Genomic analysis identifies TJP3 as a prognostic marker for radiotherapy and chemoradiotherapy in oral squamous cell carcinoma

Abstract

This study analyzed somatic mutations and differentially expressed genes in primary oral squamous cell carcinoma (OSCC) tumors to identify prognostic markers for radiotherapy (RT) and concurrent chemo-radiation therapy (CCRT). We used longitudinal whole-exome sequencing (WES) data from a patient who developed radiation-induced sarcoma (RIS) after CCRT with public datasets (TCGA and GSE41613) to identify genes associated with 24-month overall survival (OS). Candidates were further refined using Cox regression for OS. Twenty-two prognostic genes were identified and validated in WES data, functioning as indicators of RT and CCRT resistance. In patients with poor outcomes within 24 months post-RT, CLTCL1, RGL3 and H3C12 showed significant prognostic value in multivariate Cox analyses, while SLC35G5, GARS1 and JPH3 were identified as prognostic markers for CCRT. Limited overlap was observed between RT and CCRT, with TJP3 being the only common gene. RNA-seq data from the GSE217645 revealed that TJP3 exhibited significantly altered expression between radiosensitive and radioresistant cell lines (p = 0.037). RT-related genes were predominantly found in primary tumors, whereas CCRT-related genes were distributed across primary tumor, inflamed granulation tissue, inflammatory myofibroblastic proliferation and RIS. These findings provide insights into molecular mechanisms underlying RT and CCRT resistance, supporting future directions in precision medicine.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-05025-w.

Introduction

Oral squamous cell carcinoma (OSCC) is one of the ten most common cancers worldwide and a major cause of mortality from oral diseases. The main treatment for OSCC is surgical resection, often followed by adjuvant treatments. The indication for adjuvant treatment, such as postoperative radiotherapy (PO RT) with or without adjunctive chemotherapy, is determined by histopathological data of resected tissue, including status of margins, vascular or neural invasion, bone invasion, extracapsular extension, and pTNM staging¹. RT plays a pivotal role in treating OSCC, especially for patients with high-risk pathological features. In such cases, PO RT or postoperative concurrent chemoradiation therapy (PO CCRT) has been shown to improve overall survival. Initially used for inoperable cases, CCRT is now commonly applied to high-risk patients after surgery. Additionally, CCRT can serve as a primary treatment option to preserve organ function².

However, in a retrospective study involving 191 patients with stage III-IVb OSCC treated with PO RT or CCRT, disease recurrence was observed in 75 patients (39%). The 5-year disease-free survival rate was 52%, indicating that 48% of the patients experienced disease recurrence within 5 years³. The variable response to RT or CCRT among patients highlights the need for precision medicine to optimize treatment strategies. Precision medicine, often based on a patient’s genetic profile, enables targeted therapies tailored to individual patients, maximizing efficacy while minimize adverse effects⁴. Therefore, identifying effective biomarkers to predict treatment outcome is important.

The mechanisms underlying therapeutic resistance and cancer recurrence after RT or CCRT are the important parts of precision medicine. Contributing factors include cancer cell radioresistance, insufficient radiation dose, acquired drug resistance, advanced disease stage at diagnosis and influence from the tumor microenvironement⁵. In non-small cell lung cancer, patients with mutations in radiation resistance genes (KEAP1, NFE2L2, STK11 and PIK3CA) exhibited significantly higher locoregional failure rates after PO RT, suggesting the need for alternative oncological approaches⁶. In head and neck squamous cell carcinoma (HNSCC), radiation-resistant cells demonstrated increased DNA methylation compared to radiation-sensitive cells. Pathways related to differentially methylated and expressed genes in radioresistant versus radiosensitive HNSCC tumors of The Cancer Genome Atlas (TCGA) included PAK signaling, IL-6, 9, 10, 12, 15 signaling and LXR/RXR activation⁷. Four gene methylation signature ( ZNF10, TMPRSS12, ERGIC2, and RNF215) and five hub survival genes (POU4F1, IL34, HLF, CBS, and RNF165) were also associated with RT outcomes in TCGA HNSCC^8,9. However, these studies did not distinguish between RT and CCRT, which may involve different biological mechanisms.

