Tertiary lymphoid structures correlate with reduced recurrence risk and enhanced antitumor immunity in esophageal squamous cell carcinoma with pathologic non-complete response to neoadjuvant chemoimmunotherapy

Abstract

Background

The majority of patients with locally advanced esophageal squamous cell carcinoma (ESCC) undergoing neoadjuvant chemoimmunotherapy (nCIT) failed to achieve pathologic complete response (pCR), had high risk of postoperative recurrence and lacked prognostic biomarkers. Tertiary lymphoid structures (TLS) are organized aggregates of immune cells and have the potential to regulate antitumor immune response. This study aimed to investigate the prognostic value and immune profile of TLS in non-pCR ESCC.

Methods

We first analyzed clinicopathological features, recurrence events, and survival outcomes according to TLS status. Subsequently, based on the single-cell sequencing data, we analyzed the differences in the infiltration level, functional status and interaction mode of immune cells based on TLS status. The expression pattern of signature genes and the spatial localization of key immune cell subsets were verified through bulk RNA sequencing and multiplex immunohistochemistry.

Results

The TLS(+) group demonstrated a lower likelihood of postoperative recurrence and superior survival rates relative to the TLS(−) group. The key immune cell subsets responsive to immunotherapy were enriched in the TLS(+) group, and the immune cells in the TLS(+) group showed a functional state of high activation and low exhaustion. Multiplex immunohistochemistry and cell–cell communication analysis suggested that tumor reactive T cells were spatially colocalized with B cells and antigen presenting cells in TLS and exhibited high interaction potential. In the TLS(+) group, we also identified precursor exhausted T cells and long-lived plasma cells with tumor reactivity and matured affinity. The presence of TLS correlated with enhanced synergistic interaction, activation and maturation of immune cells, suggesting a potential role in shaping in situ antitumor immunity.

Conclusions

TLS status was the independent predictor of postoperative recurrence in non-pCR ESCC. TLS status correlated with the composition, functional state, and interaction patterns of immune cells. Specialized immune niches existed in non-pCR ESCC with TLS, potentially contributing to antitumor immune responses.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40164-026-00748-6.

Background

Although neoadjuvant chemoradiotherapy followed by esophagectomy significantly improved the survival rate over surgery alone among locally advanced esophageal squamous cell carcinoma (ESCC), more than 40% of patients experienced postoperative recurrence [1, 2]. Therefore, more effective neoadjuvant therapy is urgently needed to improve treatment response and long-term survival. Based on the success of immunotherapy as the first-line therapy for unresectable advanced ESCC [3], the application of immunotherapy has gradually expanded to neoadjuvant setting for resectable locally advanced ESCC [4].

Pathological complete response (pCR) was defined as no evidence of residual tumor in resected tissue samples after neoadjuvant therapy [5]. Patients who achieved pCR had better survival outcomes than patients with residual disease [1, 5–9]. Recently published meta-analysis indicated that nearly 70% of patients with locally advanced ESCC failed to achieve pCR after neoadjuvant chemoimmunotherapy (nCIT) [4, 10]. Non-pCR patients represented the majority of patients receiving nCIT and had high risk of postoperative recurrence. Previous nCIT studies mainly focused on treatment-related toxicity, surgical complications, and pathologic response rate [4, 10]; however, the clinical outcomes and immune landscape of non-pCR patients remained poorly understood, underscoring the need to decipher the immune microenvironment to investigate survival variability and identify novel therapeutic targets in this clinically challenging population.

Tertiary lymphoid structures (TLS) are organized aggregates of immune cells that emerge in settings of malignancy or chronic inflammation [11–13]. In multiple solid tumors, TLS had the potential to regulate antitumor immune response [11]. The presence of TLS was associated with favorable survival outcomes in early-stage patients undergoing upfront surgery and better response to immunotherapy in patients with locally advanced or metastatic disease [11]. ESCC is an inflammation-associated malignancy characterized by a heterogeneous immune microenvironment, extensive lymphatic drainage and marked genomic instability [14, 15], in which immune surveillance and local immune architecture, such as TLS, may contribute to disease control. In addition, lymphocyte infiltration and proinflammatory cytokine expression are frequently observed in ESCC lesions [16], potentially creating a favorable microenvironment for TLS formation. Therefore, ESCC represents a relevant disease context to evaluate the clinical and biological implications of TLS. However, despite extensive TLS research in other solid tumors, the roles of TLS in ESCC, particularly in the non-pCR population after nCIT, remain poorly defined.

In this study, we comprehensively characterized the clinical relevance and immunological landscape of TLS in non-pCR ESCC after nCIT. We first analyzed clinicopathological features, recurrence events, and survival outcomes according to TLS status to determine the clinical relevance of TLS in this specific cohort (Fig. 1A). By integrating multi-omics datasets, we further delineated the infiltration level, functional status and intercellular communication of immune cells based on TLS status, thereby providing new insights into the TLS-associated immune ecosystem in non-pCR ESCC.

Fig. 1.

Clinical significance and single-cell profiling of TLS in non-pCR ESCC. A Study overview. B Representative H&E and mIHC staining of TLS maturation stages. C Comparison of recurrence risk in different TLS subgroups of clinical cohort A. D–E Survival curves of overall survival (D) and recurrence-free survival (E) among different TLS subgroups of clinical cohort A. F UMAP visualization of captured cells. G Fractions of immune cells in different TLS subgroups. H Gene Ontology annotation of differentially expressed genes between different TLS subgroups.

Methods

Study population

Clinical cohort

Two independent clinical cohorts were included in this study. Clinical cohort A comprised patients from two high-volume centers and was used to evaluate the predictive value of TLS status for postoperative recurrence and survival outcomes. Clinical cohort B, derived from an independent third center, was used to externally confirm the robustness of the predictive value of TLS.

Specifically, clinical cohort A comprised 178 locally advanced (cT1b-2N1-3M0 or cT3-4aN0-3M0) ESCC patients who received nCIT followed by radical esophagectomy from December 2019 to December 2022 at Fudan University Shanghai Cancer Center and The Affiliated Hospital of Qingdao University. Eligibility criteria were as follows: (1) histologically confirmed ESCC; (2) without a prior history of cancer; (3) receiving 2–4 cycles of anti–PD-1 immunotherapy (Camrelizumab/Pembrolizumab/Tislelizumab) plus platinum based doublet chemotherapy; (4) R0 resection without perioperative mortality; (5) with residual tumor after nCIT.

Clinical cohort B comprised 48 locally advanced ESCC patients who received nCIT followed by radical esophagectomy from December 2021 to February 2023 at Ruijin Hospital, Shanghai Jiao Tong University School of Medicine. The eligibility criteria of clinical cohort B were consistent with those of clinical cohort A.

Single-cell cohort

Single-cell sequencing cohort comprised 7 locally advanced ESCC patients who received nCIT followed by radical esophagectomy from September 2021 to July 2022 at Fudan University Shanghai Cancer Center (Table S7). The eligibility criteria of single-cell sequencing cohort were consistent with those of clinical cohort. Post-treatment residual esophageal tumors were obtained and dissociated into single-cell suspensions. Subsequently, CD45⁺ immune cells were sorted and sent for single-cell RNA sequencing (scRNA-seq) combined with single-cell T/B cell receptor sequencing (scTCR/BCR-seq).

