Integrated peripheral blood multi-omics profiling identifies immune signatures predictive of neoadjuvant PD-1 blockade efficacy in head and neck squamous cell carcinoma

Background

Neoadjuvant PD-1 inhibitor therapy has shown promise in locally advanced head and neck squamous cell carcinoma (HNSCC), but only a subset of patients achieves major pathological responses. Liquid biopsy, the analysis of tumor-derived biomarkers in readily accessible bodily fluids (primarily blood), offers significant advantages over traditional tissue biopsies for predicting cancer treatment outcomes. The aim of this study is to develop a predictive model for neoadjuvant PD-1 therapy response in HNSCC patients using exclusively liquid biopsy approaches-namely, peripheral blood immune profiling (CyTOF) and plasma cytokine panels (Olink).

Methods

In a prospective trial involving 50 HNSCC patients treated with neoadjuvant tislelizumab plus chemotherapy, peripheral blood samples were collected pre- and post-treatment. Immune cell subsets were analyzed by mass cytometry (CyTOF), and circulating protein markers were quantified via a 92-plex targeted proteomics panel (Olink). Multimodal features were integrated into a predictive model using logistic regression.

Results

Baseline immune profiles differed significantly between responder (RD) and non-responder (NRD): RD showed higher frequencies of CD103⁻CD8⁺ central memory T cells (Tcm, c03) and elevated plasma interleukins (IL-5, IL-13), whereas NRD had more CD28⁻TIGIT^highcPARP⁻CD8⁺ terminally differentiated effector memory CD45RA⁺ T cells (Temra, c17) and higher levels of chemokines (CCL3, CCL4) and MMP7. Neoadjuvant therapy reactivated both subsets, evidenced by downregulation of PD-1 and increased expression of activation markers (e.g., CD38) and cytotoxic mediators (e.g., granzyme B and interferon γ). A multimodal predictive model incorporating CD8⁺T cell subsets (c03, c17) and plasma biomarkers (IL-5, MMP7) demonstrated superior predictive accuracy (AUC = 0.9219).

Conclusions

Integrated peripheral immune profiling enables robust, noninvasive prediction of neoadjuvant PD-1 blockade efficacy in HNSCC. The identified immune cell subsets and plasma biomarkers provide a clinically applicable framework for early response stratification and personalized immunotherapy, supporting liquid biopsy as a viable platform for clinical decision-making.

Trial registration Chinese Clinical Trial Registry, clinical trial number CHiCTR2200056354, 04 February 2022, https://www.chictr.org.cn/showproj.html?proj=151364.

Graphical Abstract

Head and neck squamous cell carcinoma (HNSCC) is the seventh most common malignancy worldwide, with nearly 900,000 new cases and 450000 deaths reported annually [1, 2]. Over 60% of patients are diagnosed at an advanced local stage (III-IV) in China, with a five-year survival rate of less than 50% [3, 4]. While multimodal therapies combining surgery and chemoradiation improve locoregional control, postoperative recurrence and distant metastasis remain predominant drivers of therapeutic failure [5,6,7]. Emerging as a paradigm-shifting strategy, neoadjuvant programmed death-1 (PD-1) inhibitors have demonstrated transformative potential in locally advanced HNSCC by reactivating antitumor immunity and inducing complete or major pathological responses [8,9,10]. However, clinical studies have shown that only 13~20% of patients derive significant benefit from PD-1 blockade monotherapy, and approximately 35% benefit from PD-1 blockade combined with chemotherapy. The absence of reliable predictive biomarkers contributes to significant uncertainty in treatment selection [8, 11,12,13,14]. This interpatient heterogeneity not only exacerbates healthcare resource allocation inefficiencies but also exposes non-responders to immune-related adverse events (irAEs) [15, 16]. Therefore, there is an urgent need to identify potential biomarkers for accurately selecting patients who will benefit from neoadjuvant therapy, thus avoiding ineffective treatments and their associated side effects and economic burden [17].

Current gold-standard biomarkers, including programmed death-ligand 1 (PD-L1) combined positive score (CPS) and tumor mutational burden (TMB), exhibit limited predictive utility due to tumor spatiotemporal heterogeneity and the non-reproducibility of invasive biopsies [18, 19]. Furthermore, the complex immunosuppressive network within the tumor microenvironment (TME) may dynamically influence therapeutic responses via the peripheral immune system, while clinical laboratory science (CLS) professionals play a pivotal role in advancing detection technologies (including liquid biopsy) to enhance diagnostic efficiency[20]. Traditional tissue biopsies are unable to capture this systemic regulatory process comprehensively [21]. In this context, peripheral blood, as one of the most commonly utilized biological samples for liquid biopsy, possesses advantages such as sustainability, ease of repeated collection, minimal invasiveness, and the ability to dynamically reflect comprehensive immune profiling, tumor heterogeneity, and disease progression. It is increasingly recognized as a promising approach for identifying predictive biomarkers [1, 22]. Among the key research areas are circulating immune cell subsets and soluble immune factors [18].

Mass cytometry (CyTOF), utilizing metal isotope-labeled antibodies, enables simultaneous detection of over 40 parameters at the single-cell level, providing high-resolution insights into the phenotypic heterogeneity and functional status of peripheral blood immune cell subsets [23, 24]. Using CyTOF, Rochigneux’s group finds that pre-therapy ICOS⁺CD4⁺T cell frequencies predict pembrolizumab response in non-small cell lung cancer (NSCLC) [25]. Olink^® targeted proteomics employs proximity extension assay (PEA) technology to quantify 92 immunomodulatory proteins (e.g., chemokines, cytokines) in plasma with a sensitivity that exceeds conventional ELISA by over 1000-fold [26, 27]. Leveraging such multiplexed protein assays, several studies have established circulating protein signatures associated with immune treatment efficacy [28]. While integrating CyTOF and Olink overcomes single-omics limitations by synergistically combining high-dimensional cellular phenotyping with ultrasensitive proteome profiling. And parallel analysis of isogenic samples maximizes biospecimen utility and eliminates technical variability. Machine learning-driven multidimensional integration identifies cross-modal biomarkers and elucidates resistance mechanisms. This systems biology framework correlates immune cell dynamics with soluble mediators, transcending predictive boundaries of single-omics to guide immunotherapy response prediction and microenvironment modulation in precision oncology.