This study aimed to identify somatic mutations and differentially expressed genes in primary OSCC tumors associated with prognosis following RT or CCRT. We compared somatic mutation profiles between patients with favorable and poor outcomes based on 24-month overall survival (OS), and further refined candidate prognostic genes using data from TCGA and GSE41613. One identified gene, TJP3, was validated using public in vitro dataset (GSE217645). In addition, longitudinal whole exome sequencing (WES) was performed on a patient who developed radiation induced sarcoma (RIS) after CCRT to investigate temporal molecular alterations associated with radiation resistance.

Results

Patient characteristics

To evaluate the impact of genetic mutations on the efficacy of RT and CCRT in patients with OSCC, we analyzed DNA mutations in a patient diagnosed with OSCC in September 2020. This patient received CCRT, was subsequently diagnosed with RIS in February 2024, and later underwent surgical resection. To assess overall survival (OS) in relation to RT- and CCRT-associated genes, univariate and multivariate Cox regression analyses were performed using OSCC cohorts from TCGA and GSE41613 (Tables S1 and S2). After excluding cases with incomplete data, multivariate analysis of 188 TCGA patients identified AJCC pathological stage, lymphovascular invasion, perineural invasion, and treatment modality as independent prognostic factors (Table S3). To further validate prognostically significant genes associated with both RT and CCRT, RNA-seq data from the GSE217645 dataset were analyzed. This dataset includes three replicates each of radiosensitive HNSCC cell lines and their respective isogenic radioresistant counterparts.

Prognostic DNA mutations and RNA expression profiles associated with radiotherapy response

Among 297 OSCC patients in the TCGA cohort, 95 received postoperative RT, of whom 26 experienced poor OS within 24 months. Fisher’s exact tests were performed to compare patients with poor versus favorable outcomes, followed by univariate Cox regression analysis on prognosis-associated genes (p < 0.05, n = 95; Table S4). Six genes harboring DNA mutations—CLTCL1, NFATC1, TJP3, ARHGEF19, ENPP7, and RGL3—were significantly associated with OS. Their mutation profiles are shown in Fig. 1a. Among them, NFATC1, TJP3, and ARHGEF19 remained statistically significant in multivariate Cox analyses of DNA mutations. Additionally, ENPP7 and TJP3 showed significant associations in multivariate Cox analyses of RNA expression using an independent validation dataset (Fig. S1). Notably, CLTCL1 and RGL3 demonstrated consistent prognostic significance across both DNA and RNA levels in multivariate analyses within the TCGA cohort (Table 1).

Fig. 1.

(a) A comparative analysis using Fisher’s exact test (p < 0.05) was performed on somatic DNA mutations in 26 patients with poor overall survival (OS) within 24 months following surgery and radiotherapy (RT), compared to 69 patients with more favorable outcomes. Univariate Cox regression analysis of OS identified six significantly prognostic genes (Table S4). The mutation patterns of these genes were visualized in patients who received RT. Among them, NFATC1, TJP3, ARHGEF19, and ENPP7 remained significant in multivariate Cox proportional hazards models after adjusting for other prognostic variables. Additionally, CLTCL1 and RGL3 showed consistent prognostic significance in both DNA and RNA multivariate Cox analyses (Table 1). (b) RNA differential expression analysis (q < 0.05, |log₂FC| > 1) was conducted in the same cohort of post-surgical RT patients. Subsequently, two genes, CES5A and H3C12,were identified as significantly associated with OS in univariate Cox regression analysis (Table S4). Among them, H3C12 demonstrated significance in both RNA expression and DNA mutation levels in multivariate Cox models (Table 1). MROH9 also showed a significant association with poor prognosis after adjustment for clinical covariates. Furthermore, CES5A and DEFB1 exhibited prognostic significance in independent validation cohorts of OSCC.

Table 1.

Multivariate cox hazard regression models of prognostically significant DNA and RNA genes associated with radiotherapy and chemo-radiotherapy.

	CLTCL1a		RGL3a
Variable	DNA	RNA	DNA	RNA
Gene	3.2(0.001b)	1.5(0.08)	2.6(0.01b)	0.3(0.05)
Age	1(0.1)	1(0.04b)	1(0.15)	1(0.04b)
AJCC pathologic stage
Stage 1 and 2 vs. 3 and 4	2.8(0.01b)	2.7(0.01b)	2.6(0.02b)	2.5(0.02b)
Lymphovascular invasion
Absence vs. presence	1.5(0.08)	1.6(0.07)	1.7(0.04b)	1.7(0.03b)
Perineural invasion
Absence vs. presence	1.9(0.01b)	1.8(0.02b)	1.9(0.01b)	1.7(0.04b)
Treatment modality
Surgery vs. surgery + RT	0.5(0.02b)	0.6(0.06)	0.5(0.03b)	0.6(0.1)
Surgery vs. surgery + CCRT	0.7(0.18)	0.7(0.19)	0.7(0.2)	0.7(0.32)

