green, missense variants; purple, splicing variants; orange, frameshift; red, truncating mutations; yellow, in-frame mutations; pink, amplification; blue, loss. The TMB of ESCCs before treatment was compared between patients who achieved cCR (n = 10) and those who did not (n = 17) (b). c,d, RNA extracted from 27 pretreatment ESCCs was subjected to RNA-seq. Gene expression in non-cCR ESCC cases before treatment was compared to that in cCR ESCCs by performing GSEA with HALLMARK gene sets (c). Top and bottom show the results achieved by analyzing the expression of epithelial–mesenchymal transition-related genes and TGF-β signaling-related genes, respectively. Volcano plots analyzing enriched differential genes between non-cCR ESCC and cCR esophageal cancer samples (d). In brief, treatment are shown. Red shows patients who achieved cCR (cCR cases); blue shows patients who did not achieve cCR (non-cCR cases). e, Expressions of MMP1 (left) and MMP13 (right) in the ESCCs of cCR cases (red) and non-cCR cases (black) pretreatment (cCR, n = 11 patients; non-cCR, n = 17 patients) were evaluated. Summaries of the TPM values of MMP1 (left) and MMP13 (right) in the ESCCs of cCR cases (red) and non-cCR cases (black) are shown. Bars indicate mean. P values were calculated by unpaired two-tailed t-test, revealing significant differences in MMP1 (P = 0.0458) and MMP13 (P = 0.0111), but not TMB (P = 0.86) between the cCR and non-cCR cases (b,e). P values were calculated using the likelihood ratio test (d). The differences with P ≤ 0.05 were considered significant. TPM, transcripts per million.
Source data
Next, we evaluated the immunological phenotypes of TILs by FCM and mIHC (Fig. 4a–d and Extended Data Fig. 5a–c). According to FCM analysis, PD-1 expression in CD8+ T cells was significantly higher in patients who achieved cCR than in those who did not (n = 18) (Fig. 4a), whereas other parameters before treatment were not significantly associated with patient outcomes (n = 18) (Fig. 4b–d and Extended Data Fig. 5a,b). Additionally, we performed mIHC analysis of the specimens before treatment and found that intratumor CD8+CD3+ T cells and PD-1+CD8+CD3+ T cells were significantly more abundant in patients who achieved cCR than in those who did not, and FOXP3+CD8−CD3+ T cells were significantly more abundant in patients who did not achieve cCR than in those who did (n = 40) (Extended Data Fig. 5c). Accordingly, FCM analysis of PBMCs showed that the proportion of eTreg cells among CD45+ cells pretreatment was significantly higher in patients who did not achieve cCR than in those who did, although the differences in other parameters were insignificant (n = 16) (Extended Data Fig. 6a–c). Due to the small sample size of patients whose PBMCs were analyzed by FCM, we further evaluated the results of the complete blood count (CBC) and found no significant differences in the counts of white blood cells (WBCs), neutrophils, lymphocytes or the neutrophil-to-lymphocyte ratio (NLR) before treatment between patients who achieved cCR and those who did not (n = 49) (Extended Data Fig. 7).
TILs extracted from ESCCs during treatment (pretreatment (cCR, n = 6 patients; non-cCR, n = 12 patients), after dCRT (cCR, n = 7 patients; non-cCR, n = 11 patients), and after atezolizumab treatment (cCR, n = 7 patients; non-cCR, n = 11 patients)) were subjected to FCM. a–d, Representative histograms (left) and summaries (right) of the frequencies of intratumor immune cells, including PD-1+CD8+CD3+ T cells (a), PD-1+FOXP3highCD45RA−CD4+CD3+ T cells (b), CTLA-4+FOXP3highCD45RA−CD4+CD3+ T cells (c) and CD80+HLA-DR+CD11c+CD45+ cells (d), during treatment (pretreatment, after dCRT and after atezolizumab treatment) are shown according to treatment response. Gray, unstained control; red, patients who achieved cCR (cCR cases); black, patients who did not achieve cCR (non-cCR cases). Bars, mean; P values were calculated using two-way ANOVA. The Holm–Sidak test was performed for multiple comparisons. The differences with P ≤ 0.05 were considered significant. Significant differences between the cCR and non-cCR cases were found for PD-1+CD8+CD3+ T cells pretreatment (P = 0.01), after dCRT (P = 0.007), and after atezolizumab (P = 0.01); for PD-1+ eTreg cells after dCRT (P = 0.02) but not pretreatment (P = 0.63) nor after atezolizumab (P = 0.63); for CTLA-4+ eTreg cells after dCRT (P = 0.02) and after atezolizumab (P = 0.01) but not pretreatment (P = 0.45); and for CD80+HLA-DR+CD11c+ cells after atezolizumab (P = 0.0004) but not pretreatment (P = 0.92) nor after dCRT (P = 0.30).
