mRNA hypertranscription is associated with poor prognosis and tumor progression in HNSCC

To investigate whether mRNA hypertranscription contributes to malignant phenotypes in HNSCC, we performed scRNA-seq on 12 surgical samples, resulting in 116,855 cells that passed stringent quality controls (Supplementary Table 1 and Supplementary Fig. 1a, b). Epithelial cell populations were identified on the basis of the expression of EpCAM, KRT8, and KRT18 (Supplementary Table 2, Supplementary Fig. 1c–f and Fig. 1a, b). From these epithelial populations, we identified and reclustered 18,798 aneuploid malignant cells by inferring genomic copy number variations via copyKAT analysis (Fig. 1c, d). To assess mRNA hypertranscription, we quantified UMI counts per cell, which reflect the number of captured mRNA molecules. After normalization to account for sequencing depth and technical variation using the method introduced by Cao et al.16, normalized total UMI counts serve as a reliable proxy for global transcriptional output at the single-cell level. Normalized total UMI counts were strongly positively correlated with total RNA abundance, raw UMI counts, and the number of genes detected per cell, supporting their use as robust proxies for global transcriptional output at single-cell resolution (Supplementary Fig. 2). Consistent with its global nature, hypertranscription was detectable across multiple cell types in the tumor microenvironment, while malignant epithelial cells exhibited the highest overall levels of transcriptional output (Supplementary Fig. 3). Accordingly, cells were stratified into low, medium, and high hypertranscription groups on the basis of tertile distribution (n = 6266 per group), revealing marked transcriptional heterogeneity across distinct clusters (Fig. 1e). Previous studies have suggested that the total number of genes expressed at the single-cell level may better predict cellular phenotype and developmental potential than individual gene markers or pathways.35 Building on these findings, we analyzed gene expression features across 18,798 tumor cells and reported a strong positive correlation between the number of expressed genes and normalized total UMI counts (R = 0.85, p < 0.0001; Fig. 1f), indicating that mRNA hypertranscription may be linked to cellular differentiation status. CytoTRACE analysis further demonstrated that compared with cells in the medium and low groups, cells in the high hypertranscription group had significantly higher scores, indicating a less differentiated state (p < 0.001; Fig. 1g). The differentially expressed genes in the high- hypertranscription group were significantly enriched in pathways related to protein synthesis (RPS23 and RPL35A), cytoskeletal organization (MYL12A/B and TMSB10), stress response (HSPE1, S100A6, and S100A11), and inflammation/immunity (LYZ and TGFBI; Fig. 1h). Consistently, GO and Reactome enrichment analyses revealed coordinated activation of developmental and SUMOylation-related pathways, reflecting transcriptionally and posttranscriptionally regulated programs likely orchestrated by context-specific transcription factors (Supplementary Fig. 4). Furthermore, we observed significant upregulation of the vast majority (48/50) of tumor-associated hallmark pathways, which correlated positively with increased mRNA hypertranscription levels in cancer cells (Supplementary Table 3 and Fig. 1i). To further validate the clinical relevance of tumor hypertranscription, we assessed its status in 210 HNSCC patients from the TCGA-HNSC cohort using the TmS metric (Supplementary Fig. 5).16 After additionally adjusting for HPV status and other established clinicopathological covariates, tumor hypertranscription remained an independent adverse prognostic factor for overall survival in patients with HNSCC (Supplementary Fig. 6). These findings reinforce the prognostic value of hypertranscription while extending prior work by incorporating HPV-related disease heterogeneity.16

Fig. 1

mRNA hypertranscription is associated with poor prognosis and tumor progression in HNSCC. a UMAP projection of all single cells from 12 HNSCC samples, annotated by major cell types. b Dot plot showing the average expression and detection frequency of canonical marker genes used for cell type identification. c Aneuploid and diploid cell prediction using copyKAT analysis. d UMAP of reclustered epithelial tumor cells (n = 18,798), defining 11 subpopulations (epi_0-epi_10). e mRNA hypertranscription status (low, medium, and high) overlaid on the tumor UMAP. f Scatter plot showing the correlation between the normalized total UMI count and the number of expressed genes. g Box plot comparing the CytoTRACE scores among the hypertranscription groups; higher scores reflect lower differentiation. Statistical significance was assessed using the Wilcoxon rank-sum test (all p < 0.001). h Heatmap of differentially expressed genes across mRNA hypertranscription groups. i Heatmap showing enrichment scores of the top 20 hallmark pathways across transcriptional states

