Combining Mendelian randomization and multiomics data to determine the causal relationship between systemic inflammation and head and neck squamous cell carcinoma Authors: Ke Yang, Guixin Zhu, Yanyan Li, Jie Huang, Yanfei Zhang, Fei Sun, Zhuomin Sha, Yun Zhang, Xuepeng Chen Published: 2026 ABSTRACT AbstractBackgroundHead and neck squamous cell carcinoma (HNSCC), the seventh most common cancer worldwide, is strongly associated with chronic inflammation. While chronic inflammation is often localized, systemic inflammation represents a generalized, systemic state of immune activation characterized by elevated circulating cytokines and changes in immune cell populations. However, the causal relationship between systemic inflammation and HNSCC remains elusive. This study employed Mendelian randomization (MR) to address this knowledge gap, using genetic variants as instrumental variables to estimate potential causal effects, thus minimizing confounding and reverse causation.MethodsWe applied two-sample Mendelian randomization (TSMR) and two-step mediation analyses to explore the potential causal relationships between HNSCC, circulating immune cell phenotypes, and inflammatory proteins. Potential confounders were assessed using the Open Targets Genetics platform, and sensitivity analyses were performed to ensure the robustness of our findings. The MR results were further assessed using transcriptomic and proteomic datasets.ResultsTSMR revealed CXCL11, CD40, and interleukin-10 receptor subunit beta (IL10RB) as potential protective factors against HNSCC, whereas cystatin D (CST5) was associated with increased susceptibility. HNSCC is causally associated with multiple immune phenotypes, including T cells, B cells, natural killer (NK) cells, dendritic cells, and regulatory T cells (Tregs). Mediation analyses revealed several key mediators that may influence inflammatory pathways. Data from transcriptomic and proteomic cohorts supported significant expression changes in these genes and proteins, which correlated with patient prognosis. Furthermore, single-cell RNA sequencing revealed CXCL-mediated monocyte–T-cell interactions, elevated Treg populations, and altered expression of the CXCL11 receptor.ConclusionsThis study suggests a bidirectional causal relationship between systemic inflammation and HNSCC, with CXCL11 and specific immune cell phenotypes serving as central mediators. These findings provide mechanistic insights into the HNSCC immunoinflammatory network and highlight potential biomarkers for early detection and targets for immunotherapeutic intervention.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12935-026-04359-3.Keywords: Head and Neck Neoplasms, Mendelian Randomization Analysis, Inflammation, Computational Biology, Immune System Background Head and neck squamous cell carcinoma (HNSCC) constitutes approximately 90% of head and neck cancers and is the seventh most common malignancy worldwide [1, 2]. Its pathogenesis is intrinsically linked to chronic inflammation, which is driven by persistent immune cell activation, sustained cytokine release, and oxidative stress [3]. HNSCC typically develops within an inflammatory microenvironment rich in leukocytes, cytokines, and chemokines [4, 5]. While chronic inflammation involves a localized immune response, systemic inflammation involves a generalized state of immune activation marked by elevated levels of circulating cytokines and acute-phase proteins. These mediators not only sustain local inflammation but also promote genomic instability, clonal selection, and tumour progression [6]. Furthermore, prolonged inflammatory signalling may dysregulate systemic immune homeostasis, affect distant tissues, and facilitate the formation of pre-metastatic niches‍ [6]. Consequently, alterations in circulating immune cell subsets and inflammatory proteins are increasingly recognized as both drivers and indicators of cancer-associated immunosuppression and progression [7]. Despite increasing recognition of the interplay between inflammation and tumorigenesis, the causal relationship between systemic immune-inflammatory profiles and HNSCC remains poorly understood. Mendelian randomization (MR) leverages genetic variants as instrumental variables (IVs) to infer causality between exposures and outcomes [8]. By utilizing the random assortment of alleles at conception, MR functions as a natural experiment analogous to a randomized controlled trial, thus effectively minimizing confounding factors and mitigating reverse causation [8]. In this study, we applied MR to investigate the bidirectional causal relationships among blood immunocyte phenotypes, circulating inflammatory proteins, and HNSCC. Specifically, we assessed whether genetically predicted immune-inflammatory traits influence HNSCC susceptibility and, conversely, whether genetic liability to HNSCC alters systemic inflammatory profiles. To ensure robustness, MR findings were integrated with transcriptomic and proteomic evaluation. This multiomics approach aims to elucidate the inflammatory mechanisms underlying HNSCC and identify novel biomarkers and therapeutic targets for clinical intervention. Materials and methods Study design The workflow for this study is illustrated in Fig. 1. First, a bidirectional two-sample MR (TSMR) was performed to investigate the causal relationship between HNSCC and immunocyte phenotypes or inflammatory proteins. A two-step mediation analysis was subsequently employed to identify key mediators. Finally, bulk RNA-seq databases, single-cell sequencing data and proteomic databases were utilized to further explore the TSMR results. Experimental design and overview of the study workflow. A) Key assumption of the experiment. B) Main sources of data. (C) Main methods used in Mendelian randomization and verification. Assumption 1: Single nucleotide polymorphisms (SNPs) show robust associations with exposure; Assumption 2: SNPs are not associated with confounders; Assumption 3: SNPs only affect results from exposure. SNPs: single nucleotide polymorphisms, TSMR: two-sample Mendelian randomization, IVW: inverse variance weighted Data sources Genome-wide association study (GWAS) summary statistics for HNSCC were obtained from the EMBL-EBI database (GCST012235; 2,497 cases and 2,928 controls) and validated using the FinnGen R10 dataset (2,281 cases and 314,193 controls) [9, 10]. Data for 91 circulating inflammatory proteins (n = 14,824) were sourced from the EMBL-EBI database (GCST90274758-GCST90274848) [11]. Furthermore, GWAS data for 731 immunocyte phenotypes (n = 3,757) were collected from the EMBL-EBI database (GCST90001908-GCST90002121) (Additional file 1) [12]. For proteomic analysis, pGWAS data (n = 10,840) from the UK Biobank Pharma Proteomics Project (UKB-PPP) were used [13]. Instrumental Variables (IVs) Selection The selection of IVs adhered to the three core assumptions: correlation, independence, and exclusivity. Given the exploratory nature of this study, single nucleotide