The phase Ib study evaluating the safety and feasibility of definitive chemoradiotherapy combined with an anti-PD-1 antibody, camrelizumab, as the first-line therapy in locally advanced ESCC (ClinicalTrials.gov NCT03671265) (4). Specifically, camrelizumab (SHR1210, Jiangsu Hengrui Medicine Co. Ltd., China) was given on day 1 of every 2-week period from the beginning of radiotherapy up to 32 weeks, concurrently with radiotherapy for 6 weeks, and with chemotherapy for 4 weeks (4). The exploratory endpoints of this phase Ib study were local and systematical immune characteristics, and potential predictive biomarkers for combination treatment outcome.

Baseline (n = 20) and on-treatment (after 40 Gy radiation, n = 18) tumor biopsies were collected (Additional file 1: Table S1 ). Deep biopsy samples of tumor tissues were collected under endoscopic ultrasonographic guidance (18, 19) and made into formalin-fixed paraffin-embedded (FFPE) tissue blocks.

This study was conformed to the ethical principles outlined in the Declaration of Helsinki, and the protocol was approved by the institutional review board and ethics committee at Tianjin Medical University Cancer Institute & Hospital (E2018142). All patients provided written informed consent to participate. This study was registered on ClinicalTrials.gov (NCT03671265).

To dynamically monitor the tumor immune microenvironment at baseline and during the combination, serial FFPE slides of the biopsy specimens were stained by using tyramide signal amplification (TSA)-based multiplex immunofluorescence assay method. The multi-color immunofluorescence staining was automatically performed in Bond III automated stainer (Leica, USA). The TSA 5-color kit (#D110051-50T) and TSA 670 (#D110016-100T) were bought from Yuanxibio, China. The stanning panel was as follows: Anti-PD-L1 (#13684, CST, 1:800)/TSA 570, anti-panCK (#GM351507, Gene Tech, 1:6)/TSA 520, anti-CD11c (#45581, CST, 1:300)/TSA 620, anti-CD68 (#GM087602, Gene Tech, ready-to-use)/TSA 670. In the first staining cycle, FFPE slides were immersed in xylene to remove paraffins on the slides. Transfer slides to 100%, 95%, 70%, and 50% alcohol, respectively. Perform antigen retrieval to unmask the antigenic epitope by using microwave treatment in optimal buffer as recommended. Add blocking buffer onto the sides. Drain off blocking buffer from the slides and apply appropriately diluted primary antibody. Add HRP-conjugated second antibody. Then, add fluorescent TSA reagent. Microwave treatment was applied to remove the first antibodies deposited and the staining process is repeated for a subsequent target. The process is repeated until all targets have been labeled. In the last steps before imaging, add 4’,6-diamidino-2-phenylindole (D1306; Thermofisher) to visualize cell nuclei and apply a cover slip. The slides were ready to image.

A whole slide scan was performed for each fluorescence-stained slide using a digital microscopy scanner Pannoramic MIDI tissue imaging system (3DHISTECH Ltd., Hungary). Because both tumor cells and normal epithelial cells have positive CK expression, it is hard to distinguish these two cell types in immunofluorescence staining. To exclude the normal epithelial cells in analysis, we applied Hematoxylin and Eosin (H&E) staining in the tissue sections after finishing the fluorescence scan. Images were analyzed by Indica Halo software (Indica Labs, UK). Two independent blinded pathologists performed histologic evaluation and supervised to split the tumor and stromal compartments by using Halo software. Cells were phenotyped into the following subsets: DC (CD11c⁺), macrophage (CD68⁺), tumor cell (CK⁺), and PD-L1⁺ subpopulations of these cells.

Immune cell infiltration was evaluated as the number of cells per slide, in the tumor compartment, stromal compartment, or total viable tissue area of the slides, respectively. To evaluate the spatial relationship between immune cells and tumor cells, the distance between each tumor cell and its nearest neighbor immune cells was measured.

To investigate the tumor mutation at baseline, the biopsy specimens from 14 patients before the combination were sequenced by using FoundationOne CDx (F1CDx) and FDA-approved 324-gene panel assay conducted by DIAN (Hangzhou Lab) with licensed technologies, to assess the tumor mutation burden (TMB) (4, 20).