SPI1 is a key transcription factor associated with mRNA hypertranscription in HNSCC

Our subsequent efforts focused on identifying TFs associated with HNSCC mRNA hypertranscription and demonstrating their correlation with this enhanced mRNA hypertranscriptional state. TF activity inference analysis identified SPI1 as the most highly activated TF in hypertranscriptional tumors (Fig. 2a) on the basis of a weighted-sum scoring approach leveraging TF-target annotations from the CollecTRI database, a comprehensive resource integrating curated regulatory interactions from 12 distinct sources.36 Notably, downstream target genes of SPI1 were predominantly upregulated in cells with high global mRNA transcriptional activity (Supplementary Fig. 7), suggesting a functional role for SPI1 in driving the hypertranscriptional phenotype. Consistent with its activation status, SPI1 expression was significantly elevated in tumor tissues compared with adjacent normal tissues in bulk RNA-seq data from the TCGA HNSCC cohort (Supplementary Fig. 8). Clinically, patients with high SPI1 activity had markedly poorer overall survival (Fig. 2b). Furthermore, after adjustment for clinicopathological variables, HPV status, and tumor microenvironment-related scores (stromal, immune, and ESTIMATE), high SPI1 activity remained independently associated with poor overall survival in patients with HNSCC (Fig. 2c).

Fig. 2

SPI1 is associated with the hypertranscriptional phenotype in HNSCC. a TF activity inference identifying SPI1 as the top candidate. Scatter plot of TF activity scores (weighted t statistic) versus significance. b Kaplan‒Meier analysis showing that patients with high SPI1 activity exhibit significantly worse overall survival. P values were calculated using the log-rank test. c Forest plot derived from an expanded multivariate Cox regression model. High SPI1 activity remained an independent predictor of poor prognosis after we adjusted for clinicopathological factors (age, sex, stage, smoking/alcohol history, tumor subsite, and HPV status) and tumor microenvironment characteristics (stromal, immune, and ESTIMATE scores). d UMAP plots of single cells illustrating graded SPI1 activity (left) and mRNA hypertranscription status (right). e Heatmap of representative differential genes across low-, medium-, and high-SPI1-activity groups, showing distinct gene expression modules associated with SPI1 activation. f Scatter plot showing a positive correlation between SPI1 activity and normalized total UMI counts (R = 0.42, p < 0.0001). g Violin plots of the SPI1 activity across tertile-defined low, medium, and high mRNA hypertranscription groups (n = 6266 cells per group; p < 0.0001). h Heatmap of the top 20 hallmark pathways enriched in cells with high SPI1 activity. i Rank-ordered distribution of TF activity t statistics from an independent single-cell HNSCC dataset (GSE234933)

To investigate the cellular underpinnings of this association, we evaluated SPI1 activity in our scRNA-seq dataset. SPI1 activity was positively correlated with mRNA expression across all major cell types (Supplementary Table 4). Given our primary interest in delineating the relationship between SPI1 activity and hypertranscription within malignant cells, all subsequent analyses were restricted to the aneuploid tumor cell population. UMAP projections revealed marked spatial colocalization between high SPI1 activity and mRNA hypertranscription (Fig. 2d). Differential expression analysis further revealed representative marker genes across low, medium, and high SPI1 activity states (Fig. 2e). Quantitatively, SPI1 activity displayed a strong positive correlation with normalized total UMI counts (R = 0.42; p < 0.0001; Fig. 2f) and increased progressively across hypertranscription tertiles (Fig. 2g), reinforcing its link to elevated transcriptional output at the single-cell level. Consistently, analysis of HNSCC cell lines from the Cancer Cell Line Encyclopedia (CCLE) revealed a strong positive correlation between SPI1 regulon-based activity and global transcriptional output (Pearson r = 0.706, P < 0.001; Supplementary Fig. 9). Transcriptomic profiling of cells stratified by SPI1 activity revealed that high-SPI1 cells exhibited activation of many of the same pathways observed in hypertranscriptional cells, especially those linked to RNA processing and nuclear receptor signaling (Supplementary Fig. 10). Notably, the AUCell scores of cancer hallmark pathways revealed that most hallmark pathways were significantly enriched in the high-SPI1-activity group (41/50, all adjusted p < 0.05; Fig. 2h; Supplementary Table 5), indicating that SPI1 is involved in widespread oncogenic reprogramming. To strengthen these findings, we analyzed an independent single-cell dataset comprising 187,399 cells from 51 HNSCC patients (Supplementary Fig. 11a, b). This external validation confirmed the correlation between SPI1 activity and mRNA hypertranscription (Supplementary Fig. 11c–e, Fig. 2i) and recapitulated the transcriptional enrichment of oncogenic pathways in cells with high SPI1 activity (Supplementary Fig. 11f, g). Together, these multilevel analyses establish SPI1 as a key transcriptional regulator associated with mRNA hypertranscription in HNSCC.