polymorphisms (SNPs) were selected using a gradient of significance thresholds: p < 5 × 10− 8, p < 5 × 10− 7, p < 1 × 10− 6, and p < 5 × 10− 6. Independent SNPs were retained after linkage disequilibrium (LD) clumping (r² <0.01, window = 10,000 kb). SNPs with an F-statistic < 10 were excluded. To screen for potential confounding factors, we used the Open Targets Genetics database and applied a threshold of p < 1 × 10− 5 [14]. Additional file 2 summarizes the key parameters and outcomes of instrumental variable selection for each exposure category analysed in the study. Additional file 3 provides details on the traits removed as confounders. Additionally, we verified that the IVs were not directly associated with the outcome using the PheWAS database within the Open Targets Genetics platform (p < 5 × 10− 8) [15]. TSMR and sensitivity analyses Five MR methods were employed, with inverse variance weighted (IVW) serving as the primary approach [16]. Sensitivity analyses included MR-PRESSO, the MR‒Egger intercept test, Cochran’s Q test, and leave-one-out analysis (Additional files 4–8) [16]. Steiger filtering was applied to address reverse causality. Analyses were performed using the “TwoSampleMR” and “MR-PRESSO” R packages, adhering to the STROBE-MR guidelines for the presentation of findings (Additional file 9) [17]. Functional enrichment Significant inflammatory proteins identified via TSMR were subjected to Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) enrichment analyses using the “clusterProfiler” R package. Mediation analyses Mediation effects were quantified using a two-step MR framework, and the validity of these assumptions was assessed within the MR framework (Fig. 1). Mediation effects were computed by multiplying β1 (exposure to the mediator) by β2 (effect of mediator on the outcome). The total effect (β0) was derived from the initial TSMR analysis, and the mediated proportion was calculated as (β1*β2)/β0 [18]. The delta method via the “Rmediation” package was used to calculate 95% confidence intervals (CIs). In cases where the direct and mediated effects showed opposing directions, the relationship was classified as inconsistent mediation (suppression effect). Bulk RNA sequencing data analysis Bulk RNA sequencing data from The Cancer Genome Atlas Program (TCGA) and Gene Expression Omnibus (GEO) database (GSE143950, GSE156178, GSE196038, GSE213862, and GSE246050) were analysed using the “DESeq2” package to compare HNSCC samples with those from the healthy population. Timer 2.0 was used to explore the relationship between HNSCC prognosis and gene expression and/or immunocyte infiltration [19]. Animal model Male C3H/HeJ mice (6 weeks old, 18 ~ 22 g; n = 6 per group) were purchased from Charles River (China). The procedures were approved by the Institutional Animal Care and Use Committee of the Hangzhou Institute of Medicine, Chinese Academy of Sciences. The mice were randomized into tumour or control groups; outcome assessors were blinded to group allocation. Under isoflurane anaesthesia, tumour-bearing mice received an orthotopic injection of 1 × 106 mycoplasma-free SCC7 cells in 20 µL of PBS into the right ventrolateral tongue, whereas control animals received 20 µL of PBS alone. Body weight and tongue tumour dimensions were recorded daily, and the mice were closely monitored for signs of distress or weight loss exceeding 20%. On day 14 post-implantation—or upon reaching humane endpoints—animals were re-anaesthetised, and blood was collected via cardiac puncture immediately prior to euthanasia. Single-cell RNA sequencing of peripheral blood cells was performed by Shanghai Biotechnology Corporation (China), and the data were analysed by our research team. Single-Cell RNA Sequencing (scRNA-seq) and data analysis The single-cell data (GSE139324) included 63 samples of paired peripheral blood mononuclear cells and tumour-infiltrating immune cells from HNSCC patients and healthy donors [20]. Data processing, dimensionality reduction, and cell clustering were performed using the Seurat R package (v5.1.0) [21]. The quality control filters were set at 250–5000 features, a maximum read count of 500–20,000, and a mitochondrial content < 20%. Data were normalized via SCTransform and integrated using the CCA workflow. Principal component analysis (PCA) was conducted on the top 5000 variable features. An elbow plot indicated that the first 20 principal components were optimal; consequently, these components were utilized for constructing the nearest neighbour plot and performing Louvain cluster analysis with a resolution of 1.0. Dimensionality reduction and clustering visualization were performed using the RunPCA and RunUMAP functions within Seurat. Cell types were annotated via CellMarker and MonacoImmuneData. Cell‒cell communication and lineage trajectories were modelled using CellChat and Monocle 3 [22–26]. Murine single-cell sequencing data were analysed in an identical manner, except that the cells were annotated using ImmGen. Proteomic data analysis To support the TSMR findings, pGWAS data for C-C motif chemokine 20 (CCL20), C-X-C motif chemokine 11 (CXCL11), interleukin-10 receptor subunit beta (IL10RB), cystatin D (CST5), CD27, CD28, CD33, CD3D, CD3E, CD3G, the CD40L receptor (CD40), CD80, and CD45 were obtained from UKB-PPP [13]. Using these data, we performed TSMR analyses consistent with the primary methodology using GCST012235 and FinnGen R10. Moreover, the Clinical Proteomic Tumour Analysis Consortium (CPTAC) and the Cancer Proteogenomic Data Analysis Site (cProcite, https://cprosite.ccr.cancer.gov/) were utilized to compare protein expression between HNSCC tumour and adjacent normal tissues. Statistical analysis A significance threshold of p < 0.05 was used. Odds ratios (ORs) with 95% confidence intervals were reported (where an OR > 1 indicated increased risk and an OR < 1 indicated a protective effect). To mitigate the risk of Type I errors arising from high-dimensional testing (91 inflammatory proteins and 731 immune phenotypes), multiple-testing correction was implemented using the Benjamini–Hochberg false discovery rate (FDR) method. Associations with a PFDR < 0.05 were considered robust discovery-level findings. Recognizing that stringent FDR thresholds may be overly conservative given the modest sample size of the HNSCC GWAS (2,497 cases) and potentially lead to Type II errors (false-negatives), we employed a tiered evidentiary framework to prioritize candidates. Beyond FDR-corrected significance, a causal relationship was classified as ‘high confidence’ if it satisfied a composite of the following internal consistency criteria: (1) the inclusion of at least three independent SNPs as IVs at one or more thresholds; (2) a consistent direction of the effect across multiple MR methods; (3) successful passage of sensitivity analyses (MR-PRESSO, MR‒Egger, leave-one-out); and (4) stability across multiple IV selection thresholds. All the ‘high-confidence’ relationships that meet