Mature tertiary lymphoid structures linked to HPV status and anti-PD-1 based chemoimmunotherapy response in head and neck squamous cell carcinoma

ABSTRACT

Mature tertiary lymphoid structures (TLSs) are immune aggregates associated with immune checkpoint blockade (ICB) responses in various cancers, yet their role in chemoimmunotherapy response in head and neck squamous cell carcinoma (HNSCC) remains unclear. By analyzing TCGA-HNSC transcriptomic data and pathology slides, we identified an immune subtype enriched in TLSs, predominantly in HPV-positive tumors, which correlated with favorable immunotherapy response. Single-cell and spatial transcriptomics further revealed distinct TLS compositions, with mature TLSs enriched in germinal center B cells, follicular helper T cells, and resident memory CD8 T cells, while immature TLSs contained FCRL4+ B cells and peripheral helper T cells. Multispectral immunohistochemistry, flow cytometry, and ELISA validated these findings. Notably, neoadjuvant chemoimmunotherapy promoted mature TLS formation. These results suggest that TLS maturity correlates with HPV status and response to anti-PD-1-based chemoimmunotherapy, providing insights for potential therapeutic strategies in HNSCC.

Introduction

According to global cancer statistics in 2018, head and neck cancer is the seventh most common cancer in the world, resulting in 890,000 new cases and 450,000 deaths per year.¹ Head and neck cancer consist of head and neck squamous cell carcinoma (HNSCC) from the oral cavity, pharynx and larynx. Epidemiological studies have shown that the causative factors of HNSCC include tobacco, alcohol, environmental pollutants and viral infections (such as human papillomavirus, HPV). Infection with high-risk HPV causes a substantial and rising proportion of these tumors, originating mainly in the oropharynx. HPV-positive oropharyngeal cancer is a type of HNSCC with unique biological characteristics. Most importantly, patients with HPV-positive oropharyngeal cancer have a significantly better prognosis than patients with HPV-negative HNSCC²; this difference may be related to its unique tumor microenvironment (TME).

Immunotherapy has been reported to be more effective in HPV-positive patients. A meta-analysis that included 11 studies focusing on first-line treatment of recurrent/metastatic (R/M) HNSCC showed that HPV-positive patients had a 1.29-fold higher likelihood of responding to immunotherapy compared to HPV-negative patients; however, this difference was not statistically significant (95% CI: 0.85–1.96, p = 0.24). Overall survival (OS) was reported to be almost twice as long in HPV-positive patients compared to HPV-negative patients, but this observation should be interpreted with caution given the limited statistical power and the heterogeneity of included studies.³ Regarding neoadjuvant immunotherapy, the CheckMate 358 trial demonstrated that both radiographic and pathologic response rates were higher in HPV-positive compared to HPV-negative patients with resectable HNSCC.⁴ However, it is important to note that immunotherapy responsiveness in head and neck cancer (HNC) is not exclusively linked to HPV status. Many HPV-negative HNC patients also respond well to immune checkpoint blockade (ICB), and therefore, the use of p16 or HPV-DNA status alone, or even in combination, as a selection criterion for immunotherapy remains controversial.

Increasing evidence show that cancer cells do not alone play a role but, together with extracellular matrix and mesenchymal cells, form the TME to carry out functions. There are many immune and nonimmune cells in the TME, which create a pro-inflammatory, immunosuppressive and pro-angiogenic microenvironment. Cancer cells can adapt to survive and proliferate in this environment and thereby escape immune surveillance. Generally, effective adaptive antitumor immunity occurs in secondary lymphoid organs. However, studies on the TME have shown that antitumor immunity occurs not only in the secondary lymphoid organs but also occur directly in the organized cell population in the tumor area. This cell population is called the tertiary lymphoid structure(TLS).⁵ In the TLS, adjacent tumor antigens are presented to T cells by follicular dendritic cells (FDCs) in situ, which activate, proliferate and differentiate T cells and B cells and finally form effector memory T helper cells, effector memory killer cells, memory B cells and plasma cells that produce antibodies. In most solid tumors, the presence of TLS is associated with a better prognosis, suggesting that they can induce long-term antitumor immune responses.⁶ TLSs can be divided into immature TLSs, which are lymphoid aggregates lacking germinal centers, and mature TLSs, which resemble secondary lymphoid organs and contain germinal centers with follicular dendritic cells.⁷ Recent studies have shown that mature TLSs can predict immune checkpoint inhibitor (ICB) therapy efficacy in solid tumors independently of PD-L1 expression.⁸ Mature TLSs also serve as critical niches for tumor-specific immune responses.⁹

In this study, we investigated the cellular composition and spatial organization of TLSs in HNSCC using single-cell RNA sequencing (scRNA-seq) and multispectral immunohistochemistry (mIHC). While previous studies have established the prognostic and predictive value of TLSs in melanoma and several other solid tumors,^8,10 TLSs remain insufficiently characterized in HNSCC. Existing research has primarily focused on the association between HPV status and TLS presence in HNSCC,^11,12 but the predictive role of TLSs in patients receiving neoadjuvant chemoimmunotherapy has not been reported. To address this gap, we conducted a comprehensive analysis of mature and immature TLSs in HNSCC, identifying key immune cell subsets – such as germinal center B cells, follicular helper T cells, and resident memory CD8+ T cells – in mature TLSs, and validating these findings through mIHC and flow cytometry. Furthermore, we analyzed paired samples before and after neoadjuvant anti-PD-1-based chemoimmunotherapy and observed changes suggestive of treatment-induced TLS maturation. This study adds new evidence on the role of TLSs in HNSCC patients receiving neoadjuvant chemoimmunotherapy and provides mechanistic insights into TLS maturation and composition, highlighting their relevance as potential biomarkers and therapeutic targets.

Materials & methods

TCGA data processing

First, the RNA-seq data (count values) of HNSCC cohort were downloaded from the Cancer Genome Atlas (TCGA) website (https://cancergenome.nih.gov/). Second, the data were merged using Perl software. RNA-seq profiles were normalized using edgeR package. Clinical information was downloaded from the Cancer Genome Atlas (TCGA) website (https://cancergenome.nih.gov/).

Five representative immune gene sets obtained from the literature¹³ were used to cluster HNSCC tumor samples. The 5 representative signatures are: “CSF1_response” for activation of macrophages/monocytes (referred to throughout text and figures as “Macrophage”), “LIexpression_score” representing overall lymphocyte infiltration, and dominated by B and T cell signatures (referred to throughout text and figures as “Lymphocyte”), TGF-β response “TGFB_score_21050467” (referred to throughout the text and figures as “TGF-beta”), “Module3_IFN_score” representing IFN-γresponse (referred to throughout the text and figures as “IFN-gamma”), and wound healing “CHANG_CORE_SERUM_RESPONSE_UP” (referred to throughout the text and figures as “Wound healing”). The gene sets were listed in Supplemental Data 1. First, the GSVA R package was implemented to calculate ssGSEA scores within the HNSCC tumor samples. Then, the EM clustering algorithm in the mclust R package was used to perform unsupervised clustering on the sample ssGSEA scores. Finally, the category K value that maximizes the Bayesian information criterion (BIC) was chosen, and 4 clusters were obtained. Heatmaps were plotted by heatmap2 from the gplots R package.

Immune cell type enrichment analysis was conducted by using xCell software. xCell is a novel gene signature-based method that can conduct cell type enrichment analysis from gene expression data to infer 64 immune and stromal cell types. Here, the lymphatic immune cells were mainly analyzed. According to the comparison of different groups, the average value of the cell type enrichment score was statistically analyzed. The histograms were generated in R using the heatmap.2 function from the gplots package. The Kruskal–Wallis test (multi-group) or Wilcoxon test (two groups) was used to analyze the difference in immune gene enrichment scores in immune subtypes, to obtain a p value of significant difference and to calculate the calibrated FDR value.

Kaplan–Meier survival analysis and log-rank test were performed to estimate the prognostic impact of immune subtypes on overall survival (OS) and diseases free survival (DFS). Samples with missing values in the features of interest or the outcomes were excluded from the analysis. The clinical features of these patients are provided in Supplemental Data 2.