SPI1 is associated with the malignant phenotypes of HNSCC cells in vitro

To define the role of SPI1 in HNSCC progression, we systematically evaluated its expression, subcellular localization, and functional effects in a panel of representative HNSCC cell lines and the immortalized human keratinocyte line HaCaT. RT‒qPCR and western blot analyses demonstrated that SPI1 was abundantly expressed in CAL27 and CAL33 cells but remained low in SAS and FaDu cells (Fig. 3a, b and Supplementary Fig. 12). Immunofluorescence staining and nuclear–cytoplasmic fractionation followed by Western blot analysis demonstrated that SPI1 was predominantly localized to the nucleus, with a minor fraction detectable in the cytoplasm (Fig. 3c and Supplementary Fig. 13). We next performed loss-of-function experiments by generating stable SPI1 knockdown lines in CAL27 and CAL33 cells using shRNA. The knockdown efficiency was confirmed at both the transcript and protein levels (Fig. 3d and Supplementary Fig. 14). Functionally, SPI1 silencing significantly reduced cell proliferation (Fig. 3e) and impaired colony-forming ability (Fig. 3f, g). Consistent with the impaired proliferative capacity, the results of the EdU incorporation assays revealed a significant decrease in DNA synthesis following SPI1 knockdown, indicating reduced S-phase entry (Fig. 3h, i). Additionally, SPI1 knockdown markedly suppressed cell migration and invasion, as determined by Transwell assays (Fig. 3j–l). In parallel, western blot analysis revealed a reduction in RB phosphorylation without appreciable changes in total RB levels in SPI1 knockdown cells (Supplementary Fig. 15), which is consistent with attenuated cell cycle progression. Moreover, SPI1 depletion increased apoptosis, as assessed by Annexin V/7-AAD staining and TUNEL assays (Supplementary Figs. 16 and 17). To complement these findings, we overexpressed SPI1 in SAS and FaDu cells using lentiviral vectors (Fig. 3m and Supplementary Fig. 14), resulting in the opposite phenotype: increased proliferation (Fig. 3n), enhanced clonogenicity (Fig. 3o, p), and elevated motility and invasion (Fig. 3q, s). These comprehensive gain- and loss-of-function assays demonstrate that SPI1 plays a critical role in maintaining the malignant phenotype of HNSCC cells in vitro.