the above conditions are listed in Table 1. These ‘high-confidence’ relationships were further subjected to multiomics cross-assessment using independent proteomic and transcriptomic datasets to enhance their biological plausibility. Table 1Summary of high-confidence causal relationships between circulating inflammatory proteins, immune cell phenotypes, and HNSCCExposureOutcomen of SNPsp value for IVWp value for Cochran’s Q p value for MR-Egger interceptp value for MR-PRESSOThresholdTD DN (CD4-CD8-) ACHNSCC130.01003180.05300860.76888220.07225.00E-06TD DN (CD4-CD8-) ACHNSCC30.00071840.18260050.3194832NA5.00E-08Naive CD4+ %T cellHNSCC60.00201770.87276360.65833680.88061.00E-06Naive CD4+ %T cellHNSCC20.03052810.9220428NANA5.00E-07Interleukin-10 receptor subunit βHNSCC100.04657970.19615070.75904640.4035.00E-07Interleukin-10 receptor subunit βHNSCC30.01164230.92974520.8716735NA5.00E-08IgD+ CD38br ACHNSCC50.02568640.89090290.45486670.89741.00E-06IgD+ CD38br ACHNSCC50.02568640.89090290.45486670.89625.00E-07HLA DR on DCHNSCC160.02291720.71901160.82494690.72581.00E-06HLA DR on DCHNSCC270.0170390.90144330.64896610.90665.00E-06HLA DR on DCHNSCC160.02217580.71529460.81565360.71765.00E-07HLA DR on DCHNSCC130.01134310.58956550.47656630.595.00E-08FSC-A on HLA DR+ NKHNSCC50.0219530.36375460.63548790.3855.00E-08Cystatin DHNSCC180.00438630.46190940.08032050.47021.00E-06Cystatin DHNSCC330.00549350.63916980.15318770.66165.00E-06Cystatin DHNSCC180.00932990.56699660.52145980.54985.00E-08C-X-C motif chemokine 11HNSCC60.00103560.8194890.48280140.8071.00E-06C-X-C motif chemokine 11HNSCC170.008810.62728940.74926720.65565.00E-06C-X-C motif chemokine 11HNSCC60.00103560.8194890.48280140.81045.00E-07C-X-C motif chemokine 11HNSCC30.01092230.5948560.6341424NA5.00E-08CD80 on monocyteHNSCC40.00301410.72628860.79348360.76425.00E-08CD62L- monocyte ACHNSCC50.01083490.94293550.76834710.95281.00E-06CD62L- monocyte ACHNSCC40.01450780.87834690.66593240.9015.00E-07CD62L- monocyte ACHNSCC30.01957760.76422380.66248NA5.00E-08CD45RA on naive CD8brHNSCC50.02165670.44671140.54889790.46745.00E-08CD45 on granulocyteHNSCC40.01292860.3614460.85695140.41821.00E-06CD45 on granulocyteHNSCC20.00304760.3566758NANA5.00E-07CD40L receptorHNSCC70.03702530.65370230.7246690.7041.00E-06CD40L receptorHNSCC140.00712060.87599780.5458410.8895.00E-06CD40L receptorHNSCC80.02886120.62790370.18651290.47265.00E-07CD40 on CD14- CD16+ monocyteHNSCC150.04968110.66369810.49615380.69225.00E-08CD3 on T cellHNSCC150.00321590.99270390.36611240.99441.00E-06CD3 on T cellHNSCC250.01736820.91409870.63512360.88745.00E-06CD3 on T cellHNSCC110.00269150.97928280.73728390.9855.00E-07CD3 on NKTHNSCC60.04329220.1526890.64341580.19681.00E-06CD3 on NKTHNSCC50.03818660.13292210.41535020.19225.00E-07CD3 on EM CD4+HNSCC110.02348130.88695920.24531570.89681.00E-06CD3 on EM CD4+HNSCC80.01454770.86208440.3755790.86765.00E-07CD27 on sw memHNSCC230.01485640.67847730.70406740.69621.00E-06CD27 on sw memHNSCC180.00983470.48053470.86806610.50865.00E-07CD27 on sw memHNSCC130.03588860.26978670.99567770.33265.00E-08CD27 on memory B cellHNSCC140.01930830.54904220.92214430.58581.00E-06CD27 on memory B cellHNSCC140.01930830.54904220.92214430.5945.00E-07CD27 on memory B cellHNSCC100.03492360.30034550.99811220.38185.00E-08CD27 on IgD+ CD24+HNSCC180.00851930.49715840.89271950.53421.00E-06CD27 on IgD+ CD24+HNSCC180.00851930.49715840.89271950.52625.00E-07CD27 on IgD+ CD24+HNSCC120.01537810.33182910.96131050.39065.00E-08CD27 on IgD- CD38dimHNSCC230.01892140.56507150.59879040.5911.00E-06CD27 on IgD- CD38dimHNSCC180.02207860.31179440.71569920.36625.00E-07CD27 on IgD- CD38dimHNSCC130.0059550.35616340.86323660.36665.00E-08CD27 on IgD- CD38-HNSCC110.03562960.41053170.4882240.45985.00E-08CD27 on CD24+ CD27+HNSCC240.0321140.38654910.28501060.41541.00E-06CD27 on CD24+ CD27+HNSCC210.03762220.25251380.35989850.3035.00E-07CD27 on CD24+ CD27+HNSCC150.00344620.44949090.6269940.4535.00E-08Activated & resting Treg ACHNSCC60.02938150.44447630.77273020.4841.00E-06Activated & resting Treg ACHNSCC40.00915350.50431630.52379750.61985.00E-07HNSCCNaive-mature B cell AC60.01388780.55477670.70900960.67461.00E-06HNSCCNaive-mature B cell AC50.01850680.45890090.97849910.60645.00E-07HNSCCNaive-mature B cell %lymphocyte60.01042550.8194990.66379780.88121.00E-06HNSCCNaive-mature B cell %lymphocyte140.03002970.59901950.23939890.6245.00E-06HNSCCNaive-mature B cell %lymphocyte50.01750750.7377820.84320970.83625.00E-07HNSCCLymphocyte AC60.04988270.60190180.47756320.67961.00E-06HNSCCLymphocyte AC140.0363760.44541790.97030140.50065.00E-06HNSCCIgD+ CD38dim AC60.02200120.5676770.54430590.6031.00E-06HNSCCIgD+ CD38dim AC50.01143440.62885440.92627470.74585.00E-07HNSCCIgD+ CD38dim %lymphocyte60.02730680.80483530.5175230.79241.00E-06HNSCCIgD+ CD38dim %lymphocyte50.01661590.86590.95439790.91765.00E-07HNSCCIgD+ CD38br %lymphocyte60.03584410.08176330.38941980.16961.00E-06HNSCCIgD+ CD38br %lymphocyte50.00287630.22188230.96795840.4255.00E-07HNSCCIgD+ CD24- AC60.01029890.63794570.54349630.70541.00E-06HNSCCIgD+ CD24- AC50.00975330.60679290.88091330.75185.00E-07HNSCCIgD+ CD24- %lymphocyte60.00780120.81265270.48097510.83821.00E-06HNSCCIgD+ CD24- %lymphocyte50.00950860.78118510.70689130.8615.00E-07HNSCCIgD+ AC60.02296560.81745060.49728790.8611.00E-06HNSCCIgD+ AC50.03203140.73666670.6070490.79285.00E-07HNSCCIgD+ %Lymphocyte60.0144880.87690410.35715920.881.00E-06HNSCCIgD+ %Lymphocyte50.02828060.79213660.35286810.75765.00E-07HNSCCGranulocyte %leukocyte30.00803260.92911020.7677287NA1.00E-06HNSCCGranulocyte %leukocyte60.02715680.66626230.34636650.66245.00E-06HNSCCCD45 on Im MDSC60.02824430.77686610.58542970.80321.00E-06HNSCCCD45 on Im MDSC140.0116210.72147290.36555950.7575.00E-06HNSCCCD45 on CD33dim HLA DR-60.02380980.30848660.25580290.37441.00E-06HNSCCCD45 on CD33dim HLA DR-50.00425850.42727960.67677560.52125.00E-07HNSCCCD40 on monocytes60.02754890.94050260.89739110.95981.00E-06HNSCCCD40 on monocytes50.04459580.8902280.91335350.9415.00E-07HNSCCCD33 on CD33br HLA DR+ CD14dim60.00989570.94079650.99187590.96721.00E-06HNSCCCD33 on CD33br HLA DR+ CD14dim130.01948660.94360110.50411690.80125.00E-06HNSCCCD33 on CD33br HLA DR+ CD14dim50.04889570.87941660.86555550.92165.00E-07HNSCCCD3 on NKT60.02762770.69140290.35470370.67681.00E-06HNSCCCD3 on NKT140.01790840.92030370.86634020.9335.00E-06HNSCCCD3 on NKT50.01112770.84055990.80574590.8625.00E-07HNSCCC-C motif chemokine 2060.02019390.292660.87335260.37961.00E-06HNSCCC-C motif chemokine 2050.027240.22350980.92976610.33485.00E-07SNPs: Single nucleotide polymorphisms, IVW: Inverse variance weighted, MR: Mendelian randomization, MR-PRESSO: Mendelian Randomization Pleiotropy RESidual Sum and Outlier, HNSCC: Head and neck squamous cell carcinoma, AC: Absolute count, DC: Dendritic cell, DN: Double negative, EM: Effector memory, FSC-A: Forward scatter area, HLA DR: Human leukocyte antigen-DR, Im MDSC: Immature myeloid-derived suppressor cell, NK: Natural killer cells, NKT: Natural killer T cells, sw mem: Switched memory, TD: Terminally