Therefore, we conducted a prospective multi-omics study in a neoadjuvant PD-1 inhibitor-treated HNSCC cohort. By integrating CyTOF-based high-dimensional immunophenotyping with Olink plasma proteomic profiling of pre- and post-treatment samples, we identified predictive cellular and plasma biomarkers closely associated with neoadjuvant immunotherapy response. The key discoveries of this study include the identification of two peripheral immune cell clusters, CD103⁻CD8⁺ central memory T cells (Tcm, c03) and CD28⁻TIGIT^highcPARP⁻CD8⁺ terminal effector memory CD45RA⁺ T cells (Temra, c17), as predictive biomarkers, as well as several associated circulating factors (such as interleukin-5 [IL-5], matrix metalloproteinase-7 [MMP7]). The multimodal prediction model combining cellular and molecular markers, outperforms single indicators and significantly improves the ability to discriminate neoadjuvant immunotherapy outcomes, providing a powerful basis for personalized immunotherapy decision-making in HNSCC.

Study design and patient samples

The clinical trial (clinical trial number CHiCTR2200056354) was designed to evaluate efficacy of neoadjuvant therapy combining tislelizumab with chemotherapy in patients diagnosed with primary head and neck squamous cell carcinoma (HNSCC). The study was approved by the Ethics Committee of Peking University School and Hospital of Stomatology (No. PKUSSIRB-202170179 and No. PKUSSIRB-202276072). Totally, fifty patients diagnosed with primary HNSCC who received neoadjuvant therapy at Peking University School and Hospital of Stomatology between January 2022 and December 2023 were enrolled. Inclusion criteria comprised: (a) age ≥ 18 years; (b) pathologically confirmed stage III-IV HNSCC according to the 8th edition of the American Joint Committee on Cancer (AJCC); (c) locally resectable tumors as assessed by experienced oral and maxillofacial surgeons; (d) no previous anticancer therapy; and (e) an Eastern Cooperative Oncology Group (ECOG) performance status of 0~1. Exclusion criteria included: (a) prior treatment with immune checkpoint inhibitors such as anti-PD-1/PD-L1 or anti-CTLA-4 drugs; (b) having a history of severe allergies or autoimmune diseases requiring systemic treatment (such as autoimmune hepatitis); (c) presence of severe cardiovascular diseases, such as uncontrolled arrhythmias or myocardial infarction within the past six months; (d) presence of major organ failure or infectious diseases; (e) diagnosis of other malignancies within the past five years; (f) having a history of substance abuse of psychotropic drugs and being unable to quit or having mental disorders. Written informed consent was obtained from all participants.

Patients underwent two cycles of neoadjuvant therapy, consisting of tislelizumab (200 mg) combined with cisplatin (75 mg) and albumin-bound paclitaxel (260 mg/m²), administered every three weeks. Imaging evaluation (CT or MRI) was performed 2~3 weeks post-therapy to guide surgical resection with margins of approximately 1.0~1.5 cm around residual tumors. Based on postoperative pathological results, patients were classified into in responders (RD) and non-responders (NRD). The RD group was defined as patients whose postoperative histopathological examination revealed no or minimal residual tumor cells (generally ≤ 10%), with primary outcomes assessed as pathological complete response (pCR) and major pathological response (MPR). The NRD group consisted of patients whose tumor tissues exhibited no significant necrosis or reduction, with the proportion of viable tumor cells showing no marked decrease or even progression. Peripheral blood samples were collected from all patients both pre- and post-therapy for subsequent analyses.

Plasma sample collection

The collected 4 mL venous blood samples were centrifuged at 1700 rpm for 30 min at room temperature (25 °C) to separate plasma. The obtained plasma samples were stored at − 80 °C and thawed subsequently for proteomic analysis.

Separation of PBMC samples

PBMCs were isolated via red blood cell (RBC) lysis. Briefly, a 10X RBC lysis buffer (Biolegend, CA, USA) was diluted to a 1X working solution with deionized water, then mixed with blood samples at a ratio of 1:10 and incubated with gentle rotation at room temperature for 10~15 min in the dark. Subsequently, PBMCs were washed twice with phosphate-buffered saline (PBS) containing with 1% bovine serum albumin (BSA; Biosharp, Anhui, China). The PBMCs were resuspended in Gibco fetal bovine serum (FBS, Thermo Fisher Scientific, MA, USA) containing 10% dimethyl sulfoxide (DMSO) and cryopreserved in liquid nitrogen for long-term storage.

Olink protein quantification

A total of 42 patients (RD group, n = 27; NRD group, n = 15) were randomly selected from all enrolled subjects for targeted plasma proteomics analysis of plasma samples pre- and post-neoadjuvant therapy.

Plasma proteins were measured using the Olink^® Target 96 Immuno-Oncology assay (Olink Proteomics AB, Uppsala, Sweden) according to the manufacturer's instructions. The proximity extension assay (PEA) technology, as previously described by Assarsson’s group [27], enabled simultaneous measurement of 92 proteins from 1 μL of plasma. Briefly, oligonucleotide-labeled antibody probes bound to target proteins, and proximal hybridization of paired probes initiated a DNA polymerase-mediated extension reaction to generate unique PCR target sequences. Subsequent nucleic acid amplification and quantitative analysis were performed using the Signature Q100 microfluidic real-time PCR platform (LC-Bio Technology Co., Ltd., Hangzhou, China).

Internal amplification controls and inter-plate controls were included for quality control and normalization to minimize intra- and inter-assay variability. Results were expressed as normalized protein expression (NPX) values, log2-transformed arbitrary units positively correlated with protein abundance. All validation data for the assays, including detection limits and intra- and inter-batch precision, can be obtained from the manufacturer's website (www.olink.com). A list of all the proteins included in the Oncology II assay is also available (Supplementary Table 1).

CyTOF analysis of PBMC samples

A total of 32 subjects (RD group, n = 18; NRD group, n = 14) were randomly selected from all enrolled patients for CyTOF analysis of PBMC samples pre- and post-neoadjuvant therapy. Among them, 24 patients also underwent paired plasma proteomic profiling.

Sample Preparation: Resting PBMCs were filtered through a 40 μm cell strainer to achieve single-cell suspensions. For each sample, 2.5 × 10⁶ cells were incubated with 0.5 μM palladium (Pd) in PBS (5 min, RT) for live and dead discrimination. Cells were then fixed with 1.6% paraformaldehyde (10 min, RT), followed by 30-min incubation at room temperature with CD45 antibodies conjugated to platinum isotopes (194Pt⁻, 195Pt⁻, 196Pt⁻, and 198Pt⁻) for cell barcoding (Supplementary Table 2). A total of 6 × 10⁵ barcoded cells per sample were pooled and equally divided for surface and intracellular antibody staining.