Fig. 3

SPI1 regulates proliferation, survival, migration, and invasion in HNSCC cells in vitro. a Representative Western blot analysis showing the SPI1 protein levels in the normal human keratinocyte cell line HaCaT and the head and neck squamous cell carcinoma cell lines SAS, FaDu, CAL27, and CAL33. β-actin was used as a loading control. b Quantification of SPI1 protein expression normalized to that of β-actin and expressed relative to that in HaCaT cells. Data are presented as the mean ± SD from independent experiments. Statistical significance was assessed using one-way ANOVA; P < 0.05 is indicated by an asterisk. c Immunofluorescence staining showing the nuclear localization of SPI1 in CAL27 and CAL33 cells; the right panels show line-scan intensity profiles. Scale bar, 20 μm. d Western blot confirmation of the SPI1 knockdown efficiency in CAL27 and CAL33 cells using two independent shRNAs. e Cell viability was measured by a CCK-8 assay at 72 h posttransduction in control and SPI1-knockdown cells (n = 3 independent replicates). Quantification (f) and representative images (g) of colony formation assays showing the clonogenic potential of CAL27 and CAL33 cells after SPI1 knockdown (n = 3 independent replicates). h, i EdU incorporation assay for cell proliferation. h Representative immunofluorescence images showing EdU incorporation in CAL27 and CAL33 cells transduced with shNC or SPI1-targeting shRNAs (scale bar = 50 μm); i quantification of EdU-positive cells (n = 3 independent replicates). j Representative images from Transwell migration and invasion assays in CAL27 and CAL33 cells with or without SPI1 knockdown (scale bar = 200 μm). Quantification of migrated and invaded CAL27 (k) and CAL33 (l) cells (n = 3 independent replicates). m Western blot showing SPI1 overexpression in SAS and FaDu cells. n CCK-8 assay showing enhanced proliferation in SPI1-overexpressing SAS and FaDu cells (n = 3 independent replicates). Colony formation assay results showing increased clonogenic capacity following SPI1 overexpression (o), with corresponding representative colony images shown in (p) (n = 3 independent replicates). q Representative transwell migration and invasion assay images of SPI1-overexpressing SAS and FaDu cells (scale bar = 200 μm). Quantification of migrated and invaded SAS (r) and FaDu (s) cells following SPI1 overexpression (n = 3 independent replicates). In (d–l), the shNC group indicates cells transduced with a lentiviral vector carrying nontargeting shRNA. In (m–s), the NC group indicates cells transduced with an empty lentiviral vector as the control for SPI1 overexpression. The data are presented as the mean ± SD. Statistical significance was assessed using one-way ANOVA followed by Tukey’s post hoc test for multiple group comparisons or two-tailed unpaired Student’s t test for two-group comparisons. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

SPI1 modulates HNSCC tumor growth and malignant phenotypes in vivo

To investigate the functional significance of SPI1 in vivo, we established subcutaneous xenograft models using CAL33 cells stably transduced with two independent SPI1-targeting shRNAs. Compared with shNC, SPI1 knockdown markedly suppressed tumor growth, as evidenced by reduced tumor volume over time and significantly decreased tumor weight at the experimental endpoint (Fig. 4a–c). Histological examination by H&E staining revealed a more dispersed and less compact tumor architecture in the SPI1-deficient xenografts (Fig. 4g). To specifically assess SPI1 expression in tumor epithelial cells, we performed multiplex immunofluorescence costaining for SPI1, EpCAM, and Ki-67. Quantitative analysis demonstrated a reduction in the proportion of SPI1⁺ cells among EpCAM⁺ tumor cells following SPI1 knockdown (Fig. 4d, g). Consistent with impaired tumor growth, multiplex immunofluorescence further revealed a significant decrease in the fraction of Ki-67⁺ cells within the EpCAM⁺ tumor cell compartment in the SPI1-knockdown xenografts (Fig. 4e, g), indicating attenuated proliferative activity. In parallel, TUNEL staining revealed a marked increase in the number of apoptotic cells in tumors lacking SPI1 expression (Fig. 4f, g), suggesting that tumor cell apoptosis increased upon depletion of SPI1. To validate these findings in an independent gain-of-function setting, we generated xenograft models using SAS and FaDu cells with stable SPI1 overexpression. In both models, compared with control tumors, tumors with increased SPI1 expression had significantly increased tumor volumes, elevated endpoint tumor weights, and accelerated tumor growth kinetics (Fig. 4h–m and Supplementary Fig. 18). Multiplex immunofluorescence staining revealed a robust increase in the proportion of SPI1⁺ cells within the EpCAM⁺ tumor cell compartment in SPI1-overexpressing SAS and FaDu xenografts (Fig. 4n–p). Concomitantly, SPI1 overexpression was associated with a significant increase in the fraction of Ki-67⁺ cells among EpCAM⁺ tumor cells (Fig. 4q–s), indicating enhanced proliferative activity in vivo. Taken together, these data demonstrate that SPI1 promotes HNSCC tumor growth and sustains malignant phenotypes in vivo by increasing tumor cell proliferation and survival.