differentiated, Treg: Regulatory T cells, br: bright Summary of high-confidence causal relationships between circulating inflammatory proteins, immune cell phenotypes, and HNSCC SNPs: Single nucleotide polymorphisms, IVW: Inverse variance weighted, MR: Mendelian randomization, MR-PRESSO: Mendelian Randomization Pleiotropy RESidual Sum and Outlier, HNSCC: Head and neck squamous cell carcinoma, AC: Absolute count, DC: Dendritic cell, DN: Double negative, EM: Effector memory, FSC-A: Forward scatter area, HLA DR: Human leukocyte antigen-DR, Im MDSC: Immature myeloid-derived suppressor cell, NK: Natural killer cells, NKT: Natural killer T cells, sw mem: Switched memory, TD: Terminally differentiated, Treg: Regulatory T cells, br: bright Results High-confidence bidirectional causal relationships between inflammatory proteins (CXCL11, CD40, IL10RB, CST5, and CCL20) and HNSCC, as suggested by TSMR Mendelian randomization analysis revealed 10 inflammatory proteins that are likely causally linked to HNSCC. KEGG and GO enrichment analyses revealed that these proteins are involved primarily in cytokine and chemokine pathways, Toll-like receptor pathways, lymphocyte and neutrophil chemotaxis, MAPK cascade regulation, cell responses to chemokines and interleukin-1 (IL-1), and calcium ion homeostasis (Additional file 10 A–B). Within this investigative framework, CXCL11 showed a robust causal association that survived Benjamini–Hochberg FDR correction (PFDR = 0.047), supporting its role as a primary discovery-level finding (Additional file 11). Furthermore, three additional proteins (CD40, IL10RB, and CST5) were prioritized as high-confidence candidates (Fig. 2A, Additional file 4). Specifically, CXCL11, CD40, and IL10RB were identified as potential protective factors, whereas CST5 was associated with an increased risk of developing HNSCC. High-confidence results from bidirectional two-sample Mendelian randomization (TSMR) of head and neck squamous cell carcinoma (HNSCC) and circulating inflammatory protein or immunocyte phenotypes. A) TSMR of inflammatory proteins in HNSCC (left) and reverse MR of HNSCC on inflammatory proteins (right). B) TSMR of immune cell phenotypes in HNSCC (left) and reverse MR of immune cell phenotypes in HNSCC (right). IL10RB: IL-10 receptor subunit beta, CXCL11: C-X-C motif chemokine 11, CST5: Cystatin D, CCL20: C-C motif chemokine 20, sw mem: switched memory, AC: absolute count, NKT: natural killer T, NK: natural killer cells, Tregs: regulatory T cells, TD: terminal differentiated, DN: double negative, EM: effector memory, DC: dendritic cells, Im MDSC: immature myeloid-derived suppressor cells Additionally, HNSCC was causally linked to 14 circulating inflammatory proteins. KEGG/GO analysis revealed associations with leukocyte migration and chemotaxis, positive regulation of cell development, and cytokine and chemokine receptor binding and activity (Additional file 10 C–D). Notably, the causal effect of HNSCC on reducing CCL20 levels was identified as a high-confidence association (Fig. 2A, Additional file 5). High-confidence bidirectional causal relationships between specific immunocyte phenotypes (T/B/NK Subsets, DCs, Monocytes, and Tregs) and HNSCC, as indicated by TSMR Fifty-eight types of circulating immunocytes are potentially causally associated with HNSCC, with 20 showing high confidence (Fig. 2B, Additional file 6). These high-confidence phenotypes include T cells, B cells and NK cells (TBNK, 3 types), conventional dendritic cells (cDC, 3 types), B cells (7 types), maturation stages of T cells (4 types), monocytes (1 type) and Tregs (2 types). Potential protective factors for HNSCC included absolute cell counts (ACs) of IgD+ CD38bright (br) cells, CD62L- monocyte AC, FSC-A on HLA-DR + NK cells, CD3 on effector memory (EM) CD4 + T cells, CD3 on T cells, and HLA-DR on DCs. Notably, “br” stands for “bright” (high expression), and “dim” stands for “dim” (low expression), which are flow cytometry terms that describe the intensity of protein expression. In contrast, the risk factors included CD27 on 6 types of B cells, CD80 on monocyte, 3 types of maturation stages of T cells, CD40 on CD14- CD16 + monocytes, CD3 on natural killer T (NKT) cells, CD45 on granulocytes, and activated and resting Tregs ACs. The positive association between activated Tregs and HNSCC risk was also supported by the FinnGen R10 database (Additional file 12). Conversely, HNSCC is potentially causally linked to 50 immunocyte phenotypes, with 16 showing high confidence (Fig. 2B, Additional file 7). HNSCC is associated with reduced levels of CD3 on NKT cells, CD45 on CD33dim HLA DR-, granulocyte %leukocytes, and CD45 on immature myeloid-derived suppressor cells (MDSCs). Furthermore, HNSCC is a potential risk factor for AC lymphocytes, CD33 on CD33br HLA DR+ CD14dim cells, CD40 on monocytes, and nine B-cell phenotypes. Specific immune mediators linking HNSCC and systemic immunity, as indicated by mediation analysis The results of the mediation analysis indicated that CD3 on EM CD4 + and CD62L- monocyte ACs might mediate the potential protective effect of CXCL11 on HNSCC. CD27 on B cells inconsistently mediated the protective effect of IL10RB on HNSCC. CD40 appeared to facilitate the protective effect of CD27 on B cells in HNSCC. CXCL11 inconsistently facilitated the potential promotional effect of CD3 on NKT cells in HNSCC. CST5 inconsistently facilitated the protective effect of terminal differentiated (TD) double-negative (DN) T-cell AC on HNSCC (Fig. 3). The quantification of the mediating effects is shown in Fig. 4, where CD40 appears to significantly facilitate the protective effect of CD27 on B-cell subsets, whereas CXCL11 acts through monocytes and EM CD4 + T cells to reduce the risk of HNSCC. Schematic of the mediation analysis framework. IL10RB: IL-10 receptor subunit beta, CXCL11: C-X-C motif chemokine 11, CST5: Cystatin D, AC: Absolute count, NKT: Natural killer T, TD: Terminal differentiated, DN: Double negative, EM: Effector memory Results of the two-step mediation analysis. sw mem: Switched memory, AC: Absolute count, EM: Effector memory, HNSCC: Head and neck squamous cell carcinoma Analysis of transcriptomic datasets supports the biological plausibility of causal findings Analysis of data from the TCGA and several other databases revealed differential gene expression between HNSCC tumour tissues and normal tissues for CD3D, CD27, CD28, CD40, CD40 ligand (CD40LG), CD70, CD80, cytotoxic T-lymphocyte-associated protein 4 (CTLA4), CXCL11, CXCR3, CCL20, CST5, and IL10RB (Fig. 5A). In peripheral blood, differential expression of CD40, CD45, and CXCL11 was observed (Fig. 5A). Analysis of high-throughput sequencing data in HNSCC. A) Bulk RNA-seq analysis of high-confidence factors in tumour tissues and peripheral blood from HNSCC patients versus normal controls. B) Cell clustering from single-cell sequencing. C) Comparison of cell type proportions across groups by single-cell sequencing. D–F) Analysis of cell‒cell communication using CellChat. G) Correlations between gene expression and HNSCC prognosis. H–I) Association of gene expression with immune cell infiltration and patient prognosis. The grey colour indicates nonsignificance (p > 0.05). CD40LG: CD40 ligand, CTLA4: Cytotoxic T-lymphocyte-associated protein 4, CXCL: C-X-C motif chemokine, CCL20: C-C motif chemokine 20, CST5: Cystatin D, IL10RB: IL-10 receptor subunit beta, HNSCC: Head and neck squamous cell carcinoma, pDCs: Plasmacytoid dendritic cells, mDCs: Myeloid dendritic cells, NK: Natural killer cells, Tregs: Regulatory T cells, TE: Terminal effector, EM/CM: Effector Memory/Central Memory, Th cell: T helper cells, HPV: Human papillomavirus Single-cell sequencing data from the GSE139324 dataset revealed distinct clustering of immune cell types in HNSCC patients, including higher white blood cell counts and increased proportions of lymphocytes, T cells, Tregs, and granulocytes (Fig. 5B–C). CXCL signalling was predominantly observed between intermediate monocytes and Tregs and between EM/central memory (CM) CD8 + T cells, while CD40 signalling was evident between T helper (Th) cells and B cells and monocytes (Fig. 5D–E). Extensive communication among T cells, monocytes, B cells, cDCs, and MDSCs was detected through CD45 signalling (Fig. 5F). Previous MR findings revealed an average effect across HNSCCs, which could obscure subtype-specific associations. When TIMER 2.0 was used for prognostic assessment, we stratified HNSCC into HPV-positive (HPV+) and HPV-negative (HPV-) subtypes. The results of the analysis suggested that CST5 is associated with poor prognosis in HPV+ HNSCC patients, whereas the prognostic profiles of the HPV + and HPV- subtypes were distinct for the expression of CCL20 (Fig. 5G). CD27, CD3, and CD45 were linked to a more favourable HNSCC prognosis, and the effects of CXCL11 were also consistent when combined with T-cell levels (Fig. 5G–H). Additionally, CD40 expression was associated with improved prognosis when combined with B-cell or T-cell levels (Fig. 5I). Single-cell trajectory analysis using Monocle 3 suggested concordant gene expression dynamics during T and B-cell differentiation. With respect to the differentiation of naive T cells into Th cells, the CD45 and CD27 levels tended to increase, whereas the CXCR3 and CCL20 levels were markedly upregulated at the terminal stage (Additional file 13 A). During differentiation into Tregs, the expression of CD27 and CD45 initially decreased before increasing, with that of CXCR3 increasing in the later stages (Additional file 13B). During the transition to EM/CM CD8 + T cells, the CD27 and CD45 levels again fluctuated upwards, whereas the CXCR3 level initially increased but then returned to baseline (Additional file 13 C). With respect to naive B cells differentiating into memory/activated B cells, the expression of CD27 and CD45 increased steadily, that of CXCR3 increased at terminal differentiation, and that of CD40 increased early, followed by a return to baseline (Additional file 13D). However, these findings should be interpreted as providing biological plausibility and mechanistic context rather than definitive causal validation. These tissue-level and cellular expression patterns serve as complementary evidence that aligns with genetic instruments, and further independent genetic replication remains necessary to confirm the circulating causal relationships identified by our MR framework. Murine single-cell sequencing provides a complementary context for causal relationships Single-cell sequencing analysis of whole blood from a mouse model of tongue squamous cell carcinoma and a healthy group revealed 15 cell clusters (Additional file 14 A). The tumour-bearing mice had increased proportions of granulocytes and Tregs (Additional file 14B). B cells showed intercellular communication with various immune cells through the CXCL, CD45, and CD22 pathways (Additional file 14 C–E). The tumour-associated alterations included decreased CD3 expression on CD4/CD8 T cells and increased CD33 expression on macrophages (Additional file 14 F). In addition, the expressions of CD45 and CD40 on B cells; CD45 on granulocytes, NK cells and monocytes; IL10RB and CD40 on macrophages; and CD45, IL10RB, and CD27 on NK cells were significantly lower in the tumour group (Additional file 14 F). Proteomic analysis supports a broader dimension of regulation MR analysis of pGWAS using the GCST012235 and FinnGen R10 databases supported the finding that CXCL11 is a potential protective factor for HNSCC. Specifically, genetically predicted higher levels of this protein are associated with a decreased risk of developing malignancy. The study further utilized genetic instrumental variables to assess the potential protective effects of IL10RB and CD40, whereas CST5 was identified as a potential risk factor, whereby a genetic predisposition to higher CST5 levels correlated with an increased probability of HNSCC (Additional file 15). To explore the directionality of these associations, bidirectional MR suggested that HNSCC acts as a causal “factor” for reduced CD3 expression (Additional file 15). Within this causal framework, this signifies that a genetic liability for HNSCC leads to a secondary reduction in CD3 levels, representing the potential causal impact of the disease state on circulating immune cell phenotypes, likely via cancer-associated immunosuppression. To complement these MR-derived causal estimates, proteomic data revealed the upregulation of CD3, CD40, and CD45 in HNSCC tumour tissues, while the expressions of IL10RB and CD33 were downregulated (Additional file 16). Discussion Cancer is a systemic disease intricately linked to systemic inflammation [7]. Immunocytes and inflammatory proteins play dual roles in tumorigenesis—suppressing malignant transformation while also facilitating tumour progression through immune evasion and chronic inflammatory signalling. This study explores the intricate bidirectional causal relationships between HNSCC and various immunocyte phenotypes and inflammatory proteins, providing insights into potential circulating biomarkers and immunotherapeutic targets. Circulating inflammatory proteins as causal factors for HNSCC risk Circulating levels of CXCL11, IL10RB and CD40 serve as protective factors against HNSCC. CXCL11, a selective ligand of CXCR3, facilitates communication between monocytes and T cells, thus supporting antitumour effects through the CXCL11/CXCR3 axis [27]. IL-10RB, a component of the IL-10 receptor complex, is essential for intratumoral CD8 + T-cell proliferation and is implicated in antitumour immunity [28, 29]. CD40 signalling—particularly when it is analysed together with T or B-cell infiltration—is associated with a favourable prognosis, likely via DC activation, cytotoxic T-cell stimulation and NK cell responses, which clear tumour cells and provide long-term immunity [30]. Conversely, CST5 emerged as a risk factor. As a member of the cystatin superfamily, CST5 modulates extracellular matrix degradation, although its specific role in HNSCC remains unexplored [31]. Additionally, circulating levels of CCL20, a proinflammatory chemokine that mobilizes dendritic cell precursors and may inhibit tumour metastasis, appear to be suppressed in HNSCC [32]. Immune cell phenotypes associated with HNSCC susceptibility Our analysis also revealed causal associations between specific immunocyte phenotypes and HNSCC. Activated/resting Tregs and naive CD4 + T cells were identified as risk factors. Tregs, which are recruited by chemokines, inhibit antitumour immunity [33]. Naive CD4 + T cells share TCR clonotypes with tumour-infiltrating Tregs, suggesting peripheral recruitment and intratumoral differentiation [34]. DN T-cell AC is also a risk factor for HNSCC. DN T cells are present in oral squamous cell carcinoma and may differentiate into CD4 + and CD8 + T cells [35]; however, their function in the development of HNSCC remains to be fully elucidated. Other risk factors included CD45 expression on granulocytes, CD45RA expression on naive CD8br T cells, and CD80 expression on monocytes. Extensive communication among immunocytes via the CD45 pathway was identified, although its exact role needs to be further explored. Despite the known costimulatory role of CD80 in antitumour immunity, its dual function remains a subject of debate [36]. CD27 on B cells and CD3 on NKT cells were also identified as factors that promote disease. CD27 facilitates plasma cell differentiation and antibody production, whereas CD3-TCR signalling is critical for NKT cell effector function [37, 38]. In contrast, HLA DR on DCs and CD3 expression on (EM CD4+) T cells are protective factors against HNSCC. These markers support antigen presentation and T-cell activation [39]. Additional protective factors include IgD+ CD38br B-cell AC, CD62L- monocyte AC and FSC-A on HLA DR + NK cells. IgD+ CD38br B cells are pre-germinal centre cells that can leave the bloodstream, enter secondary lymphoid organs, and initiate humoral immunity following antigen recognition [40]. CD62L is a cell adhesion molecule that helps tether leukocytes to the luminal vessel wall, initiating their migration from the bloodstream to extravascular inflammatory sites [41]. During transendothelial migration, monocytes often shed CD62L, suggesting that CD62L- monocytes are likely in an intermediate stage of transition from the bloodstream to peripheral inflammatory sites [41]. As the FSC-A value increases in NK cells, their overall size increases, suggesting a possible correlation with NK cell activation and proliferation [42]. Moreover, HLA DR + NK cells are capable of cytokine production, degranulation, and prompt proliferation upon stimulation and may also participate in antigen presentation to CD4 + and CD8 + T cells [43]. Alterations in systemic immune profiles induced by HNSCC HNSCC also exerts a causal effect on circulating immunocyte phenotypes. HNSCC increases the counts of circulating B cells, including CD24-CD38-/dim (naive mature) and IgD+ (CD24- or CD3br/dim) B cells, likely reflecting the activation of adaptive immunity [44]. Furthermore, HNSCC is a risk factor for CD40 expression on monocytes, which can trigger T-cell activation and differentiation, thus exerting antitumour effects [45]. HNSCC decreased the levels of circulating CD45 on CD33dim HLA DR- myeloid cells and immature MDSCs. CD33dim HLA DR- myeloid cells, including immature MDSCs, possess immunosuppressive properties and are massively expanded in the tumour environment [46]. Reduced CD45 expression may facilitate immune recognition and the clearance of MDSCs [47]. In contrast, HNSCC is a factor for CD33 expression on CD33br HLA DR+ CD14dim myeloid cells and a protective factor against CD3 expression on NKT cells. Strong CD33 expression suggests an immature myeloid cell phenotype linked to immune suppression [48]. CD3 stimulation enhances Th2 cytokine secretion from human NKT cells and preserves antitumour cytotoxicity [49]. Notably, HNSCC was also associated with reduced granulocyte proportions. Granulocytes in cancer are diverse and plastic and may exhibit antitumour or protumour properties at different stages [50], but the ultimate role of granulocytes in HNSCC remains unclear. Key signalling axes and mediation pathways in HNSCC pathogenesis CD40 plays a mediating role in the promotion of HNSCC by CD27 on B cells. In this context, “promotion” refers to genetic susceptibility to tumour initiation and the acceleration of malignant progression. Additionally, a combined analysis of CD40 expression and the infiltration of B cells or T cells can predict the prognosis of HNSCC patients. CD40 signalling primarily originates from Th cells and is directed towards B cells, with CD40 expression fluctuating during B-cell differentiation, while CD27 expression steadily increases. The effect of CD27 on B cells in HNSCC may be influenced by Th cells via the CD40 pathway. Single-cell sequencing data analysis suggested that the CXCL signalling pathway is predominantly communicated from monocytes to EM T cells, with fluctuations in CXCR3 receptor expression observed during EM T-cell differentiation. Moreover, the combined analysis of CXCL11 expression and T-cell infiltration revealed a difference in HNSCC prognosis. These findings suggest that the interaction between monocytes and EM T cells through the CXCL11 pathway is closely related to HNSCC development and progression. Together, the CD40 and CXCL11 signalling axes appear central to HNSCC pathogenesis. Strengths, limitations, and future research This study has several notable strengths. A key strength is the integration of large-scale GWAS data from multiple cohorts, which enhances the robustness of causal estimates through TSMR. The application of multiple MR methods for validation, combined with complementary transcriptomic, proteomic, and single-cell analyses, as well as the development of a prognostic model, provides a comprehensive multiomics perspective on the relationship between systemic inflammation and HNSCC. Nevertheless, several limitations should be acknowledged. First, the genetic data used are restricted to European populations, which limits the generalizability of our findings to other ethnic groups. Second, while our discovery of CXCL11 withstood stringent FDR correction (PFDR = 0.047), many other associations were identified at the nominal significance level. The relatively modest sample size of the primary HNSCC GWAS cohort (2,497 cases and 2,928 controls) poses inherent constraints on statistical power. This limitation increases the potential for false-positive findings and the “winner’s curse”—where effect sizes may be overestimated—particularly when the high-dimensional space of 731 immune phenotypes and 91 inflammatory proteins is navigated. To mitigate the risk of Type I errors without incurring excessive Type II errors (false-negatives) or inflating effect sizes due to the winner’s curse, we categorized these as “high-confidence” findings only if they showed strict internal consistency across all sensitivity analyses and were supported by orthogonal multiomics evidence. While this “biological triangulation” significantly enhances the plausibility of our prioritized candidates, large-scale replication in independent, high-powered HNSCC cohorts remains essential to further substantiate putative causal associations. Third, while we strictly filtered instrumental variables on the basis of the F statistic (F > 10), the risk of weak instrument bias cannot be entirely ruled out. In a high-dimensional MR framework at this scale, such bias may still influence the stability of the results despite standard filtering. Furthermore, the prognostic model was based on tumour tissue gene expression data from the TCGA, which may not fully reflect the expression profiles of circulating proteins and immune cells in peripheral blood. Fourth, although multivariable MR (MVMR) is a robust approach to account for horizontal pleiotropy, its application to the entire dataset was not feasible because of the extreme dimensionality and potential collinearity among the 822 immune signatures. Consequently, while our mediation analyses revealed key inflammatory pathways, these specific mediating effects should be further refined in targeted MVMR models or functional experiments. Finally, while MR provides evidence consistent with causality, further functional intervention studies are needed to verify the molecular mechanisms through which factors such as CXCL11 and CD40 potentially modulate the HNSCC microenvironment. Conclusions This study suggests a complex and bidirectional relationship between HNSCC and systemic immune-inflammatory status, highlighting the systemic nature of this disease. Immune-inflammatory responses appear to contribute to HNSCC onset and progression and may serve as potential prognostic markers and therapeutic targets. Supplementary Information Supplementary material 1. Additional file 1. Case definitions and exclusion criteria in the included GWAS datasets Supplementary material 2. Additional file 2. Overview of GWAS data sources and instrumental variable selection criteria for the Mendelian randomization analyses Supplementary material 3. Additional file 3. Confounders removed in the Mendelian randomization Supplementary material 4. Additional file 4. Two-sample Mendelian randomization (MR) analysis of the causal effects of circulating inflammatory proteins on HNSCC. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 5. Additional file 5. Two-sample Mendelian randomisation (MR) analysis of the causal effects of HNSCC on circulating inflammatory proteins. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (green, p<5×10-7) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 6. Additional file 6. Two-sample Mendelian randomization (MR) analysis of the causal effects of immune cell phenotypes on HNSCC. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 7. Additional file 7. Two-sample Mendelian randomization (MR) analysis of the causal effects of HNSCC on immune cell phenotypes. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (green, p<5×10-7) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 8. Additional file 8. Examples of leave-one-out analysis. A) Leave-one-out analysis of TSMR analysis of CXCL11 expression on HNSCC at p<5×10-7. B) Leave-one-out analysis of TSMR analysis of HNSCC on CCL20 at p<5×10-7. C) Leave-one-out analysis of TSMR analysis of activated and resting Tregs AC on HNSCC at p<5×10-7. D) Leave-one-out analysis of TSMR analysis of HNSCC on naive-mature B cell AC at p<1×10-6. Supplementary material 9. Additional file 9. Strengthening the Reporting of Observational Studies in Epidemiology using the Mendelian Randomization (STROBE-MR) checklist. Supplementary material 10. Additional file 10. KEGG and GO enrichment analyses of significant MR results. A) GO analysis of MR significance results for the causal effect of inflammatory factors on HNSCC. B) KEGG analysis of MR significance results for the causal effect of inflammatory factors on HNSCC. C) GO analysis of MR significance results for the causal effect of HNSCC on inflammatory factors. D) KEGG analysis of MR significance results for the causal effect of HNSCC on inflammatory factors. Supplementary material 11. Additional file 11. Summary of Mendelian randomization (MR) analysis and multiple-testing correction results. p: the nominal p value derived from the inverse variance weighted method. PFDR: adjusted p value after Benjamini–Hochberg false discovery rate (FDR) correction. Supplementary material 12. Additional file 12. Validation results using the FinnGen R10 dataset in bidirectional Mendelian randomization. Supplementary material 13. Additional file 13. Pseudotime trajectory analysis of the scRNA-seq data. Supplementary material 14. Additional file 14. Single-cell trajectory analysis of whole blood samples from a mouse tongue squamous cell carcinoma model. A–B) Cell clustering and composition. C–E) Intercluster communication analysed by CellChat. F) Differential gene expression in tumour samples. Supplementary material 15. Additional file 15. Results of MR analysis between the pGWAS and HNSCC databases. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 16. Additional file 16. Proteomic analysis of high-confidence mediators: differential expression between HNSCC tumour tissues and adjacent normal tissues. Supplementary material 17. Additional file 17. Approval Letter by Medical Ethics Committee of Shaoxing Stomatological Hospital. Acknowledgements The participants and investigators of the original GWAS databases, TCGA databases, GEO databases, UKB-PPP databases and CPTAC are appreciated [9–13, 19]. Abbreviations Absolute Count B-Cell Receptor Bright Canonical Correlation Analysis C-C Motif Chemokine C-C Motif Chemokine 20 Conventional Dendritic Cells Confidence Interval Central Memory Clinical Proteomic Tumour Analysis Consortium Cystatin D C-X-C Motif Chemokine C-X-C Motif Chemokine 11 C-X-C Motif Chemokine Receptor 3 Dendritic Cells Double Negative European Bioinformatics Institute Effector Memory European Molecular Biology Laboratory Forward Scatter-Area Benjamini‒Hochberg False Discovery Rate Gene Expression Omnibus Gene Ontology Genome-Wide Association Studies Human Leukocyte Antigen Head and Neck Squamous Cell Carcinoma Human Papillomavirus Immunoglobulin D Interleukin-10 Receptor Subunit Beta Immature Myeloid-Derived Suppressor Cells Instrumental Variables Inverse Variance Weighted Kyoto Encyclopedia of Genes and Genomes Mitogen-Activated Protein Kinase Myeloid-Derived Suppressor Cells Mendelian Randomization Natural Killer Cells Natural Killer T Cells Odds Ratio Phosphate-Buffered Saline Principal Component Analysis Phenome-Wide Association Study Pharma Proteomics Project Proteome Genome-Wide Association Study RNA Sequencing Single-Cell RNA Sequencing Single-Nucleotide Polymorphism Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization T, B, NK Cells The Cancer Genome Atlas Terminal Effector T helper Cells Tumour Immune Estimation Resource Regulatory T Cells Two-Sample Mendelian Randomization UK Biobank UK Biobank Pharma Proteomics Project Uniform Manifold Approximation and Projection Author contributions Ke Yang: Conceptualization, Formal analysis, Methodology, Software, Visualization, Writing – original draft; Guixin Zhu: Data curation, Investigation, Methodology, Validation, Visualization, Writing – review and editing; Yanyan Li: Validation, Visualization, Writing – review and editing; Jie Huang: Data curation, Writing – review and editing; Yanfei Zhang: Data curation, Writing – review and editing; Fei Sun: Data curation, Writing – review and editing; Zhuomin Sha: Writing – review and editing; Yun Zhang: Writing – review and editing; Xuepeng Chen: Conceptualization, Funding acquisition, Supervision, Writing – review and editing. Funding This study was supported by the National Natural Science Foundation of China (Nos. 81400511 and 82020108011), the Zhejiang Provincial Natural Science Foundation of China (LY18H140001), and the R&D Program of the Stomatology Hospital of Zhejiang University School of Medicine (RD2022JCEL04 and RD2022JCYL10). Data availability All GWAS summary statistics used in this study are publicly available from the EMBL-EBI and FinnGen R10 databases. Single-cell RNA-seq datasets were acquired from the GEO databases. The pGWAS data was from the UK Biobank Pharma Proteomics Project (UKB-PPP). To ensure reproducibility, we have deposited the analytical R scripts and harmonized datasets in GitHub repository: https://github.com/KEERYREEK/Mendelian\_HNSCC\_IF\_IC. Declarations Ethics approval and consent to participate The animal experiments in this study were approved by the Hangzhou Institute of Medicine, Chinese Academy of Sciences (Approval No. AP2024-07-0088). In addition, all the datasets used were obtained from publicly available databases and complied with the Declaration of Helsinki, and the use of these data was approved by the Medical Ethics Committee of Shaoxing Stomatological Hospital (Approval No. SL202500701; Additional file 17). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Ke Yang, Guixin Zhu and Yanyan Li contributed equally to this work. References Associated Data Supplementary Materials Supplementary material 1. Additional file 1. Case definitions and exclusion criteria in the included GWAS datasets Supplementary material 2. Additional file 2. Overview of GWAS data sources and instrumental variable selection criteria for the Mendelian randomization analyses Supplementary material 3. Additional file 3. Confounders removed in the Mendelian randomization Supplementary material 4. Additional file 4. Two-sample Mendelian randomization (MR) analysis of the causal effects of circulating inflammatory proteins on HNSCC. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 5. Additional file 5. Two-sample Mendelian randomisation (MR) analysis of the causal effects of HNSCC on circulating inflammatory proteins. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (green, p<5×10-7) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 6. Additional file 6. Two-sample Mendelian randomization (MR) analysis of the causal effects of immune cell phenotypes on HNSCC. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 7. Additional file 7. Two-sample Mendelian randomization (MR) analysis of the causal effects of HNSCC on immune cell phenotypes. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (green, p<5×10-7) to increasingly lenient (purple, p<5×10-6) thresholds. The traits shown in bold denote associations that meet the pre-specified “high-confidence” criteria. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 8. Additional file 8. Examples of leave-one-out analysis. A) Leave-one-out analysis of TSMR analysis of CXCL11 expression on HNSCC at p<5×10-7. B) Leave-one-out analysis of TSMR analysis of HNSCC on CCL20 at p<5×10-7. C) Leave-one-out analysis of TSMR analysis of activated and resting Tregs AC on HNSCC at p<5×10-7. D) Leave-one-out analysis of TSMR analysis of HNSCC on naive-mature B cell AC at p<1×10-6. Supplementary material 9. Additional file 9. Strengthening the Reporting of Observational Studies in Epidemiology using the Mendelian Randomization (STROBE-MR) checklist. Supplementary material 10. Additional file 10. KEGG and GO enrichment analyses of significant MR results. A) GO analysis of MR significance results for the causal effect of inflammatory factors on HNSCC. B) KEGG analysis of MR significance results for the causal effect of inflammatory factors on HNSCC. C) GO analysis of MR significance results for the causal effect of HNSCC on inflammatory factors. D) KEGG analysis of MR significance results for the causal effect of HNSCC on inflammatory factors. Supplementary material 11. Additional file 11. Summary of Mendelian randomization (MR) analysis and multiple-testing correction results. p: the nominal p value derived from the inverse variance weighted method. PFDR: adjusted p value after Benjamini–Hochberg false discovery rate (FDR) correction. Supplementary material 12. Additional file 12. Validation results using the FinnGen R10 dataset in bidirectional Mendelian randomization. Supplementary material 13. Additional file 13. Pseudotime trajectory analysis of the scRNA-seq data. Supplementary material 14. Additional file 14. Single-cell trajectory analysis of whole blood samples from a mouse tongue squamous cell carcinoma model. A–B) Cell clustering and composition. C–E) Intercluster communication analysed by CellChat. F) Differential gene expression in tumour samples. Supplementary material 15. Additional file 15. Results of MR analysis between the pGWAS and HNSCC databases. The legend colours indicate different levels of instrumental variable (IV) selection stringency, progressing from more stringent thresholds (grey, p<5×10-8) to increasingly lenient (purple, p<5×10-6) thresholds. Het P represents the P value from Cochran’s Q test for heterogeneity; Ple P denotes the P value of the MR‒Egger intercept test for directional pleiotropy; MR-PRESSO P indicates the P value from the MR-PRESSO test. Supplementary material 16. Additional file 16. Proteomic analysis of high-confidence mediators: differential expression between HNSCC tumour tissues and adjacent normal tissues. Supplementary material 17. Additional file 17. Approval Letter by Medical Ethics Committee of Shaoxing Stomatological Hospital. Data Availability Statement All GWAS summary statistics used in this study are publicly available from the EMBL-EBI and FinnGen R10 databases. Single-cell RNA-seq datasets were acquired from the GEO databases. The pGWAS data was from the UK Biobank Pharma Proteomics Project (UKB-PPP). To ensure reproducibility, we have deposited the analytical R scripts and harmonized datasets in GitHub repository: https://github.com/KEERYREEK/Mendelian\_HNSCC\_IF\_IC.