Surface Antibody Staining: Fc receptors were blocked using Human TruStain FcX (BioLegend, USA) to reduce non-specific binding. The Maxpar^® X8 metal-conjugated antibody (Standard BioTools, USA) cocktail was prepared in the tube at the concentration derived from a previously established titration. Cell suspensions were incubated with 100 μL surface antibody cocktails (Supplementary Table 3) with shaking (30 min, RT). The surface antibody panel verification was shown in Supplementary Fig. 1A.

Intracellular Antibody Staining: Secondary fixation and permeabilization was achieved using nuclear antigen staining buffer (Standard BioTools, USA) for 30 min at RT. Intracellular antibody cocktail (100 μL; Supplementary Table 4) were applied under identical conditions. The intracellular antibody panel verification was shown in Supplementary Fig. 1B.

Cell ID Staining: Nucleic acids were labeled with ¹⁰³Rh intercalator (Standard BioTools, USA) via overnight incubation at 4 °C.

Data Acquisition and Analysis: Stained cells were resuspended in deionized water containing 10% EQ four element calibration beads (Standard BioTools, USA) and adjusted to 5 × 10⁵ ~ 1 × 10⁶ cells/mL. Data acquisition was performed using the Helios 6.7 CyTOF system (Standard BioTools, USA). High-dimensional data analyses were conducted using FlowJo v10.8 software (FlowJo, LLC, OR, USA) in combination with the Cytobank cloud-based platform (Beckman Coulter Life Sciences, CA, USA) and OMIQ cloud-based platform (OMIQ, Inc., MA, USA).

Statistical analysis

Differentially expressed proteins (DEPs) were identified using the “Olink® Analyze” R package with a significance threshold of p < 0.05. All other statistical analyses were performed using GraphPad Prism software (version 6.02, GraphPad Software, CA, USA). Data were presented as mean ± standard error of the mean (SEM) or mean ± standard deviation (SD), with error bars derived from a minimum of three independent biological replicates. All samples were processed under consistent platform and batch conditions, with no significant batch effects observed; thus, correction was not required. Collinearity assessment, feature selection, and cross-validation were performed in R using appropriate statistical packages. Pearson correlation coefficients were calculated to assess the association between groups. Statistical significance was determined using t-tests. The diagnostic efficacy of the biomarkers was evaluated by constructing receiver operating characteristic (ROC) curves using a dedicated ROC analysis package. A p-value ≤ 0.05 was considered statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Clinical characteristics of the study cohort

The study cohort comprised 50 patients with HNSCC was stratified into RD (n = 29) and NRD (n = 21) group based on therapeutic response. The mean age of the entire cohort was 57 years (from 34 to 77), with RD exhibiting a marginally higher age distribution compared to NRD (59 vs. 55). Male predominance was observed (92%), with no significant differences in gender distribution. High prevalence rates of smoking (76%) and alcohol consumption (78%) were identified, both being numerically elevated in RD (79% vs. 71% for smoking; 83% vs. 71% for alcohol).

Anatomical analysis demonstrated the oropharynx as the most frequent primary site (34%), with significantly higher oropharyngeal involvement in RD (38% vs. 29%), whereas NRD exhibited increased proportions of tongue (29% vs. 24%) and floor-of-mouth (29% vs. 17%) lesions. Tumor staging revealed T4 classification in 74% of cases. Nodal staging (N0/N + : 52%/48%) demonstrated balanced distribution across groups. Comprehensive clinical characteristics are summarized in Table 1.

Comparative analysis of peripheral immune landscapes in HNSCC patients with differential neoadjuvant therapy responses

A comprehensive immunophenotypic analysis was conducted using CyTOF on 64 PBMC samples obtained from 32 patients with HNSCC pre- and post-neoadjuvant therapy. Two customized 39-marker panels were designed to delineate cellular composition and functional states in HNSCC, comprising 26 shared markers and 13 panel-specific markers: Panel 1 focused on surface proteins, while Panel 2 targeted cytokine secretory profiles (Supplementary Table 3 and 4).

High-dimensional data visualization via t-distributed stochastic neighbor embedding (t-SNE) resolved 21 distinct immune clusters (Fig. 1A, Supplementary Figs. 2, 3, 4 and 5A, Supplementary Table 5), which were categorized into seven major subsets: B cells, CD4⁺T cells (subdivided into naïve, central memory [Tcm], effector memory [Tem], terminal effector memory CD45RA⁺ [Temra], and regulatory T cells [Treg]), CD8⁺T cells (including tissue-resident memory [Trm], exhausted [Tex], naïve, Tcm, Tem, and Temra subsets), gamma delta T cells (γδ T), natural killer cells (NK), granulocytes, dendritic cells (myeloid dendritic cell [mDC] and plasmacytoid dendritic cell [pDC] subsets), and monocytes (classical [cMo], intermediate [intMo], and non-classical [ncMo] subtypes).

Immunophenotypic Landscape of PBMCs in HNSCC. A t-SNE plot illustrating manual cluster annotation of major immune cell subsets in PBMCs. B Comparative frequencies of major PBMC immune subsets in RD (n = 18) and NRD (n = 14) groups pre- and post-neoadjuvant therapy. Data are presented as mean ± SEM. C Dual-axis density plots depicting immune subset distributions across PBMC populations; right panel highlights temporal dynamics of CD8⁺T cell density distributions in RD and NRD groups following neoadjuvant intervention. D Changes of CD38 expression (mean signal intensity) within PBMC immune subsets following neoadjuvant therapy in RD and NRD groups. Statistical significance was determined using paired t-tests B and D. *p < 0.05, **p < 0.01. PBMC peripheral blood mononuclear cells, HNSCC head and neck squamous cell carcinoma, RD responders, NRD non-responders, SEM standard error of the mean

Full size image

Quantitative comparison of major subset frequencies between RD and NRD groups revealed no statistically significant differences except for a post-therapy reduction in B-cell proportions within the RD cohort (Fig. 1B). Further stratification of immune lineage distributions demonstrated heterogeneity in cluster frequencies in RD and NRD groups pre- and post-neoadjuvant therapy. t-SNE density plots highlighted pretreatment compositional disparities between groups, with dynamic post-treatment redistribution observed in RD versus NRD samples, particularly within CD8⁺T cell populations (Fig. 1C, Supplementary Fig. 5B). Marker expression profiling identified significantly elevated CD38 levels across major immune subsets in both groups post-treatment (Fig. 1D, Supplementary Fig. 5C). These findings collectively indicate that, although the frequency changes of major peripheral immune cell subsets post-neoadjuvant therapy were not significant, there were notable alterations in the frequency and the expression of activation marker of their clusters, particularly within the CD8⁺T cell compartment.