Tumor immune dysletion and exclusion (TIDE) analysis

To address the limitations of conventional biomarkers for immune checkpoint blockade (ICB) response prediction, we implemented the computational framework TIDE.¹⁴ This algorithm quantifies two distinct immune evasion mechanisms through transcriptomic profiling:

T Cell Dysfunction Score: Computed using expression levels of T cell exhaustion markers (PDCD1, CTLA4, LAG3, HAVCR2) normalized against cytotoxic effector genes (CD8A, GZMB, PRF1). The score reflects the ratio of inhibitory receptor signaling to functional cytotoxic potential.

Immune Exclusion Score: Derived from stromal barrier signatures (COL1A1, ACTA2, TGFB1) and immunosuppressive myeloid cell markers (CD163, S100A8/9), capturing physical and cellular mechanisms limiting T cell infiltration.

The composite TIDE score was calculated as: TIDE = (Z-normalized Dysfunction Score) + (Z-normalized Exclusion Score)

For the TCGA-HNSC cohort, RNA-seq data (FPKM) were processed through the TIDE web portal (http://tide.dfci.harvard.edu/) with default parameters. Higher scores indicate predominant immune evasion mechanisms, predictive of ICB resistance.

Human tissue samples

The 150 treatment-naïve HNSCC samples (Supplemental Data 3) were collected prior to any systemic or local therapy, including surgery, radiotherapy, chemotherapy, or immunotherapy. For the 24 neoadjuvant anti-PD-1-treated HNSCC pre-treatment samples (Supplemental Data 4, from NCT04718415, NCT04807140, NCT05582265 and NCT05980702), these were biopsy samples collected before the initiation of neoadjuvant therapy. pCR (pathological complete response) is defined as no residual tumor cells. Major pathological response (MPR) includes patients with ≤ 10% residual tumor cells. Pathological partial response (pPR) is defined as ≤ 50% of the tumor area containing residual tumor cells. Pathological no response (pNR) refers to patients with > 50% residual tumor cells. The MPR group consists of both pCR and MPR cases, while the non-MPR group includes pPR and pNR cases.

This study was conducted in accordance with the Declaration of Helsinki and was approved by Ethical Committee of Sun Yat-Sen Memorial Hospital (SYSEC-KY-KS-2021–109). Given the retrospective nature of this study, the requirement for informed consent was waived by he Ethical Committee of Sun Yat-Sen Memorial Hospital, as all data were anonymized and posed minimal risk to participants.

Immunohistochemistry (IHC)

Formalin-fixed paraffin-embedded (FFPE) sections were obtained from the pathology department, where initial p16 immunohistochemistry (IHC) was performed using the CINtec® Histology Kit (Roche, Basel, Switzerland) as part of standard diagnostic procedures.

To further validate p16 status, sections were subjected to a secondary immunostaining procedure. Briefly, FFPE sections were deparaffinized, rehydrated, and subjected to antigen retrieval using Tris-EDTA buffer (pH 8.0). Endogenous peroxidase activity was quenched before incubation with CDKN2A/p16INK4A antibody (1:100; Abcam, #ab108349, Cambridge, UK) or CD20 antibody (1:400; Cell Signaling Technology, #48750, Danvers, MA, USA) at 4 °C overnight. The following day, the sections were incubated with secondary antibodies and stained using DAB detection kits (Genetech, #GK6005, Shanghai, China).

To ensure the reliability of p16 status determination, only cases where both CINtec®-based and Abcam-based staining results were concordant (i.e., both positive or both negative) were classified as p16-positive or p16-negative. Sections were scanned using a Nikon microscope and analyzed with ImageJ software.

Multispectral IHC

FFPE sections were deparaffinized, rehydrated, and subjected to antigen retrieval by sodium citrate (pH 6.0) or Tris-EDTA (pH 8.0). After blocking with goat serum for 10 minutes at room temperature, the sections were incubated with a primary antibody. Then the sections were washed three times for 2 minutes each in 0.02% Tris buffered saline Tween (TBST), followed by incubated with the horseradish peroxidase – conjugated secondary antibody (Panovue, #10217100050, Beijing, CN) for 10 minutes at room temperature. After washed again with TBST, the sections were incubated with fluorophore working solution for 10 minutes at room temperature. The above information describes one cycle. After each cycle, antigen retrieval, blocking, primary antibody incubation, secondary antibody incubation, and fluorescent dye incubation were repeated. Following the completion of all the cycles, the sections were incubated with DAPI at room temperature for 5 minutes. PPD 520 Fluorophore, PPD 570 Fluorophore, PPD 620 Fluorophore, PPD 650 Fluorophore, PPD 690 Fluorophore and DAPI (all from Panovue, #10217100050, Beijing, CN) were applied to each antibody. Multispectral images were acquired by the PerkinElmer Vectra platform at 200-fold magnification. The following primary antibodies were used: anti-CD4 (1:1000; Abcam, #ab133616, Cambridge, UK), anti-CD8α (1:200; Cell Signaling Technology, #70306, Danvers, MA, USA), anti-CD20 (1:400; Cell Signaling Technology, #48750, Danvers, MA, USA), anti-CD21 (1:250; Abcam, #ab75985, Cambridge, UK), anti-BCL6 (1:1500; Abcam, #ab172610, Cambridge, UK), anti-CD103 (1:1000; Abcam, #ab129202, Cambridge, UK), anti-CXCL13 (1:1000; Abcam, #ab246518, Cambridge, UK), anti-CXCR5 (1:5000; Abcam, #ab254415, Cambridge, UK), anti-PD1 (1:500, Abcam,#ab52587, Cambridge, UK) and anti-FCRL4 (1:500; Abcam,# ab239076, Cambridge, UK).

Flow cytometry

CD8⁺ T cells were separated from healthy donor peripheral blood using EasySep™ Human CD8⁺ T Cell Isolation Kit (STEMCELL, #17953, Vancouver, BC, Canada) according to manufacturer’s instructions. For TGF-β treatment, 2x10⁶/ml CD8⁺ T cells were treated with 2 ng/ml TGFβ-1 (PeproTech, #100–21, Cranbury, NJ, USA) and CD3/CD28 MicroBeads (Miltenyi, #130–050–101 and #130–093–247, Bergisch Gladbach, Germany) (TGF-β group) or CD3/CD28 MicroBeads (Miltenyi, #130–050–101 and #130–093–247, Bergisch Gladbach, Germany) (NC group). Following antibodies were used for flow cytometry: anti-CD8-FITC (Biolegend, #301005, San Diego, CA, USA), anti-CD103-BV421 (Biolegend, #350214, San Diego, CA, USA), anti-CXCL13-APC (Thermos Fisher, #17–7981–82, Waltham, MA, USA).

Single-cell RNA sequencing data processing

Eight publicly available HNSCC scRNA-seq datasets were obtained from various studies.^15–22

Sample preparation, scRNA-seq, and library construction of In-house data

Eight tumor biopsy samples were collected during routine clinical procedures and stored in MACS Tissue Storage Solution (Miltenyi, Germany) as the discovery cohort. The samples were enzymatically dissociated using the gentleMACS Tumor Dissociation Kit (Miltenyi) at 37°C for 60 minutes, following the manufacturer’s instructions. Dissociated cells were filtered through a 40 μm strainer (Biosharp, China), centrifuged at 300 × g for 10 minutes, treated with red blood cell lysis buffer, and washed twice with PBS (Gibco, USA) supplemented with 2% FBS (ExCell, China) before resuspension.