Fig. 4: SPI1 modulates tumor growth, proliferation, and apoptosis in HNSCC xenograft models.

a Representative images of xenograft tumors derived from CAL33 cells transduced with nontargeting shRNA (shNC) or two independent SPI1-targeting shRNAs (n = 6 biologically independent mice per group). b Growth curves of CAL33 xenografts with or without SPI1 knockdown (n = 6 biologically independent mice per group). c Tumor weights at the endpoint in CAL33 xenografts (n = 6 biologically independent mice per group). d Quantification of the percentage of SPI1⁺ cells among EpCAM⁺ tumor cells in CAL33 xenografts on the basis of immunofluorescence analysis (n = 6 independent tumors). e Quantification of the percentage of Ki-67⁺ cells among EpCAM⁺ tumor cells in CAL33 xenografts on the basis of immunofluorescence analysis (n = 6 independent tumors). f Quantification of apoptotic cell percentages in CAL33 xenografts by TUNEL staining (n = 6 independent tumors). g Representative histological and immunofluorescence images of CAL33 xenograft tumors. Hematoxylin and eosin (H&E) staining is shown in the top row. Immunofluorescence images show costaining of SPI1 (magenta) and EpCAM (yellow), Ki-67 (green) and EpCAM (yellow) with DAPI (blue), and TUNEL-positive apoptotic cells (green) with DAPI (blue). SPI1/EPCAM and Ki-67/EPCAM images were derived from the same multiplex-stained section and identical field, with channels displayed separately for visualization. Scale bar, 100 μm. h Representative images of xenograft tumors derived from SAS cells with SPI1 overexpression or control cells (n = 6 biologically independent mice per group). i Tumor weights at the endpoint in SAS xenografts (n = 6 biologically independent mice per group). j Growth curves of SAS xenografts with or without SPI1 overexpression (n = 6 biologically independent mice per group). k Representative images of xenograft tumors derived from FaDu cells with SPI1 overexpression or control cells (n = 6 biologically independent mice per group). l Tumor weights at the endpoint in the FaDu xenografts (n = 6 biologically independent mice per group). m Growth curves of FaDu xenografts with or without SPI1 overexpression (n = 6 biologically independent mice per group). Quantification of the percentage of SPI1⁺ cells among EpCAM⁺ tumor cells in SAS (n) and FaDu (o) xenografts on the basis of immunofluorescence analysis (n = 6 independent tumors). Representative immunofluorescence images showing the costaining of SPI1 (magenta) with EpCAM (yellow) and Ki-67 (green) with EpCAM (yellow) in SAS (p) and FaDu (q) xenografts; DAPI (blue). SPI1/EPCAM and Ki-67/EPCAM images were derived from the same multiplex-stained section and identical field, with channels displayed separately for visualization. Scale bar, 100 μm. Quantification of the percentage of Ki-67⁺ cells among EpCAM⁺ tumor cells in SAS (r) and FaDu (s) xenografts on the basis of immunofluorescence analysis (n = 6 independent tumors). In (a–g), the shNC group indicates cells transduced with a lentiviral vector carrying nontargeting shRNA. In (h–s), the NC group indicates cells transduced with an empty lentiviral vector as the control for SPI1 overexpression. In (b, j, m), the data are presented as the mean ± SEM. Statistical significance was determined using one-way ANOVA followed by Tukey’s post hoc test for multiple group comparisons or two-tailed unpaired Student’s t test for two-group comparisons. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; ns, not significant