RD and NRD groups exhibited similar alterations of circulating immune cells abundance and plasma proteins expression

Subsequent investigations focused on CD8⁺T cell alterations to comprehensively understand the immunological effects of neoadjuvant therapy in the peripheral circulation of HNSCC patients. Using the unsupervised clustering algorithm phenograph (UMAP) analysis on CD8⁺T cells, 20 immune cell clusters were identified and annotated based on the standardized average expression heatmap of surface or intracellular markers (Fig. 2A and B, Supplementary Fig. 6A, Supplementary Table 6). These clusters included Trm, Tex, naïve T, Tcm, Temra, and Tem cells. Tex subpopulations were further stratified by TCF-1 and CD69 expression into three exhaustion states: precursor exhausted (Tpex, PD-1⁺TCF-1⁺), intermediate transitional (Tex-int, PD-1⁺TCF-1⁻CD69⁻), and terminal exhausted (Tex-term, PD-1⁺TCF-1⁻CD69⁺) subsets. These CD8⁺T cell clusters exhibited considerable heterogeneity between different samples (Supplementary Fig. 6B). Paired pre- and post-treatment analysis of CD8⁺T cell clusters in RD and NRD groups revealed almost identical changing trends, with clusters 03, 05, 06, 15, and 19 (c03, c05, c06, c15 and c19) demonstrating significant alterations (Fig. 2C and D). The results indicated that the frequencies of CD8⁺Tcm (c03) and CD8⁺Tem (c06 and c19) were significantly elevated post-neoadjuvant therapy compared to pre-therapy, while Tpex (c05) and Tex-term (c15 and c12) frequencies decreased, concomitant with Tex-int (c08). RD group exhibited more pronounced cluster frequency changes compared to NRD expansion (Fig. 2E, Supplementary Fig. 6C).

Differential analysis of peripheral blood CD8⁺T cell clusters pre- versus post-neoadjuvant therapy. A Heatmap illustrating clustering and annotation of CD8⁺T cell clusters based on the mean expression levels of each marker. B UMAP plot depicting annotated CD8⁺T cell clusters. C, D Volcano plots demonstrating differential abundance analysis of CD8⁺T cell clusters pre- versus post-neoadjuvant therapy in RD and NRD groups. Red indicates clusters significantly increased after therapy, blue represents clusters significantly decreased after therapy, and orange denotes clusters with minor changes post-therapy. The vertical dashed lines at log2(FC) = − 0.58 and 0.58 correspond to fold change thresholds of 1.5-fold downregulation and upregulation, respectively. The horizontal dashed line at -log10(p-value) = 1.3 indicates the significance threshold (p = 0.05). E Frequency changes of cluster 03, 05, 06, 15, and 19 pre- versus post-neoadjuvant therapy in RD and NRD patient groups. Statistical analysis was performed using paired t-tests C–E. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. UMAP uniform manifold approximation and projection, RD responders, NRD non-responders, FC fold change

Full size image

Plasma proteomic profiling identified therapy-induced modulation of tumor-immunomodulatory proteins. Volcano plot analysis demonstrated significant post- therapy increases in IL-8, C–C motif chemokine 3 (CCL3), CCL4, CCL20, tumor necrosis factor (TNF), and monocyte chemoattractant protein 3 (MCP-3), with concurrent ARG1 downregulation. The changes in plasma protein expression were highly consistent between the RD and NRD groups, although the RD group exhibited more accentuated alteration magnitudes (Fig. 3A). Additionally, GO and KEGG enrichment analyses of post-treatment plasma proteins also yielded similar results. GO analysis revealed that, compared to pre-therapy, the differentially expressed proteins (DEPs) in both RD and NRD groups post-neoadjuvant therapy were predominantly enriched in the inflammatory response pathway (biological process), extracellular space and extracellular region (cellular component), and protein binding (molecular function) (Fig. 3B, Supplementary Fig. 7A). KEGG analysis indicated that the functions of the DEPs were associated with the following pathways: necroptosis (cellular processes), cytokine-cytokine receptor interaction (environmental information processing), pathways in cancer (human diseases), and IL-17 signaling pathway (organ systems), with the cytokine-cytokine receptor interaction pathway showing the most significant enrichment (Fig. 3C, Supplementary Fig. 7B).

Differential expression and enrichment analysis of plasma immunooncological proteins pre- and post-neoadjuvant therapy. A Volcano plot analysis of differentially expressed immunooncological proteins pre- versus post-neoadjuvant therapy phases in RD and NRD groups. Red dots indicate significantly upregulated proteins; blue dots indicate significantly downregulated proteins; gray dots represent proteins without significant differential expression. The vertical dashed lines at log2(FC) = − 0.58 and 0.58 correspond to fold change thresholds of 1.5-fold downregulation and upregulation, respectively. The horizontal dashed line at -log10(p-value) = 1.3 indicates the significance threshold (p = 0.05). Statistical significance was determined by paired t-tests. B GO enrichment analysis of DEPs pre- versus post-neoadjuvant therapy in RD and NRD groups, categorized by biological processes, cellular components, and molecular functions. The x-axis represents enriched functional categories, while the y-axis shows the number of proteins enriched in the corresponding pathways. C KEGG enrichment analysis of DEPs identified pre- versus post-neoadjuvant therapy in RD and NRD groups. The vertical axis denotes enriched KEGG pathways, and the horizontal axis represents the ratio of the number of DEPs enriched in the current pathway to the total identified DEPs. The color gradient of bubbles indicates the significance levels of enrichment across different pathways. RD responders, NRD non-responders, FC fold change, GO gene ontology, DEPs differentially expressed proteins, KEGG kyoto encyclopedia of genes and genomes

Full size image

Collectively, neoadjuvant therapy predominantly modulated peripheral CD8⁺T cell subset dynamics (Tex, Tcm and Tem) and plasma immunoproteome profiles (IL-8 or CCL chemokines) in HNSCC. While RD and NRD groups exhibited comparable immunomodulatory trends, the former demonstrated quantitatively enhanced responses, suggesting differential response thresholds rather than distinct mechanistic pathways.