Single-cell RNA sequencing (scRNA-seq) libraries were prepared according to standard protocols. Briefly, after an additional PBS wash with 0.04% BSA (Invitrogen, USA), the cell density was assessed, and ~2 × 10⁵ cells were loaded onto the 10x Genomics GemCode Single-Cell Instrument to generate Gel Bead-in-Emulsion (GEMs). cDNA synthesis and library preparation were performed using the Chromium Next GEM Single Cell 3’ Kit v3.1 (10x Genomics), incorporating barcoded primers with an Illumina® R1 sequence, a 16-nt 10x Barcode, a 10-nt Unique Molecular Identifier (UMI), and a poly-dT primer. Silane magnetic beads were used to purify cDNA, which was amplified by qRT-PCR for sequencing. The libraries were sequenced on an Illumina HiSeq X-Ten platform, generating 150 bp paired-end reads.

scRNA-seq data processing

We used Scanpy to analyze scRNA-seq data. Briefly, cells with low complexity libraries ( < 500 genes), likely dying or apoptotic cells ( >10% of transcripts are derived from the mitochondria) and cells with high-complexity libraries ( >5000 genes) were removed. Genes expressed fewer than 10 cells were also excluded. After quality control, following the standard protocol of Scanpy, the count data were normalized and the logarithm-transformed. Top 2000 highly variable genes were selected for downstream analysis. After completing principal component analysis, we selected 50 principal components to compute the neighborhood relations. Representative markers were used to annotate clusters. The distribution preferences of cell clusters were calculated using odds ratio (OR). Ligand-receptor analysis was performed using cellphondb and cellchat according to the instructions. Transcription factor analysis was conducted by decoupleR according to instructions.

Spatial transcriptomic analysis

A HNSCC spatial transcriptomic dataset (GSE181300) was obtained from GEO.²³ The spatial transcriptomic data was analyzed by Scanpy. To evaluate the cellular components in TLSs, Tangram was used to align scRNA-seq data to spatial transcriptomic data. TLS area was identified using reported markers (IGHA1, IGHG1, IGHG2, IGHG3, IGHG4, IGHGP, IGHM, IGKC, IGLC1, IGLC2, IGLC3, JCHAIN, CD79A, FCRL5, MZB1, SSR4, XBP1, TRBC2, IL7R, CXCL12, LUM, C1QA, C7, CD52, APOE, PTLP, PTGDS, PIM2, DERL3).²⁴

Spatial proximity analysis

To measure the spatial relationship between CD103⁺CD8⁺T cells and CD20⁺B cells, 2 representative 200-fold fields of view with or without TLSs were captured by PerkinElmer Vectra platform and processed by HALO (lndica Labs). The XY locations of CD103⁺CD8⁺T objects and CD20⁺B objects were generated, and the distance from CD103⁺CD8⁺T cells to the nearest CD20⁺B objects was calculated.

ELISA

Fresh tumor tissue samples were obtained from three patients with HNSCC during surgery at the Department of Oral and Maxillofacial Surgery at Sun Yat-sen Memorial Hospital (SYSMH), (Guangzhou, China). Fresh tissues were washed 3 times with cold phosphate-buffered saline (PBS) containing 1% FBS before being minced into small pieces. Specimens were dissociated using a Tumor Dissociation Kit (Miltenyi Biotec, 130–095–929) according to the manufacturer’s instructions. Cell suspensions were then filtered through a 100 μm cell strainer and a 70 μm cell strainer, washed once with PBS, and resuspended in cell staining buffer. Single-cell suspensions were immunostained with anti-CD3-PE (BD), anti-CD4-APC (eBioscience), anti-CD8-FITC (Biolegend), anti-CD103-BV421 (Biolegend). CD103⁻CD8⁺T cells and CD103⁺CD8⁺ T cells were sorted by Beckman MoFlo EQs. Sorted T cells remained unstimulated or were activated either with a stimulation cocktail containing phorbol myristate acetate (PMA, 40.5 mmol/L) and ionomycin (670 mmol/L, 500* dilution, Invitrogen, 00–4970–93). Supernatant concentrations of soluble CXCL13 protein were determined using a commercially available Human BLC ELISA kit (RayBio, ELH-BLC-1) according to the manufacturer’s instructions. The concentration of CXCL13 in the supernatant was caculated from a standard curve, which was generated using the respective recombinant protein.

Statistical analysis

GraphPad Prism 9.0 for Windows (GraphPad Software, Inc.) was used to conduct statistical analyses. Data analyses were conducted by either Mann–Whitney U test or unpaired t-test for two-group comparisons or by one-way analysis of variance (ANOVA) for multiple-group comparisons. Kaplan–Meier method and log-rank test were used to analyze the survival of patients. A p < 0.05 was considered statistically significant.

Results

Identification of a unique immune subgroup from TCGA mining

First, 5 representative immune-related gene sets were used to cluster the TCGA-HNSC tumor samples, and 500 HNSCC samples were calculated and clustered into 4 immune subtypes (Supplemental Figure S1a). As shown in Figure 1a, each of the four immune subtypes had its own unique immune characteristics. C1 (n = 152) showed a strong TGF-β response, C2 (n = 131) reflected a high IFN-γ response; C3 (n = 108) showed a high wound healing-associated score and C4 (n = 109) was characterized by lymphocyte infiltration (Supplemental Data 2). These immune subtypes were associated with different prognoses. As illustrated in Figure 1b, patients in the C4 subtype group had a better prognosis (OS, p = 0.03; DFS, p = 0.01). Immune cell type enrichment analysis was performed to assess immune cell infiltration across the four subtypes using transcriptomic data. Figure 1c shows that both C4 and C2 subtypes had higher CD4+ and CD8+ T cell infiltration scores, with C4 exhibiting the highest B cell infiltration. Specifically, the C4 subtype displayed the greatest infiltration scores for memory B cells, naïve B cells, plasma B cells, pro B cells, and class-switched memory B cells (Supplemental Figure S1b). Additionally, higher B cell infiltration was significantly associated with a better prognosis (p < 0.001) (Supplemental Figure S1c), which may explain the superior outcomes in the C4 subtype. Given that the high levels of CD4⁺ T cell, CD8⁺ T cell and B cell in the C4 subtype were characteristic of TLS, we evaluated TLS presence in all four subtypes. A 9-gene TLS transcription score¹⁰ was calculated, and the C4 subtype had the highest TLS transcription score (Figure 1d). Moreover, a deep learning model²⁵ was applied to analyze TCGA pathology slides, where the TLS HE score was defined as the ratio of segmented TLS area to total tissue area. A representative pathology slide from the C4 subtype was shown in Figure 1e. Similarly, C4 subtype showed the highest TLS HE score (Figure 1f). The TLS transcription score was significantly correlated with the TLS HE score in TCGA-HNSC cohort (p < 0.0001) (Supplemental Figure S1d), confirming the consistency of these two methods. These findings suggest that the C4 subtype is characterized by high TLS infiltration.

Figure 1.

Identification of a unique immune subgroup from TCGA mining.

(a) The characteristics of 5 gene signatures within 4 immune clusters. (b) The survival analysis within 4 immune subsets. Left, overall survival; Right, disease-free survival. (c) The enrichment of immune cells within 4 immune subgroups. (d) The TLS-transcription score within 4 immune subgroups. (e) Representative image of a HE staining pathology slide from TCGA-HNSC C4 group. (f) The TLS-HE score within 4 immune subgroups. (g) The distribution of HPV (-) and HPV (+) HNSCC patients within the 4 immune clusters. HPV (-) represents HPV-negative, while HPV (+) represents HPV-positive. (h) The TLS-HE score in HPV-negative and HPV-positive group. (i) The TIDE score within 4 immune subgroups. (j) The correlation between TLS-HE score and TIDE score. *p < 0.05; **p < 0.01; ***p < 0.001

Numerous studies^26,27 have demonstrated that HPV-positive HNSCC patients had higher B cell infiltration levels, and we analyzed the HPV status of the immune subsets. As shown in Figure 1g, analysis of the distribution of 102 HNSCC patients with known HPV status revealed that most HPV-positive HNSCC patients were in the C4 subtype (p < 0.001). Additionally, HPV-positive HNSCC exhibited a higher TLS HE score (p < 0.01) (Figure 1h). Previous studies have showed that the responders to anti-PD-1 therapy had higher levels of TLS infiltration.¹⁰ The TIDE algorithm, which evaluates T cell dysfunction and exclusion to predict cancer immunotherapy response, revealed that the C4 subgroup had the lowest TIDE score, indicating a higher sensitivity to immunotherapy (Figure 1i). Further analysis demonstrated a significant negative correlation between TIDE score and TLS HE score (p = 0.02), indicating that HNSCC patients with more TLSs are more responsive to immunotherapy. Our previous study showed HPV-positive HNSCC patients had lower levels of tumor mutation burden (TMB).²⁸ To further explore the relationship between TLSs, TMB, and HPV status, we examined TLS transcription scores in HNSCC patients. Interestingly, HNSCC patients with higher TMB had lower TLS transcription scores (Supplemental Figure S1e). While patients with higher TMB are expected to generate more neoantigens and activate the adaptive immune system, they exhibited fewer TLSs, suggesting that antigen-presenting machinery may play a critical role in TLS formation. In conclusion, our data identified a unique immune subtype in HNSCC characterized by high TLS infiltration, which is associated with HPV-positive status and better immunotherapy response.