Source data
In conclusion, these data suggest that driver gene mutations, EMT phenotypes with high expression of MMP1 and MMP13, and abundant Treg cells in the TME or blood contribute to the evasion of the antitumor response by effector T cells, thereby causing resistance to combination therapy with dCRT and atezolizumab. These factors may serve as valuable negative predictive biomarkers for this treatment strategy.
ESCC analysis after dCRT based on treatment response
We compared the immunological status of patients who did and did not achieve cCR immediately after dCRT and before atezolizumab treatment. Transcriptomic analysis of ESCCs (n = 25) immediately after dCRT showed no significant differences between patients who did and did not achieve cCR (Extended Data Fig. 4a–c). FCM analysis of TILs immediately after dCRT and before atezolizumab treatment showed that PD-1 expression in CD8+ T cells was significantly higher in patients who achieved cCR than in those who did not, whereas PD-1 and CTLA-4 expression in eTreg cells were significantly lower in patients who achieved cCR than in those who did not (n = 18) (Fig. 4a–c). No significant differences were observed in other parameters evaluated in the FCM analysis of TILs between patients who did and those who did not (Extended Data Fig. 5a,b). FCM analysis of PBMCs (n = 17) and CBC analysis (n = 49) performed immediately after dCRT showed no significant differences between patients who did and did not achieve cCR (Extended Data Fig. 6a–c and Extended Data Fig. 7).
These data indicate that the activation of tumor-infiltrating eTreg cells after dCRT was significantly associated with resistance to atezolizumab treatment. Furthermore, the balance between PD-1 expression in effector T cells and eTreg cells in the TME immediately before atezolizumab treatment predicted the efficacy of atezolizumab39.
Residual ESCC characteristics after combination therapy
We evaluated the genomic and immunological features of ESCCs after treatment to investigate the resistance mechanisms after combination therapy with dCRT and sequential atezolizumab. DEG analysis of ESCCs after atezolizumab treatment was performed to identify genes significantly associated with non-cCR (Fig. 5a). We identified IL1A and IL1B among the DEGs in the residual ESCCs of patients who did not achieve cCR (n = 25) (Fig. 5a and Extended Data Fig. 8a). During the treatment course, IL1A gene expression was not significantly elevated in patients who achieved cCR or in those who did not (n = 25) (Fig. 5b). In contrast, IL1B expression was significantly upregulated only in patients who did not achieve cCR (n = 25) (Fig. 5c).
a, Volcano plots analyzing enriched differential genes between non-cCR ESCC and cCR esophageal cancer samples after combination therapy are shown. Red, patients who achieved cCR (cCR cases); blue, patients who did not achieve cCR (non-cCR cases). P values were calculated using the likelihood ratio test. b, IL1A expression in the ESCCs of cCR cases (left) and non-cCR cases (right) during treatment (pretreatment (cCR, n = 11 patients; non-cCR, n = 17 patients), after dCRT (cCR, n = 10 patients; non-cCR, n = 15 patients) and after atezolizumab treatment (cCR, n = 10 patients; non-cCR, n = 15 patients)) was evaluated. Summaries of the TPM values of IL1A in the ESCCs of cCR cases (left) and non-cCR cases (right) are shown. c, IL1B expression in the ESCCs of cCR cases (left) and non-cCR cases (right) during treatment (pretreatment (cCR, n = 11 patients; non-cCR, n = 17 patients), after dCRT (cCR, n = 10 patients; non-cCR, n = 15 patients) and after atezolizumab treatment (cCR, n = 10 patients; non-cCR, n = 15 patients)) was evaluated. Summaries of the TPM values of IL1B in the ESCCs of cCR cases (left) and non-cCR cases (right) are shown. Gray, pretreatment; red, after dCRT; blue, after atezolizumab (b,c). P values were calculated by one-way ANOVA. For multiple comparisons, Holm‒Sidak’s test was performed. Differences with P ≤ 0.05 were considered statistically significant. Significant differences were found in IL1A for cCR cases (P = 0.00007) but not non-CCR cases (P = 0.45) and in IL1B for non-cCR cases (P = 0.02) but not cCR cases (P = 0.67). d, Gene expression in ESCCs after dCRT was compared to that in ESCCs before treatment by performing GSEA with the indicated gene sets in non-cCR cases. Plots of the enrichment of IL-1R pathway-related genes (left top), granulocyte pathway-related genes (left middle), CTLA-4 pathway-related genes, the iCAF signature (right top) and the CAF signature (right bottom) are shown.