Multimodal data integration identifies determinants of response to neoadjuvant therapy

To investigate the determinants of response to neoadjuvant therapy, a multimodal integrative analysis of pre-therapy PBMC and plasma proteomic data was performed. Plasma protein expression was first evaluated using the Olink platform, with hierarchical clustering heatmaps demonstrating distinct differences in protein expression profiles between RD and NRD groups (Fig. 4A). Volcano plot analysis identified five tumor-immunomodulatory proteins exhibiting significant differential expression between the two groups (Fig. 4B). Pre-therapy plasma levels of IL-5 and IL-13 were significantly upregulated in RD, whereas CCL3, CCL4, and MMP7 were downregulated (Fig. 4C).

Differential analysis of plasma proteins and peripheral blood CD8⁺T cell clusters in RD versus NRD groups pre-neoadjuvant therapy. A Heatmap depicting the expression of immunooncological plasma proteins in RD and NRD groups. B Volcano plot illustrating differential expression of 92 immunooncological proteins in RD versus NRD pre-neoadjuvant therapy. Red dots indicate significantly upregulated proteins, blue dots indicate significantly downregulated proteins, and gray dots represent proteins with no significant difference in expression. C Scatter plots of plasma proteins with the most significant differential expression between RD and NRD groups, including IL-5, IL-13, CCL3, CCL4, and MMP7. D Volcano plot demonstrating differential abundance analysis of CD8⁺T cell clusters between RD and NRD groups. Red dots indicate clusters significantly more abundant in the RD group, while orange dots indicate clusters enriched in the NRD group, and gray dots represent clusters with no significant difference. E Frequencies of CD8⁺Tcm (c03) and CD8⁺Temra (c17) in RD and NRD groups. The vertical dashed lines at log2(FC) = − 0.58 and 0.58 correspond to fold change thresholds of 1.5-fold downregulation and upregulation, respectively. The horizontal dashed line at -log10(p-value) = 1.3 indicates the significance threshold (p = 0.05). Statistical significance was determined by unpaired t-tests B–E. *p < 0.05, **p < 0.01. RD responders, NRD non-responders, FC fold change

Full size image

Differential abundance analysis of pre-therapy CD8⁺T cell clusters via volcano plots identified clusters 03 (c03, Tcm) and 17 (c17, Temra) as significantly divergent between groups (Fig. 4D). RD exhibited higher Tcm (c03) and lower Temra (c17) frequencies compared to NRD. Functional marker profiling demonstrated that Tcm (c03) displayed reduced expression of CD103, CD69, and CD38 relative to other Tcm clusters, indicating a quiescent phenotype but with highly responsive to antigen re-exposure. Conversely, Temra (c17) exhibited diminished CD28 and cleaved PARP (cPARP) expression alongside elevated TIGIT levels, consistent with an immunosuppressive state. These findings collectively suggest that NRD patients exhibited a more pronounced immunosuppressive state in the pre-therapy peripheral circulation, with reduced responsiveness and a plasma proteomic milieu conducive to tumor immune evasion. The observed phenotypic and proteomic disparities likely underpin impaired responsiveness to neoadjuvant therapy in NRD.

Reactivation of Tcm (c03) and Temra (c17) and interactions with different immune subsets in antitumor response

To elucidate the functional dynamics of CD103⁻CD8⁺Tcm (c03) and CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17) during neoadjuvant therapy, their phenotypic evolution and inter-cluster interactions were systematically interrogated. Post-therapy analysis revealed that CD103⁻CD8⁺Tcm (c03) exhibited significant downregulation of the immunosuppressive marker PD-1, concurrent with upregulated expression of activation markers (HLADR, CD38), cytotoxic effector molecules (Granzyme B), and memory-associated receptors (CD45RO, CD103) compared to baseline (Fig. 5A, Supplementary Fig. 8A). This phenotypic shift indicated functional reactivation of CD8⁺Tcm (c03), potentiating its antitumor responses and immunological memory. Similarly, CD8⁺Temra (c17) demonstrated marked post-therapy reductions in exhaustion markers (PD-1, LAG3 and TIGIT) alongside enhanced activation (HLADR and CD38), and cytokine production (IFN-γ) (Fig. 5B, Supplementary Fig. 8A). Notably, c17 frequencies increased in RD but declined in NRD post-therapy (Supplementary Fig. 6C).

CD8⁺Tcm (c03) and CD8⁺Temra (c17) changes and interaction with other CD8⁺T cell clusters. A, B Expression changes of CD8⁺Tcm (c03) and CD8⁺Temra (c17) markers in RD and NRD groups following neoadjuvant therapy. C, D Network heatmap and scatter plots illustrating correlations between the abundance of CD8⁺Tcm (c03) and CD8⁺Temra (c17) pre-neoadjuvant therapy and the FC value of CD8⁺T cell clusters following therapy, with regression lines indicating significant covariation (Pearson's r). Statistical analysis was performed using paired t-tests A, B and D. *p < 0.05, **p < 0.01., ***p < 0.001, ****p < 0.0001. RD responders, NRD non-responders, FC fold changes

Full size image

A correlation analysis was performed to examine the relationship between the pre-therapy abundances of the CD103⁻CD8⁺Tcm (c03) and CD28⁻TIGIT^highcPARP⁻ CD8⁺Temra (c17) subsets and the dynamic changes in the frequencies of CD8⁺T cell clusters during neoadjuvant therapy (Fig. 5C and D, Supplementary Fig. 8B). Alterations in cellular subsets are quantitatively represented by “fold changes” (FC), which are operationally defined as the ratio of post- versus pre-therapy frequencies for individual CD8⁺T cell cluster. Significant negative correlations were observed between the baseline CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17) levels and FC values of both c17 (Temra) and c06 (Tem) clusters. In contrast, FC c06 (Tem) showed positive correlations with FC c19 (Tem) and FC c17 (Temra), suggesting a synergistic interaction between the Tem (c06 and c19) and Temra (c17) subsets during neoadjuvant therapy. Additionally, the pre-therapy abundance of CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17) was positively associated with FC c05 (Tpex) and FC c15 (Tex-term), indicating that patients with lower baseline frequencies of Temra (c17) showed more pronounced reductions in the Tpex (c05) and Tex-term (c15) clusters after neoadjuvant therapy. This reduction was correlated with superior therapeutic outcomes, suggesting that a lower starting frequency of Temra (c17) may favor a more effective therapeutic response (Fig. 2E). Differential regulatory patterns were identified among Temra subsets: FC c02 (Temra) showed significant negative correlations with FC c05 (Tpex) and FC c15 (Tex-term), but a significant positive correlation with FC c08 (Tex-int), and these correlations were notably stronger compared to those observed for FC c17 (Temra). These findings suggest functional heterogeneity within the Temra cell populations, with c02 (Temra) playing a more prominent role in modulating Tex subsets. Specifically, c02 (Temra) may antagonize Tpex and Tex-term (c05 and c15) while promoting Tex-int (c08).