Mature tertiary lymphoid structure was enriched in HPV-positive HNSCC tumors and anti-PD-1 based immunochemotherapy responders

Recent studies have demonstrated that mature TLSs (germinal center like structures containing follicular dendritic cells) can predict ICB therapy efficacy in solid tumors independently of PD-L1 expression,⁸ and mature TLSs were key niches of tumor-specific immune responses.⁹ To further investigate this, we assessed mature TLS infiltration in 150 treatment-naïve HNSCC samples (91 HPV-negative and 59 HPV-positive), and 24 biopsy samples from HNSCC patients (all HPV-negative) receiving anti-PD-1 based neoadjuvant chemoimmunotherapy (17 non-MPR and 7 MPR) from Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University (Supplemental Data 3 and 4). As illustrated in Figure 2a and Supplemental Figure S2a, p16 immunohistochemical staining (black box in the upper left corner) was used to distinguish HPV-negative HNSCC (left) from HPV-positive HNSCC (Right). Through mIHC staining for CD4, CD8, CD20, and CD21, we identified that cell aggregates, characterized by a central B cell population resembling a germinal center and follicular dendritic cells (FDCs), surrounded by CD4⁺ T cells and CD8⁺ T cells are mature TLSs. Supplemental Figures S2b,c show the calculation of cell density and TLS density using Inform. The number of mature TLSs per mm² could be calculated by identifying mature TLSs in the panoramic tissue, using a Phenochart circle drawing, and calculating the tissue area. The number of B cells, CD4⁺T cells, and CD8⁺T cells were calculated by identifying cells by Inform, segmenting cells, and using specific fluorescent pattern. As shown in Figure 2b, in HPV-positive HNSCC, the density of CD4⁺ T cells (p < 0.001, Cohen’s d = 0.549), CD8⁺ T cells (p = 0.001, Cohen’s d = 0.530) and B cells (p < 0.001, Cohen’s d = 0.673) was significantly higher, and more mature TLSs were formed (p = 0.005, Cohen’s d = 0.446). In the cohort of HNSCC patients undergoing neoadjuvant chemoimmunotherapy, patients achieving a pathological complete response (pCR) or major pathological response (MPR) were classified as the MPR group, while those with a pathological partial response (pPR) or no response (pNR) were classified as the non-MPR group (Figure 2c). We found in MPR group, the densities of CD4⁺ T cells (p = 0.020, Cohen’s d = 0.844), CD8⁺ T cells (p = 0.011, Cohen’s d = 1.160) and B cells (p = 0.013, Cohen’s d = 1.021) were significantly higher, and more mature TLSs were formed (p = 0.020, Cohen’s d = 0.853) (Figure 2d). Our results showed that mature TLSs were enriched in HPV-positive HNSCC and in patients responding to neoadjuvant chemoimmunotherapy.

Figure 2.

Mature tertiary lymphoid structure was enriched in HPV-positive HNSCC tumors and anti-PD-1 based immunochemotherapy responders.

(a) Upper, representative images of CD20 staining of HPV (-) and HPV (+) HNSCC. Upper left boxes show the p16 staining which distinguishes HPV (-) and HPV (+) HNSCC. Black arrows represent CD20-positive cell aggregates. Lower, representative images of CD20, CD21, CD8, CD4, DAPI multispectral immunochemistry staining. Left, scattered infiltration of immune cells; Right, cell aggregates featuring TLSs. (b) Comparison of TLS, CD20, CD4, CD8 densities between HPV (-) and HPV (+) HNSCC. (c) Upper, representative images of HE staining of non-MPR and MPR group. Lower, representative images of CD20, CD21, CD8, CD4, DAPI multispectral immunochemistry staining. (d) Comparison of TLS, CD20, CD4, CD8 densities between non-MPR and MPR group. *p < 0.05; **p < 0.01; ***p < 0.001

Single-cell RNA sequencing analysis of HNSCC

To characterize the cellular composition of mature TLS, we combined our in-house single-cell RNA sequencing (scRNA-seq) data with seven published scRNA-seq datasets.^15–21 As illustrated in Figure 3a, 338451 CD45⁺ immune infiltrating cells from 138 patients with HNSCC were clustered. Using graph-based clustering, we identified four major clusters based on characteristic markers of different cell types, including T cell, NK cells, B cells, myeloid cells. The four major clusters were further divided into 36 minor clusters including naïve CD4 T cell, memory CD4 T cells, follicular helper T cells (Tfh), follicular helper T cells/T helper cell 1 (Tfh/Th1), cytotoxic CD4 T cells, T helper cell 17 (Th17), regulatory CD4 T cells (Treg), naïve CD8 T cells, central memory CD8 T cells, effecter CD8 T cells, HSP⁺ CD8 T cells, ISG15⁺ CD8 T cells, resident memory CD8 T cells, CD16+ NK, CD103+ NK, IFIT3+ NK, XCL1+ NK, naïve B cells, activated B cells, FCRL4+ B cells, HSP+ B cells, germinal center B cells, plasma cells, C1QC+ macrophages, CXCL9+ macrophages, FCN1+ macrophages, FOLR2+ macrophages, IL1B+ macrophages, SPP1+ macrophages, monocytes, neutrophils, conventional dendritic cell 1 (cDC1), conventional dendritic cell 2 (cDC2), conventional dendritic cell 3 (cDC3), plasmacytoid dendritic cell (pDC) and mast cells based on characteristic markers. Our previous results showed that mature TLSs were enriched in both HPV-positive patients and in responders to neoadjuvant chemoimmunotherapy. Studies have shown that the responders to anti-PD-1 therapy had a higher level of mature TLS infiltration.⁸ To further decipher the cellular composition of mature TLSs, we compared the distribution of cell types between HPV-positive and HPV-negative patients (HPV- (n = 84), HPV+ (n = 54)), as well as between responders and non-responders to anti-PD-1 therapy (from GSE226620, Non-responder (n = 6), Responder (n = 4)), and patients with major pathological response (MPR) and patients with non-MPR to anti-PD-1 based neoadjuvant chemoimmunotherapy (from our in-house scRNA-seq cohort, Non-MPR (n = 3), MPR (n = 5)) (Figure 3b). As shown in Figure 3c, activated B cells, germinal center B cells, Tfh and resident memory CD8 T cells were enriched in both HPV-positive patients (from merged 8 scRNA-seq datasets) and responders to anti-PD-1 therapy (from GSE226620), as well as in patients who responded to the combination of anti-PD-1 and chemotherapy (from our in-house scRNA-seq cohort). These findings indicate that these cell types are likely critical components of mature TLSs, irrespective of whether the response occurs following chemoimmunotherapy or immunotherapy alone. Notably, correlation analysis revealed that these four clusters were highly correlated across HNSCC patients (Figure 3d). Spatial transcriptomic analysis showed that activated B cells, germinal center B cells, Tfh and resident memory CD8 T cells were localized in mature TLS region (Supplemental Figure S3a). Furthermore, cell interaction analysis showed that ligand-receptor interactions among these four cell clusters were stronger in HPV-positive patients and anti-PD-1 responders (Supplemental Figure S3b). These results indicated that activated B cells, germinal center B cells, Tfh and resident memory CD8 T cells are key components in mature TLSs.