Validation cohort

We obtained additional tumor samples (validation cohort) following the same eligibility criteria to verify the expression pattern of signature genes and the spatial distribution of key cell subsets. Validation cohort comprised 30 locally advanced ESCC patients who received nCIT followed by radical esophagectomy from September 2021 to October 2022 at Fudan University Shanghai Cancer Center. Seven patients from single-cell sequencing cohort were included in the validation cohort. Post-treatment residual esophageal tumors were obtained and sent for bulk RNA-seq and multiplex immunohistochemistry (mIHC) analyses.

This study was approved by the institutional review board (IRB #20875697P).

TLS evaluation

Serial FFPE tissue sections (4 µm thickness) of residual esophageal tumors from clinical cohort, single-cell sequencing cohort, and validation cohort were collected. Hematoxylin & Eosin (H&E) staining and multiplex immunohistochemistry (mIHC) staining of CD3, CD20, CD21, and CD23 were performed to identify TLS. Two experienced pathologists (Drs. Shen and Hou) independently evaluated the maturity of TLS distributed in original tumor bed and invasive margin (≤ 1 mm). The discrepancies were resolved through discussion. TLS maturation stages were classified as follows (Fig. 1B): (1) lymphoid aggregates (LA), characterized by ill-defined aggregates of lymphocytes without CD21/CD23 signal; (2) primary follicle-like TLS (pTLS), characterized by well-defined round-shaped aggregates of lymphocytes with CD21 signal but not CD23 signal; (3) secondary follicle-like TLS (sTLS), characterized by well-defined round-shaped aggregates of lymphocytes with germinal center formation (CD21⁺CD23⁺) [11, 17, 18]. According to the maturity of TLS, patients were divided into three subgroups: (1) TLS(−) group: patients lacking pTLS and sTLS; (2) pTLS group: patients with at least one pTLS but no sTLS; (3) sTLS group: patients with at least one sTLS. It was reported that LA represented the very early stage of TLS formation and could not induce effective immune response [19, 20], and its infiltration level did not affect the prognosis of ESCC [21]. Therefore, patients with only LA were classified into TLS(−) group in this study.

Follow-up

Patients were advised to undergo chest and abdominal CT scans, neck ultrasound, and upper gastrointestinal series every three months for the initial two years, followed by evaluations every six months until recurrence or death. Distant metastasis was identified as the presence of cancer in the lung, liver, bone, brain, pleura, or peritoneum. Locoregional recurrence included cancer reappearance in the esophagus, anastomotic site, and cervical/mediastinal/abdominal lymph nodes [22]. Recurrence was established on radiologic, histologic, or cytologic evidence. Detailed classifications of recurrence events were in accordance with previous studies [1, 22, 23]. Overall survival (OS) was calculated as the time from surgery to death due to any cause, while recurrence-free survival (RFS) was calculated as the duration from surgery to either recurrence or death from any cause.

Single-cell sequencing

Fresh post-treatment residual esophageal tumors from single-cell sequencing cohort (n = 7) were obtained and stored in MACS Tissue Storage Solution (Miltenyi Biotec) until processing (within 1 h). Samples were minced into small pieces on ice and enzymatically digested. After digestion, samples were sieved through a 40 µm cell strainer and red blood cells were lysed. CD45⁺ immune cells were sorted by flow cytometry. The scRNA-seq and scTCR/BCR-seq libraries were generated via the 10 × Genomics Chromium Controller Instrument and Chromium Single Cell 5’ library & gel bead kit, along with the V(D)J enrichment kit (10 × Genomics) following the recommended protocols. The quality of final libraries was evaluated using the Qubit 4 Fluorometer (Thermo Fisher Scientific) and Bioanalyzer 2200 (Agilent). All libraries were sequenced on a 150 bp paired-end run.

Single-cell data processing and clustering

Using fastp with default settings, we performed adaptor sequence filtering and eliminated low-quality reads. Subsequently, reads were aligned to the human reference genome (GRCh38 Ensemble v104) using CellRanger (v6.1.1) to generate feature-barcode matrices. Seurat (v4.4.0) was used for subsequent single-cell data quality control, dimensional reduction and clustering analysis. The following quality control metrics were used to filter out low-quality cells: (1) 200 < number of detected genes < 6000; (2) mitochondrial reads < 15%; (3) ribosomal reads < 50%; (4) hemoglobin reads < 3%; (5) genes expressed in ≥ 3 cells. Identification and removal of suspected doublets were processed by DoubletFinder (v2.0.4) [24]. SCTransform (v0.4.1) was used for data normalization and variance stabilization. Dimension reduction was performed via RunPCA function. Sample batch effects were corrected using RunHarmony function. The FindNeighbors and FindClusters functions were used for clustering, and the RunUMAP function was used for dimension reduction and 2D visualization. When clustering all cells, B cells, CD8⁺ T cells, CD4⁺ T cells and myeloid cells, the value of the PC parameter in FindNeighbors function was set at 20, 15, 15, 15 and 10 respectively, and the value of the resolution parameter in FindClusters function was set at 0.7, 0.7, 1.0, 1.0 and 1.0 respectively.

Cell type annotation

For precise annotation of cell types and states, we employed the FindAllMarkers function (min.pct = 0.25, logfc.threshold = 0.25) to detect differentially expressed genes (DEG) across distinct cell populations. According to FindAllMarkers results, CellMarker database and published studies, marker genes were selected to annotate cell types. ClusterProfiler (v4.10.0) was used to visualize functional profiles of DEG of different subclusters [25].

scTCR/BCR data analysis

The single-cell TCR/BCR sequencing data were analyzed using CellRanger (v6.1.1) with the human VDJ reference database from 10 × Genomics. The scTCR/BCR-seq data and scRNA-seq data were combined using scRepertoire (v1.12.0) [26]. Any cell barcode with an NA value in at least one of the TCR/BCR chains and any cell barcode with more than two TCR/BCR chains were removed. In order to analyze somatic hypermutation (SHM) of B cells, we first used CreateGermlines.py tool in Change-O (clip.med.yale.edu/changeo) to reconstruct germline sequence of BCR. Subsequently, the SHM rate for each BCR sequence was determined with the observedMutations function of Shazam (v1.1.0). Clonotypes shared with 2–5 cells, 6–20 cells and > 20 cells were further determined as small, medium and large clonotypes respectively.

Pseudotime analysis

The developmental pseudotime trajectories of CD8⁺ exhausted T cells and plasma cells were inferred by Monocle2 (v2.26.0). The differentialGeneTest function was used to choose genes that define a cell's progress. The pseudotime trajectory was constructed after dimension reduction and cell ordering. To complement pseudotime analysis, the cellular differentiation states and stemness were estimated by CytoTRACE (v0.3.3) [27].

Intercellular communication analysis

Intercellular communication analyses were performed using CellChat (v1.6.1) and NicheNet (v2.0.4) [28, 29]. We created CellChat objects in the TLS(−) and TLS(+) groups respectively, and then integrated the two datasets to compare the potential of intercellular interactions and the probability of receptor-ligand communications. NicheNet could analyze the potential of specific ligands of sender cells in regulating the expression of target genes in receiver cells.

Gene signature calculation

We collected previously reported gene signatures reflecting immune cell functional status and evaluated the signature scores using AddModuleScore function (Table S6). Based on the scRNA-seq data of this study, gene signatures of some key cell subsets were defined as top 15 highly expressed genes in each subset compared to other subsets within their primary immune compartments (Table S6).