RNA sequencing reveals the molecular pathways regulated by SPI1 in HNSCC

To elucidate the transcriptional programs regulated by SPI1 in HNSCC cells, we performed bulk RNA sequencing followed by TF activity inference and pathway enrichment analyses in both the SPI1 knockdown and overexpression models. In the loss-of-function setting, SPI1 knockdown in CAL33 cells led to a significant reduction in inferred SPI1 TF activity (Fig. 5a) and coordinated remodeling of the TF activity landscape (Fig. 5b). At the pathway level, gene set enrichment analysis (GSEA) revealed a broad attenuation of oncogenic and inflammation-associated transcriptional programs in SPI1-deficient cells, indicating global suppression of protumorigenic signaling upon loss of SPI1 expression (Supplementary Fig. 19). In the gain-of-function setting, enforced SPI1 expression in SAS cells resulted in a marked increase in SPI1 TF activity (Fig. 5c). Consistent with this regulatory activation, SPI1 overexpression was accompanied by coordinated upregulation of epithelial–mesenchymal transition (EMT) and invasion-associated genes, as reflected by increased expression of representative mesenchymal markers (Fig. 5d). Hallmark pathway analysis further demonstrated that SPI1 overexpression activated multiple oncogenic signaling programs related to inflammation, growth signaling, and cellular stress responses (Fig. 5e). Complementary functional enrichment analyses of SPI1-induced genes revealed pathways involved in cytoskeletal remodeling and cell motility, which is consistent with the enhanced invasive potential (Supplementary Fig. 20). Together, these results demonstrate that SPI1 acts as a central transcriptional regulator in HNSCC cells, suppressing oncogenic programs upon depletion while promoting EMT-associated transcriptional reprogramming and malignant signaling upon overexpression.

Fig. 5

Transcriptional reprogramming and phenotypic alterations driven by SPI1 perturbation in HNSCC cells. a Violin plot showing inferred SPI1 TF activity in CAL33 cells transduced with shNC or shSPI1 #1. b Heatmap displaying the top differentially active TFs between shNC and shSPI1 #1 inferred by decoupleR. The red arrow highlights the specific downregulation of SPI1 activity. c Violin plot confirming significantly increased SPI1 TF activity in SAS cells upon SPI1 overexpression (OE) compared with that in the negative control (NC). d Heatmap showing the expression profiles of key EMT- and invasion-related markers (e.g., MMP9, VIM, and CDH2) in SAS-NC versus SAS-OE cells. The side bar indicates the log2-fold change (log2FC). e Gene set enrichment analysis (GSEA) of cancer hallmark pathways in SPI1-overexpressing cells. Red circles indicate upregulated pathways, while blue circles indicate downregulated pathways. Dot size represents statistical significance (−log10 FDR). *, P < 0.05

SPI1 promotes mRNA hypertranscription in HNSCC

To directly assess whether SPI1 modulates global transcriptional activity in HNSCC, we employed two complementary strategies: 5-ethynyl uridine (EU) incorporation to measure nascent RNA synthesis and quantification of total mRNA content normalized to cell number. The EU is a uridine analog that labels all newly transcribed RNA, providing a sensitive readout of global transcription.37 Furthermore, to specifically evaluate mRNA output independent of noncoding RNA contributions, we further performed poly(A)-selected mRNA purification followed by quantification adjusted for cell count. In CAL27 and CAL33 cells, stable knockdown of SPI1 led to a marked reduction in EU fluorescence intensity (Fig. 6a, b) and a significant decrease in per-cell mRNA abundance (Fig. 6c), indicating impaired transcriptional output. Conversely, SPI1 overexpression in SAS and FaDu cells significantly increased EU incorporation and mRNA levels per cell (Fig. 6d–f), establishing that SPI1 is necessary to drive bulk mRNA synthesis.