These findings demonstrate that neoadjuvant therapy restores functional competence in CD8⁺Tcm (c03) and CD8⁺Temra (c17 and c02) subsets, enabling synergistic interactions with CD8⁺Tex (c05, c08 and c15) and CD8⁺Tem (c06 and c19) cells to orchestrate antitumor immunity.

Integrated both cell cluster and protein biomarkers from peripheral blood enhance predictive capacity for neoadjuvant therapy outcomes

As shown in above, we identified CD8⁺T cell clusters (c03 and c17) and differentially expressed proteins (IL-5, IL-13, CCL3, CCL4, MMP7) could distinguished RD and NRD group who underwent neoadjuvant therapy. When analyzed as standalone predictors, CD103⁻CD8⁺Tcm (c03) and CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17) demonstrated area under the curve (AUC) values of 0.7321 and 0.7460, respectively (Fig. 6A). Plasma proteins IL-5, IL-13, CCL3, CCL4, and MMP7 exhibited individual AUCs of 0.7407, 0.7235, 0.7753, 0.7136, and 0.6790, respectively (Fig. 6B).

ROC analysis of CD8⁺T cell clusters and plasma proteins in RD and NRD groups. A Predictive capacity to classify neoadjuvant therapy RD of CD8⁺Tcm (c03) and CD8⁺Temra (c17). B Predictive capacity to classify neoadjuvant therapy RD of five differentially expressed proteins (IL-5, IL-13, CCL3, CCL4, MMP7). C Network heatmap depicting the interaction relationships between differential CD8⁺T cell clusters (c03 and c17) and differentially expressed proteins (IL-5, IL-13, CCL3, CCL4, MMP7). D Predictive capacity to classify neoadjuvant therapy RD of differentially cluster-protein combinations (c03, c17, IL-5 and MMP7). ROC Receiver operating characteristic, AUC area under the curve, RD responders, NRD non-responders

Full size image

A correlational analysis was performed on paired samples from 24 patients to investigate inter-parameter relationships across datasets. Positive correlations were identified between CD103⁻CD8⁺Tcm (c03) and IL-5, while negative correlations were observed with MMP7. Conversely, CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17) exhibited a positive correlation with MMP7 (Fig. 6C). Receiver operating characteristic (ROC) analysis was subsequently conducted for the parameters CD103⁻CD8⁺Tcm (c03), CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17), IL-5, and MMP7. A multivariate logistic regression model integrating both cellular and proteomic datasets achieved an optimized AUC of 0.9219 (Fig. 6D), enhancing predictive accuracy. These findings establish the combinatorial signature of CD103⁻CD8⁺Tcm (c03), CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17), IL-5, and MMP7 as a reliable predictive classifier of neoadjuvant therapy outcome in HNSCC [27,28,29,30].

Biomarker validity

Accurate prediction of neoadjuvant PD-1 blockade efficacy in HNSCC is essential for clinical optimization and translational research. While PD-1 inhibitors demonstrate therapeutic breakthroughs, marked response heterogeneity limits durable clinical benefits in most patients. Neoadjuvant immunotherapy theoretically enables tumor downstaging and micrometastasis clearance through preoperative immune activation, yet precise patient stratification remains critical to minimize toxicity and optimize resource allocation. Current biomarkers such as PD-L1 CPS, TMB, HPV status, and CD8⁺T cell spatial patterns show predictive potential but lack standalone clinical utility. Integrative multi-omics approaches combining CyTOF and Olink may improve prognostic model accuracy, and patient stratification enhances surgical resectability and supports rational immunotherapy combinations clinically.

In this study, we establish a predictive model centered on CD103⁻CD8⁺Tcm (c03), CD28⁻TIGIT^highcPARP⁻Temra (c17), IL-5 and MMP7 through multi-omics analysis. The integrated cellular-proteomic predictive model (AUC = 0.9219) demonstrated superior performance compared to current clinical standards, including PD-L1 CPS (AUC = 0.65–0.75) [29] and TMB (AUC = 0.55–0.70) [30], with three key advantages: pre-therapy predictive capacity (avoiding futile therapy), minimal invasiveness (single blood draw for advanced patients), and cost-effectiveness (4-parameter panel). The biomarker panel exhibited clear biological relevance: elevated CD103⁻CD8⁺Tcm (c03) levels indicated enhanced immune memory reservoirs responsive to PD-1-driven T cell clonal expansion; reduced CD28⁻TIGIT^highcPARP⁻ CD8⁺Temra (c17) abundance reflected reversible exhaustion states predictive of therapeutic sensitivity; and the IL-5/MMP7 ratio served as a quantitative metric for systemic immune activation-suppression homeostasis. These findings not only provide a molecular basis for personalized combinatorial therapies but also advance liquid biopsy from descriptive profiling to clinical decision-making tools. Therefore, future integration of AI-driven multi-omics and dynamic biomarker monitoring will advance HNSCC immunotherapy toward personalized precision oncology [31,32,33,34].

Biological mechanisms

Plasticity of CD8⁺T cell subsets determines therapeutic response

This study provided compelling multidimensional evidence supporting the critical immunological mechanisms by which PD-1 blockade mediates antitumor effects through reshaping the peripheral immune landscape. The therapeutic effectiveness of PD-1 inhibitors is fundamentally contingent upon the phenotypic and functional plasticity of CD8⁺T cell subsets, particularly the dynamic equilibrium between Tcm and Temra. Based on the findings of this study, patients in the RD group exhibited significant expansion of CD103⁻CD8⁺Tcm (c03) and CD28⁻TIGIT^highcPARP⁻ CD8⁺Temra (c17), along with upregulated expression of activation markers CD38 and HLA-DR following PD-1 inhibitor therapy. These observations align with the results reported by Hui Wei et al. [35].