Bulk RNA-seq data processing

Frozen post-treatment residual esophageal tumors from validation cohort (n = 30) were obtained and total RNA was extracted by RNeasy kit (Qiagen). Libraries were constructed using the TruSeq Stranded mRNA Library Prep Kit (illumina) following the recommended protocols. The quality of final libraries was evaluated using the Qubit 4 Fluorometer (Thermo Fisher Scientific) and Bioanalyzer 2200 (Agilent). All libraries were sequenced on a 150 bp paired-end run. DEG in different TLS subgroups were identified by limma (v3.58.1). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed by clusterProfiler (v4.10.0). Gene signature scores of key immune cell subsets were evaluated by ssGSEA method in GSVA (v1.50.0).

Multiplex immunohistochemistry (mIHC) analyses

Serial FFPE tissue Sects. (4 µm thickness) of residual esophageal tumors from the validation cohort were collected and further processed for mIHC assay to determine the infiltration level and spatial localization of key immune cell subsets. The mIHC assay was performed using PANO multiplex IHC Kit (Panovue) following the recommended protocols. After applying various primary antibodies in succession, the secondary antibodies conjugated with horseradish peroxidase were incubated, and the tyramide signal was amplified. Following the labeling of all the antigens, nuclei were stained with DAPI. Primary antibodies included anti-CD3 (Cell Signaling Technology; Cat.# 85061), anti-CD20 (Cell Signaling Technology; Cat.# 74332), anti-CD21 (Abcam; Cat.# ab227668), anti-CD23 (Abcam; Cat.# ab254162), anti-CD4 (Abcam, Cat.# ab133616), anti-CD8 (Abcam, Cat.# ab178089), anti-FOXP3 (Cell Signaling Technology, Cat.# 98377s), anti-PD-1 (Abcam, Cat.# ab52587), anti-CD68 (Abcam, Cat.# ab955), anti-CD14 (Abcam, Cat.# ab183322), anti-VEGFA (Abcam, Cat.# ab213244), anti-CXCL13 (Abcam, Cat.# ab246518), anti-CD1C (Abcam, Cat.# ab156708), and anti-KI67 (Abcam, Cat.# ab16667). Slides were scanned using the TissueFAXS Spectra (TissueGnostics) and the data analysis was performed using HALO (v3.2) and StrataQuest (v7.1) softwares. Spatial assessments were conducted with the HALO Spatial Analysis module. Nearest-neighbor analysis was applied to calculate the mean distance between two different cell populations, while proximity analysis was used to quantify the proportion of cells located within a defined radius (e.g., 25 μm) of another cell [30].

Statistical analyses

We utilized the previously described Ro/e value to estimate the enrichment or depletion of each immune cell cluster in TLS subgroups [31]. A Ro/e value > 1 indicated enrichment, whereas a Ro/e value < 1 indicated depletion of cells in a specific category. The quantitative data were compared by Wilcoxon rank-sum test, Student's t-test, One-Way ANOVA, and Kruskal–Wallis test. The categorical data were compared by Chi-square test and Fisher’s exact test. Pearson correlation analysis was used to evaluate the correlation between different cell subsets. Risk factors for postoperative recurrence were identified using logistic regression analyses. Kaplan–Meier methods estimated OS and RFS, and log-rank tests evaluated the survival difference. Independent prognostic factors were identified by Cox proportional hazards models. R (v4.2.3) software, SPSS (v26.0) software and GraphPad Prism 9 software were used for statistical analyses and plotting.

Results

Clinical significance of TLS in non-pCR ESCC

A total of 178 non-pCR patients were included in the clinical cohort A (Fig. 1A; Table S1). The proportions of patients in TLS(−) group, pTLS group and sTLS group were 26.4% (47/178), 27.5% (49/178) and 46.1% (82/178), respectively (Table S1; Fig. 1B). There were no significant differences in ypTNM stage distribution, lymphovascular invasion status and perineural invasion status between different TLS subgroups (Table S1).

After a median follow-up of 32 months, 73 patients (41.0%) experienced recurrences. Of these, 45 (25.3%) had locoregional recurrence alone, 12 (6.7%) had distant metastasis alone, and the remaining 16 (9.0%) presented with both locoregional and distant recurrence (Table S2). Locoregional recurrence most frequently involved the lymph nodes, while distant metastasis most often occurred in the lung. The survival rate of recurrent patients was significantly lower than that of non-recurrent patients (Fig. S1A). The 2-year cumulative incidence of recurrence in TLS(−) group, pTLS group and sTLS group was 68.5%, 31.5% and 23.1%, respectively (Fig. 1C; P < 0.001). Compared with TLS(−) group, pTLS group (P < 0.001) and sTLS group (P < 0.001) had significantly lower risk of recurrence, but there was no significant difference in recurrence risk between the pTLS and sTLS groups (P = 0.358). On univariate logistic regression analysis (Table S3), metastatic lymph node count (P = 0.026) and TLS status (P < 0.001) were predictive of postoperative recurrence. Multivariate logistic regression (Table S3) indicated that TLS status was the independent predictor of postoperative recurrence (P = 0.012). In Cox proportional hazards models, TLS status remained an independent prognostic factor for OS and RFS (Table S4-5).

The 2-year OS rate in TLS(−) group, pTLS group and sTLS group was 50.2%, 79.2% and 85.9%, respectively (Fig. 1D; P < 0.001). The 2-year RFS rate in TLS(−) group, pTLS group and sTLS group was 31.5%, 63.1% and 70.0%, respectively (Fig. 1E; P < 0.001). Compared with TLS(−) group, pTLS group (P = 0.001) and sTLS group (P < 0.001) had significantly improved RFS, but there was no significant difference in RFS between the pTLS and sTLS groups (P = 0.396).

Considering that the recurrence risk and survival rate of pTLS group and sTLS group were similar, we categorized pTLS group and sTLS group together as TLS(+) group in the subsequent analysis.

A total of 48 non-pCR patients were included in the clinical cohort B. Pathological evaluation identified 15 TLS(−) and 33 TLS(+) cases. Survival analysis demonstrated that TLS(+) patients exhibited a lower risk of recurrence (P < 0.001, Fig. S1B) and improved survival (RFS, P < 0.001, Fig. S1C; OS, P < 0.001, Fig. S1D), thereby confirming the robustness of the predictive value of TLS.

Single-cell profiling of immune cells in non-pCR ESCC

We collected post-treatment residual esophageal tumor samples from 7 ESCC patients undergoing nCIT followed by esophagectomy (single-cell sequencing cohort). Subsequently, CD45⁺ immune cells were sorted and sent for scRNA-seq and scTCR/BCR-seq analyses (Table S7; Fig. 1A; F). According to pathological evaluation, 3 patients were categorized into the TLS(−) group and 4 patients were categorized into the TLS(+) group (including pTLS and sTLS groups). Following quality control, we successfully captured single-cell transcriptomic profiles for 52,010 immune cells, encompassing T cells, B/plasma cells, myeloid cells, and mast cells (Figs. 1F; S1E-F). Notable batch effects were not detected across different patients or TLS subgroups (Fig. S1G-H). We observed that B/plasma cells and mast cells were enriched in the TLS(+) group, while myeloid cells were enriched in the TLS(−) group (Figs. 1G; S1I).

A total of 30 patients were included in the validation cohort of this study. After pathological evaluation, 10 patients were categorized into the TLS(−) group and 20 patients were categorized into the TLS(+) group (including pTLS and sTLS groups). We performed differential gene expression analysis on the bulk RNA-seq data of the validation cohort (Fig. 1A). The TLS(+) group highly expressed immune cell marker genes and chemokine genes (Fig. S1J). GO enrichment analysis showed that up-regulated genes were mainly involved in the processes of immune receptor rearrangement and immune cell activation (Fig. 1H). KEGG enrichment analysis showed that up-regulated genes were mainly involved in cell adhesion, antigen presentation, antibody production and other pathways (Fig. S1K). The above analysis suggested that the infiltration level and functional status of immune cell subsets under different TLS subgroups might be different.

TLS status correlated with the differentiation, longevity, and affinity maturation of B cell repertoires

The roles of T cells in tumor immune microenvironment have always been the research focus, while B cells are usually ignored. Recent studies indicated that B cells were the major component of TLS and mainly facilitated humoral immunity [11]. Meanwhile, the infiltration level of B cells was associated with the response to immunotherapy and survival rates in diverse cancers [32]. Sub-clustering the B-lineage cells revealed naïve B cells (Naive B), memory B cells (Mem B), germinal center B cells (GC B), IgA⁺ plasma cells (IgA⁺ PC), and IgG⁺ plasma cells (IgG⁺ PC) subsets (Figs. 2A; S2A) [33]. Naive B and Mem B cells exhibited elevated expression of MHC class II genes (Fig. S2B). The proportion of Mem B cells was increased in the TLS(+) group, while the proportion of IgA⁺ PC and IgG⁺ PC was increased in the TLS(−) group (Fig. 2B-C).

Fig. 2.

Distinct patterns of B cell proliferation and differentiation by TLS status. A UMAP visualization of B cell subsets. B Fractions of B cell subsets in different TLS subgroups. C Distribution of B cell subsets across different TLS subgroups estimated by Ro/e score. Ro/e values assess the enrichment or depletion of each immune cell subset. D Comparison of clonal expansion across TLS status in each B cell subset. The proportion of each clonotype helps visualize repertoire structure. E BCR clonotype sharing across B cell subsets. Chord diagram reveals unique and shared clones between B cell clusters. The thickness of the arc represents the extent of BCR clonotype sharing. F Comparison of clonal expansion between unique and shared BCRs in Mem B/PC. G Scatter plots showing frequencies of shared and unique clonotypes in memory B and plasma cells. Dot color indicates clonotype feature, and dot size reflects total number. The BCR repertoire overlap across B cell subsets in TLS(+) group (H) and TLS(−) group (I). ****, P < 0.0001

The B cell receptor (BCR) allows B cells to recognize foreign antigens and mediates B cells activation. Analyzing the characteristics of BCR is crucial for understanding the functional role of B cells under different environments [34]. A total of 5,508 B cells had paired scRNA-seq and scBCR-seq data (Fig. S2C). We observed that Naive B cells were mainly composed of single BCR clonotypes, while Mem B, GC B, IgA⁺ PC and IgG⁺ PC had increased proportion of expanded BCR clonotypes (Fig. S2D). Mem B cells, GC B cells and PC in the TLS(+) group demonstrated greater clonal expansion than those in the TLS(−) group (Fig. 2D).

We observed substantial BCR clonotype sharing between Mem B, GC B and PC, while clonotype sharing between other B cells was relatively rare (Fig. 2E). Shared clonotypes between Mem B cells and PC were significantly more expanded than unique clonotypes that only existed in either Mem B cells or PC (Fig. 2F). Moreover, high-frequency clonotypes in Mem B cells and PC were predominantly derived from shared clonotypes (Fig. 2G). We speculated that shared high-frequency clonotypes might represent potentially tumor-reactive clonotypes. We also found that the degree of clonotype sharing between Mem B cells, GC B cells and PC in the TLS(+) group was higher than that in the TLS(−) group (Fig. 2H, I), which indicated that B cells in TLS(+) group might stay in an active differentiation state.

Repeated cycles of clonal expansion and somatic hypermutation (SHM) promoted the affinity maturation and selection of B cells. Strong antigen-specific affinity was generally indicated by a high SHM frequency [35]. SHM rates were quantified using scBCR-seq data (Fig. 3A). As expected, naive B cells exhibited the lowest SHM frequency and IgA/IgG⁺ PC exhibited the highest SHM frequency (Fig. S3A), indicating an improved affinity for tumor antigens during the differentiation of B cells. In tumor microenvironment, PC could secrete tumor antigen-specific antibodies and activate antibody dependent cytotoxicity or phagocytosis [11]. The infiltration level of PC was closely related to survival outcomes and response to immunotherapy in multiple cancers [36]. To further analyze the differentiation trajectory and phenotype of PC, we conducted pseudotime analysis of PC. We observed that PC in the TLS(−) group were mainly distributed in the early stage of pseudotime trajectory, while PC in the TLS(+) group were mainly distributed in the end stage of pseudotime trajectory (Fig. 3B). This suggested that the functional status of PC in different TLS subgroups might be different.

Fig. 3.

Developmental trajectory of plasma cells. A UMAP visualization of SHM rate in B cell subsets. The color indicates the SHM level. B Distribution of plasma cells based on TLS status along with the pseudotime. C The temporal expression profile of LLPC signature genes along pseudotime in the developmental trajectory of plasma cells. D Expression of LLPC gene signature across different TLS subgroups in validation cohort. E Dynamic pattern of cytotrace score along pseudotime in the trajectory of plasma cells. F Comparison of plasma cell cytotrace score across different TLS subgroups. G Comparison of plasma cell SHM rate across different TLS subgroups. *, P < 0.05; ****, P < 0.0001

Along the pseudotime trajectory, the expression of long-lived plasma cells (LLPC) feature genes gradually increased (Fig. 3C), indicating that PC could differentiate from short-lived plasma cells (SLPC) to LLPC. LLPC feature genes were mainly associated with antibody affinity maturation, PC activation, and inhibition of PC apoptosis [37]. Analysis of bulk RNA-seq data from the validation cohort revealed significantly elevated LLPC signature scores in the TLS(+) group compared to the TLS(−) group (Fig. 3D). LLPC represented unique subsets of PC that continuously secreted high affinity antibodies and were vital for maintaining immune memory. Their survival depended on specific microenvironments (e.g. secondary lymphoid organs) [38]. The roles of LLPC in anti-infectious immunity had been widely reported [38], but their roles in tumor microenvironment remained unclear. LLPC were enriched in the TLS(+) group, and TLS, as an ectopic lymphoid structure, might promote PC differentiation and maintain longevity phenotype. We believed that LLPC had the potential to secrete antibodies that bound with tumor antigens with high affinity and therefore improved anti-tumor humoral immunity. We also evaluated the differentiation state and stemness of PC using CytoTRACE [27]. A higher CytoTRACE score indicated a lower degree of differentiation. We found that the CytoTRACE score gradually decreased along with the PC pseudotime trajectory (Figs. S3B, 3E). The CytoTRACE score of PC was notably reduced in the TLS(+) group compared to the TLS(−) group (Fig. 3F), and the SHM frequency of PC was significantly elevated in the TLS(+) group (Fig. 3G). This further confirmed that well differentiated LLPC with high affinity accumulated in the TLS(+) group.

Mem B cells in tumor microenvironment could differentiate into PC under antigen stimulation, and Mem B cells could also present antigens and further activate other T cell subsets [34]. In this study, Mem B cells were the most common B cell subset and were enriched in the TLS(+) group (Fig. 2B-C). The percentage of switched IgG/IgA⁺ Mem B in TLS(+) and TLS(−) groups was similar (Fig. S3C). Clonally expanded Mem B cells exhibited a markedly increased SHM frequency in the TLS(+) versus TLS(−) groups (Fig. S3D), which indicated that clonally expanded Mem B cells from TLS(+) group represented potential tumor-reactive B cells with strong antigen-specific affinity.

Activated tumor-reactive CD8⁺ T cells were enriched in the TLS(+) group.

T cells are the major component of tumor microenvironment and play a crucial role in mediating immunotherapy response. Exploring the immunobiology and clinical significance of different T cell subsets could improve the understanding of their functions in anti-tumor immunity. Sub-clustering the CD8⁺ T cells revealed tissue resident memory T cells (CD8_Trm), memory T cells (CD8_Tm), exhausted T cells (CD8_Tex), effector memory T cells (CD8_Tem), heat shock protein T cells (CD8_HSP), proliferating T cells (CD8_Prolif), natural killer T cells (CD8_NKT), interferon-stimulated gene T cells (CD8_ISG) and aging T cells (CD8_Aging) subsets (Fig. 4A–D) [39, 40]. CD8_HSP cells, characterized by high expression of heat shock gene (Fig. 4D), were associated with immunotherapy resistance [39]. CD8_Aging cells, characterized by high expression of NEAT1 and MALAT1 (Fig. 4D), were enriched in aging and immunocompromised populations [41]. The enrichment of the above two CD8⁺ T cell subsets in the TLS(−) group might promote immunosuppressive microenvironment formation (Fig. 4B, C). Meanwhile, the majority of CD8⁺ T cell subsets in the TLS(−) group showed downregulated effector/chemokine signature gene expression and upregulated exhaustion signature gene expression (Fig. S4A-C), which further suggested the existence of immunosuppressive microenvironment in the TLS(−) group.

Fig. 4.

Enrichment of tumor-reactive CD8⁺ T cells in TLS(+) group. A UMAP visualization of CD8⁺ T cell subsets. B Fractions of CD8⁺ T cell subsets in different TLS subgroups. C Distribution of CD8⁺ T cell subsets across different TLS subgroups estimated by Ro/e score. Ro/e values assess the enrichment or depletion of each immune cell subset. D Bubble plot showing expressions of selected marker genes for CD8⁺ T cell subsets. Dot color indicates mean expression level of marker gene, and dot size reflects the fraction of cells expressing the gene. E Fractions of expanded clonotypes in CD8⁺ T cell subsets. The proportion of each clonotype helps visualize repertoire structure. F Distribution of CD8_Tex based on TLS status along with the pseudotime. G Expression of Tpex gene signature across different TLS subgroups in validation cohort. H Comparison of proportion of CXCL13⁺CD8⁺ T cells based on TLS status in validation cohort. Representative mIHC staining images of CXCL13⁺CD8⁺ T cells in TLS(+) group (I) and TLS(−) group (J). **, P < 0.01; ****, P < 0.0001

The T cell receptor (TCR) is essential to almost every facet of T cell function and endows T cell with antigen specificity. TCR repertoire profiling makes it possible to characterize clonal dynamics, cellular phenotype, and tumor reactivity of diverse T cell subsets [42]. A total of 11,393 CD8⁺ T cells had paired scRNA-seq and scTCR-seq data (Fig. S4D). Although CD8⁺ T cells subsets exhibited remarkable clonal expansion in this study (Fig. 4E), not all tumor infiltrating CD8⁺ T cells possessed tumor-specific reactivity. Recent studies showed that non-exhausted CD8⁺ T cells, distinguished by lack of co-inhibitory receptors (e.g. PD-1, TIGIT, LAG-3, CTLA-4) expression, only recognized cancer-unrelated viral epitopes originating from endemic viruses including influenza, cytomegalovirus, and Epstein-Barr virus [43]. Because of their incapacity to identify tumor antigens, non-exhausted CD8⁺ T cells were regarded as bystander T cells [43]. By contrast, exhausted CD8⁺ T cells (CD8_Tex) expressing co-inhibitory receptors demonstrated reactivity against tumor antigens and were regarded as tumor-reactive T cells [44]. In this study, CD8_Tex cells showed a higher abundance in the TLS(+) group with significant clonal expansion (Fig. 4B, C; Fig. 4E), indicating that TLS(+) group might exhibit activated anti-tumor immunity.

To further track the dynamics and phenotypes of CD8_Tex cells, we performed pseudotime analysis. We found that CD8_Tex cells from TLS(+) group were uniformly distributed in the early and end stage of pseudotime trajectory, while CD8_Tex cells in the TLS(−) group were mainly distributed in the end stage of pseudotime trajectory (Fig. 4F). This suggested that the functional status of CD8_Tex cells in different TLS subgroups might be different. Consistent with the reported differentiation process from precursor exhausted T cells (Tpex) to terminally exhausted T cells (Ttex) [45, 46], the expression of stemness marker (TCF7) gradually decreased along the pseudotime trajectory, while the expression of exhaustion markers (ENTPD1, HAVCR2, LAG3, PDCD1) gradually increased (Fig. S4E). We classified CD8_Tex cells in the early trajectory (pseudotime ≤ 5) and end trajectory (pseudotime > 5) into Tpex and Ttex cells, respectively (Fig. 4F; Fig. S4F). We found that Tpex and Ttex clusters identified in this study highly expressed progenitor and terminally exhausted gene signatures respectively (Fig. S4G-H), which further validated the accuracy of trajectory classification. Tpex cells were identified as the key subset responsive to immunotherapy with cytotoxic and self-renewal potential [45]. We found that Tpex cells were enriched in the TLS(+) group (Fig. 4F; Fig. S4F), and bulk RNA-seq analysis revealed a markedly elevated progenitor gene signature score in the TLS(+) group compared to the TLS(−) group (Fig. 4G), indicating that TLS presence correlated with T cell differentiation trajectories characterized by preserved Tpex stemness and attenuated exhaustion programs.

Previous studies reported that CXCL13 could serve as a marker of tumor-reactive T cells, and the infiltration level of CXCL13⁺CD8⁺ T cells was related to immunotherapy response [46]. The pseudotime analysis of this research also revealed that CXCL13 was highly expressed at the initial stage of CD8_Tex cells differentiation (Fig. S4E). We then analyzed the spatial localization of CXCL13⁺CD8⁺ T cells in validation cohort using mIHC (Fig. 4H–J). CXCL13⁺CD8⁺ T cells were predominantly localized at the outer layer of TLS and nearby area (Fig. 4I; blue stars in HE image indicated TLS). In addition, TLS(+) patients (n = 20) showed increased abundance of CXCL13⁺CD8⁺ T cells than TLS(−) patients (n = 10; Fig. 4H). The above analyses showed that TLS might be the site for maturation of tumor-reactive CD8⁺ T cells and activated tumor-reactive CD8⁺ T cells could further disseminate into the tumor bed and exert anti-tumor immune responses.

Co-localization of CD4_Tph with B cells within TLS correlated with enhanced immune responses.

Sub-clustering the CD4⁺ T cells revealed regulatory T cells (CD4_Treg), memory T cells (CD4_Tm), peripheral helper T cells (CD4_Tph), central memory T cells (CD4_Tcm), and proliferating T cells (CD4_Prolif) subsets (Fig. 5A–D) [39, 40]. CD4_Treg cells were abundant in tumors and were associated with disease progression, metastases evolution, and immunotherapy resistance [47]. The enrichment of CD4_Treg cells in the TLS(−) group might promote the formation of immunosuppressive microenvironment (Fig. 5B, C). Meanwhile, the majority of CD4⁺ T cells subsets in the TLS(−) group showed downregulated chemokine signature gene expression and upregulated exhaustion signature gene expression (Fig. S5A, B), which further suggested the existence of immunosuppressive microenvironment in the TLS(−) group.

Fig. 5.

Co-localization and interaction of CD4_Tph with B cells within TLS. A UMAP visualization of CD4⁺ T cell subsets. B Fractions of CD4⁺ T cell subsets in different TLS subgroups. C Distribution of CD4⁺ T cell subsets across different TLS subgroups estimated by Ro/e score. Ro/e values assess the enrichment or depletion of each immune cell subset. D Bubble plot showing expressions of selected marker genes for CD4⁺ T cell subsets. Dot color indicates mean expression level of marker gene, and dot size reflects the fraction of cells expressing the gene. E Comparison of clonal expansion across different TLS subgroups in CD4_Tph and CD4_Treg. The proportion of each clonotype helps visualize repertoire structure. (F-G) CD4⁺ T cell compositions of the top 10 enriched clonotypes in TLS(+) group (F) and TLS(−) group (G). H Interaction potential between CD4_Tph and B cell subsets in different TLS subgroups. The thickness of the arrow represents the interaction potential. Representative images of mIHC staining of CD4_Tph and GC B cells in TLS(+) group (I) and TLS(−) group (J). ns, non-significant; ***, P < 0.001

A total of 9,333 CD4⁺ T cells had paired scRNA-seq and scTCR-seq data (Fig. S5C). We observed that CD4_Prolif, CD4_Tcm and CD4_Tm cells were mainly composed of single TCR clonotypes, while CD4_Treg and CD4_Tph cells had increased proportion of expanded TCR clonotypes (Fig. S5D). Several studies have identified CD4_Tph cells as tumor reactive CD4⁺ T cells and major cell type responding to immunotherapy [46, 48, 49]. In this study, CD4_Tph cells were enriched in TLS(+) group with stronger degree of clonal expansion than TLS(−) group (Fig. 5B, C, E). Moreover, the top 10 most expanded clonotypes in TLS(+) group were mainly mapped to CD4_Tph, while the top 10 most expanded clonotypes in TLS(−) group were mainly mapped to CD4_Treg (Fig. 5F, G). Previous studies showed that CD4_Tph cells were capable of producing CXCL13, attracting B cells, and facilitating the development of TLS [50]. Our analysis revealed that GC B cells showed the strongest predicted interaction potential with CD4_Tph cells, and interaction potential between CD4_Tph cells and each B cell subset was significantly elevated in TLS(+) group compared to TLS(−) group (Fig. 5H). Multi-immunofluorescence assays showed that CD4_Tph cells (CD4⁺ CXCL13⁺) co-localized with GC B cells (CD20⁺ KI67⁺) within the TLS (Fig. 5I). By contrast, TLS(−) group lacked structured CD4_Tph and GC B cells aggregates (Fig. 5J). Indeed, the TLS(+) group had shorter average nearest-neighbor distances between CD4_Tph and GC B cells than the TLS(−) group (Fig. S5E-F). Moreover, we found that a significantly higher percentage of GC B cells were located closer (within 25 μm) to CD4_Tph in the TLS(+) group compared with the TLS(−) group (Fig. S5G). The above analyses indicated that the presence of TLS correlated with enhanced synergistic interaction, expansion, and activation of tumor-reactive immune cells.

The phenotypes of myeloid cells were associated with TLS status

Myeloid cells were major component of immune microenvironment and played a crucial role in regulating tumor progression and angiogenesis. However, the functional diversity of myeloid cells has not been fully elucidated. Sub-clustering myeloid cells revealed S100A8⁺ neutrophil (Neut_S100A8), IFIT⁺ neutrophil (Neut_IFIT), VEGFA⁺ monocyte (Mono_VEGFA), TIMP1⁺ monocyte (Mono_ TIMP1), mature dendritic cell (mDC), mast cell (Mast), C1QC⁺ macrophage (Macro_C1QC), and type-2 conventional dendritic cell (cDC2) subsets (Fig. 6A–C, Fig. S6A) [51]. The majority of myeloid cell subsets in the TLS(−) group showed upregulated expression of genes associated with angiogenesis (Fig. 6D). VEGFA was considered as one of the most important regulators of tumor angiogenesis and it could promote angiogenesis, increase vascular permeability, inhibit DC maturation and induce immunosuppressive microenvironment [52]. Mono_VEGFA cells highly expressed VEGFA and were enriched in TLS(−) group, indicating the pro-angiogenic property in TLS(−) group (Fig. 6B, C, Fig. S6A). Moreover, we observed that Macro_C1QC cells in TLS(−) group showed higher expression of SPP1 (Fig. 6E), which was associated with angiogenesis, tumor metastasis, and decreased survival [53]. Taken together, these results suggested that myeloid cells in TLS(−) group had the potential to promote angiogenesis and induce immunosuppressive microenvironment.

Fig. 6.

Distinct myeloid cell phenotypes across TLS subgroups. A UMAP visualization of myeloid cell subsets. B Fractions of myeloid cell subsets in different TLS subgroups. C Distribution of myeloid cell subsets across different TLS subgroups estimated by Ro/e score. Ro/e values assess the enrichment or depletion of each immune cell subset. D Expression of angiogenesis gene signature across different TLS subgroups in each myeloid cell subset. E Volcano plots showing differentially expressed genes of Macro_C1QC in different TLS subgroups. F Expression of activation gene signature across different TLS subgroups in each dendritic cell subset. G Expression of migration gene signature across different TLS subgroups in each dendritic cell subset. H Expression of antigen presentation gene signature across different TLS subgroups in each dendritic cell subset. I Expression of M2 macrophage gene signature across different TLS subgroups in Macro_C1QC. J Expression of phagocytosis gene signature across different TLS subgroups in Macro_C1QC. K Correlations of distinct myeloid cell subsets with Tpex in cellular abundance. L Heatmap showing the regulatory potential of the prioritized ligands in Macro_C1QC driving the anti-tumor and progenitor potential of CD8_Tex. Color indicates regulatory potential. M Representative mIHC staining images of CD8_Tex and macrophage in TLS(+) group. ns, non-significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001

Dendritic cells have long been the focus of immunotherapy due to their roles in the initiation of antigen-specific immunity. We found that cDC2 and mDC cells were enriched in the TLS(+) group and exhibited more activated phenotypes with improved migratory and antigen presentation capacity as compared with the TLS(−) group (Fig. 6B, C, F–H). Macrophages could display distinct functional properties (M1/M2) depending on specific tissue environments [54]. We observed that Macro_C1QC cells in TLS(−) group skewed toward immunosuppressive pro-tumoral M2 phenotype (Fig. 6I), while Macro_C1QC cells in TLS(+) group highly expressed phagocytosis related genes (Fig. 6J). Collectively, these results indicated that TLS status correlated with phenotypic properties and functional patterns of dendritic cells and macrophages.

Immune-activated niches existed in the TLS(+) group

As hinted by recent research emphasizing the relevance of myeloid cells in shaping immunotherapy response and promoting T cell immunity [55], we investigated the interactions of myeloid cells with tumor-reactive T cells. We observed that Macro_C1QC cells exhibited the strongest correlation with Tpex cells in terms of cellular abundance (Fig. 6K), and this trend was verified by bulk RNA-seq data (Fig. S6B). The interaction potential of CD8_Tex cells with Macro_C1QC and cDC2 cells was significantly elevated in the TLS(+) group (Fig. S6C). NicheNet analysis predicted that Macro_C1QC cells could activate the anti-tumor potential and inhibit the exhaustion program of CD8_Tex cells by inducing the expression of FOS, JUN, and GADD45B (Fig. 6L). Noteworthily, FOS and JUN were experimentally validated as master regulators of terminal differentiation and functional capacity of T lymphocytes [56, 57], and GADD45B was crucial for activating anti-tumor potency of T lymphocytes by upregulating IFN-γ expression [58]. We additionally validated the spatial juxtaposition of CD8_Tex cells and macrophages within TLS by mIHC (Fig. 6M).

cDC2 and Macro_C1QC cells exhibited high correlations with CD4_Tph cells in cellular proportions (Fig. 7A), and this trend was verified by bulk RNA-seq data (Fig. 7B). CellChat analysis revealed that cDC2 cells in TLS(+) group recruited and activated CD4_Tph cells through MIF − (CD74 + CXCR4), CD40 − CD40LG, and CXCL16 − CXCR6 ligand-receptor pairs, while cDC2 cells in TLS(−) group interacted with CD4_Tph cells through LGALS9 − HAVCR2 pairs (Fig. 7C). It was reported that LGALS9 − HAVCR2 axis mediated the apoptosis and immunosuppression of T cells [59]. Multi-immunofluorescence assays showed that CD4_Tph cells (CD4⁺ CXCL13⁺) co-localized with cDC2 cells (CD1C⁺) within TLS (Fig. 7D). By contrast, TLS(−) group lacked colocalized CD4_Tph and cDC2 aggregates (Fig. 7E). Quantitative spatial analysis confirmed that CD4_Tph cells were located closer to cDC2 cells in the TLS(+) group compared with the TLS(−) group (Fig. S6D-E). It was worth noting that cDC2 was also distributed in the B cell aggregates of TLS (Fig. 7D). As professional antigen-presenting cells, cDC2 had the potential to present antigen and further activate CD4_Tph and B cells.

Fig. 7.

Immune-activated niche in TLS(+) group. A Correlations of distinct myeloid cell subsets with CD4_Tph in their cellular proportion. B Scatterplot showing correlations of cDC2 with CD4_Tph in validation cohort based on their signature gene expressions. C Ligand-receptor interactions between myeloid cells and CD4_Tph in different TLS subgroups. Dot color indicates communication probability, and dot size reflects p-value. Representative mIHC staining images of cDC2 and CD4_Tph in TLS(+) group (D) and TLS(−) group (E). F Correlations of distinct myeloid cell subsets with CD4_Treg in their cellular proportion. G Scatterplot showing correlations of Mono_VEGFA with CD4_Treg in validation cohort based on their signature gene expressions. H Ligand-receptor interactions between myeloid cells and CD4_Treg in different TLS subgroups. Dot color indicates communication probability, and dot size reflects p-value. Representative mIHC staining images of Mono_VEGFA and CD4_Treg in TLS(−) group (I) and TLS(+) group (J)

Mono_VEGFA and Neut_IFIT cells exhibited high correlations with CD4_Treg cells in cellular proportions (Fig. 7F), and this trend was verified by bulk RNA-seq data (Fig. 7G). CellChat analysis revealed that Mono_VEGFA and Neut_IFIT cells in TLS(−) group recruited CD4_Treg cells through CCL3 − CCR5 and CCL4 − CCR5 ligand-receptor pairs (Fig. 7H). Multi-immunofluorescence assays demonstrated a higher abundance of Mono_VEGFA and CD4_Treg cells in TLS(−) group compared to TLS(+) group (Fig. 7I, J). Quantitative spatial analysis confirmed that CD4_Treg cells were located closer to Mono_VEGFA cells in the TLS(−) group compared with the TLS(+) group (Fig. S6F-G).

Taken together, these data revealed that antigen presenting cells and tumor reactive T cells were enriched in TLS(+) group and were spatially colocalized within TLS. Antigen presenting cells in the TLS(+) group had the potential to activate tumor reactive T cells and inhibit their exhaustion program, while the immunosuppressive myeloid cells in the TLS(−) group might induce the exhaustion of T cells and recruit CD4_Treg. The immune-activated niche in TLS(+) non-pCR ESCC was characterized by orchestrated cross-talk between tumor reactive immune cells within TLS (Fig. 8).

Fig. 8.

Summary of immune profile in different TLS subgroups

Discussion

Neoadjuvant immunotherapy aimed to promote systemic immune responses to tumor-specific antigens, potentially enhancing treatment compliance, improving resectability, and eradicating micro-metastasis. Pathological response was closely related to postoperative survival and had become the primary endpoint of several neoadjuvant clinical trials [4]. The underlying mechanisms and predictive biomarkers of response to immunotherapy remained inadequately understood. The non-pCR patients represented a subgroup with poor prognosis, distinct immune microenvironmental features, and relatively high prevalence, but historically received limited attention [4]. Recent studies suggested the central role of TLS in adaptive immunity and lymphocyte maturation [11]. ESCC is an inflammation-associated malignancy with a heterogeneous immune microenvironment [14, 15, 60], where immune surveillance and local immune architecture such as TLS may influence disease control. The frequent presence of lymphocytic infiltration and proinflammatory cytokines in ESCC further suggests a microenvironment permissive for TLS formation [16]. Previous TLS-related studies in multiple solid tumors were mainly conducted in either treatment-naive or metastatic settings, while our study specifically focused on clinically challenging non-pCR ESCC. By integrating multi-center clinical outcomes with multi-omics profiling data, our study provided novel insights into the clinical significance and immune profile of TLS in non-pCR ESCC.

Based on the clinical cohort analysis, we found that TLS status was the independent predictor of postoperative recurrence, and TLS(+) group exhibited reduced recurrence risk and improved survival in non-pCR ESCC. We believe that these findings have practical implications for postoperative management and future therapeutic development for non-pCR ESCC. While nCIT had the potential to enhance survival for locally advanced ESCC, the outcomes of poor responders with recurrent disease were not optimistic. Considering that effective treatment strategies for patients with recurrent disease are currently limited, induction of TLS represents an innovative approach to enhance antitumor immunity and thereby reduce recurrence risk, as further supported by our multi-omics analyses. Moreover, TLS status might help refine risk stratification and follow-up planning, as TLS(−) patients faced higher recurrence risk and might benefit from closer surveillance or intensified adjuvant therapy. In addition, the recurrence patterns observed after nCIT provided valuable clues for optimizing future neoadjuvant systemic or local treatment strategies. Together, these points emphasized the clinical relevance of our study and further laid the groundwork for subsequent multi-omics investigations.

Based on single-cell sequencing data, we found that there were significant differences in the composition, functional status and interaction mode of immune cells under different TLS subgroups. The key cell subsets responsive to immunotherapy were enriched in the TLS(+) group, and immune cells in the TLS(+) group showed a functional state of high activation and low exhaustion. The variation in responses to immunotherapies stimulated broad interest in understanding how cellular states and antigen specificities of T cells evolved with respect to immune response. Tumor-reactive exhausted T cells expressed co-inhibitory receptors and demonstrated reactivity against tumor antigens [44]. Meanwhile, exhausted T cell population might include T cells at various states of response to antigen stimulation. Recent studies revealed that the success of lung cancer immunotherapy was fostered by the activation and expansion of pre-existing Tpex with self-renewal and cytotoxic potential [45]. However, immunotherapy was unable to effectively reinvigorate the functionality of Ttex at terminal phase of dysfunction [45, 46]. This study found that Tpex was also present in patients with incomplete response to nCIT and the abundance of Tpex was associated with TLS status. TLS presence correlated with T cell differentiation trajectories characterized by preserved Tpex stemness and attenuated exhaustion programs. While the majority of landmark immunotherapy studies largely focused on boosting the CD8⁺ T cells, the anti-tumour potency of CD4⁺ helper T cells has been increasingly emphasized. CD4_Tph cells have been identified as tumor-reactive T cell subsets enriched in responders to immunotherapy [46, 48]. In this study, CD4_Tph cells were enriched in the TLS(+) group with stronger degree of clonal expansion as compared with TLS(−) group. Moreover, CD4_Tph cells exhibited the strongest predicted interaction potential with GC B cells through CXCL13 − CXCR5 ligand-receptor pairs. Multiplex immunohistochemistry and cell–cell communication analysis suggested that tumor reactive CD8⁺/CD4⁺ T cells were spatially colocalized with B cells and antigen presenting cells within TLS and had high interaction potential. The spatial co-localization and synergistic interactions of tumor-specific immune cells within TLS were indicative of the generation of adaptive immune responses.

For a long time, the roles of T cells in the control of tumor development have been widely recognized, while B cells have been considered as bystanders in tumor microenvironment. In recent years, several studies pointed out that B cells were closely related to the response to immunotherapy and the formation of TLS [11]. Patil et al. reported that the presence of B and plasma cell signatures was associated with improved outcomes in lung cancer patients who underwent immunotherapy, whereas such association was not found in those administered docetaxel [36]. Meanwhile, the prognostic significance of plasma cell was not influenced by CD8⁺ T cell infiltration, which suggested an independent and positive effect of humoral immunity on the response to immunotherapy. Various microenvironmental factors and cellular alterations may have contributed to the phenotypic heterogeneity of plasma cells. LLPC resided in the secondary lymphoid organs and matured in a T cell‐dependent manner, providing durable and protective immunity against multiple pathogens [38]. However, the roles of LLPC in tumor microenvironment remained unclear. We found that well differentiated LLPC with high affinity were enriched in the TLS(+) group. We believed that LLPC had the potential to secrete antibodies that bound with tumor antigens with high affinity, which could further activate effector mechanisms such as antibody dependent cytotoxicity or phagocytosis and ultimately improved anti-tumor humoral immunity. We also revealed that the degree of clonotype sharing between Mem B, GC B and PC in TLS(+) group was higher than that in TLS(−) group, which indicated that B cells from TLS(+) group might stay in an active differentiation state. Collectively, these data highlighted that TLS status correlated with the selection, amplification, longevity persistence, and affinity maturation of B cell repertoires. Due to the high dropout rate, scRNA-seq datasets failed to capture interleukin transcripts, and therefore the expression of IL10, crucial marker gene to characterize regulatory B cells (Bregs), was not informative. Bregs were not identified in this study and other single-cell studies [33]. The phenotypes and roles of Bregs warrant further research.

The complexity of cellular crosstalks and functional properties of myeloid cells made it difficult to understand the roles of various myeloid cells subsets in tumor progression. In this study, we found that antigen presenting cells were enriched in the TLS(+) group with upregulated expression of activation, migration, and antigen presentation related genes. Meanwhile, antigen presenting cells in the TLS(+) group were spatially colocalized with tumor-reactive T cells and were predicted to have the potential to inhibit T cells exhaustion program. Spatial co-localization of immune cells provided key insights into the functional organization of the tumor immune microenvironment [61]. Quantitative assessment of spatial relationships provided a comprehensive understanding of how immune cells orchestrated responses in situ [30, 62]. The crosstalk between tumor-specific immune cells would foster a diversified and intensified immune response. The morphological characteristics and cellular compositions of TLS were similar to those of secondary lymphoid organs [11]. However, unlike secondary lymphoid organs, TLS were not encapsulated. The lack of physical delimitation also facilitated increased exposure to labile or low concentrations antigens and pro-inflammatory molecules, conferring considerable advantages to anti-tumor immunity. By contrast, immunosuppressive and pro-angiogenic myeloid cells were enriched in TLS(−) group, and were predicted to induce the exhaustion of T cells and recruit CD4_Treg. Moreover, TLS(−) group lacked functional and organized tumor-reactive immune cell aggregates, which further hindered the activation of immune responses. In summary, these results revealed the cellular composition and dynamic interaction of immune cells stratified by TLS status.

This study has several limitations. First, the small sample size and observational nature of the study design limited our ability to definitively establish the prognostic significance of TLS, and long-term follow-up data from ongoing phase III nCIT trials would further verify our findings. Second, the mechanisms of TLS formation remained unclear and there was a lack of reliable method to induce the formation of TLS in preclinical ESCC model [63]. Therefore, we failed to thoroughly decipher the underlying mechanism of TLS in boosting immune response. Third, additional biomarkers, such as circulating tumor DNA, also hold promise for recurrence risk stratification and warrant validation in future prospective studies. Moreover, large-scale spatial omics data in future studies would further refine and strengthen these findings. Furthermore, the relatively limited number of pCR cases restricted a comprehensive evaluation of TLS status in the pCR population and its association with recurrence risk. We also acknowledge the inherent limitations of pCR as a surrogate for long-term survival.

Conclusion

In conclusion, through multi-center clinical analysis integrated with multi-omics profiling, this study delineated the prognostic relevance and immunological landscape of TLS in non-pCR ESCC. TLS status was consistently associated with postoperative recurrence risk and survival, as well as distinct immune cell composition, functional states, and interaction patterns. The spatial co-localization and synergistic interaction of tumor-specific immune cells within TLS were indicative of the activation of adaptive immunity. Collectively, our findings provided a framework for understanding immune heterogeneity in non-pCR ESCC and highlighted TLS as a promising biomarker for risk stratification and a potential target for future therapeutic strategies (Fig. 8).

Author contributions

Yang Wo; Tong Lu; Yihua Sun: Conceptualization, Methodology, Investigation, Supervision, Writing—original draft, Writing—review & editing. Zijiang Yang; Xiongfei Li; Zheyi Wang; Yizhou Peng: Conceptualization, Methodology, Writing—review & editing. Xuxia Shen; Feng Hou; Wenjie Jiao: Investigation, Writing—review & editing.

Funding

This work was supported by the National Natural Science Foundation of China (82172744), Science and Technology Commission of Shanghai Municipality (21Y11913700), Shandong Provincial Natural Science Foundation (ZR2025QC1933Z), Health Commission of Shandong Province (202504080911), and QDFY Research Foundation (QDFYQN2024211).

Data availability

Single-cell sequencing data in this study have been deposited at the Genome Sequence Archive at the National Genomics Data Center (https://ngdc.cncb.ac.cn/gsa-human/). Sample information and accession number were listed in Table S7. Additional information will be available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the institutional review board (IRB #20875697P). Written informed consent to participate was obtained from all patients.

Consent for publication

All authors have agreed with publishing this manuscript.

Competing interests

The authors declare no competing interests.