Fig. 6

SPI1 promotes mRNA hypertranscription in HNSCC through a time-dependent mechanism. Representative images (a) and quantification (b) of EU incorporation in CAL27 and CAL33 cells after stable knockdown of SPI1 using two independent shRNAs (n = 3 independent replicates). Red: EU-labeled nascent RNA; blue: DAPI-stained nuclei. Scale bar, 100 μm. c Relative per-cell mRNA yield in CAL27 and CAL33 cells, determined after mRNA purification and normalized by cell number (n = 3 independent replicates). Representative EU-stained images (d) and quantification (e) of SAS and FaDu cells with or without SPI1 overexpression (n = 3 independent replicates). f Relative per-cell mRNA yield in SPI1-overexpressing SAS and FaDu cells, based on poly(A)-selected RNA quantification (n = 3 independent replicates). g Time-course RT‒qPCR analysis of SPI1 expression in Dox-inducible Tet-ON SAS and FaDu cells following doxycycline treatment (n = 3 independent replicates). The data were normalized to that of GAPDH and expressed relative to 0 h. h Immunoblot analysis of the SPI1 and c-MYC protein levels. Representative images (i) and quantification of the fluorescence intensity (j) of cells with incorporated EU at the indicated time points after Dox induction (n = 3 independent replicates). The data are presented as the mean ± SD. Scale bars, 50 μm. In (a–c), the shNC group indicates cells transduced with a lentiviral vector carrying nontargeting shRNA. In (d–f), the NC group indicates cells transduced with an empty lentiviral vector as the control for SPI1 overexpression. Statistical significance was determined using one-way ANOVA followed by Tukey’s post hoc test or an unpaired two-tailed Student’s t test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

To define the kinetics of this transcriptional amplification, we developed a doxycycline-inducible Tet-ON system in SAS and FaDu cells. Upon Dox exposure, the SPI1 mRNA and protein levels increased in a time-dependent manner, peaking at 24 h (Fig. 6g, h). EU fluorescence likewise increased progressively, with a detectable upregulation beginning at 3–6 h and peaking at 24 h (Fig. 6i, j), confirming the time-resolved activation of transcription. To rule out potential leakiness inherent to doxycycline-inducible systems, we further implemented a Shield-1-regulated destabilization domain (DD)-SPI1 system to achieve tighter posttranslational control of SPI1 protein abundance. Shield-1 treatment resulted in time-dependent stabilization of the DD-SPI1 protein and a concordant increase in EU incorporation in SAS and FaDu cells (Supplementary Fig. 21), independently corroborating that acute SPI1 induction is sufficient to increase globax bh transcriptional output. Notably, although MYC target gene sets were enriched upon SPI1 overexpression, MYC protein levels were not elevated under these conditions (Fig. 6h), suggesting the activation of MYC-associated transcriptional programs without direct MYC upregulation.

ChIP and RNA-seq analyses revealed that SPI1 widely activates the cell transcriptional program

Given that SPI1 is an important transcription factor in cellular development, we hypothesized that SPI1 may promote global mRNA output in HNSCC through its intrinsic transcriptional regulatory activity. To directly map SPI1 chromatin engagement, we performed ChIP-seq in SPI1-overexpressing SAS cells using two independent ChIP-seq replicates with matched input controls. Across two independent SPI1-OE ChIP-seq replicates, compared with their corresponding input controls, immunoprecipitated (IP) samples presented substantially higher FRiP scores, indicating robust and reproducible SPI1 chromatin enrichment and overall high ChIP-seq data quality (Supplementary Table 6). ChIP analysis of SAS-overexpressing cells revealed that SPI1 binding sites were highly enriched around transcription start sites (TSSs), with symmetric peak signals within ±1.5 kb flanking regions, indicating preferential promoter occupancy (Fig. 7a, b and Supplementary Fig. 22). SPI1 binding events were evenly distributed across all chromosomes without overt regional bias (Fig. 7c). Among these genes, SPI1 directly binds to the transcription sites of 7250 protein-coding genes, indicating its important role in inducing the transcription program in HNSCC cells (Supplementary Table 7). To define SPI1-dependent transcriptional programs, we conducted RNA sequencing of SAS cells with enforced SPI1 expression versus vector controls. Differential expression analysis revealed that SPI1 overexpression upregulated a large set of genes (Fig. 7d). GSEA demonstrated significant enrichment of hallmark oncogenic pathways, prominently hypoxia and glycolysis, in SPI1-overexpressing cells (Fig. 7e and Supplementary Table 8). Integration of the RNA-seq and ChIP-seq data revealed 2154 direct target genes, highlighting its role as a direct transcriptional activator (Fig. 7f–h). To summarize network-level organization, we constructed a protein‒protein interaction (PPI) network from the SPI1-regulated gene set and performed module detection with functional enrichment; these module-level results are provided in the Supplementary Information (Supplementary Fig. 23 and Supplementary Table 9). To validate the transcriptional activity of SPI1, we also performed a dual-luciferase reporter assay, which revealed time-dependent increases in the firefly/Renilla ratio upon doxycycline-induced SPI1 expression in SAS cells, confirming its activation via its cognate motif (Fig. 7i). Together, these results identify SPI1 as a broad transcriptional activator that drives pro-oncogenic programs in HNSCC.