Mechanistically, this immunological remodeling reflects the reactivation of a previously rested or exhausted CD8⁺T cell compartment following the alleviation of PD-1-mediated inhibitory signaling. PD-1 blockade mitigates suppression of tumor-specific T cells, facilitating their re-entry into the cell cycle and acquisition of a highly activated phenotype characterized by co-expression of HLA-DR and CD38. Notably, proliferating PD-1⁺CD8⁺T cells under PD-1 blockade therapy frequently exhibit this activated effector profile, indicative of enhanced cytotoxic potential and immune effector functionality [36]. Moreover, immune checkpoint inhibition appears to preferentially target Tpex cells, promoting their differentiation into both long-lived memory and highly cytotoxic effector phenotypes, rather than reactivating terminally exhausted populations [37]. These findings elucidate the concurrent elevation in Tcm and Temra frequencies-the former constitutes a reservoir of self-renewing, proliferative lymphocytes essential for sustained immune surveillance, while the latter (identified by CD45RA re-expression) represents terminally differentiated, highly cytotoxic effector cells primed for immediate tumoricidal activity. We also further validated the above dynamic evolution of Tpex, tex-term, Tcm and Temra through pseudotime trajectory analysis, as presented in Supplementary Figs. 9 and 10.

Importantly, the emergence of CD8⁺T cells co-expressing HLA-DR and CD38 has been repeatedly associated with improved therapeutic outcomes across multiple malignancies. Patients who exhibit robust CD8⁺T cell activation—evidenced by substantial expansion of HLA-DR⁺CD38⁺ subsets during the initial phase of PD-1 blockade therapy—frequently experience enhanced tumor regression and greater clinical benefit [38]. Similarly, an increased frequency of peripheral Temra cells has been correlated with favorable treatment responses and prolonged survival in diverse oncological contexts, including non-small cell lung cancer and melanoma [36, 39]. Collectively, the concurrent enrichment of CD8⁺Tcm and Temra subsets, together with the elevated expression of activation markers CD38 and HLA-DR, delineates a key immunodynamic mechanism through which PD-1 blockade therapy restores antitumor T cell competency, bridging rapid effector- mediated tumor clearance with the generation of durable immunological memory.

Furthermore, Tcm cells maintain stem cell-like self-renewal capacity through high expression of transcription factor TCF-1 and can rapidly differentiate into effector cells following PD-1 blockade, thus driving persistent and effective antitumor immune responses [40]. Their oxidative phosphorylation (OXPHOS)-dependent metabolic profile enables mitochondrial functional recovery post PD-1/PD-L1 inhibition, consequently facilitating proliferation and migration [41]. Our data revealed that CD103⁻CD8⁺Tcm subset exhibited PD-1 downregulation and granzyme B and CD38 upregulation post-therapy, indicating their conversion toward an effector phenotype and enhanced antigen recognition and cytotoxic capabilities. Furthermore, CCR7-mediated lymph node homing ensures continuous antigenic stimulation of Tcm, and patients with higher baseline Tcm proportions demonstrated superior treatment responsiveness [42], consistent with our findings.

Conversely, despite high cytotoxic potential (Perforin⁺, Granzyme B⁺), the functionality of Temra cells is restricted by a terminal exhaustion phenotype (PD-1^highTIM-3^high) and irreversible epigenetic modifications such as DNA methylation [43]. The absence of TCF-1 expression and hyperactivation of transcription factors like BATF in Temra cells impede their reverse differentiation into memory phenotypes. Preclinical studies indicate selective depletion of Temra enhances PD-1 inhibitor efficacy, suggesting that Temra cells indirectly limit therapeutic efficacy via pro-inflammatory cytokine (e.g., IFN-γ)-mediated recruitment of regulatory T cells (Treg) [44, 45]. Consistently, our data revealed that patients with a lower baseline frequency of CD28⁻TIGIT^highcPARP⁻CD8⁺Temra cells displayed superior responses, further confirming their negative correlation with therapeutic outcomes. Collectively, the expansion potential of Tcm and functional limitations of Temra jointly shape the therapeutic response to PD-1 inhibitors, indicating targeted modulation of their plasticity as a potential strategy for enhancing therapeutic efficacy.

Plasma proteomic microenvironment modulates immune response pattern

This study identified a pre-therapy plasma proteomic signature (IL-5 and IL-13 high, CCL3, CCL4 and MMP7 low) in RD group, suggesting a dynamic equilibrium between immune pre-activation and immunosuppression alleviation that forms the critical microenvironmental basis underpinning neoadjuvant therapeutic efficacy. Specifically, IL-5 potentiates antitumor immunity by activating eosinophils and promoting their tumor infiltration, directly mediating cytotoxic effects via granule proteins such as eosinophil peroxidase (EPO), and indirectly enhancing CD8⁺T cell functionality through improved antigen presentation [46, 47]. IL13 induces macrophage M1 polarization via STAT6 signaling, augmenting antigen-presenting capacity while concurrently suppressing TGF-β secretion from cancer-associated fibroblasts (CAFs), thus attenuating differentiation of Tregs and recruitment of myeloid-derived suppressor cells (MDSCs) [48, 49]. Clinical studies further corroborate that high expression of IL5 and IL13 in non-small cell lung cancer patient pre-therapy strongly correlate with improved objective responses to PD-1 inhibitors [50].

Additionally, reduced expression of chemokines CCL3 and CCL4 mitigates Treg trafficking to the tumor site and enhances T cell activation [51]. Lowered MMP7 expression alleviates extracellular matrix (ECM) degradation and inhibits TGF-β activation, promoting CD8⁺T cell infiltration and mitigating functional exhaustion [48, 52]. Indeed, clinical cohorts of melanoma patients exhibiting lower MMP7 expression displayed significantly prolonged progression-free survival following PD-1 blockade therapy [53], underscoring the critical role of stromal remodeling in immunotherapeutic responses.