Statistical significance between groups was compared using non-parametric two-sided Mann–Whitney U tests for two independent samples or Wilcoxon Signed-Rank tests for paired samples, and correlations were evaluated assuming a non-Gaussian distribution (Spearman correlation) unless otherwise indicated. OS was defined as the time from inclusion until death from any cause or the last date of follow-up time. Progression-free survival (PFS) was defined as the time from inclusion until the date of objective disease progression or death from any cause in the absence of progression. The Kaplan–Meier analysis was used to estimate OS and PFS. Differences in survival were compared with log-rank tests. The best cutoff of Kaplan–Meier survival analysis was calculated by the Youden index of the ROC curve.

All analyses were performed using SPSS v.25.0 (STATA, College Station, TX, USA). Reported p values were two-sided, and the significance level was set at 0.05. Survival curves and summary graphs were performed using GraphPad Prism v.8.0. The data cutoff date for all analyses was May 1, 2021.

We used multiplex immunofluorescence to identify DCs and macrophages in the tumor microenvironment ( Figures 1A–G ). A total of 36 scanned slides were finally included in analysis except for two baseline slides without tumor tissues, including 18 baseline and 18 on-treatment specimens, with 16 matched pairs at these two time points (Additional file 1: Table S1 ).

At the updated data cutoff date of May 1, 2021, the median follow-up duration was 26.6 months (95% CI 24.3 to 29.0). Thirteen patients were alive, and 11 patients were free of progressive disease (Additional file 1: Table S2 ). The OS and PFS ranged from 8.2 to 31.4 months and from 3.9 to 31.4 months, respectively. We initially analyzed the association between the total CD11c⁺ DCs, CD68⁺ macrophages (included in both tumor and stromal compartments), and the clinical outcome. However, the results of Kaplan–Meier analysis showed that neither the total DCs nor the total macrophages were associated with patient survival.

To investigate if these APCs located in different tissue compartments contributed to the combination treatment outcome, we then separated DCs and macrophages according to their location in the tumor or stromal compartment (Additional file 1: Table S3 ). We found that the high level of DCs located in the tumor compartment (defined as tumor DCs) at baseline, but not the DCs located in the stromal compartment (defined as stromal DCs), was associated with improved OS (p = 0.040, Figure 1H ). High level of tumor macrophages at baseline had a tendency to be correlated with better OS (p = 0.054, Figure 1I ). On the contrary, a high level of stomal DCs during the combination was related to poor PFS (p = 0.018, Figure 1J ). These results demonstrated that the DCs and macrophages located in the tumor compartment played an important role in antitumor response in ESCC patients receiving combined chemoradiotherapy and PD-1 blockade.

Using multi-immunofluorescence assay, we could identify the PD-L1 expression on tumor cells, DCs, and macrophages simultaneously (4) ( Figures 1C–G ). Of the total PD-L1 expressed cells, the median percentages of tumor cells, DCs, and macrophages were 30.48%, 28.54%, and 15.44%, individually (Additional file 1: Figure S1A ). PD-L1⁺ tumor cells decreased significantly (30.48% vs. 5.46%, p = 0.008) after the combination (Additional file 1: Figures S1B, C ).

Because of the close associations between the survival and APCs in the tumor compartment in our above finding, we here focused on PD-L1 expression on the APCs in the tumor compartment. The median percentage of DCs in the tumor compartment was 3.524% (95% CI, 2.754%–6.684%) at baseline, and increased to 11.394% (95% CI, 8.295%–23.439%) during combination treatment (Additional file 1: Figure S2A ). The median proportions of PD-L1^- and PD-L1⁺ DCs in the tumor compartments were 2.297% (95% CI, 1.424%–3.927%) and 0.685% (95% CI, 0.425%–3.663%) at baseline ( Figure 2A ), and 4.579% (95% CI, 3.149%–16.728%) and 5.160% (95% CI, 3.322-8.896%) during the combination ( Figure 2A ), respectively. The PD-L1 expression on the tumor DCs showed great variability in individuals both at baseline and during the combination treatment ( Figure 2B ). The percentage of PD-L1^- DC was higher than that of PD-L1⁺ DCs in baseline tumor compartments (74.73% vs. 25.27%, p = 0.048, Figure 2B ). While this difference disappeared during the combination treatment ( Figure 2B ). The Kaplan–Meier analysis showed that high levels of PD-L1^- DCs in both baseline and on-treatment tumor compartments were associated with improved OS (baseline, p = 0.011, Figure 2C ; on-treatment, p = 0.042, Figure 2D ). However, the PD-L1⁺ DCs in baseline or on-treatment tumor compartments were not related with survival (Additional file 1: Figure S3 ).

The median percentage of macrophages in the tumor compartment was 2.156% (95% CI, 1.786%–4.850%) at baseline, and increased to 5.822% (95% CI, 4.917%–11.415%) during combination treatment (Additional file 1: Figure S2B ). The median proportions of PD-L1^- and PD-L1⁺ macrophages in tumor compartments were 1.464% (95% CI, 1.043%–2.669%) and 0.790% (95% CI, 0.469%–2.455%) at baseline ( Figure 2E ), which elevated to 3.043% (95% CI, 2.382%–5.290%) and 2.376% (95% CI, 1.694%–6.964%) during the combination ( Figure 2E ). Similar to DCs, tumor macrophages also exhibited heterogenous PD-L1 expression inter-individuals ( Figure 2F ). Patients with high level of PD-L1^- macrophages in baseline tumor compartments had longer PFS (p = 0.032, Figure 2G ). Patients having high level of PD-L1^- macrophages in on-treatment tumor compartments had better OS (p = 0.018, Figure 2H ) and PFS (p = 0.028, Figure 2I ). These results suggested that the PD-L1^- DCs and macrophages in the tumor compartments promoted antitumor efficacy of chemoradiotherapy combined with PD-1 blockade in ESCC.

We also investigated the association between dendritic cells and macrophages in the tumor compartment. We found loose correlation between dendritic cells and macrophages in the on-treatment tumor compartment (Spearman coefficient 0.484, p = 0.042, Additional file 1: Figure S4A ). The close relationship was observed between PD-L1⁺ dendritic cells and PD-L1⁺ macrophages in both baseline and on-treatment tumor compartment (Spearman coefficient 0.899 and 0.905, p < 0.001, Additional file 1: Figures S4B, C ).

As both the compartment distribution and PD-L1^- or PD-1⁺ APCs inconsistently contributed to the outcome of the combination treatment, we next quantified the dynamical spatial relationship between tumor cells and these APC subpopulations, respectively. By using spatial multi-immunofluorescence analysis, we identified the coordinate position of the cells of each tissue section, and measured the distances from each tumor cell to the nearest PD-L1^- or PD-L1⁺ DCs or macrophages. The index of nearest distance was defined as the average of the closest distances from all tumor cells to the neighbors of each tissue section ( Figures 3A–F ). The distance from tumor cells to PD-L1⁺ DCs exhibited much more variability compared with that from tumor cells to PD-L1^- DCs ( Figure 3G ). Under the combination treatment, PD-L1^- DCs moved nearer to tumor cells compared with the corresponding ones at baseline (p = 0.012). The PD-L1^- DCs also located closer to tumor cells than the PD-L1⁺ DCs during the combination treatment (p = 0.048) ( Figure 3G and Table 1 ). Similarly, a higher variability was found in the distance from tumor cells to the PD-L1⁺ macrophages ( Figure 3H ). The on-treatment PD-L1^- macrophages were closer to tumor cells than the baseline PD-L1^- macrophages and the on-treatment PD-L1⁺ macrophages ( Figure 3H and Table 1 ). Kaplan–Meier analysis showed that patients with PD-L1^- macrophages nearer to tumor cells during the combination treatment had better OS (p = 0.018, Figure 3I ) and PFS (p = 0.013, Figure 3J ). These results showed that PD-L1^- DCs and macrophages locating adjacently to the tumor cells provided them spatial advantage to participate in the antitumor immune response under the combination treatment.

To further explore the interaction between tumor cells and the DCs and macrophages, we next assessed their spatial relationship by dividing the tumor cells and APCs into PD-L1^- and PD-L1⁺ subpopulations, respectively ( Figures 4A–D ). Firstly, we calculated the distance from each PD-L1^- tumor cells to the nearest PD-L1^- and PD-L1⁺ DCs. We found that the PD-L1^- DCs located closer to the PD-L1^- tumor cells compared with PD-L1⁺ DCs both at baseline and during the combination treatment (baseline, p = 0.008; on-treatment, p = 0.016, Figure 4E and Table 2 ). After the combination treatment, the PD-L1^- DCs further moved nearer to the PD-L1^- tumor cells than those at baseline (p = 0.015, Figure 4E and Table 2 ). Next, we analyzed the distance from PD-L1⁺ tumor cells to PD-L1^- and PD-L1⁺ DCs. The distance from the PD-L1⁺ tumor cells to PD-L1^- or PD-L1⁺ DCs did not change significantly under the combination treatment. However, the PD-L1⁺ DCs located closer to PD-L1⁺ tumor cells than PD-L1^- DCs during the combination treatment (p = 0.010, Figure 4F and Table 2 ).

In analyzing the distance from PD-L1^- tumor cells to PD-L1^- and PD-L1⁺ macrophages ( Figures 5A–D ), we found that the PD-L1^- macrophages were farther away from the PD-L1^- tumor cells compared with PD-L1⁺ macrophages at baseline (p = 0.005, Figure 5E and Table 2 ). PD-L1^- macrophages migrated closer to PD-L1^- tumor cells (p = 0.039), while PD-L1⁺ macrophages moved away from PD-L1^- tumor cells (p = 0.026) during the combination treatment ( Figure 5E and Table 2 ). Consequently, opposite to the distances at baseline, PD-L1^- macrophages got closer to PD-L1^- tumor cells than PD-L1⁺ macrophages during combination treatment (p = 0.039, Figure 5E and Table 2 ). We then analyzed the distance from PD-L1⁺ tumor cells to PD-L1^- and PD-L1⁺ macrophages. The distance from PD-L1⁺ tumor cells to PD-L1^- macrophages as well as to PD-L1⁺ macrophages did not alter under the combination treatment ( Figure 5F and Table 2 ). However, PD-L1⁺ macrophages located nearer to PD-L1⁺ tumor cells than PD-L1^- macrophages during the combination treatment (p = 0.048, Figure 5F , and Table 2 ).

Accordingly, of the APC subsets located relative to PD-L1^- tumor cells, PD-L1⁺ macrophages were the nearest at baseline, while PD-L1^- DCs and macrophages turned to the nearest during the combination treatment. Inconsistently, PD-L1⁺ DCs and macrophages were the nearest APC subsets to PD-L1⁺ tumor cells both before and under the combination treatment ( Figure 6 ). These results further proved the distinct spatial pattern of APC subpopulations in ESCC, which was also modulated by chemoradiotherapy combined with PD-1 blockade.

The results of Kaplan–Meier analysis showed that shorter distances from PD-L1⁺ tumor cells to PD-L1^- DCs and to PD-L1^- macrophages at baseline were associated with worse PFS and OS, respectively (p = 0.034, p = 0.003, Figures 7A, B ). On the contrary, shorter distances from PD-L1⁺ tumor cells to PD-L1⁺ DCs and PD-L1^- macrophages during the combination treatment were both correlated with better OS (p = 0.023, p = 0.018, Figures 7C, D ). Shorter distances from PD-L1^- tumor cells to PD-L1^- macrophages during the combination treatment also predicted improved OS (p = 0.018, Figure 7E ) and PFS (p = 0.008, Figure 7F ). These results elicited that the close interaction between tumor cells and APCs with different PD-L1 expression contributed to divergent outcome of chemoradiotherapy combined with PD-1 blockade in ESCC.

Lastly, to explore the tumor-derived factors that might affect the distribution of DCs and macrophages, we evaluated the association between TMB and the distance of these APCs to tumor cells. We found that higher TMB was associated with shorter distance between macrophages and tumor cells (p = 0.001, Figure 8A ), especially between macrophages and PD-L1^- tumor cells (p = 0.001, Figure 8B ). Similarly, higher TMB was correlated with shorter distance between DCs and PD-L1^- tumor cells (p = 0.029, Figure 8C ), as well as between DCs and PD-L1⁺ tumor cells (p = 0.049, Figure 8D ). The association between high TMB and far distance of the APCs to tumor cells was not observed. These results indicated that high TMB would result in closer distribution of APCs to tumor cells.