Fig. 7

Integrative transcriptomic and epigenomic analyses of SPI1 function in HNSCC. a Heatmaps and averaged ChIP-seq signal profiles showing SPI1 occupancy around transcription start sites (TSSs) in SPI1-overexpressing SAS cells, with SPI1 ChIP-seq Rep1 and its corresponding input control shown separately. b Aggregated metagene profiles illustrating normalized SPI1 ChIP-seq read density relative to the TSS across two independent SPI1-OE ChIP replicates and matched input samples. c Circos plot showing the genome-wide chromosomal distribution of SPI1 ChIP-seq binding peaks. d Volcano plot of DEGs identified by RNA-seq following SPI1 overexpression. DEGs were defined using an FDR < 0.05 and a |log2FC| > 1. e Representative gene set enrichment analysis (GSEA) plots showing hallmark pathways enriched in SPI1-overexpressing cells. f Venn diagram depicting the overlap between genes upregulated upon SPI1 overexpression and genes annotated with SPI1 ChIP-seq binding peaks. g Bubble plot showing Gene Ontology (GO) biological process enrichment of genes coregulated by SPI1 transcriptional activation and direct chromatin binding; the dot size indicates the gene count, and the color represents the adjusted P value. h De novo motif analysis of SPI1 ChIP-seq binding peaks. i Dual-luciferase reporter assay measuring SPI1-dependent transcriptional activity in a doxycycline-inducible system over time (n = 3 independent experiments; mean ± SD; one-way ANOVA with Tukey’s post hoc test)

Elevated SPI1 expression is correlated with poor prognosis in clinical HNSCC cohorts

To assess the clinical significance of SPI1, we performed IHC staining in two independent HNSCC patient cohorts from West China Hospital. In the laryngeal cancer cohort (n = 90; Supplementary Table 10), SPI1 protein levels were significantly higher in tumor tissues than in adjacent nontumor epithelium, as measured by the IRS (Fig. 8a). Kaplan‒Meier survival analysis revealed that high SPI1 expression was associated with poor overall survival (Fig. 8b). Notably, multivariate Cox regression analysis demonstrated that SPI1 remained a significant independent prognostic factor after adjusting for clinical stage, indicating that its prognostic impact was not confounded by tumor burden (Fig. 8c). Similar patterns were observed in the hypopharyngeal cancer cohort (n = 81; Supplementary Table 11 and Supplementary Fig. 24). SPI1 protein levels were again significantly elevated in tumor tissues (Fig. 8d), and higher expression correlated with poorer survival outcomes (Fig. 8e). Multivariate Cox regression further confirmed the independent prognostic value of the SPI1 in this cohort (Fig. 8f). To explore whether this expression pattern extends beyond laryngeal and hypopharyngeal cancers, we additionally assessed SPI1 expression in a small cohort of oropharyngeal cancer specimens, including HPV-positive cases. SPI1 immunoreactivity was significantly greater in tumor tissues than in peritumoral tissues and tended to be consistent with that observed in laryngeal and hypopharyngeal cancers (Supplementary Fig. 25). Together, these findings establish SPI1 as a clinically relevant biomarker of poor prognosis in HNSCC patients, with potential utility for risk stratification across subtypes.

Fig. 8

Elevated SPI1 expression is associated with poor prognosis in laryngeal and hypopharyngeal cancer patients. a Validation of SPI1 expression in a local cohort of laryngeal cancer patients. b Kaplan‒Meier survival curve and c multivariate Cox forest plot for laryngeal cancer patients. d Validation of SPI1 expression in a local cohort of hypopharyngeal cancer patients. e Kaplan‒Meier survival curve and f multivariate Cox forest plot for hypopharyngeal cancer patients. For Kaplan‒Meier analyses (b, e), follow-up was truncated at 1000 days to ensure stable risk set sizes and avoid overinterpretation of sparse late follow-up