Figure 3.

Single-cell RNA sequencing analysis of HNSCC.

(a) UMAP plot of 338,451 CD45⁺ immune infiltrating cells from 138 patients clustered into 4 major clusters and 36 subclusters. (b) OR (odds ratio) analysis of subcluster distribution. (c) Venn plot showing the cell type enriched in HPV-positive patients, anti-PD-1 therapy responders and anti-PD-1 plus chemotherapy responders. (d) Heatmap showing the correlation between cell subclusters. *p < 0.05; **p < 0.01; ***p < 0.001

Identification and validation of germinal center B cells in mature TLS

To explore the cellular lineage and function of the B cell subpopulations associated with mature TLSs, we clustered B cell into 6 subpopulations compromising: naïve B cells (IGHD, IGHM), activated B cells (CD69, CD83), FCRL4+ B cells (FCRL4, ITGAX, DUSP4), HSP+ B cells (HSPA1A, HSPA1B), germinal center B cells (RGS13, SUGCT, NEIL1, LMO2, GMD5, MARCKSL1), plasma cells (JCHAIN, MZB1, XBP1) using representative markers (Figure 4a,b). FCRL4+ B cells and HSP+ B cells were activated B cells with distinct features. FCRL4+ B cells, which highly expressed FCRL4, ITGAX and DUSP4, were previously known as atypical memory B cells or age-associated B cells in infection, autoimmune diseases and aging mice.²⁹ Recent studies have shown this subset are prevalent in tumor microenvironment,^7,30 indicating a potential role in tumor immunity. HSP+ B cells, which express high levels of heat shock proteins HSPA1A and HSPA1B, exhibit a stressed phenotype. As illustrated in Figure 4c, germinal center B cells were more abundant in HPV-positive patients (from merged 8 scRNA-seq datasets), anti-PD-1 therapy responders (from GSE226620) and neoadjuvant chemoimmunotherapy responders (from our in-house scRNA-seq cohort). Additionally, we observed that multiple ligands and receptors related to TLS formation, such as LTBR, LTA, CCL17, IL21R, and TNFRSF8, were highly expressed on germinal center B cells, indicating their crucial role in TLS development (Figure 4d). mIHC analysis confirmed the presence of BCL6+CD20+ germinal center B cells within mature TLSs (Figure 4e). These cells were significantly enriched in HPV-positive patients (p < 0.001) and responders to neoadjuvant chemoimmunotherapy (p = 0.010) (Figure 4f). As FCRL4+ B cells also highly expressed ligands and receptors related to TLS formation (Figure 4d), and cell interaction analysis showed that germinal center B cells and FCRL4+ B cells had higher interaction scores with other immune cells among B cell subsets (Supplemental Figure S4a), we further investigated the transcription profile and location of FCRL4+ B cells. As shown in Supplemental Figure S4b, CollecTRI³¹ based regulon analysis showed the activation of BCL6 in germinal center B cells and BHLHE40 activation in FCRL4+ B cells, which was consistent with previous study.³⁰ As shown in Supplemental Figure S4c, pathway analysis showed significant enrichment of proliferation, apoptosis and DNA change in germinal center B cells, which was consistent with the function of germinal center. B cells underwent proliferation after activation, somatic hypermutation, antibody affinity maturation, and B cells with lower affinity underwent apoptosis. However, FCRL4+ B cells showed enrichment of pathways related to antigen processing and presentation and T cell activation, suggesting their role in antigen presentation. Next, We explored the localization of FCRL4+ B cells and found that in immature TLSs (lymphoid aggregates), FCRL4+ B cells were centrally located, whereas in mature TLSs (germinal center-like structures containing follicular dendritic cells), they were found at the periphery (Supplemental Figure S4d), which was consistent with previous finding.⁷ This suggests that FCRL4+ B cells may play a key role in the early stages of TLS formation by presenting antigens to T cells, and as TLSs mature, these cells migrate to the periphery. The higher infiltration of FCRL4+ B cells in HPV-positive tumors may explain the increased TLS formation in these patients. However, we did not observe a significant difference in FCRL4+ B cell infiltration between responders and non-responders to anti-PD-1 therapy (Figure 4c). This suggests that only mature TLSs are predictive of the response to anti-PD-1 therapy, which was consistent with previous studies.^8,9 In summary, our results demonstrate that germinal center B cells are a critical component of mature TLSs, while FCRL4+ B cells are primarily located in immature TLSs.

Figure 4.

Identification and validation of germinal center B cells in mature TLS.

(a) UMAP plot of 6 B cell subclusters. (b) Dotplot showing the representative markers in 6 B cell subclusters. (c) Stacked bar graph showing the proportion of 6 B cell subclusters among B cells. The figures on the bars indicate the proportion of B cell subclusters. (d) Matrix plot showing the expression of ligands and receptors involved in TLS formation in B cell subclusters. (e) Representative image of germinal center B cells in mature TLS by multispectral immunochemistry staining. Red arrows indicate BCL6⁺CD20⁺ germinal center B cells. (f) Comparison of the density of BCL6⁺CD20⁺ cells between HPV (-) and HPV (+) HNSCC, non-MPR and MPR group. *, p < 0.05; **, p < 0.01; ***, p < 0.001

Identification and validation of Tfh in mature TLS

To investigate the cellular lineage and function of the T cell subpopulations associated with mature TLSs, we first clustered T cell into CD8 T cells and CD4 T cells (Figure 5a). CD4 T cells were clustered into seven subpopulations compromising: naïve CD4 T cells (CCR7, SELL, TCF7), memory CD4 T cells (CD69, GPR183, IL7R), Tfh cells (CXCL13, PDCD1, TOX, CXCR5), Tfh/Th1 cells (IFNG, GZMA, GZMB, GZMK, GZMH), cytotoxic CD4 T cells (NKG7, PRF1), Th17 cells (IL17A, IL17F), and Tregs (FOXP3, IL2RA), based on representative markers (Figure 5b,c). As shown in Figure 5d, Tfh cells were found in greater proportions in HPV-positive patients (from merged 8 scRNA-seq datasets), anti-PD-1 therapy responders (from GSE226620) and neoadjuvant chemoimmunotherapy responders (from our in-house scRNA-seq cohort). Furthermore, several ligands and receptors involved in TLS formation, including IL6ST, CD40LG, and LTB, were highly expressed on Tfh cells, indicating their key role in TLS formation (Figure 5e). Single-cell clonotype analysis revealed that Tfh cells had a higher proportion of expanded clones ( >5 clones) in anti-PD-1 responders than in non-responders, suggesting that Tfh activation plays a crucial role in anti-PD-1 therapy-mediated antitumor immunity (Figure 5f). mIHC analysis confirmed the presence of BCL6+CD4+ Tfh cells in mature TLSs (Figure 5g). These BCL6+CD4+ Tfh cells were significantly enriched in HPV-positive patients (p < 0.001) and responders to neoadjuvant chemoimmunotherapy (p = 0.005) (Figure 5h). Additionally, we identified a CD4+ T cell subset, Tfh/Th1, which expressed the highest levels of CXCL13, a key chemokine in TLS formation (Figure 5c). Tfh/Th1 cells co-expressed Tfh-related genes (PDCD1, TOX) and Th1-related genes (IFNG, GZMA, GMZB), but had low CXCR5 expression, resembling peripheral helper T cells (Tph) previously described in autoimmune diseases.³² Given the strong CXCL13 secretion capacity of Tph cells, we hypothesize that Tph may play a critical role in the early stages of TLS formation. As shown in Supplemental Figure S5a, Tfh cells were enriched in pathways related to B cell activation, whereas Tfh/Th1 (Tph) cells were enriched in pathways related to B cell chemotaxis. This suggests that Tph cells may function primarily in the early phase of TLS formation by recruiting B cells, and as TLSs mature, Tph cells are replaced by Tfh cells, which then promote B cell activation. As shown in Supplemental Figure S5b, CollecTRI³¹ based regulon analysis showed the activation of BCL6 in Tfh and SOX4 activation in Tfh/Th1 (Tph), which was consistent with previous study.³² As shown in Supplemental Figure S5c, cell interaction analysis indicated that Tfh/Th1 (Tph) cells had the strongest interaction with germinal center B cells through the CXCL13-CXCR5 axis, demonstrating the robust recruitment capacity of Tph cells. Additionally, Tfh cells can secrete CXCL13 and interact with CXCR5+ germinal center B cells, while germinal center B cells also secrete CXCL13 to interact with CXCR5+ Tfh cells. Next, we explored the location of Tph, we found that CD4+PD-1+CXCR5- Tph cells were located within immature TLSs, whereas in mature TLSs, the majority of CD4+ T cells were CD4+PD-1+CXCR5+ Tfh cells (Supplemental Figure S5d). Taken together, our results demonstrate that Tfh cells are essential components of mature TLSs, while Tfh/Th1 (Tph) cells are primarily localized in immature TLSs.

Figure 5.

Identification and validation of Tfh in mature TLS.

(a) UMAP plot of CD4 T cells and CD8 T cells. (b) UMAP plot of 7 CD4 T cell subclusters. (c) Dotplot showing the representative markers in 7 CD4 T cell subclusters. (d) Stacked bar graph showing the proportion of 7 CD4 T cell subclusters among CD4 T cells. The figures on the bars indicate the proportion of CD4 T cell subclusters. (e) Matrix plot showing the expression of ligands and receptors involved in TLS formation in CD4 T cell subclusters. (f) Stacked bar graph showing the proportion of clonal expansion category among CD4 T cell subclusters. The figures on the bars indicate the proportion of clonal expansion categories. (e) Representative image of Tfh in mature TLS by multispectral immunochemistry staining. Red arrows indicate BCL6⁺CD20⁺ Tfh. (g) Comparison of the density of BCL6⁺CD4⁺ cells between HPV (-) and HPV (+) HNSCC, non-MPR and MPR group. *p < 0.05; **p < 0.01; ***p < 0.001.

Identification and validation of resident memory CD8 T cell in mature TLS

To investigate the cellular lineage and function of CD8+ T cell subpopulations associated with mature TLSs, we clustered CD8+ T cells into six subpopulations: naïve CD8 T cells, central memory CD8 T cells, effecter CD8 T cells, HSP⁺ CD8 T cells, ISG15⁺ CD8 T cells, resident memory CD8 T cells using representative markers (Figure 6a,b). As shown in Figure 6c, resident memory CD8+ T cells were more prevalent in HPV- positive patients (from merged 8 scRNA-seq datasets), anti-PD-1 therapy responders (from GSE226620) and neoadjuvant chemoimmunotherapy responders (from our in-house scRNA-seq cohort). Additionally, we found that several ligands and receptors involved in TLS formation, including CXCL13, TNFSF4, CTLA4, and CXCR6, were highly expressed on resident memory CD8+ T cells, indicating their critical role in TLS development (Figure 6d). Single-cell clonotype analysis revealed that the proportion of highly expanded resident memory CD8+ T cells ( > 5 clones) was greater in anti-PD-1 responders than in non-responders, suggesting that the activation of resident memory CD8+ T cells may play a significant role in anti-PD-1 therapy-mediated antitumor immunity (Figure 6e). mIHC analysis confirmed that CD103+CD8+ resident memory T cells were located in mature TLSs (Figure 6f). These cells were enriched in HPV-positive patients (p = 0.002) and responders to neoadjuvant chemoimmunotherapy (p = 0.028) (Figure 6g). Notably, resident memory CD8+ T cells expressed high levels of CXCL13 (Figure 6b), a key chemokine in TLS formation, and we observed colocalization of CXCL13 with some resident memory CD8+ T cells (Figure 6f). Comparisons between areas (field-of-view) with TLSs and those without revealed significantly higher infiltration of resident memory CD8+ T cells in TLS-positive areas (p < 0.001) (Figure 6g). Moreover, we selected 2 representative areas and performed space proximity analysis. As shown in Supplemental Figure S6a, resident memory CD8+ T cells were significantly closer to B cells in TLS areas compared to non-TLS areas (p < 0.001). In line with these findings, in-silico receptor-ligand interaction analysis demonstrated that resident memory CD8+ T cells had the strongest interactions with germinal center B cells, particularly through the CXCL13-CXCR5 ligand-receptor pair (Supplemental Figure S6b). To confirm that resident memory CD8+ T cells express CXCL13 to recruit B cells at the protein level, we sorted CD103⁻ CD8⁺ T cells and CD103⁺ CD8⁺ T cells from tumor-infiltrating T cells (Figure 6h). ELISA showed the secretion of CXCL13 by CD103⁺ CD8⁺ T cells was significantly higher than CD103⁻ CD8⁺ T cells upon PMA stimulation (p = 0.026) (Figure 6i). As TGF-β was reported to induce CD103 expression on CD8 T cells, we evaluated the level of CXCL13 after TGF-β treatment. As shown in Supplemental 6C, TGF-β significantly increased the expression of CD103 and CXCL13 on CD8 T cells (p < 0.001). In conclusion, our results demonstrate that resident memory CD8+ T cells are an important component of mature TLSs and can secrete CXCL13, potentially playing a role in TLS formation.

Figure 6.

Identification and validation of resident memory CD8 T cell in mature TLS.

(a) UMAP plot of 6 CD8 T cell subclusters. (b) Dotplot showing the representative markers in 6 CD8 T cell subclusters. (c) Stacked bar graph showing the proportion of 6 CD8 T cell subclusters among CD8 T cells. The figures on the bars indicate the proportion of CD8 T cell subclusters. (d) Matrix plot showing the expression of ligands and receptors involved in TLS formation in CD8 T cell subclusters. (e) Stacked bar graph showing the proportion of clonal expansion category among CD8 T cell subclusters. The figures on the bars indicate the proportion of clonal expansion categories. (f) Representative image of resident memory CD8 T cells in mature TLS by multispectral immunochemistry staining. Red arrows indicate CD103⁺CXCL13⁺CD8⁺ resident memory CD8 T cells. (g) Comparison of the density of CD103⁺CD8⁺ cells between HPV (-) and HPV (+) HNSCC, non-MPR and MPR group. Comparison of the density of CD103⁺CD8⁺ cells between fields (field-of-view) with mature TLSs and without mature TLSs. (h) Sorting strategy of CD103⁻CD8⁺ T cells and CD103⁺CD8⁺ T cells from tumor infiltrating lymphocytes. (i) ELISA results showing the secretion of CXCL13 in indicated group. *p < 0.05; **p < 0.01; ***p < 0.001.

Mature TLS formation after anti-PD-1 based chemoimmunotherapy

To investigate the impact of anti-PD-1 therapy on TLS formation in HNSCC, we analyzed a scRNA-seq dataset (GSE200996) containing HNSCC samples collected before and after anti-PD-1 treatment (pre-treatment, n = 6; post-treatment, n = 19). Twenty-four cell subpopulations were identified using characteristic markers (Figure 7a,b). We found that germinal center B cells, Tfh and resident memory CD8 T cells were enriched in post-treatment samples. However, Tfh/Th1 was enriched in pre-treatment samples (Figure 7c). These results indicated that anti-PD-1 therapy may promote the maturation of TLSs. We examined the density of mature TLSs in 24 paired samples (7 pairs from the MPR group and 17 pairs from the non-MPR group) from the neoadjuvant chemoimmunotherapy cohort. While we observed an increase in the density of mature TLSs after treatment (p = 0.162), this difference was not statistically significant (Figure 7d). No significant change in TLS density was detected in the MPR group (p = 0.763), whereas a significant increase was observed in the non-MPR group (p = 0.049) (Figure 7d). As illustrated in Figure 7e, a representative biopsy sample from an HNSCC patient receiving neoadjuvant chemoimmunotherapy showed no mature TLS formation before treatment, whereas four mature TLSs were observed in the resected tumor after two cycles of treatment. However, given the limited sample size, this observation remains exploratory and requires further validation in a larger cohort. Additionally, the observed differences between pre- and post-treatment samples may be influenced by variations in sampling regions. While our findings suggest that anti-PD-1-based chemoimmunotherapy may promote TLS formation, additional studies with larger sample sizes and more standardized sampling approaches are needed to establish a definitive relationship.

Figure 7.

Mature TLS formation after anti-PD-1 based chemoimmunotherapy.

(a) UMAP plot showing 26 subclusters identified in GSE200996 dataset. (b) OR (odds ratio) analysis of subcluster distribution in pre-treatment samples and post-treatment samples. (c) Dotplot showing the representative markers in 26 subclusters. (d) Paired dotplot showing the density of mature TLSs in pre-treatment samples and post-treatment samples. (e) Representative image of mature TLSs in pre-treatment sample and post-treatment sample. Red boxes indicate 3 mature TLSs in post-treatment sample. *p < 0.05;**p < 0.01; ***p < 0.001.

Overall, we identified an immune subtype characterized by TLS in TCGA-HNSC. Patients in this subtype were mostly HPV-positive and showed a favorable response to immunotherapy. Mature TLSs were enriched in HPV-positive HNSCC tumors and anti-PD-1 based chemoimmunotherapy responders. In addition, germinal center B cells, follicular helper T cells and resident memory CD8 T cells were enriched in mature TLSs. FCRL4+ B cells and Tph were enriched in immature TLSs. Furthermore, neoadjuvant chemoimmunotherapy induced the formation of mature TLSs in HNSCC patients (Figure 8).

Figure 8.

Schematic diagram.

Unsupervised clustering based on five immune characteristics identified an immune subtype characterized by TLS in TCGA-HNSC. Patients in this subtype were mostly HPV-positive and showed a favorable response to immunotherapy. Mature TLSs were enriched in HPV-positive HNSCC tumors and anti-PD-1 based chemoimmunotherapy responders. In addition, germinal center B cells, follicular helper T cells and resident memory CD8 T cells were enriched in mature TLSs. FCRL4+ B cells and Tph were enriched in immature TLSs. Furthermore, neoadjuvant chemoimmunotherapy induced the formation of mature TLSs in HNSCC patients.

Discussion

HNSCC exhibits heterogeneity in molecular and immune features based on HPV status, tumor site, and therapy response. HPV-positive tumors tend to have an inflamed TME with higher T and B cell infiltration, while HPV-negative tumors often display immunosuppressive TMEs dominated by myeloid cells.³³ These differences underscore the need for effective immune biomarkers for patient stratification. Recent studies have highlighted TLSs as potential predictors of immunotherapy response. Our study provides comprehensive insights into TLS composition and their relevance in HNSCC immunity and therapy.

Through scRNA-seq and mIHC, we identified mature TLSs predominantly in HPV-positive tumors and in responders to anti-PD-1-based neoadjuvant chemoimmunotherapy. These TLSs were characterized by the presence of germinal center B cells, Tfh cells, and resident memory CD8+ T cells. Consistent with our findings, previous studies have also associated mature TLSs with improved prognosis in HPV-positive HNSCCs and other solid tumors.¹⁹ Regarding the formation of more mature TLSs in HPV-positive tumors, we speculate that the chronic presence of HPV antigens enhances adaptive immune activation,³⁴ facilitating TLS maturation. Supporting this, HPV E6/E7 vaccination has been shown to induce TLSs in cervical lesions.³⁵ This suggests therapeutic TLS induction may be a viable strategy not only for HPV-related tumors but also for immune-cold cancers via antigen vaccines or immune-modulatory approaches. In the scenario of anti-PD-1 based neoadjuvant chemoimmunotherapy, consistent with our findings, recent studies showed responders to chemoimmunotherapy showed higher level of TLSs in biopsy tissue prior to chemoimmunotherapy in advanced gastric cancer³⁶ and resectable non-small cell lung cancer.³⁷

Germinal center B cells and Tfh cells were found in mature TLSs. Consistent with our findings, Ruffin et al¹¹ showed that more germinal center tumor-infiltrating B cells localized in TLSs were found in HPV-positive HNSCCs. Lechner et al³⁸ reported that Tfh cells are key components of TLSs in HNSCC. The anticancer effect of TLSs in the TME is closely related to its cell composition. The germinal center, a hallmark of mature TLSs, is the core organizational structure. Many studies have shown that only mature TLSs with germinal centers predict a good prognosis.³⁹ Germinal center B cells⁴⁰ and Tfh cells⁴¹ located in the germinal center can also predict better outcomes. We speculate that germinal center B cells present antigens to CD4+ T cells and secrete tumor-specific antibodies to exert antitumor effects.

We also found resident memory CD8 T cells (CD103⁺ TRMs) in mature TLSs. Recent studies have shown that intratumoral TLSs harbor abundant CD103⁺ TRMs in spatial proximity to B cell zones, and this co-localization is associated with improved prognosis and immune activation in gastric cancer⁴² and breast cancer.⁴³ The cell phenotype of CD103⁺CD8⁺ T cells and the high expression of exhaustion-related indicators such as PDCD1, CTLA4, TIGIT and LAG3, are consistent with a type of PD-1^T T cells in non-small cell lung cancer.⁴⁴ This suggests that CD103⁺CD8⁺T cells are targets of ICB and may help restore antitumor T cell response. Edwards et al⁴⁵ showed that in immunotherapy-naïve melanoma patients, CD103⁺ tumor-resident CD8⁺T cells were enriched in anti-PD-1 responders and expanded during treatment. A recent study found HPV-specific PD-1⁺ stem-like CD8⁺T cells capable of differentiating into effector-like cells in HPV-positive HNSCCs.⁴⁶ These PD-1⁺CD8⁺T cells resembled resident memory CD8 T cells in our study. In mice, this subset provides the proliferative burst of effector-like T cells after PD-1 blockade.⁴⁷

Importantly, spatial analysis showed that CD103⁺ TRMs localized near germinal center B cells within TLSs. While it remains unclear whether GC B cells directly present antigens to TRMs, their physical proximity suggests the possibility of crosstalk. A recent study demonstrated that B cells can promote the activation and cytotoxic function of CXCL13⁺ CD103⁺ CD8⁺ TRMs through the LTα–TNFR2 axis, enhancing CXCL13 and granzyme B expression independent of classical antigen presentation.⁴⁸ These findings point to a non-cognate, cytokine-mediated mechanism through which B cells may influence TRM behavior within TLSs.

Furthermore, resident memory CD8 T cells expressed CXCL13 and expressed CXCL13 and interacted closely with B cells. We speculated that CXCL13⁺CD103⁺CD8⁺T cells could secrete CXCL13 and elicit the genesis of TLSs. Similar to our results, in ovarian cancer⁴⁹ a group of CXCL13⁺CD103⁺CD8⁺T cells involved in B cell recruitment was also found. CXCL13 is a key cytokine that recruits B cells to form TLSs. These findings suggest that resident memory CD8 T cells may enhance the efficacy of anti-PD-1 therapy not only by responding to PD-1 blockade but also by secreting CXCL13 and promoting TLS formation.

As for why TRMs – typically found in TGF-β–rich niches – reside within TLSs, recent studies suggest that TLSs themselves may represent TGF-β–enriched microenvironments. TGF-β is essential for CD103 induction and TRM differentiation by promoting integrin expression and local retention.⁵⁰ Simultaneously, TGF-β has been shown to promote Tfh cell differentiation via SATB1 silencing, thereby supporting both Tfh and TRM residency within TLSs.⁵¹ This shared reliance on TGF-β signaling may underlie the coordinated localization of TRMs and GC B cells within TLSs and contribute to their cooperative roles in sustaining anti-tumor immune responses.

In addition, we found FCRL4+ B cells and Tph within immature TLSs. FCRL4+ B cells and Tph cells were initially found in autoimmune diseases and their roles in tumor recently been investigated. A recent study showed that FCRL4+ B cells represent an extrafollicular response path, distinct from germinal center reaction.⁷ Extrafollicular response generating early, lower affinity effector antibody responses and the germinal center response generating delayed but higher affinity and longer-lasting antibody responses.⁵²

Interestingly, we found that FCRL4⁺ B cells were more abundant in HPV-positive tumors. Although the classical model proposed by Goodnow and colleagues⁵³ suggests that self-reactive B cells are excluded from germinal centers to prevent autoimmunity, our findings suggest a more nuanced explanation. While there is no direct evidence linking HPV infection to FCRL4⁺ B cell generation, studies of chronic viral infections such as HIV⁵⁴ and HCV⁵⁵ have shown that persistent viral antigen exposure promotes the expansion of FCRL4⁺ B cells, possibly through extrafollicular activation. Furthermore, viral infections can induce autoimmunity through molecular mimicry, antigen spreading, and tissue damage, potentially leading to the emergence of self-antigen – like targets that trigger tolerance mechanisms.⁵⁶ Therefore, the enrichment of FCRL4⁺ B cells in HPV-positive tumors may reflect either chronic viral stimulation or a viral infection – induced self-tolerance response. Since our current dataset does not provide antigen specificity, future single cell B cell receptor (BCR) sequencing will be necessary to clarify whether these B cells recognize viral or self-antigens.

In terms of Tph, Rao et al.⁵⁷ reported more Tph than Tfh cells were found adjacent to B cells in areas outside of lymphoid aggregates. These findings indicate that FCRL4+ B cells and Tph cells may play important roles in the early phases of TLS formation by functioning outside the lymphoid follicle. Additionally, similar to the role of TGF-β in inducing CD103+ CD8+ T cells, previous studies have reported TGF-β can induce FCRL4+ B cells⁵⁸ and Tph.⁵⁹ These findings suggest that TGF-β may play a crucial role in inducing immature TLSs, although the key regulators that promote TLS maturation remain to be identified.

Furthermore, we found increased mature TLSs formed after anti-PD-1 based chemoimmunotherapy, consistent with previous studies in bladder cancer,⁶⁰ esophageal squamous cell carcinoma⁶¹ and in advanced gastric cancer.³⁶ In our study, we did not observe a significant difference in TLS density between pre- and post-treatment in the MPR group, possibly due to the small sample size (8 paired samples) or the elimination of tumor antigens by neoadjuvant chemoimmunotherapy, which may reduce sustained antigen stimulation and lead to TLS degeneration.⁶² Interestingly, we found increased TLS density in the non-MPR group after neoadjuvant chemoimmunotherapy. This suggests that neoadjuvant chemoimmunotherapy may release tumor antigens and elicit an antitumor immune response in this group. Additionally, administering more cycles of neoadjuvant chemoimmunotherapy may be a potential option for certain HNSCC patients, as they can form mature TLSs after treatment. Identifying more precise biomarkers will be crucial for stratifying these patients.

While our study demonstrates that neoadjuvant anti-PD-1-based chemoimmunotherapy is associated with increased TLS maturation, the underlying mechanisms merit further investigation. Immune checkpoint blockade may alleviate T cell exhaustion and enhance the activity of Tfh cells,^63–65 which are pivotal for germinal center formation and TLS organization. Strengthened interactions between Tfh cells and germinal center B cells may further contribute to TLS maturation.⁶⁶ Recent study also indicates that anti-PD-1 therapy promotes the activation of conventional type 1 dendritic cells (cDC1s) within the tumor microenvironment.⁶⁷ These activated cDC1s play a central role in the formation, maintenance, and germinal center development of TLSs by facilitating chemokine production, antigen presentation, and coordination of Tfh – B cell interactions.⁶⁸ Additionally, immune checkpoint blockade has been shown to induce the formation of high endothelial venules (HEVs), which serve as structural platforms for TLS assembly and enable efficient lymphocyte recruitment into tumors.⁶⁹ In parallel, chemotherapy-induced tumor cell death enhances antigen release and presentation, potentially amplifying local immune activation and triggering de novo TLS formation in tumors previously lacking these structures.^70,71 Together, these findings suggest that the synergistic effects of anti-PD-1 therapy and chemotherapy may not only promote tumor regression but also remodel the tumor immune microenvironment to favor the development and maturation of TLSs.

This study provides important insights into the role of TLSs in HNSCC: Firstly, we provide a detailed analysis of B cell and T cell populations within TLSs, identifying key players such as germinal center B cells, Tfh cells, and resident memory CD8+ T cells, which contribute to TLS-mediated anti-tumor immunity. Secondly, we show that mature TLSs are enriched in HPV-positive tumors and responders to anti-PD-1-based neoadjuvant chemoimmunotherapy, supporting their potential as predictive biomarkers. Thirdly, we propose that resident memory CD8+ T cells may play a role in TLS formation through CXCL13 secretion, offering potential new targets for TLS-inducing therapies. The observation of FCRL4+ B cells and Tph cells in immature TLSs suggests their involvement in the early stages of TLS formation. Lastly, our study supports therapeutically inducing TLS formation in immune-cold tumors may offer new avenues for improving responses to checkpoint blockade and personalized immunotherapy strategies. These findings provide new perspectives on the role of TLSs in HNSCC tumor immunity and highlight their potential utility as biomarkers and therapeutic targets.

Funding Statement

This work was supported by the Joint Funds of the National Natural Science Foundation of China [U21A20381], the General Funds of the National Natural Science Foundation of China [82373452], the Guangdong Natural Science Funds for Distinguished Young Scholar [2022B1515020061], the Guangdong Basic and Applied Basic Research Foundation [2021A1515220138], the Guangzhou Basic Research Program Jointly Funded by Municipal Schools (Institutes) [202201020367], the Fundamental Research Funds for the Central Universities, Sun Yat-sen University [16ykpy10], he Fundamental Research Funds for the Central Universities, Sun Yat-sen University [19ykzd20].

Disclosure statement

No potential conflict of interest was reported by the author(s).

Ethics approval and consent to participate

This study was approved by the Ethical Committee of Sun Yat-Sen Memorial Hospital (SYSEC-KY-KS-2021–109), and the informed written consent of all participants was obtained for research with the collection of tissue and blood samples. This study was conducted in accordance with the ethical principles of the Declaration of Helsinki.

Data availability statement

Published scRNA-seq datasets and the spatial transcriptomic dataset used in this study are publicly available. Specifically, the TCGA Pan-Cancer expression and clinical data are accessible via the Xena database (Batch effects normalized mRNA data, Pan-Cancer Atlas Hub) at https://xenabrowser.net/. The in-house scRNA-seq dataset generated and analyzed during the current study, along with other supporting data, are available from the corresponding author upon reasonable request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/2162402X.2025.2528109.

Abbreviations

TLS –

Tertiary Lymphoid Structure

ICB –

Immune Checkpoint Blockade

HNC –

head and neck cancer

HNSCC –

Head and Neck Squamous Cell Carcinoma

mIHC –

Multispectral Immunohistochemistry

ELISA –

Enzyme-Linked Immunosorbent Assay

TCGA –

The Cancer Genome Atlas

HPV –

Human Papillomavirus

DFS –

Disease-Free Survival

OS –

Overall Survival

TIDE –

Tumor Immune Dysfunction and Exclusion

TMB –

Tumor Mutation Burden

scRNA-seq –

Single-Cell RNA Sequencing

NK –

Natural Killer Cell

cDC –

Conventional Dendritic Cell

pDC –

Plasmacytoid Dendritic Cell

FDC –

Follicular Dendritic Cell

MPR –

Major Pathological Response

pCR –

Pathological Complete Response

pNR –

Pathological No Response

pPR –

Pathological Partial Response

Tfh –

Follicular Helper T Cell

Tph –

Peripheral Helper T Cell

TME –

Tumor Microenvironment