Source data
IL-1β induces myeloid-derived suppressor cells (MDSCs) in the TME40. Furthermore, upon irradiation, IL-1 polarizes cancer-associated fibroblasts (CAFs) toward an inflammatory phenotype and triggers oxidative DNA damage, thereby predisposing inflammatory CAFs (iCAFs) to p53-mediated therapy-induced senescence, resulting in treatment resistance41. Consistent with this, GSEA showed that in patients who did not achieve cCR, IL-1R-related gene expression, granulocyte-related gene expression, iCAF signature and CAF signature were enriched in residual ESCCs immediately after dCRT compared to specimens before treatment (Fig. 5d). When comparing the gene expression of specimens from patients who achieved cCR with those who did not after atezolizumab treatment, IL-1R-related gene expression, granulocyte-related gene expression and iCAF and CAF signatures were enriched in the ESCCs of patients who did not achieve cCR (Extended Data Fig. 8b). Furthermore, our transcriptomic analysis showed that the MDSC-related ssGSEA scores (MDSC_ShirleyLiu scores) of patients who did not achieve cCR were significantly higher than those of patients who achieved cCR after atezolizumab treatment (n = 25) (Extended Data Fig. 8c). We also compared the gene expression of immune-related molecules (Extended Data Figs. 4a and 5c) and found that the expression of NECTIN2 and PVR, which are major ligands of TIGIT42, was significantly higher in patients who did not achieve cCR than in those who achieved cCR after atezolizumab treatment (Extended Data Fig. 4b). Additionally, we analyzed the FCM data from TILs after atezolizumab treatment (n = 18) (Fig. 4a–e and Extended Data Fig. 5a,b). PD-1 expression in CD8+ T cells from patients who achieved cCR was significantly higher than that in CD8+ T cells from patients who did not achieve cCR (Fig. 4a), whereas CTLA-4 expression in eTreg cells from patients who achieved cCR was significantly lower than that in eTreg cells from patients who did not achieve cCR (Fig. 4c). The expression of CD80 in the APCs of patients who achieved cCR was higher than that in those who did not, although this difference was not significant (Fig. 4d and Extended Data Fig. 5b). Consistent with the FCM data, GSEA showed that CTLA-4-related gene expression increased in patients who did not achieve cCR during the treatment course (Fig. 5d and Extended Data Fig. 8d). These data aligned with a previous report suggesting that IL-1 increases the suppressive function of Treg cells in the inflammatory TME and that intratumoral IL1R1+ Treg cells are highly activated43. Additionally, although no significant differences were found in the FCM analysis of PBMCs, the CBC analysis determined that the neutrophil counts and NLR post-atezolizumab treatment were significantly higher in patients who did not achieve cCR than in those who did (Extended Data Fig. 7). Based on the CBC and tumor tissue sample analysis results, residual ESCCs refractory to combination therapy with dCRT and atezolizumab may cause systemic inflammatory responses in some patients.
Collectively, we suggest that after sequential combination therapy with dCRT and atezolizumab, residual ESCCs could employ an immunosuppressive mechanism in eTreg cells involving the recruitment of iCAFs and MDSCs and upregulation of CTLA-4 expression via IL-1β (Extended Data Fig. 8e). These insights into resistance mechanisms may facilitate the development of novel and effective treatment strategies for ESCC.
Discussion
This phase II proof-of-concept clinical trial evaluated the safety, efficacy and predictive biomarkers of definitive CRT followed by the anti-PD-L1 antibody atezolizumab in patients with unresectable locally advanced ESCC. We selected cases of purely unresectable locally advanced ESCC, mainly clinical stage IVA or IVB, and observed a promising 42.1% cCR rate of 65.8% 1-year OS rate. The Chinese phase II EC-CRT-001 trial indicated a 62% cCR rate and a 78.4% 1-year OS rate in patients treated concurrently with the anti-PD-1 antibody toripalimab; however, only 62% of inoperable patients and 38% of patients with clinical stage IVA were included because patients who refused radical surgery or had cervical esophageal cancer were eligible44.
The synergistic efficacy of combination treatment with CRT followed by atezolizumab was strongly suggested by our translational analysis and the detected predictive signals were potential biomarkers. Moreover, because CRT is a standard treatment option for patients with early stage locally advanced ESCC, the application of sequential combinations of CRT and atezolizumab is potentially expandable to patients with ESCC at all stages and the results of our trial may guide standard ESCC treatment modifications.
Although preclinical studies have shown that CRT improves the therapeutic effect of ICI therapy15,16,17,45, concurrent CRT and ICI therapy increases the risk of serious adverse events (AEs)46 and may decrease the number of immune-activated lymphocytes from adjacent lymph nodes. This clinical trial involved dCRT followed by sequential atezolizumab treatment. In this study, dCRT not only killed tumor cells, but also increased the response to IFNs via enhancement of the cGAS–STING and/or RIG-I pathways16,17,20,21,22,23, resulting in the activation of APCs and effector T cells, thereby inducing an antitumor immune response. In contrast, our analysis of ESCCs during treatment showed that the expression of suppressive immune-checkpoint molecules, such as PD-L1, in tumor tissues was upregulated after dCRT, as shown in previous studies15,16,17. In addition, dCRT increased the expression of CD80, CD86 and PD-L1 in APCs, which is consistent with preclinical studies15,47. These results strongly support our hypothesis that PD-L1 inhibition after the completion of dCRT therapy can further improve the antitumor immune response and, thus, the therapeutic response.
In this study, we evaluated whether TMB and PD-L1 expression are predictive biomarkers of dCRT followed by sequential atezolizumab treatment for ESCC; however, neither TMB nor PD-L1 expression correlated with clinical outcomes. We have previously reported that the balance of PD-1 expression between CD8+ T cells and Treg cells in the TME is a promising biomarker for PD-1 blockade39. Analysis of patient samples pre-combination treatment with dCRT and sequential atezolizumab revealed high and low infiltration of PD-1+CD8+ T cells and Treg cells in the TME, respectively. Additionally, the low proportion of eTreg cells among CD45+ cells in PBMCs was associated with a therapeutic response. Furthermore, analysis of specimens immediately after dCRT treatment and before atezolizumab treatment revealed that PD-1 expression in tumor-infiltrating CD8+ T cells was significantly positively correlated with the therapeutic response, whereas PD-1 expression in tumor-infiltrating Treg cells was significantly negatively correlated with the therapeutic response, consistent with the results of our previous study29,39,48. In some patients in the current study, PD-1 expression in eTreg cells in the TME was elevated after dCRT and PD-1 blockade enhanced suppressive activity, including increased CTLA-4 expression and, thus, dampened antitumor immune responses. This suggests the importance of combining Treg cell-targeted therapies, including anti-CTLA-4, anti-CD25 or anti-CCR8 antibodies11,49,50.
We performed genomic analysis of treatment-resistant ESCCs and found that they harbored genetic abnormalities, including EGFR, PIK3CA, PTEN or KEAP1 mutations or MYC amplification, before treatment. Previous studies have shown that driver gene mutations in cancers are associated with resistance to immunotherapy28,29,30,31,32,33. Our transcriptome analysis of ESCCs before treatment revealed that the ESCCs exhibiting the EMT phenotype with high expressions of MMP1 and MMP13 and/or upregulated TGF-β signaling were likely to be resistant to sequential combination treatment with dCRT and atezolizumab. These results corroborate those of previous studies suggesting that cancers exhibiting the EMT phenotype are resistant to both cancer immunotherapy and CRT34,35,38. High expression levels of MMP1 and MMP13 in tumor samples can be used as negative predictive biomarkers.
Further analysis of the post-treatment samples revealed several possible resistance mechanisms. First, in patients who showed resistance to this combination of dCRT and sequential atezolizumab, the gene signature of the IL-1R pathway was upregulated after combination with dCRT and sequential atezolizumab, possibly causing systemic inflammatory responses. A previous report suggested that after irradiation, IL-1 not only polarizes CAFs toward the inflammatory phenotype, but also triggers oxidative DNA damage, resulting in chemoradiotherapy resistance41. Furthermore, IL-1β reportedly induces MDSCs in the TME40, causing resistance to cancer immunotherapy51. In this study, abundant IL-1β in patients who did not achieve cCR could possibly induce MDSCs in the TME after dCRT, suppressing antitumor immune responses. Second, CTLA-4 expression in eTreg cells in the TME increased, possibly inhibiting antitumor immune responses by downregulating CD80/CD86 (ref. 52). IL-1 strengthens the suppressive activities of Treg cells in the inflammatory microenvironment43.
Finally, high expression of PVR and NECTIN2 may inhibit effector T cell activation via the TIGIT pathway42, resulting in resistance to combination therapy. The addition of inhibitors of IL-1R41, CTLA-4 (refs. 53,54) or TIGIT55 in combination with dCRT and sequential atezolizumab may be considered for patients with ESCC who are refractory to this treatment. Further clinical trials are warranted to develop these novel therapies. The randomized phase III SKYSCRAPER-07 trial (NCT04543617) comparing 1 year of atezolizumab plus anti-TIGIT antibody (tiragolumab), atezolizumab plus placebo, and double placebo after platinum-based dCRT in a 1:1:1 manner in patients with unresectable locally advanced ESCC is ongoing. This trial will confirm the clinical benefits of anti-PD-L1 ± anti-TIGIT blockade after dCRT for unresectable locally advanced ESCC.
Our study has several limitations. This was a nonrandomized phase II study with a relatively small sample size. Although promising cCR and 1-year PFS rates have been reported, long-term survival data have not been reported. To confirm the survival benefits of the sequential combination of dCRT and atezolizumab, the results of the randomized phase III SKYSCRAPER-07 trial are essential.
In conclusion, atezolizumab monotherapy following dCRT resulted in a promising cCR rate and may lead to long-term survival without additional safety concerns. Tumor-infiltrating PD-1+CD8+ T cells detected by FMC and/or mIHC can serve as biomarkers of cCR. Further studies are required to generalize our findings to all stages of ESCC.
Methods
Ethical approval and consent to participate
This study was conducted in accordance with the guidelines for Good Clinical Practice of the International Council on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use and the Ethical Guidelines for Medical and Health Research Involving Human Subjects. The REporting recommendations for tumor MARKer prognostic studies (REMARK)56 and the Consolidated Standards of Reporting Trials (CONSORT)57 were followed. This study was approved by the ethics committee of each participating institution (National Cancer Center Institutional Review Board, Aichi Cancer Center Institutional Review Board, Saitama Cancer Center Review Board, Shizuoka Cancer Center Review Board and Cancer Institute Hospital of the Japanese Foundation for Cancer Research Review Board). All patients provided written informed consent before enrollment.
Study design and treatments
The EPOC1802 TENERGY trial is a multicenter phase II study that assessed the safety and efficacy of sequential combination therapy with atezolizumab following 5-FU plus cisplatin-based chemoradiotherapy (60 Gy/30 fractions without prophylactic irradiation) in patients with unresectable locally advanced ESCC without distant metastasis. All patients who met the eligibility criteria were registered sequentially. The study was registered with the University Hospital Medical Information Network (clinical trial information no. UMIN000034373).
Three-dimensional conformal radiotherapy was administered weekly for five consecutive days with a target dose of 60 Gy in 30 fractions. The gross tumor volume (GTV) was defined as the primary tumor and positive lymph nodes. The clinical target volume (CTV) was defined as the primary GTV plus a 2 cm proximal and distal margin, and the nodal CTV was equal to the corresponding GTV. For recurrent lesions, the CTV was identical to the GTV. The planning target volume was defined as the primary, nodal and recurrence CTV plus a 0.5–2 cm proximal and distal margin as well as a radial margin of 0.5–1.0 cm. Dose constraints to the organs at risk were as follows: lung V20 (percentage of the total lung volume receiving ≥20 Gy) <35%, lung V5 (percentage of the total lung volume receiving ≥5 Gy) <65%, mean heart dose <40 Gy and maximum spinal cord dose <45 Gy, spinal cord V45 (volume of the spinal cord receiving ≥45 Gy). Quality assurance for radiotherapy includes target volume determination, volume to be irradiated, dose and fractionation, and dosimetric criteria for targets and organs at risk, which must align with the trial protocol. Normal tissue dose constraints were based on the QUANTEC criteria. Radiation plans were collected and subjected to a centralized review to assess compliance with protocol requirements pretreatment.
We enrolled patients with primary locally advanced ESCC in the primary cohort (n = 40) and those with locoregionally recurrent ESCC after surgical resection in the exploratory cohort (n = 10). Seven institutions including the National Cancer Center Hospital East, National Cancer Center Hospital, Saitama Cancer Center, Cancer Institute Hospital of the Japanese Foundation for Cancer Research, Shizuoka Cancer Center, and Aichi Cancer Center participated in this study. The enrollment period was from December 2018 to March 2021 and the study was completed in January 2022.
The main inclusion criteria were histologically confirmed primary ESCC, an unresectable tumor with T4b (a tumor invading other adjacent structures such as the aorta, vertebral body or trachea) and/or regional lymph node and/or supraclavicular lymph node invasion of adjacent structures other than the esophagus, treatment with two cycles of CDDP (70 mg m−2) on day 1 plus 5-FU (700 mg m−2) on days 1–4 every 4 weeks and 60 Gy/30 fractions of radiotherapy, and the absence of esophageal perforation, esophagotracheal fistula, esophagomediastinal fistula, bleeding from arterial invasion or respiratory stenosis when CRT was completed (Supplementary Table 6). Clinical stage was assessed by members of a multidisciplinary team according to the Eighth TNM staging system of the American Joint Committee on Cancer58.
Both chemotherapy and radiation are highly qualified based on quality assurance and quality control programs. The study treatment was initiated within 6 weeks of CRT completion. Patients received 1,200 mg atezolizumab every 3 weeks for 12 months (maximum of 17 doses) or until disease progression, patient withdrawal, investigator’s decision, pregnancy or unacceptable toxicity, whichever occurred first. The study protocol is included in the Supplementary Information.
End points and assessments
The primary end point was cCR rate, as determined by the investigator’s assessment of the primary locally advanced ESCC cohort. The secondary end points were the cCR rate determined by an investigator’s assessment in the locoregionally recurrent ESCC cohort. Moreover, we assessed the cCR rate determined by central assessment, overall response rate determined by the investigator’s assessment, PFS, OS and the incidence of AEs in both cohorts as the secondary end points59.
Complete response (CR) was determined by both computed tomography (CT) and endoscopy based on the Response Evaluation Criteria in Solid Tumors v.1.1, with modifications, and the Japanese Classification of Esophageal Cancer (11th edition). CR was defined as absence of erosive changes with irregularities, ulcerative lesions, obvious elevation (including submucosal mass elevations) or histopathological malignancy. Both CT and endoscopic examinations were performed every 4 weeks until CR, following confirmation of CR after more than 4 weeks and every 12 weeks thereafter59. Before atezolizumab therapy, we performed CT and endoscopy, with no patient achieving CR immediately after CRT.
PFS was defined as the period from the date of enrollment to the date of disease progression or death from any cause, whichever occurred first. OS was defined as the period from the date of enrollment to death from any cause. We assessed AEs according to the Common Terminology Criteria for Adverse Events v.4.0 (ref. 59).
Clinical data were collected using Medidata Rave (v.2015.1.1) and Medidata Classic Rave (v.2018.2.0). The data center of the Clinical Research Support Office, National Cancer Center Hospital East (NCCHE-OCRS) confirmed patient eligibility. Data collection, analyses and interpretation were performed by the NCCHE-OCRS.
Complete blood count
The CBCs, which included WBCs, RBCs and platelet counts and the concentrations of hemoglobin, hematocrit and WBC differentials, were determined using automated hematology analyzers at each institute.
PD-L1 evaluation and biomarker analyses
Serial biopsies from the primary site and blood collections were performed at three time points (<4 weeks before treatment, <6 weeks after dCRT and 4 ± 2 weeks after the first atezolizumab dose). PD-L1 expression in the tumor cells was evaluated using the VENTANA PD-L1 (SP263) assay. Using the collected samples, we investigated the biomarkers for efficacy or resistance to the study treatments. We collected available samples, including frozen tumor tissue specimens, formalin-fixed paraffin-embedded tumor tissue specimens and frozen blood specimens, from patients enrolled in this study. We analyzed WES and RNA-seq data, TMB, phenotype of immune-competent cells using immunohistochemistry and PD-L1 status. Fresh tumor tissues and blood specimens were obtained from patients enrolled in the National Cancer Center Hospital East (n = 18). The fresh specimens were processed as described in the following sections and subjected to FCM.
WES
DNA and/or RNA were extracted from frozen tumors or blood samples at each time point using an AllPrep DNA/RNA Mini kit (QIAGEN). WES was performed on 27 patients before treatment. Exome library construction was performed using the SureSelect V6-Post kit (Agilent Technology) as described in the manufacturer’s protocol. The adaptor-ligated fragments were amplified by PCR. Paired-end sequencing (151 bp) was performed using HiSeq X (Illumina). For exome analyses, paired-end sequencing reads with masked nucleotides and a quality score of <20 were aligned to hg38 using BWA-MEM (http://bio-bwa.sourceforge.net/) and Bowtie2 (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml). Both synonymous and nonsynonymous somatic mutations were identified using our inhouse caller and two publicly available mutation callers: the Genome Analysis Toolkit (https://gatk.broadinstitute.org/hc/en-us), MuTect2 and VarScan2 (http://varscan.sourceforge.net/). Mutations were discarded if any of the following criteria were met: a total read number of <100, a variant allele frequency in the tumor samples of <0.05, a mutant read number in the germline control samples of >2 mutations detected in only one strand of the genome or a variant present in normal human genomes in either the 1000 Genomes Project dataset (https://www.internationalgenome.org/) or our inhouse database. Gene mutations were annotated using SnpEff (http://snpeff.sourceforge.net).
RNA-seq analysis
RNA-seq libraries for paired patient tumor samples (n = 28) during treatment were prepared using the TruSeq RNA Access Library Prep kit and 101 bp paired-end sequencing was performed using the Illumina platform at MACROGEN. The raw FASTQ read sequences were trimmed and filtered using Trim Galore (v.0.6.5), to remove low-quality bases. Adaptor sequences and reads mapped to rRNA genes were removed using bowtie2 (v.2.4.1). The filtered FASTQs were then aligned to the human genome (GRCh38) using STAR (v.2.7.3a). Gene expression values were calculated as either counts or transcripts per million (TPM) using Stringtie (v.2.2.1), with GENCODE (v.34) gene annotation. Normalization and differential expression analyses were performed using DESeq2 with the count data.
GSEA and ssGSEA
GSEA of gene expression data from RNA-seq was performed using clusterProfiler (v.4.4.4)60. The following gene signatures were analyzed: HALLMARK and C2 curated gene sets from MSigDB25,61, inflammatory cancer-associated fibroblast (iCAF) signature41, and CAF signature62. GSEA plots were generated using Enrichment Plot (v.1.16.2). ssGSEA was conducted using GSVA (v.1.44.5)63 with TPM profiles. The immunological gene signature of Dummer et al.64 was analyzed. ssGSEA scores were normalized using Z-scaling. These analyses were performed using R software (v.4.2.2).
Multicolor FCM analysis
Paired TILs (n = 19) and PBMCs (n = 18) were collected from patients treated at the National Cancer Center Hospital East. To collect TILs, tumor tissues were minced and treated within 72 h after biopsy with a new TIL preparation protocol using newly optimized tissue preservation reagent (tumor tissue dissociation reagent; TTPR) and TIL-isolation reagent (tumor tissue dissociation reagent; TTDR) co-developed with BD Biosciences (see details in PCT/JP2020/005991) as previously described, and the prepared cells (TILs) were subjected to analyses65. In brief, fresh tumor specimens were placed into a tube filled with TTPR solution at 2–8 °C for ≤72 h. After preservation, tumor specimens were cut into small pieces (approximately 0.5–1.0 mm per side) using scissors in a tube. TTDR was added to the tubes. The cap was tightly closed and the tube was incubated at 20–25 °C for 15 min on the nutator. After incubation, a cold stop buffer was added to the tubes. The cell suspension was passed through a cell strainer and centrifuged to isolate tumor-infiltrating immune cells. The PBMCs were isolated by density gradient centrifugation using Ficoll-Paque (GE Healthcare). FCM staining and analysis were performed as previously described. The antibodies used for FCM analyses are summarized in Supplementary Table 7. To evaluate the influence of atezolizumab treatment on the stainability of the anti-PD-L1 antibody, PBMC samples were incubated with or without atezolizumab (final concentration of 200 μg ml−1) at 37 °C for 20 min before staining with FCM antibodies. Cells (PBMCs and TILs) were washed using a washing solution and stained with surface antibodies, including the anti-CTLA-4 antibody (1:50 dilution). Intracellular staining of some molecules, including FOXP3, was performed with the corresponding antibodies (for example, anti-FOXP3 monoclonal antibody) and the Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific), according to the manufacturer’s instructions. After washing, the cells were analyzed using a FACSymphony (BD Biosciences) and FlowJo software v.10.8.1 (TreeStar). The gating strategy used in this study is shown in Extended Data Fig. 9.
mIHC
mIHC was performed on tumor tissues from 40 patients. Optimized mIHC was performed using an Opal six-color tyramide signaling amplification kit (Akoya Biosciences). Cells were stained with antibodies against CD3 (clone SP7, ab16669, Abcam, 1:200 dilution), CD8 (clone C8/144B, M710301-2, Agilent, 1:200 dilution), FOXP3 (clone 236A/E7, ab96048, Abcam, 1:100 dilution), cytokeratin (clone AE1+AE3, ARG56129, Arigo Biolaboratories, 1:200 dilution) and PD-1 (clone EPR4877(2), ab137132, Abcam, 1:50 dilution). Fluorescence signals were captured with the fluorophores Opal 520, Opal 540, Opal 570, Opal 620, Opal 650 and Opal 690. Multiplex-stained slides were obtained using the Vectra Polaris Quantitative Pathology Imaging System (PerkinElmer). The images were analyzed using inForm (v.2.4.4) image analysis software (PerkinElmer) and Spotfire software (v.12.2; TIBCO Software). Whole-slide images were analyzed using IA software (HALO v.2.3; Indica Labs). Slides were manually annotated to remove tissue artifacts (for example, folds, air bubbles and fluorescent precipitates), and segmentation and markup of individual cells were performed, reviewed and scored by two pathologists using the Indica Labs-HighPlex FL module.
Statistical analysis for translational analysis
Data analyses were performed using GraphPad Prism v.9 (GraphPad Software). Microsoft Excel 2016 (Microsoft Corporation) was used for data collection. ssGSEA scores, TPM, FCM analysis parameters and mIHC parameters were calculated using a paired one-way analysis of variance or an unpaired two-tailed t-test. The Holm–Sidak test was performed for multiple comparisons. The differences with P ≤ 0.05 were considered significant. The data distribution was assumed to be normal; however, this was not formally tested.
Statistics and reproducibility
The reported cCR rates range from 11 to 25%. In the present study, we estimated the rate of cCR threshold at 20%, based on discussions with investigators and statisticians, and regulatory science consultation by healthcare authorities. Therefore, we calculated the sample size of the primary locally advanced ESCC cohort at 38 with a cCR of 40% deemed promising and one of 20% deemed unacceptable (one-sided α, 0.05; β, 0.2). Using this calculation, we included a maximum of 40 patients with locally advanced primary ESCC in the primary cohort. We further set the planned sample size of the exploratory cohort with postoperative locoregionally recurrent ESCC at a maximum of ten patients in an exploratory manner. The cCR was determined by the investigator’s assessment and the CI was estimated using the exact binomial method. The primary end point was analyzed in the first 38 consecutive patients who were deemed eligible and received the protocol treatment at least once. Results were considered statistically significant for more than 13 confirmed cases of CR. PFS and OS were estimated using the Kaplan–Meier method. Statistical analyses were performed using the SAS software (v.9.4; SAS Institute). The data cutoff date for the analyses was 4 March 2022. No data were excluded from the analyses and the experiments were not randomized. The investigators were not blinded to the allocation during the experiments or outcome assessment. Data collection and analysis were conducted without blinding to the conditions of the experiments. As this was a single-arm phase II study, specimens were collected from patients only once at each time point and the volume of each specimen collected was small. Therefore, it was impossible to replicate the validation experiments.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Responses