Multi-omics integration reveals multilayered synergy in peripheral immunity

Multi-omics analyses uncovered a multilayered synergistic network between the IL-5, MMP7 and CD8⁺T cell subsets (Tcm and Temra) in peripheral immunity. IL-5 promotes Temra differentiation towards an effector phenotype and reduces exhaustion marker (e.g., PD-1, TIM-3) expression through eosinophil-derived IL-12 release [46, 54, 55]. Concurrently, low MMP7 expression mitigates Temra functional inhibition by suppressing TGF-β activation [48, 56]. Clinical cohorts corroborated that an IL-5^high and MMP7^low signature strongly correlates with enhanced PD-1 inhibitor efficacy [56, 57]. Tcm replenish the effector T cell pool through rapid differentiation and sustain Th1 immunity via IFN-γ or TNF-α secretion [59, 60], while Temra exert immediate cytotoxic functions, collectively generating a dynamic immunologic amplification loop [35, 61]. The IL-5 and MMP7 synergizes with Tcm and Temra spatiotemporal cooperation to overcome tumor immune evasion by activating effector immunity while counteracting MMP7-mediated stromal suppression and TGF-β signaling. Consequently, combinatorial strategies targeting IL-5 agonists or MMP7 inhibitor (e.g., Marimastat) have entered clinical trials, with preliminary data showing enhanced PD-1 inhibitor efficacy [62, 63]. Future studies should delineate dynamic microenvironmental remodeling through integrated multi-omics to achieve precision immunotherapeutic modulation.

Clinical implications and study innovations

This study demonstrates tripartite innovation across methodological, mechanistic, and clinical domains. Methodologically, the study integrated the CyTOF (single-cell resolution) and Olink (ultrasensitive proteomics) to establish a multidimensional immune profiling platform for HNSCC, coupled with machine learning-enabled predictive modeling. Mechanistically, the study revealed a multilayered synergistic network between the IL-5, MMP7 and CD8⁺T cell subsets (Tcm and Temra) in peripheral immunity, overcoming tumor immune evasion and enhancing effector of PD-1 inhibitors, and provided novel therapeutic targets to counteract PD-1 inhibitor resistance. Clinically, we developed the pre-therapy peripheral blood predictive model for HNSCC neoadjuvant immunotherapy, with baseline immune metrics as core prognostic biomarkers. These innovations collectively advance precision neoadjuvant immunotherapy through technological integration and mechanism-driven biomarker discovery.

Study limitations

Although this study yielded significant findings, there are certain limitations, including a relatively small sample size (n = 50), single-center design, the absence of tumor-level histopathological assessment, the lack of stratified analyses regarding the potential impact of sex and age on therapeutic response, and the unavailability of an independent external validation cohort, which may limit the generalizability of the findings. Future investigations should aim to replicate and validate these findings in larger, multicenter cohorts to enhance the robustness and clinical applicability of the proposed conclusions. Additionally, while CyTOF and Olink technologies offer high resolution, their complexity and cost hinder clinical use. Exploring simpler, cost-effective technologies could improve their clinical application. Lastly, this study did not fully explore the interactions between immune cells and the tumor microenvironment, which should be addressed in future research to better understand their impact on neoadjuvant immunotherapy outcomes.

Through multi-omics integration, this study delineates dynamic peripheral immune remodeling in HNSCC patients receiving neoadjuvant PD-1 blockade therapy and pioneers a predictive model (AUC = 0.9219) centered on CD103⁻CD8⁺Tcm (c03), CD28⁻TIGIT^highcPARP⁻CD8⁺Temra (c17), IL-5 and MMP7. This model transcends conventional biomarker limitations, mechanistically elucidating synergism between T cell plasticity and cytokine microenvironments, thereby providing both theoretical and translational foundations for precision neoadjuvant immunotherapy in HNSCC.

The data supporting the conclusions of this article have been provided in this article and its supplementary files. In addition, all data from this study can be obtained from the corresponding author upon reasonable request.

HNSCC:

Head and neck squamous cell carcinoma

PD-1:

Programmed death-1

irAEs:

Immune-related adverse event

CPS:

PD-L1 combined positive score

TMB:

Tumor mutational burden

TME:

Tumor microenvironment

CyTOF:

Cytometry by time of flight (mass cytometry)

NSCLC:

Non-small cell lung cancer

PEA:

Proximity extension assay

Tcm:

Central memory T cell

Temra:

Terminal effector memory RA⁺ T cell

IL:

Interleukin

MMP7:

Matrix metalloproteinase-7

AJCC:

American Joint Committee on Cancer

ECOG:

Eastern Cooperative Oncology Group

RD:

Responder

NRD:

Non-responder

pCR:

Pathological complete response

MPR:

Major pathological response

PBMC:

Peripheral blood mononuclear cell

RBC:

Red blood cell

BSA:

Bovine serum albumin

DMSO:

Dimethyl sulfoxide

PEA:

Proximity extension assay

NPX:

Normalized protein expression

DEP:

Differentially expressed protein

SEM:

Standard error of the mean

ROC:

Receiver operating characteristic

UMAP:

Unsupervised clustering algorithm phonograph

CCL:

C–C motif chemokine

TNF:

Tumor necrosis factor

MCP-3:

Monocyte chemoattractant protein 3

GO:

Gene ontology

KEGG:

Kyoto encyclopedia of genes and genomes

AUC:

Area under the curve

This work was supported by the projects of the Key Technologies Research and Development Program [grant number 2022YFC2406305] and the Clinical Medicine Plus X - Young Scholars Project of Peking University under Grant (PKU2022LCXQ19).

Authors and Affiliations

1. Institute of Medical Technology, Peking University Health Science Center, Beijing, 100191, China

   Hao Zhang & Guojun Han

2. Biomedical Engineering Department, Peking University, Beijing, 100191, China

   Hao Zhang & Guojun Han

3. Department of Prosthodontics, Peking University School and Hospital of Stomatology, Beijing, 100081, China

   Hao Zhang

4. Department of Oral and Maxillofacial Surgery, Peking University School and Hospital of Stomatology, Beijing, 100081, China

   Wenjie Wu, Meng Wang, Jie Zhang, Chuanbin Guo & Lin Wang

5. Institute of Advanced Clinical Medicine, Peking University, Beijing, 100191, China

   Guojun Han

6. Peking University international cancer institute, Peking University, Beijing, 100191, China

   Guojun Han

Contributions

Chuanbin Guo and Lin Wang conceived and supervised the study. Wenjie Wu Jie Zhang collected samples of HNSCC patients. Hao Zhang and Meng Wang performed the experiments, analyzed the data and wrote the manuscript. Lin Wang, Guojun Han revised the final version of the manuscript.

Corresponding author

Correspondence to Lin Wang.

Ethics approval and consent to participate

The study was approved by the Ethics Committee of Peking University School and Hospital of Stomatology (No. PKUSSIRB-202170179 and No. PKUSSIRB-202276072). And written informed consent was obtained from all participants.

Consent for publication

All the authors approved the publication.

Competing interests

The authors declare that they have no competing interests.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions