Spatial distribution analysis of tertiary lymphoid structures in esophageal squamous cell carcinoma predicts patient survival and response to neoadjuvant chemo-immunotherapy

Abstract

Background

The prognostic and predictive significance of tertiary lymphoid structures (TLSs) exhibits spatial specificity in various cancers. However, the spatial distribution, phenotypic characteristics of TLSs in esophageal squamous cell carcinoma (ESCC) and their impact on prognosis and prediction are not yet fully understood.

Methods

We performed multiplex immunofluorescence staining and digital image analysis on 87 untreated ESCC specimens to characterized TLSs expression and phenotypic characteristics, CD8 + T cells and PNAD+ high endothelial venules (HEVs) in different spatial regions of ESCC tissues using Panel-1 (CD20/CD21/CD23/ PNAD/CD8/Pan CK/DAPI). Panel-2 (CD20/CD3/Foxp3/DC-LAMP/Pan CK/DAPI) was used to evaluate the spatial distribution of TLS-related immune cells (CD20 + B cells, CD3 + T cells, Foxp3 + Treg cells, and LAMP+ mature DCs) within the tumor microenvironment of ESCC. Furthermore, the predictive value of TLSs and the clinical prognostic significance of TLSs at different spatial locations were assessed in an independent cohort of 15 ESCC patients who received neoadjuvant chemo- immunotherapy (NACI).

Results

In untreated ESCC, a high number/density of mature follicular TLSs (F-TLSs) in the stromal distal region (> 500 μm from the outer boundary of the tumour nests) was significantly associated with better overall survival (OS) in patients (number: p = 0.0092; density: p = 0.0268). Patients with distal high F-TLSs not only exhibited high densities of CD3 + T cells, CD8 + T cells, CD20 + B cells, LAMP + DCs, and PNAD+HEVs in the stromal regions, but this was also associated with increased CD8 + T cell infiltration within the tumour nests (p < 0.05). Additionally, it was correlated with a lower levels of Foxp3 + Treg cells in the stromal distal and tumour nests regions. In the NACI cohort, the total amount of TLSs significantly increased after treatment. The partial response (PR) group exhibited higher numbers and densities of F-TLSs post-treatment than the non-PR group (p < 0.05). High numbers/densities of total TLSs, early-TLSs, and F-TLSs in the stromal proximal region (≤ 500 μm from the outer boundary of the tumour nests) of post-treatment specimens were significantly associated with better OS (p < 0.05).

Conclusions

Our research establishes that the prognostic value of TLSs is contingent upon their spatial location and maturation status. In untreated ESCC, distal mature F-TLSs is critical prognostic indicator; however, after NACI, TLSs quantity and spatial distribution were reshaped. Mature F-TLSs predicted treatment response, while treatment-induced expansion of TLSs localised at the proximal tumour-stroma interface is a key indicator for predicting favourable long-term survival.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-07950-4.

Background

Eesophageal cancer (EC) is the seventh leading cause of cancer-related deaths worldwide, and epidemiological studies have shown that its incidence is gradually increasing [1]. Surgical resection is the foundational strategy for localised and locally advanced EC, with a 5-year survival rate of 15–39% [1, 2]. Neoadjuvant therapy (radiotherapy, chemotherapy, or a combination of both) administered before surgery offers the advantage of addressing micrometastases and improving the rate of complete resection [2]. With the increasing use of immunotherapy in clinical practice in recent years, immune checkpoint inhibitors (ICIs) have become crucial treatments for patients with EC, and programmed cell death 1 (PD-1) inhibitors combined with chemotherapy have been established as the standard first-line treatment strategy for recurrent and metastatic oesophageal squamous cell carcinoma (ESCC) [3, 4]. Currently, the use of neoadjuvant chemo-immunotherapy (NACI) has become widespread, and multiple phase I and II clinical trials have shown good efficacy in locally advanced resectable ESCC [5–9]. However, the biomarkers for predicting the efficacy of NACI in ESCC and alterations in the intratumoural immune microenvironment (IME) remain unclear.

Tertiary lymphoid structures (TLSs) are organised aggregates of multiple immune cells, including T lymphocytes, B lymphocytes, and high endothelial venules in nonlymphoid tissues [10]. This structure has been associated with better clinical outcomes and responses to immunotherapy in many human tumours such as melanoma, head and neck squamous cell carcinoma (HNSCC), and ovarian cancer [11–14]. Recent studies have indicated that the function and prognostic significance of TLSs are related to their location [15, 16]. For instance, in clear cell renal cell carcinoma (ccRCC) [16], mature peritumoural TLSs were associated with a poorer prognosis, whereas the presence of distal TLSs in patients with endometrial cancer was significantly correlated with prolonged overall survival (OS) [17]. In ESCC, three studies examined the prognostic value of TLSs [18–20], and another study reported that TLSs can predict the response to ICIs in recurrent ESCC [21]. However, most of these studies used simple haematoxylin and eosin (HE)-stained sections or single-stain immunohistochemistry (IHC) of T and B cells to determine TLS presence and maturity, which may underestimate the true number of TLSs [22, 23]. Although these observations preliminarily explored the prognostic and predictive value of TLSs, the criteria for including TLSs in the quantification were inconsistent regarding their location. For instance, Hayashi et al. [21] only counted TLSs located within 1000 μm of the tumour nest boundary, whereas Rutao Li et al. [19] considered both intratumoural and peritumoural TLSs. Therefore, it is necessary to map the detailed spatial tissue architecture based on the cellular composition, maturity, location, and functional characteristics of TLSs to further explore their clinical correlation with patients with ESCC and comprehensively understand their role in anti-tumour immune responses within the tumour microenvironment.

Multiplex immunofluorescence (MIF) imaging can quantify the expression of multiple protein markers in the same tissue slice while maintaining spatial localisation [24]. In this study, we set two multiplexed immunofluorescence panels: an opal 7-colour dyes panel-1 (CD20/CD21/CD23/PNAD/CD8/Pan CK/DAPI) was used for simultaneous analysis of TLSs expression, phenotypic characterisation, and as well as tumour-infiltrating CD8 + T cells and PNAD+ high endothelial venules in different spatial regions of ESCC tissues; and a opal 6-colour dyes panel-2 (CD20/CD3/Foxp3/DC-Lamp/Pan CK/DAPI) was used to the evaluate the spatial distribution of TLS-associated immune cells—CD20 + B cells, CD3 + T cells, Foxp3 + Treg cells, and LAMP+ mature DCs—within the tumour microenvironment of ESCC tissue. Using the open-source Qupath software [25] for digital image analysis of fluorescent whole-tissue slices, we established relatively standardised operational approaches for TLSs acquisition, quantification, and spatial location. We analysed the spatial locations of TLSs with different phenotypes to explore the relationship between TLSs with distinct phenotypes at different locations and patients with ESCC prognosis. Additionally, we investigated the spatial distribution of TLSs in patients with ESCC receiving NACI and explored the predictive and prognostic value of spatially distinct TLS phenotypes for treatment efficacy.

Materials and methods

Patient cohorts and specimens

In this study, we obtained surgically removed tumour specimens from 87 previously untreated consecutive patients with ESCC who underwent curative surgical resection at the First Medical Center of the PLA General Hospital between May 2011 and June 2012. These patients did not receive any tumour-related preoperative chemoradiotherapy or immunotherapy and received a routine cisplatin-based combination chemotherapy regimen for postoperative recurrence and progression. The patient’s mean follow-up time was 30.67 months (interquartile range [IQR] 12.7–41.47 months), and basic clinicopathological characteristics are shown in Supplementary Table S1.

Herein, we also enrolled another cohort that received PD-1 blockade pembrolizumab in combination with nab-paclitaxel and cisplatin as neoadjuvant therapy for 15 patients with resectable ESCC. Between July 2020 and March 2022, patients diagnosed with ESCC were administered two cycles of preoperative neoadjuvant therapy, and oesophagectomy for ESCC was performed within a time frame of 4–8 weeks after neoadjuvant therapy. Postoperatively, 73% (11/15) of the patients completed two cycles of adjuvant therapy with an identical regimen as the neoadjuvant therapy. In patients with recurrence and progression, standard therapeutic strategies should be implemented according to the patient’s situation. These patients will be followed-up until 14 February 2025. The radiological RECIST 1.1 criteria were used to evaluate the clinical response to neoadjuvant therapy. We collected paired tumour samples, which were baseline biopsies and surgically resected tissues, from pre- and post-NACI patients. Supplementary Table S2 presents the basic clinicopathological features of the patients.

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of the PLA General Hospital (Approval No. S2019-228-02). Informed consent for the use of relevant clinical data and tissue samples was obtained from the patients prior to surgery.

Multiplexed immunofluorescence assay (MIF)

MIF assay was performed using an OPAL™ Polaris 7-Colour Manual IHC Kit (NEL861001KT; Akoya Biosciences). Specimen preparation was performed according to the manufacturer’s instructions using a previously published procedure [26]. Opal 7 colour dyes panel-1 (CD20/CD21/CD23/PNAD/CD8/Pan CK/DAPI) and opal 6 colour dyes panel-2 (CD20/CD3/Foxp3/DC-Lamp/Pan CK/DAPI) were stained in two sequential sections from the same patient. The primary antibodies and dilutions were anti-CD3 (1:300, Abcam, clone SP7), anti-CD8 (1:100, ZSBio, clone SP16), anti-CD20 (1:350, Abcam, clone SP16), anti-CD21 (1:300, ZSBio, clone EP64), anti-CD23 (1:200, ZSBio, clone EP75), anti-Foxp3 (1:100, Abcam, clone 236 A/E7), anti-DC-Lamp (1:500, Abcam, clone EPR24265-8), anti-PNAD (1:100, BioLegend, clone MECA-79), and anti-Pan CK (1:2000, Abcam, clone, PAN-CK). Tyramide signal amplification was performed using Opal 480 (CD21), Opal 520 (CD3 and CD23), Opal 570 (CD8 and DC-Lamp), Opal 620 (Foxp3 and PNAD), Opal 690 (CD20), and Opal 780 (CK) fluorescent dyes. Nuclear counterstaining of the sections was used by spectral DAPI.

Digital image analysis of MIF

The whole-tissue section (WTS) stained with MIF was scanned using the Vectra Polaris platform on a x20 multispectral microscopy (Akoya Biosciences, USA), and the images were visualised using Akoya Phenochart software (v.1.1.0, Akoya Biosciences, USA). Quantitative digital image analysis and spatial annotation of MIF images were performed using QuPath (v.0.5.1) software [25].

For the quantitative analysis of markers, the DAPI channel for nuclear staining was selected for cell segmentation, the positive threshold of pixel fluorescence intensity was set to 15, and the nuclear area and cell parameters were the default options. In the single measurement classifier module, manually adjust the log histograms and real-time preview options to set a positive threshold for each marker: CD3 (dye-cell-positive threshold: mean MIF = 25), CD8 (dye-cell-positive threshold: mean MIF = 25), CD20 (dye-cell-positive threshold: mean MIF = 25), CD21 (dye-cell-positive threshold: mean MIF = 35), CD23 (dye-cell-positive threshold: mean MIF = 28), Foxp3 (dye-nucleus-positive threshold: mean MIF = 30), DC-Lamp (dye-cell-positive threshold: mean MIF = 30), PNAD (dye-cell-positive threshold: mean MIF = 30), and Pan CK (dye-cell-positive threshold: mean MIF = 22). A composite classifier for the MIF images was then created according to the individual classifiers for each marker. A composite classifier was used to quantify positive intensity values of each marker in each MIF image.

For the acquisition of TLS, we employed automated annotation using CD20 channel density maps, followed by manual secondary validation and edge adjustment by two experienced pathologists. During this manual review, pathologists considered the T-cell zone of TLSs [27], such as the visualised CD3 or CD8 channels. In the density map module, the object type main class performed the CD20 channel, the density radius was set to 50 μm, the density threshold was set to 25, and other parameters were the default option.

In spatial analysis, tumour nests and stromal regions were first annotated in tissue sections by Pan CK channel. Then select the outer boundary annotation line of the tumour nest as the interface, the tumour nests and stroma region of all markers were spatially annotated with a diameter of 100 μm by expand annotation module, covering all whole tissue. A hierarchical insertion approach was used for the spatial annotation of each TLS. In the annotation module, all TLSs that have been fully annotated are selected. Subsequently, all TLSs were hierarchically inserted, and the relative spatial positions were determined.

Statistical analysis

All Statistical analyses and data descriptions were performed using IBM SPSS Statistics (version 26). Data visualisation was performed using GraphPad Prism (version 8.0). Visualisation of cell proportional distribution analysis was performed using Chiplot online. Descriptions of the patients’ clinicopathological data are presented as percentages. Statistical comparisons between two or more groups were performed using two-tailed t-tests and one-way ANOVA. Correlation analysis of TLSs annotations in serial sections was performed using Pearson’s correlation coefficients. The correlation coefficient of ≥ 0.8 indicates a very strong correlation. The association between the distal mature follicular TLSs (F-TLSs) and clinicopathological features was analysed using the chi-square test. Patient survival analysis was conducted using the log-rank test and Kaplan-Meier method. All cases were stratified into high number/density and low number/density TLS groups based on an optimal cutoff value determined using the X-Tile software [28] through minimal p-value analysis. Statistical significance was done with the following conventions: *p* < 0.05, **p* < 0.01, ***p* < 0.001, and ****p* < 0.0001; non-significant (*p* > 0.05) results were denoted as ‘ns’.

Results

Strategies for quantification, spatial localization and phenotypic classification of TLSs

TLSs differ from specific structured secondary lymphoid organs (SLOs) in that they exhibit different organisational patterns, which can be simple lymphocyte aggregates or more organised structures [10, 23]. Therefore, in the TLS acquisition process, CD20 + B cells were used to identify as many aggregates as possible. Subsequently, we expanded the region around each B cell aggregate to include the surrounding T cell zone and ultimately obtained the TLSs of the WTS (Fig. 1A, left). Previous studies only considered the total cell population of TLS aggregates [29, 30]. In this study, in addition to setting the minimum radius (50 μm) of positive B cell dense aggregates when obtaining TLSs, we also set a threshold of at least 10 CD20 + positive B cells per TLS during data statistical analysis to quantify TLS. Here, we analysed the consistency of quantitative TLSs between two consecutive slices, and the results showed good agreement (Supplementary Fig. 1). In the spatial position annotation of TLS, we referred to the approach of Werner, Wagner, Simon, Glatz, Mertz, Läubli, Griss & Wagner [31] in melanoma. Specifically, using the outer boundary of the tumour nest as the reference line, extend the internal and external space of the tumour nest by a distance of 100 μm in diameter. Spatial annotation of TLSs within the tumour nest and stroma in the WTS was subsequently performed using a hierarchical insertion approach (Fig. 1A, middle). When calculating the number of TLSs in adjacent spatial regions, if the relative area of the TLSs was significantly large, TLSs intersecting multiple perimeters were assigned to the perimeter with a dominant proportional area (Fig. 1A, right).

Fig. 1.

Strategies for quantification, spatial localisation and phenotypic classification of tertiary lymphoid structures (TLSs) in whole tissue sections (WTS) stained with multiplex immunofluorescence (MIF) in patients with esophageal squamous cell carcinoma (ESCC). (A) Schematic diagram of digital image analysis for quantification and spatial annotation of TLSs in MIF-stained WTS. The left panel shows the tumour nests and stromal regions annotated based on Pan CK channels in WTS, as well as all TLSs annotated via the density of CD20-positive B cells and artificial edge adjustment. The red shaded areas represent tumour nests, the blue-gray shaded areas indicate stromal regions, and the blue elliptical outlines denote TLSs. The middle panel illustrates spatial annotation of the WTS from the tumor nest to the stromal region, with a single subregion diameter of 100 μm. The trapezoidal lines represent spatial annotation guides: yellow trapezoidal contour lines for the tumour nests region and light yellow trapezoidal contour lines for the stromal region. The right panel displays a 200 μm magnified view of the area indicated by the white square in the left/middle panels. The blue shaded TLS localized within 100 μm of the stromal region during spatial annotation. (B) Representative images of early E-TLSs and mature F-TLSs (PFL-TLSs, SFL-TLSs) in MIF staining. CD20 + B cells (red), CD21 + FDCs (cyan-blue), CD23 + GC cells (green), CD8 + T cells (yellow), PNAD+ HVEs (orange), Pan CK+ tumor cells (white), and DAPI+ nuclei (blue). FDC, follicular dendritic cell; GC, germinal center; HVEs, high endothelial venules

Based on molecular signal characterisation and structural features reported in previous studies [20, 21], TLS phenotypes were divided into early TLSs (E-TLSs) and mature follicular TLSs (F-TLSs). Early TLSs were characterised by relatively dense CD20 + lymphocyte-dominated immune cell aggregations without CD21 or CD23 expression and with scattered T cells interspersed within the aggregates (Fig. 1B, top). The F-TLSs can be further classified into two distinct subtypes: primary follicle-like TLSs (PFL-TLSs) and secondary follicle-like TLSs (SFL-TLSs). PFL-TLSs present as well-defined round or oval clusters of small lymphocytes containing PNAD+ high endothelial vessels expressing CD21 but lacking CD23. (Fig. 1B, middle), and SFL-TLSs exhibiting large round or oval follicles with definite germinal centres expressing CD23 and the variable presence of CD21. (Fig. 1B, bottom). In the quantitative statistics of PFL-TLSs and SFL-TLSs, we set a minimum threshold of no fewer than five positive cells for both CD21 + B cells and CD23 + B cells.

Spatial distribution analysis of TLSs in patients with ESCC

In surgical resection tumour tissues from 87 untreated patients with ESCC, spatial distribution analysis of TLSs was performed in both intra-tumoural regions within 1000 μm (i.e., the tumour nests) and extra-tumoural regions located within a range of 7700 μm from the outer boundary of the tumour nests (i.e., the stroma) (Fig. 2A-B). The intra- to extra-tumoural continuum was stratified into 87 consecutive sub-region with a diameter of 100 μm. Among these, 95% of the cases (83/87) had TLSs, and a total of 2,746 TLSs were detected (Supplementary Fig. 2A). To calculate the density of TLSs, we divided the number of TLS in each subregion of the tumour nest and stroma by the corresponding area of that region. This approach ensures the precise measurement of TLS distribution relative to the tissue area. No discernible TLSs were observed within the tumour nests, which is consistent with previous reports [21] (Fig. 2A). We found TLSs were mainly distributed in the stromal region within a range of 4200 μm outside the tumour nests, exhibiting a gradual decrease in both number and density with increasing distance from the tumour margin (Fig. 2B-C).

Fig. 2.

The spatial distribution of tertiary lymphoid structures (TLSs) within tumour nests and stromal regions in surgical resection samples from 87 untreated esophageal squamous cell carcinoma (ESCC) patients. (A) Representative images of the anatomical and spatial division of tumour nests and stromal regions and the distribution of TLSs in the multiplex immunofluorescence (MIF)-stained tissue section. The area enclosed by the solid yellow line represents the tumor nest region (Pan-CK positive), while the stromal components outside the tumor nest constitute the peritumoral stromal region (Pan-CK negative). The solid yellow line marks the boundary between the tumour nest and the stroma. The region within ≤ 500 μm from the outer boundary of the tumour nest is defined as the proximal stroma, while the region beyond > 500 μm from the outer boundary is defined as the distal stroma. The boundary between the proximal and distal stromal regions is indicated by a solid white line. The red ellipses in proximal stromal regions represent proximal TLS (pTLS). The yellow ellipses in distal stromal regions represent distal TLS (dTLS). The spatial division of the tumour nest region is outlined with a magenta contour, that of the proximal stromal region with a green contour, and that of the distal stromal region with a yellow contour. The cyan-blue ellipses represents TLSs. (B) and (C) show the number and density distribution of total TLSs in different subspatial regions from tumour nests and stromal regions. (D-G) display the density distribution of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in different subspatial regions respectively. (H) presents the number and proportion of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in proximal versus distal regions of the stroma. (I) shows the density of Total-TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in proximal and distal stromal regions

Phenotypic stratification was performed for each TLS of each patient. A total of 2, 370 E-TLSs and 376 F-TLSs were detected (171 PFL-TLSs and 205 SFL-TLSs) (Supplementary Fig. 2A). Next, we analysed the spatial distributions of E-TLSs, PFL-TLSs, SFL-TLSs, and F-TLSs. The E-TLSs population was mainly localised within the 1,500 μm stroma region outside the tumour nest, and the number and density showed obvious peri-tumoural accumulation (Fig. 2D; Supplementary Fig. 2B). The highest density was observed in the 100–300 μm subregion, accounting for 49.41% (1,171/2,370) of total E-TLSs. In contrast, although F-TLSs have a density peak in the stroma region within 500 μm outside the tumour nest, its overall spatial distribution was relatively dispersed (Fig. 2E; Supplementary Fig. 2C). Notably, within the substromal region beyond 500 μm to 1100 μm, F-TLS also exhibited a high relatively density (mean density: 0.029/mm², [± 0.055]), despite their abundance progressively diminishing with increasing distance from the tumour nests (Fig. 2E). PFL-TLSs and SFL-TLSs can detect up to 3500 μm from the outer boundary of the tumour nests (Fig. 2F-G, Supplementary Fig. 2D-E).

Assessment of TLSs’ spatial distribution relative to the tumour is important for the efficacy of its predictive outcomes [16, 32]. Here, based on the above spatial distribution observations and previous reports in endometrial cancer [17, 21], we further classified TLSs within the stroma into proximal TLSs (≤ 500 μm from the outer boundary of the tumour nests) and distal TLSs (> 500 μm from the outer boundary of the tumour nests) (Fig. 2A). We examined differences in the quantity and density of TLSs with different phenotypes in the proximal and distal regions. We found that proximal TLSs were mainly composed of early stage subtypes, accounting for 88.85% (1844/2073) of the total proximal TLSs (Fig. 2H, left). In contrast, the proportion of mature phenotype F-TLSs significantly increased in the distal TLSs, reaching 21.84% (147/673) (Fig. 2H, right). The densities (/mm2, [± SD]) of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in the proximal TLSs were 0.284 [± 0.241], 0.041 [± 0.055], 0.017 [± 0.029], and 0.023 [± 0.038], respectively (Fig. 2I, left). The densities of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in the distal TLSs were 0.050 [± 0.060], 0.014 [± 0.039], 0.006 [± 0.024], and 0.010 [± 0.023], respectively (Fig. 2I, right).

Prognostic value of different spatial resident TLSs and maturation status in patients with ESCC

In research on the relationship between TLSs and the prognosis of ESCC, some researchers have utilised the number of TLSs within the tumour tissue [20], whereas others have considered the median density of TLSs per unit tissue area [21]. In this study, we used the X-tile software [28] to iterate through all potential cutpoints. The value that minimized the P-value between the two groups was selected as the optimal cutoff to divide ESCC patients into high and low groups for analyzing the association between the number and density of TLSs and patient prognosis. No TLSs with a clear structure were found in the tumour nests. Therefore, we mainly studied the TLSs in different spaces of the stromal region and the prognosis of ESCC patients. We first investigated the prognostic relationship between total TLSs and TLSs with distinct maturation states (E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs) in patients with ESCC within individual stromal subregions (Fig. 3A; Supplementary Fig. 3A). We observed a prognostic difference between high and low density TLS groups for total TLSs within the 300–400 μm (p = 0.00964) and 600–800 μm (p = 0.00355, p = 0.00652) subregions. Furthermore, survival analysis within stromal subregions for different phenotypes revealed statistically significant differences: patients with high-density E-TLSs in the 300–400 μm (p = 0.00584) and 600–900 μm (p = 0.02686, p = 0.01902, p = 0.0455) subregions, and high-density F-TLSs in the 800–900 μm (p = 0.04288) subregion, showed significantly better survival outcomes compared to their respective low-density TLS groups.

Fig. 3.

Survival analysis of tertiary lymphoid structures (TLSs) with different spatial and maturation states in 87 untreated esophageal squamous cell carcinoma (ESCC) patients. (A) Log-rank test results showing p-values for overall survival (OS) in 87 patients based on the density of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs across each subregion. (B) Cumulative subregional p-values for OS in 87 patients derived from the log-rank test based on the density of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs. (C) Kaplan-Meier survival curves for OS in 87 patients based on the number of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in the distal stromal region, analyzed using the log-rank test. (D) Kaplan-Meier survival curves for OS in 87 patients based on the density of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in the distal stromal region, analyzed using the log-rank test

Subsequently, we counted the number and density of TLSs in the cumulative stromal subspace spatial regions to investigate the prognostic relationship between TLSs and patients with ESCC. We found that in the cumulative stromal regions beyond 500 μm from the outer boundary of tumour nests, patients with ESCC with high-density total TLSs showed significantly better survival compared to those with low density TLSs, with a statistically significant p-value (Fig. 3B; p < 0.05 for cumulative regions beyond 500 μm). The impact of TLSs with different phenotypes on survival in cumulative stromal regions was analysed. We found that mature F-TLSs, whether assessed by number or density in the cumulative regions beyond 600 μm away from the outer boundary of tumour nest, compared with the low number or density TLSs group, the high number or density TLSs group showed a consistent significant correlation with the prognosis of patients with ESCC (Fig. 3B; Supplementary Fig. 3B). This finding is also echoes the secondary spatial density distribution observed in our analysis of the F-TLS density patterns (Fig. 2E).

Based on the above results, we speculate that the stromal distal TLSs (> 500 μm) have a more significant impact on the survival of patients with ESCC (Fig. 3C-D, supplementary Fig. 3C-D). We further examined the prognostic relationship between different TLS phenotypes within the distal TLSs in patients with ESCC. We found that patients with ESCC with a high number or density of mature F-TLSs at the distal had better OS compared with those with a low number or density of TLSs at the distal (Fig. 3C-D, TLSs high number, p = 0.0092; TLSs high density, p = 0.0268). Here, we also analysed the correlation between stromal distal F-TLSs and clinicopathological features, and the results are shown in Supplementary Table S3-4.

Overall, through multi-level spatially stratified analysis of survival, we revealed the differential prognostic impact of TLSs based on their spatial localisation in patients with ESCC. The stromal distal TLSs demonstrated superior prognostic value, particularly as the stromal regions beyond 500 μm from tumour nests contained a higher abundance of mature F-TLSs, which showed a statistically significant correlation with improved OS.

Analysis of spatial distinct IMEs related to TLSs in ESCC patients and its relationship with TLSs

Previous studies have shown that mature TLSs shape intratumoural IME in ESCC. ESCC with mature TLSs shows higher intratumoural CD8 + T cell infiltration compared to absent mature TLSs [18]. In this study, we observed that spatially distinct TLSs with different maturation states located in the stromal region beyond the boundary of tumour nests differentially impacted the prognosis of patients with ESCC. Therefore, we hypothesised that the distribution of TLSs with different maturation states within the spatially heterogeneous IME of ESCC may exhibit functional divergence. The antitumour efficacy of TLSs is likely influenced by both their spatial localisation and maturation status in ESCC.

Using MIF staining, we simultaneously analysed the spatial distribution of CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + regulatory T cells (Tregs), LAMP+ dendritic cells (mature DCs), and PNAD+ high endothelial venules (HEVs) within both the tumour nests and stroma of the ESCC specimens (Fig. 4A). Based on the differential density distribution of TLSs, we also stratified the stromal distribution of the above-mentioned TLS-related immune cells and HEVs into proximal (≤ 500 μm) and distal (> 500 μm) regions. We observed that CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Tregs, and LAMP+ mature DCs exhibited varying densities of infiltration in both tumour nests and stroma, whereas PNAD+ HEVs were mainly located within the stroma (Supplementary Fig. 4A). Spatial stratification analysis revealed that the infiltration densities of CD20 + B cells, CD3 + T cells, CD8 + T cells, and Foxp3 + Tregs in the stromal region were significantly higher in the proximal compared to the distal (Fig. 4B).

Fig. 4.

Analysis of the spatial immune microenvironment (IME) related to tertiary lymphoid structures (TLSs) in 87 untreated esophageal squamous cell carcinoma (ESCC) patients and its relationship with TLSs. (A) Representative images of multiplex immunofluorescence (MIF) staining of TLSs-associated immune cells and high endothelial venules (HEVs). CD20 + B cells (red), CD21 + FDCs (cyan-blue), CD23 + GC cells (green), CD3 + T cells (green), CD8 + T cells (yellow), Foxp3 + Treg cells (orange), Lamp + DCs (yellow), PNAD+ HEVs (orange), Pan CK+ tumor cells (white), DAPI+ nuclei (blue). The orange arrow marks the tumor nest area, the red arrow points to the stromal proximal region, and the yellow arrow designates the stromal distal region. The solid white line indicates the boundary between the proximal and distal stromal regions. The red ellipses represents the stromal proximal follicular TLS (pF-TLS). The yellow ellipses represents the stromal distal follicular TLS (dF-TLS). Analysis of the density (B) and cellular proportions (C) of CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Treg cells, Lamp+ mature DCs, and PNAD+ HEVs in different spatial regions: the tumor nest and the stroma (both proximal and distal). (D) Analysis of the correlation between the density of total TLSs in WTS and the density of CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Treg cells, Lamp+ mature DCs, and PNAD+ HEVs in different spatial regions: the tumor nest and the stroma (both proximal and distal). (E) and (F) respectively show the analysis of the correlation between the density of E-TLSs and F-TLSs in the stroma distal region and the density of CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Treg cells, Lamp+ mature DCs, and PNAD+ HEVs in different spatial regions: the tumor nest and the stroma (both proximal and distal)

The density distribution relationship among TLSs, different immune cells, and PNAD+ HEVs in different subspatial regions was investigated. We found that in patients with ESCC with a high density of total TLSs compared to those with a low density of total TLS, CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Tregs, and LAMP3 + DCs exhibited distinct proportional clustering within tumour nests (Fig. 4C). Statistical analysis of cell densities revealed a significant difference in CD8 + T cells among patients with ESCC with a high density of total TLSs compared with those with a low density of total TLS (Fig. 4D). In the stromal region, overall, patients with a high density of total TLSs exhibited increased infiltration of CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Tregs, and LAMP3 + DCs, along with enhanced PNAD + HEV distribution (Fig. 4D). Notably, regional stratification analysis revealed that in the high-density total TLSs group, proximal CD3 + T cells and CD8 + T cells were significantly higher than their distal counterparts. Furthermore, a greater distribution of PNAD+ HEVs was observed at the proximal than at the distal (Fig. 4D). Additionally, in patients with a high density of total TLSs, the increase in Foxp3 + Tregs in both the proximal and distal did not reach statistical significance (Fig. 4D). Cellular proportion analysis showed significantly higher clustering of Foxp3 + Tregs in the low-density total TLSs group than in the high-density total TLSs group (Fig. 4C).

Our spatial survival analysis revealed that stromal distal TLSs demonstrated a better prognostic value, with stromal distal F-TLSs specifically showing a significant association with improved OS. Therefore, we explored the relationship among distal TLSs, different immune cell subsets, and PNAD+ HEVs within distinct spatial regions. We observed that in the patient group with a high number or density of distal E-TLSs compared to the group with a low number or density, although there was an obvious dense infiltration of CD8 + T cells within the stromal regions, the tumour nests did not exhibit the corresponding high-density infiltration of CD8 + T cells (Fig. 4E, Supplementary Fig. 4B). Additionally, we found that in the patient group with a high number or density of distal E-TLSs compared to the group with a low number or density, the distal stromal region had significant infiltration of Foxp3 + Tregs (Fig. 4E, Supplementary Fig. S4B). However, in the patient group with a high number or density of distal F-TLSs compared to the group with a low number or density, in addition to significant infiltration of CD20 + B cells and CD3 + T cells in the stromal region, particularly in the proximal, it is also noteworthy that the high distal F-TLSs group exhibited a higher density of CD8 + T cells not only in both the proximal and distal stromal regions but also within the tumour nests. These differences reached statistical significance (Fig. 4F, Supplementary Fig. S4C). Furthermore, in the patient group with a high number or density of distal F-TLSs, compared to the group with a low number or density, there was a significantly greater distribution of PNAD+ HEVs in both the proximal and distal stromal regions (Fig. 4F, Supplementary Fig. 4C). The results for LAMP + DCs showed slightly less robustness between groups. A higher infiltration of LAMP + DCs in the high distal F-TLSs group, compared to the low F-TLSs group, was observed only in the stromal region when using the density-based grouping, and specifically in the proximal and distal stromal region when using the number-based grouping (Fig. 4F, Supplementary Fig. 4C). Additionally, while the group with a high number or density of distal F-TLSs showed significant Foxp3 + Treg infiltration in the proximal stromal region compared to the low F-TLSs group, it exhibited lower levels of Foxp3 + Tregs in both the tumour nests and the distal stromal region (Fig. 4F, Supplementary Fig. S4C).

The spatial distribution of TLSs in patients with ESCC with NACI and the predictive value of TLSs for their therapeutic efficacy and prognosis

The spatial distribution of TLSs was examined in 15 paired ESCC tumour specimens collected pre- and post-NACI. In the pre-treatment biopsy samples, 15 consecutive subregions were identified, extending from the tumour nest to the surrounding stroma (Supplementary Fig. 5A). Phenotypic analysis of TLSs revealed a predominance of E-TLSs (46/49), primarily located within 600 μm of the outer boundary of the tumour nests (Supplementary Fig. 5B-E). After NACI, the sample was divided into 46 consecutive subregions from the tumour nest to the stroma (Supplementary Fig. S5 F). Spatial analysis revealed that high densities of both E-TLSs and F-TLSs were mainly concentrated within 1000 μm outside the tumour nests, although there was heterogeneous TLSs distribution in the compartment within 3300 μm outside the tumour nests (Supplementary Fig. 5F-J). A comparison of TLS number and density between pre- and post-treatment samples showed a significant increase in both the number and density of TLSs after treatment. In particular, the number and density of F-TLSs, PFL-TLSs, and SFL-TLSs post-treatment showed statistically significant differences (Fig. 5A, Supplementary Fig. S5 K). Given the limited tissue volume of pre-NACI biopsy specimens, TLSs were observed only in the stromal region 600 μm beyond the tumour nest margin, making it difficult to distinguish between the proximal (≤ 500 μm) and distal (> 500 μm) regions. The post-NACI treatment samples were surgically resected tissues, consistent with the samples from the untreated cohort. Therefore, here, we also examined the distribution of different phenotypic TLSs in the proximal and distal stromal regions of post-NACI samples (Fig. 5B) and compared it with that in the untreated cohort. We found that compared to the proportion of proximal E-TLSs in the untreated cohort (88.85%, Fig. 2H), the proportion of E-TLSs among proximal TLSs decreased significantly after NACI treatment, accounting for 67.28% (109/162) (Fig. 5B). Notably, compared to the proportion of proximal F-TLSs in the untreated cohort (11.05%, Fig. 2H), the proportion of F-TLSs among proximal TLSs increased significantly after NACI treatment, accounting for 32.72% (53/162) (Fig. 5B). Furthermore, we observed a significant increase in the proportion of F-TLSs among distal TLSs (51.82%, Fig. 5B) compared to 21.84% in the untreated cohort (Fig. 2H).

Fig. 5.

To evaluate the predictive value of tertiary lymphoid structures (TLSs) for efficacy and prognosis in the neoadjuvant chemo-immunotherapy (NACI) cohort. (A) Analyze the density changes of total TLSs, E-TLSs, F-TLSs, PFL-TLSs and SFL-TLSs in pre- and post-treatment samples. (B) presents the number and proportion of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in proximal versus distal regions of the stroma in the post-treatment samples. (C) show the density differences of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs between PR and non-PR groups in pre- and post-treatment samples. (D) and (E) Analysis of density changes in E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in pre- and post-treatment specimens from the PR and non-PR patient groups, respectively. (F) and (G) respectively show the analysis of density differences in CD20 + B cells, CD3 + T cells, CD8 + T cells, Foxp3 + Treg cells, LAMP+ mature DCs, and PNAD+ HVEs between the PR and non-PR group in pre- and post-treatment samples. Among these analyses, groups with less than 3 samples or a standard deviation of zero were not subject to statistical analysis but were still included in the observation. These groups will be excluded from statistical analysis but will still be visualized. (H) Kaplan-Meier survival curves for OS of 15 patients receiving NACI were plotted based on the density of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in stroma-proximal regions of post-treatment samples, using the Log-rank test. (I) Kaplan-Meier survival curves for OS of 15 patients receiving NACI were plotted based on the density of total TLSs, E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs in stroma-distal regions of post-treatment samples, using the Log-rank test

Recent studies have shown a positive correlation between mature TLSs and treatment efficacy in patients with ESCC receiving either neoadjuvant chemotherapy or neoadjuvant immunotherapy alone [29]. In this study, we investigated the association between the TLSs and the efficacy of NACI in ESCC. In the pre-treatment baseline, among patients with radiological Partial Response (PR) and non-PR, whether it was total TLSs or TLSs of different phenotypes, there was no statistically significant difference in their number and density between the two groups (Fig. 5C; Supplementary Fig. 6A). However, after treatment, both the number and density of F-TLSs were significantly increased in patients with PR compared to those in non-PR patients (Fig. 5C; Supplementary Fig. 6A). We further analyzed the association between TLSs in the proximal and distal stromal regions and the efficacy of NACI for ESCC in post-treatment samples. Compared to non-PR patients, PR patients showed increased numbers and densities of F-TLSs in both proximal and distal regions, although these differences did not reach statistical significance (Supplementary Fig. 6B and C). Comparison of TLSs number and density of TLSs between PR and non-PR patients before and after treatment revealed that in patients with PR, both the number and density of E-TLSs, F-TLSs, PFL-TLSs, and SFL-TLSs increased significantly after treatment (Fig. 5D; Supplementary Fig. 6C). In contrast, among non-PR patients, although the number of E-TLSs and F-TLSs increased post-treatment, their density showed no statistically significant difference (Fig. 5E; Supplementary Fig. 6D). In this study, we examined the IME of TLS-related cells in patients with ESCC undergoing NACI and observed that within the tumour nests, the non-PR patient group exhibited a significantly higher infiltration of Treg cells compared to the PR group at baseline, indicating an immunosuppressive microenvironmental bias (Fig. 5F). Following treatment, PR patients showed prominent infiltration of CD3 + T cells within regions proximal to the stroma (Fig. 5G).

Survival analysis was performed on a cohort of patients with ESCC who received NACI. No association was found between the TLS levels in baseline biopsy specimens and OS (Supplementary Fig. 6E). This finding is consistent with the tumour microenvironment analysis, which revealed high intratumoural Treg cell infiltration in patients exhibiting high TLSs density (Supplementary Fig. 7A). In post-treatment specimens, patients with a higher number or density of total TLSs had significantly improved survival rates compared to those with a lower number or density of total TLSs. Following phenotypic stratification, statistically significant differences were observed in both E-TLSs and F-TLSs among patient cohorts with a high number or density of total TLSs (Supplementary Fig. 6F). Spatial regionalisation analysis of TLSs revealed that compared to the group with low number and density of proximal TLSs, high number and density of total TLSs, E-TLSs, and F-TLSs in the proximal region were all significantly associated with improved patient survival, with statistically significant differences (Fig. 5H; Supplementary Fig. 6G). However, patients with a high number and density of distal TLSs showed a trend toward better survival than those with a low number and density of distal TLS; however, this difference was not statistically significant. This finding corresponded to the IME analysis across different subspatial regions (Supplementary Fig. 7B).

Discussion

TLSs are ectopic lymphoid structures formed within non-lymphoid tissues, including tumours. Antitumour immunity and prognostic impact of TLSs in different cancers vary significantly depending on their spatial distribution, maturity, and abundance within the tumour tissue. Intratumoural TLSs abundance serves as an effective prognostic marker for favourable outcomes in intrahepatic cholangiocarcinoma, whereas the presence of peritumoural TLSs is positively correlated with poor prognosis [15]. TLSs in breast cancer, hepatocellular carcinoma, and ccRCC also demonstrate dual prognostic functions based on their spatial location, although its underlying mechanisms require further elucidated [16, 33, 34]. TLSs are categorized into E-TLSs, PFL-TLSs, and SFL-TLSs. F-TLSs is a general term for follicle-like mature TLSs, including PFL-TLSs and SFL-TLSs [27, 35]. Studies have shown that the presence of PFL-TLSs is associated with favorable clinical outcomes in various cancer types, such as colon cancer [36], ESCC [20] and laryngeal squamous cell carcinoma [37]. SFL-TLSs is generally regarded as a positive prognostic indicator in nearly all cancer types studied [38, 39]. Specifically, in triple-negative breast cancer (TNBC), the presence of SFL-TLSs is correlated with improved survival and higher clinical responses to neoadjuvant therapy [40]. In HNSCC, NACI promotes the formation of mature TLSs, and the maturity level of TLSs is associated with HPV status and treatment response to PD-1-based chemoimmunotherapy [41]. This study established a standardised operational procedure for the acquisition, quantification, and spatial localisation of TLSs using MIF and spatial digital image analysis. Our spatial stratification analysis show that mature F-TLSs located in the stromal distal region above 500 μm on the outer boundary of the tumour nest have the most significant positive impact on the OS of patients with ESCC (Fig. 3C-D). This finding is consistent with observations in endometrial cancer, in which the presence of distal TLSs was associated with significantly prolonged OS [17]. Studies have shown that E-TLSs, as a new lymphocyte aggregate lacking follicular organization, its existence marks the early recruitment of T and B lymphocytes within tumor tissues [10]. PFL-TLSs represent primary follicle-like TLSs; when CD21 + FDCs appear, they provide survival signals and antigen- presenting capacity, promoting local clonal expansion and differentiation of B cells, indicating a strengthening anti-tumor immune response [10, 35]. SFL-TLSs is a fully functional subtype. When mature SFL-TLSs with germinal centers (GC) are observed in tumor tissue, they can facilitate GC reactions including somatic hypermutation, antibody class switching, and affinity maturation, indicating stronger host immunity and better clinical outcomes [27, 35]. Therefore, we speculate that the stromal distal structures located outside the tumor boundary in ESCC, especially the mature F-TLSs (including PFL-TLSs and SFL-TLSs), form the in situ “immune centers” of the ESCC tumor tissue [42]. They may continuously recruit naive lymphocytes through hev to support the GC response, and then transport effector cells and antibodies around the tumor, thereby exerting an effective anti-tumor effect and prolonging the survival period of patients [10, 35].

To establish a mechanism bridge between TLSs and patient prognosis, we conducted an in-depth analysis of the spatial distribution of TLS-associated immune cells and PNAD+ HEVs in relation to the density of different TLS subtypes. We found that the presence of distal high density F-TLSs was strongly correlated with a significant enrichment of CD3 + T cells, CD20 + B cells, and CD8 + T cells in the stromal regions, particularly in proximal region (Fig. 4F). More importantly, compared to the low F-TLSs group, the distal high F-TLSs group also exhibited high density infiltration of CD8 + T cells within the tumour nests (Fig. 4F). This indicates that mature TLSs located in the distal stroma can effectively reshape the tumor microenvironment, overcome the immune exclusion barrier of tumour nests, and promote synergistic interactions between T cells and B cells—a functional core of the GC response [43–45]. This contrast sharply with the limited immune cell infiltration observed in the patient group with only distal enrichment of E-TLSs (Fig. 4E). In this favorable immune microenvironment, LAMP3 + DCs play a critical role. Our data show that in the stromal regions grouped by density and the proximal/distal stromal regions grouped by number, the distal high F-TLSs patient group presented a trend toward increased infiltration of LAMP3 + DCs (Fig. 4F, Supplementary Fig. 4C). As the most efficient antigen-presenting cells, the enrichment of mature DCs within TLSs is essential for activating naive CD3 + T cells, especially in driving their differentiation into functional effector cells [46]. They may serve as a bridge connecting tumor antigens with T cell activation within TLSs. Furthermore, our data show that distal high F-TLSs is associated with a lower proportion of immunosuppressive cells. Although we observed an associated increase in Foxp3 + Tregs was observed in the proximal stromal regions, the infiltration level of Foxp3 + Tregs was lower in the more functionally significant tumour nests and distal stromal region in the high-density distal F-TLSs patient group (Fig. 4F, Supplementary Fig. 4C). This finding suggests that mature distal F-TLSs may effectively counteract [47] or remodel [48] the immunosuppressive network locally (particularly within the tumour nest and its distal periphery) by establishing a robust effector immune response mediated through the coordinated action of T cells (especially non-Treg subsets), B cells, and DCs, thereby maintaining a favorable balance between effector and regulatory cells. Additionally, distal high density F-TLSs is associated with a more abundant distribution of PNAD+ HEVs in the stromal regions (particularly in distal) (Fig. 4F), which provides a structural basis for the dynamic trafficking of the aforementioned immune cells. HEVs are present around TLSs, are associated with tumour infiltration by T and B cells [49, 50], and serve as the primary site for the extravasation of naïve CD8 + T cell subsets [51]. Therefore, we speculate that the structural maturity of TLSs determines their immune functional output. Mature F-TLSs located in the distal stroma function as an in-situ “synergistic immune centers”, continuously recruit naïve CD3 + T cells through surrounding HEVs, facilitating local antigen presentation by mature DCs, and supporting T cell-B cell collaboration within GC, ultimately leading to generating and deployment effector cells for anti-tumor immunity [10, 27].

Previous studies have assessed the predictive value of mature TLSs for treatment efficacy in patients with ESCC receiving either neoadjuvant chemotherapy or neoadjuvant immunotherapy alone [29, 52]. In addition, high-density TLSs in patients with recurrent ESCC treated with anti-PD-1 antibodies predicted the clinical response to anti-PD-1 antibodies and patient survival [21]. Our study further analyzed the dynamic changes and clinical significance of TLSs in ESCC patients under the synergistic treatment model of neoadjuvant chemotherapy combined with immunotherapy. Spatial analysis of NACI treatment samples revealed that after NACI treatment, the high densities regions of both E-TLSs and F-TLSs were mainly concentrated within 1000 μm outside the tumour nests. By comparing the TLS characteristics of the samples pre- and post-treatment, it was found that the number and density of TLS significantly increased after treatment, especially F-TLSs, PFL-TLSs and SFL-TLSs. This indicates that NACI not only increases the quantity of TLSs but also promotes their enrichment in the effector region near the tumor core, which may optimize the generation and delivery of effector immune cells [29, 53]. We evaluated the clinical significance of dynamic changes in TLSs pre- and post-NACI treatment. The results showed that baseline TLSs characteristics (both number and density) in pre-treatment specimens were not significantly correlated with treatment response (PR vs. non-PR). However, treatment response (PR) was closely associated with significant expansion and maturation of F-TLSs post-treatment (Fig. 5C). Further dynamic comparative analysis indicated that in patients achieving PR, the number and density of E-TLSs, F-TLSs, and their subtypes (PFL-TLSs and SFL-TLSs) showed statistically significant increases post-treatment. In contrast, although TLSs number increased in non-PR patients, no significant change was observed in their density (Fig. 5D-E). This suggests that NACI may not only inducing the formation of TLSs, but also driving them to reach a specific “density threshold” and complete the maturation process [41, 54]. Therefore, from baseline to post-treatment, the density dynamic changes and maturity evolution of TLSs (especially F-TLSs) can provide more predictive biological information compared with the static assessment at a single time point, pointing the way for the development of biomarkers using longitudinal samples.

In addition, our research found that after NACI treatment, proximal (rather than distal) TLSs showed a stronger statistical difference in patient OS. This seemingly contradicts our finding in the untreated cohort that distal F-TLSs are of core prognostic value. However, of note, in addition to significantly increasing TLS quantity, their spatial composition was also altered after NACI compared to the untreated cohort. The comparison results showed that post-treatment, the proportion of E-TLSs in the proximal stroma (≤ 500 μm) decreased markedly from 88.85% to 67.28%, while that of F-TLSs in the same region increased significantly from 11.05% to 32.72%. In parallel, the proportion of F-TLSs in the distal stroma (> 500 μm) also rose from 21.84% to 51.82% (Figs. 2H and 5B). This spatial pattern optimization characterized by “systemic maturation and proximal enrichment” not only reflects the successful remodeling of the tumor IME by NACI, but also thereby offers key insights into the underlying basis for the good prognostic value of proximal TLSs. Thus, we speculate that the mature F-TLSs induced by NACI treatment and located in proximal regions do not represent “sanctuaries” of treatment resistance or immunosuppression; rather, they may serve as effectors of treatment success and frontline sites for immune attack [21, 53]. Under untreated conditions, distal F-TLSs function as relatively independent “synergistic immune center” that is not easily affected by local tumor immunosuppression, can effectively initiate and maintain anti-tumor immunity. In contrast, NACI therapy—particularly through the synergistic effects of chemotherapy-induced immunogenic cell death and ICIs [55]—greatly reshapes the tumor microenvironment [41, 53]. This remodeling involves eliminating immunosuppressive cells (such as Tregs) near tumour nests and releasing large amounts of tumor antigens [7, 56]. This creates conditions for the formation or expansion of mature F-TLSs in the proximal stromal regions closer to tumor cells. These newly formed or expanded proximal F-TLSs, being directly exposed to high concentrations of tumor antigens, can more efficiently recruit and activate lymphocytes, and rapidly transport effector cells (such as CD8 + T cells) into tumour nests, thereby achieving effective tumor cell killing [53, 57]. The significant infiltration of CD3 + T cells observed in the proximal stromal regions of PR patients post-treatment directly exemplifies this process. Consequently, under NACI therapy, the abundance and maturity of proximal F-TLSs serve as a key indicator of whether the treatment has successfully advances the anti-tumor immune response into the tumor core [29, 53]. In contrast, although distal F-TLSs also increased following therapy, their relatively diminished association in survival analysis may reflect that in treatment-responsive patients, the intensity of the proximal immune response is sufficient to dominate the prognosis, thereby partially “masking” the statistical contribution of distal structures. This does not imply a loss of their function, as they may still play important roles in maintaining systemic immune surveillance and long-term immunologic memory [17, 21].

However, this study has several limitations. First, our NACI cohort had a relatively small sample size (n = 15), which constrained the statistical power of the cohort, especially when performing complex subgroup analyses based on multiple TLSs subtypes and spatial regions. Therefore, the findings in the NACI cohort regarding the association between the spatial distribution of TLSs and efficacy and prognosis should be regarded as preliminary and enlightening clinical observations, and their robustness still needs to be verified in larger-scale prospective cohorts. Second, regarding the predictive value of TLSs in pre-treatment biopsy samples, our findings differ from some studies in melanoma [12], sarcoma [58], and TNBC [40]. This might be caused by several factors: (1) Sampling Bias: Pre-treatment needle biopsies capture only local tumor information and may fail to represent the overall spatial heterogeneity of TLSs within the tumor, resulting in an incomplete assessment. (2) Maturity Differences: The pre-treatment TLSs we detected were mainly early-stage E-TLSs (46/49), with an insufficient number of mature F-TLSs, which are likely the key predictors of immunotherapy response. (3) Tumor Type Specificity: The IME of ESCC and its response to therapy may have unique characteristics distinct from other cancer types. Third, current research indicates that the proportion of CD138 + plasma cells is significantly associated with TLS maturity [21], with mature TLSs exhibiting a higher percentage of IgM-containing cells [29] compared to immature TLSs. However, owing to limitations in the number of markers included in the MIF panel, this study did not incorporate markers related to mature B cells and their functional states. Therefore, future investigations are necessary to further explore the relationship between mature B cells, spatial distribution of TLSs, and tumour IME.

In summary, this study, by integrating spatial localisation, maturation phenotypes, and the dynamic dimensions of therapeutic intervention, elucidates the critical role of TLSs maturity and spatial distribution in predicting ESCC prognosis and response to NACI therapy. In untreated patients, mature F-TLSs located in the stromal distal region serve as a key indicator of favorable outcomes, functioning as efficient “synergistic immune center.” In patients receiving NACI, the treatment induced the maturation and spatial distribution optimization of TLSs, particularly enriching proximal F-TLSs. This signifies the establishment of an anti-tumor immune frontline, which is associated with improved treatment response and survival. Despite limitations such as sample size, our findings provide new insights into the functional heterogeneity of TLSs in ESCC and their dynamic evolution under combination therapy, highlighting the potential value of integrating TLSs spatial and maturity features into future biomarker development and immunotherapeutic strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ESCC

Esophageal squamous cell carcinoma

NACI

Neoadjuvant chemo- immunotherapy

PD-1

Programmed cell death 1

ICIs

Immune checkpoint inhibitors

TLSs

Tertiary lymphoid structures

IME

Immune microenvironment

OS

Overall survival

WTS

Whole-tissue section

MIF

Multiplexed immunofluorescence

HEVs

High endothelial venules

Author contributions

Lingxiong Wang: Conceptualization, Writing - review & editing, Writing - original draft, Visualization, Validation, Supervision, Software, Methodology, Investigation, Formal analysis, Data curation. Jinfeng Li: Investigation, Visualization, Validation, Software, Methodology. Yanyun Ao: Investigation, Validation, Supervision, Resources. Yanju Yu: Methodology, Data curation, Jinzhao Zhai: Methodology, Formal analysis. Yingjie Zhang: Methodology, Software. Liangliang Wu: Investigation, Software. Jianqing Hao: Formal analysis. Yanyan Hu: Supervision. Qiong Wang: Conceptualization, Writing - review & editing, Validation, Fan Yin: Conceptualization, Writing - review & editing, Resources, Investigation. Tianyi Liu: Conceptualization, Project Administration, Writing - review & editing, Data curation, Formal analysis.

Funding

None.

Data availability

All data generated from the use or analysis in this study are available by the corresponding author upon reasonable request, except the patients data.

Declarations

Ethics approval and consent to participate

The study conformed to the ethical approval processes and was approved by the Institutional Review Committee of Chinese PLA General Hospital (Approval No. S2019-228-02). Before the surgery, informed consent from patients regarding relevant clinical data and tissue samples was obtained

Competing interests

The authors declare that they have no conflicts of interest.