	H3C12a		JPH3a
Variable	DNA	RNA	DNA	RNA
Gene	18.5(0.001b)	2.3(0.001b)	3.2(0.01b)	0.4(0.02b)
Age	1(0.1)	1(0.13)	1(0.08)	1(0.02b)
AJCC pathologic stage
Stage 1 and 2 vs. 3 and 4	2.8(0.01b)	2.1(0.05)	2.9(0.01b)	2.6(0.01b)
Lymphovascular invasion
Absence vs. presence	1.6(0.06)	1.6(0.06)	1.6(0.07)	1.7(0.03b)
Perineural invasion
Absence vs. presence	1.8(0.02b)	2.3(0.002b)	2(0.01b)	1.8(0.02b)
Treatment modality
Surgery vs. surgery + RT	0.5(0.03b)	0.5(0.02b)	0.5(0.02b)	0.6(0.1)
Surgery vs. surgery + CCRT	0.7(0.19)	0.6(0.1)	0.6(0.12)	0.9(0.61)

	SLC35G5a		GARS1a
Variable	DNA	RNA	DNA	RNA
Gene	2.2(0.001b)	0.5(0.048b)	2.6(0.01b)	1.5(0.06)
Age	1(0.14)	1(0.05)	1(0.12)	1(0.04b)
AJCC pathologic stage
Stage 1 and 2 vs. 3 and 4	2.6(0.01b)	2.3(0.03b)	2.7(0.01b)	2.6(0.01b)
Lymphovascular invasion
Absence vs. presence	1.7(0.03b)	1.5(0.08)	1.7(0.04b)	1.7(0.04b)
Perineural invasion
Absence vs. presence	2.1(0.003b)	1.9(0.01b)	1.7(0.03b)	1.8(0.02b)
Treatment modality
Surgery vs. surgery + RT	0.6(0.05)	0.6(0.12)	0.5(0.01b)	0.5(0.04b)
Surgery vs. surgery + CCRT	0.7(0.23)	0.7(0.19)	0.6(0.12)	0.7(0.21)

All six genes are reported as hazard ratios (p-values) in multivariate Cox analyses, adjusted for other prognostic factors (n = 188 and 185, respectively).

AJCC American Joint Committee on Cancer, RT radiotherapy, CCRT concurrent chemotherapy and radiotherapy, HR hazard ratio.

^a The statistical significance of fitted model of Cox proportional hazard ratio was calculated by likelihood ratio test (p < 0.00001).

^b p < 0.05, multivariate Cox regression analysis.

Among the 95 patients who received RT, differentially expressed genes (DEGs) based on RNA expression (FDR q < 0.05, |log₂FC| > 1) and significantly associated with OS in univariate Cox analyses (p < 0.05, n = 92; Table S4) included MROH9, H3C12, CES5A, and DEFB1 (Fig. 1b). The MROH9 DNA mutation was also significant in multivariate Cox analysis. Among the DEGs, CES5A and DEFB1 showed significant associations with OS in multivariate Cox analyses using the validation dataset (Fig. 1b). Additionally, H3C12 showed consistent prognostic significance across both DNA and RNA levels in multivariate Cox analyses within the TCGA cohort (Table 1).

All six DNA and four RNA prognostic genes associated with poor outcomes following RT were also detected in the somatic mutation profile of the OSCC patient who later developed RIS (Fig. 2a). In this case, CLTCL1 appeared as a splice donor variant, ARHGEF19 as an upstream variant, and ENPP7 and H3C12 as missense variants. In inflamed granulation tissue, RGL3 exhibited a missense variant and NFATC1 an intronic variant. Additionally, DEFB1 was identified as an upstream variant in this RIS case. Among the DNA mutation-related pathways associated with poor prognosis within 24 months post-RT, only the TGF-β signaling pathway showed statistically significant enrichment (Fig. 2b, Table S5).

Fig. 2.

(a) In the TCGA cohort of oral squamous cell carcinoma (OSCC) patients, six significant DNA mutations and four prognostic RNA expression markers were identified following radiotherapy. These ten genes exhibited a distribution pattern consistent with somatic mutations observed in radiation-induced sarcoma cases from the present study, resembling a Venn diagram. Notably, one patient diagnosed with OSCC in September 2020, who received postoperative chemo-radiotherapy and later developed radiation-induced sarcoma in February 2024, also exhibited inflamed granulation tissue on biopsy in December 2021. Additionally, protein-level mutations corresponding to the six significant DNA-altered genes were visualized. (b) A pathway-level analysis of DNA mutations was conducted in the TCGA OSCC cohort, comparing 26 patients with poor overall survival (OS) outcomes and 69 with favorable outcomes within 24 months post-radiotherapy. The analysis revealed significant enrichment of mutations in the TGF-β signaling pathway (p = 0.042; Table S5). Specifically, TGF-β pathway mutations were detected in 46% (12/26) of patients with poor OS outcomes, compared to 23% (16/69) of those with favorable outcomes.

Prognostic DNA mutations and RNA expression profiles associated with chemo-radiotherapy response

Among the 297 patients, 83 received postoperative CCRT, of whom 32 experienced poor OS within 24 months. Fisher’s exact tests were used to compare patients with poor versus favorable outcomes, followed by univariate Cox regression analyses of significantly associated genes (p < 0.05, n = 83; Table S6). Eight genes—TJP3, ANO8, EVPL, MAP3K2, RAB2A, SLC35G5, GARS1, and WDR59—were significantly associated with OS, with their DNA mutation profiles shown in Fig. 3a. Among these, TJP3, ANO8, EVPL, MAP3K2, and RAB2A remained significant in multivariate Cox analyses. In addition, SLC35G5 and GARS1 showed prognostic significance at both the DNA and RNA levels in multivariate Cox models (Table 1). TJP3, EVPL, WDR59, and MAP3K2 were also significantly associated with OS in multivariate RNA analyses in the validation cohort (Fig. S1).

Fig. 3.

(a) Fisher’s exact test (p < 0.05) was employed to identify significant somatic DNA mutations among 32 oral squamous cell carcinoma (OSCC) patients with poor overall survival (OS) outcomes within 24 months following chemo-radiotherapy and 51 patients with more favorable outcomes. Eight prognostically significant genes were subsequently identified through univariate Cox regression analyses of OS (Table S6). Visualization of DNA mutation patterns in post-chemo-radiotherapy patients revealed that ANO8, EVPL, MAP3K2, and RAB2A remained significant in multivariate Cox proportional hazards models after adjusting for other clinical covariates. Additionally, SLC35G5 and GARS1 demonstrated consistent prognostic significance in both DNA and RNA multivariate Cox analyses (Table 1). (b) Differential RNA expression analysis (q < 0.05, |log₂FC| > 1) was performed using the same cohort of post-surgical chemo-radiotherapy patients. Among these, four genes with significant associations with OS were identified through univariate Cox regression analyses (Table S6). PPME1 and CAPN8 were found to be significant in multivariate Cox proportional hazards models after adjustment for other prognostic factors. JPH3 showed significance at both the RNA expression and DNA mutation levels in multivariate Cox analyses (Table 1). Furthermore, JPH3 and EMID1 were validated as significant prognostic biomarkers in independent OSCC datasets.

DEGs significantly associated with OS (FDR q < 0.05, |log₂FC| > 1; p < 0.05, n = 83; Table S6) included CAPN8, PPME1, DPYSL5, JPH3, and EMID1 (Fig. 3b). Among these, PPME1, CAPN8, and DPYSL5 retained significance in both univariate and multivariate Cox models. JPH3 was significantly associated with OS at both DNA and RNA levels (Table 1), while JPH3 and EMID1 remained significant in multivariate RNA analyses in the validation cohort (Fig. 3b).

All eight DNA and five RNA genes associated with poor outcomes following CCRT were consistently observed in the somatic mutation profile of the RIS-developing OSCC patient (Fig. 4a). Specifically, SLC35G5 and JPH3 were detected as missense variants; ANO8 as both missense and synonymous; CAPN8 as a downstream variant; and PPME1 as synonymous. Additional genes appeared as intronic variants. In inflamed granulation tissue, GARS1 was detected as a splice donor variant; MAP3K2 as a missense variant; and CAPN8 and JPH3 as downstream variants. In inflammatory myofibroblastic tissue, WDR59 was identified as both stop-gained and synonymous, JPH3 again as downstream, EVPL as synonymous, and DPYSL5 as intronic.

Fig. 4.

(a) Eight DNA mutations and five RNA expression markers with significant prognostic relevance were identified in the TCGA oral squamous cell carcinoma (OSCC) cohort following chemo-radiotherapy. These 13 genes exhibited distribution patterns similar to those observed in radiation-induced sarcoma cases in the present study. One patient, diagnosed with OSCC and treated with chemo-radiotherapy in 2020, subsequently underwent surgery for radiation-induced sarcoma in February 2024. Intermediate biopsies conducted in December 2021 and March 2022 revealed inflamed granulation tissue and inflammatory myofibroblastic proliferation, respectively. Protein-level mutation patterns corresponding to the significant DNA-altered genes were also visualized. (b) Gene set enrichment analysis (GSEA) was conducted using RNA expression data from 32 OSCC patients with poor overall survival (OS) outcomes and 51 with favorable outcomes. Significant enrichment was observed in the following Hallmark pathways: MYC Targets V1, Oxidative Phosphorylation, Coagulation, MTORC1 Signaling, Mitotic Spindle, G2M Checkpoint, and E2F Targets (FDR q < 0.05).

To explore biological pathways associated with RNA expression in poor-outcome CCRT cases, gene set enrichment analysis (GSEA) was performed (Fig. 4b). Significant enrichment was observed in multiple Hallmark pathways, including MYC targets v1, oxidative phosphorylation, coagulation, MTORC1 signaling, mitotic spindle, G2M checkpoint, and E2F targets (FDR q < 0.05).

Development of a prognostic gene signature for radiotherapy and chemo-radiotherapy and validation of TJP3 in HNSCC cell lines

Of the 10 prognostic genes associated with poor outcomes after RT, all were detected in the somatic mutation profile of the RIS patient. All genes except ENPP7 were previously reported in COSMIC as head and neck cancer-associated. Similarly, among the 13 prognostic genes associated with poor CCRT outcomes, all except GARS1, MAP3K2, RAB2A, and PPME1 were COSMIC-listed. TJP3 was frequently mutated in both RT and CCRT poor-outcome cases, and its higher tissue expression, as reported in the Human Protein Atlas, correlates with favorable prognosis¹⁰. Drug–gene interactions involving the identified genes are presented in Fig. S2.

To predict 24-month survival following RT or CCRT, multivariate logistic regression models were constructed using 10 and 13 candidate genes, respectively. As shown in Table 2, the models demonstrated strong predictive performance, with area under the curve (AUC) values of 0.89 for RT and 0.94 for CCRT. Gene selection was guided by stepwise Akaike Information Criterion (AIC), identifying CLTCL1, ENPP7, TJP3, and H3C12 as significant in the RT model, and EVPL, SLC35G5, ANO8, TJP3, and EMID1 in the CCRT model. Notably, TJP3 was consistently significant in both.

Table 2.

Multivariate logistic regression analyses of oral squamous cell carcinoma patients.

Gene	Estimate	Standard error	Odds ratio	95% CI Lower	95% CI upper	P-value
Genes associated with radiotherapy (RT) response
CLTCL1	2.860	1.000	17.461	2.830	161.915	0.004a
ENPP7	2.412	1.068	11.152	1.504	115.082	0.024a
TJP3	2.529	1.127	12.535	1.577	150.578	0.025a
ARHGEF19	1.657	1.000	5.242	0.717	41.088	0.098
H3C12	1.404	0.659	4.069	1.168	16.174	0.033a
MROH9	1.204	0.643	3.332	0.977	12.626	0.061
CES5A	−1.523	1.051	0.218	0.019	1.361	0.147
Genes associated with concurrent chemoradiotherapy (CCRT) response
EVPL	−4.824	1.766	0.008	0.0001	0.143	0.006a
SLC35G5	2.257	0.870	9.555	1.947	64.274	0.010a
ANO8	−2.966	1.180	0.0515	0.004	0.390	0.012a
TJP3	3.810	1.563	45.161	3.157	2099.978	0.015a
GARS1	3.339	1.883	28.201	1.285	2797.726	0.076
MAP3K2	2.924	1.846	18.620	0.650	1422.276	0.113
RAB2A	18.295	2475.443	88200000.0	< 0.0001		0.994
EMID1	−4.732	2.178	0.009	< 0.0001	0.246	0.030a
CAPN8	1.466	0.896	4.332	0.805	30.214	0.102

Multivariate logistic regression models were developed to predict 24-month overall survival in patients treated with radiotherapy (RT) or concurrent chemoradiotherapy (CCRT), based on genes significantly associated with prognosis in each treatment group. Gene selection was performed using Stepwise inclusion and exclusion guided by the Akaike information criterion (AIC). The final models demonstrated strong predictive performance, with an area under the curve (AUC) of 0.89 for the RT group and 0.94 for the CCRT group.

CI Confidence interval.

^a p < 0.05, multivariate logistic regression analyses.

To evaluate the association between TJP3 expression and radiotherapy response in vitro, RNA-seq data from the GSE217645 dataset were analyzed. This dataset includes three replicates each of two radiosensitive HNSCC cell lines and their respective isogenic radioresistant counterparts (each with three subclones). Clustered heatmap analysis was performed, and one outlier sample was excluded. Differential expression analysis revealed that TJP3 exhibited significantly altered expression between radiosensitive and radioresistant groups (p = 0.037; |log₂FC| = 1.589).

Biological pathways shared between oral squamous cell carcinoma and radiation-induced sarcoma

The OSCC patient was diagnosed in September 2020 and received surgery followed by CCRT. In February 2024, the patient developed secondary RIS, which was surgically resected. Whole-exome sequencing of OSCC and RIS tissues revealed distinct and overlapping DNA mutation profiles (Fig. 5a). Mutations exclusive to OSCC were enriched in chromosome organization, protein modification, cellular processes, and metabolism (Fig. 5b). Mutations shared between OSCC and RIS were associated with carboxylic acid biosynthesis, oxoacid metabolism, CMP catabolism, and mesenchymal cell proliferation (Fig. 5c). Mutations specific to RIS were linked to osteoblast differentiation, bone development, icosanoid secretion, and amino acid homeostasis (Fig. 5d).

Fig. 5.

(a) The patient in the present study, diagnosed with oral squamous cell carcinoma (OSCC) in September 2020 and treated with chemo-radiotherapy, subsequently underwent surgery for radiation-induced sarcoma in February 2024. Whole-exome sequencing revealed DNA mutation patterns with overlapping and distinct features, resembling a Venn diagram. (b) Biological processes associated with DNA mutations identified exclusively in the primary OSCC tissue included chromosome organization, protein modification, general cellular processes, and metabolic processes. (c) Genes mutated in both the primary OSCC and radiation-induced sarcoma tissues were enriched in pathways related to carboxylic acid biosynthesis, oxoacid metabolic processes, CMP catabolism, and mesenchymal cell proliferation. (d) DNA mutations identified exclusively in the radiation-induced sarcoma tissue were associated with osteoblast differentiation, bone development, icosanoid secretion, and intracellular amino acid homeostasis.

The primary OSCC was pathologically staged as T4aN1M0. To provide context, somatic mutations were compared with those of 13 TCGA patients with the same staging. GO analysis of mutations unique to this patient revealed enrichment in metabolic regulation, DNA damage checkpoint control, endosomal transport, organonitrogen compound metabolism, nuclear chromosome segregation, and negative regulation of JAK–STAT signaling (Fig. 6). Molecular function analysis highlighted NAD⁺-dependent aldehyde dehydrogenase activity, catalytic activity, and ATP-dependent DNA damage sensing. Enriched cellular components included intracellular structures and mitochondria. KEGG analysis revealed associations with metabolic pathways, arginine and proline metabolism, and ribosomal function.

Fig. 6.

The patient in the present study was diagnosed with primary oral squamous cell carcinoma (OSCC) at an AJCC pathological stage of T4aN1M0. Within the TCGA OSCC cohort, 13 patients with the same pathological stage (T4aN1) were identified, and their somatic mutation profiles were compared with that of the present case. This analysis revealed DNA mutations uniquely present in the OSCC case that subsequently progressed to radiation-induced sarcoma. These mutations were functionally annotated and visualized based on (a) biological processes, (b) molecular functions, (c) cellular components, and (d) KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways.

Discussion

The significant prognostic genes identified in TCGA OSCC patients after RT or CCRT were validated using WES data from a patient who developed RIS, functioning as indicators of RT and CCRT resistance. This highlights that even patients diagnosed with the same oral cancer can have different CCRT prognoses based on individual genes, underscoring the role of precision medicine.

In patients with poor outcomes regarding overall survival within 24 months following RT, CLTCL1, NFATC1, TJP3, ARHGEF19, ENPP7and RGL3 were mutated and TGF- β signaling pathway was significant. In addition, MROH9, H3C12, CES5A and DEFB1 were differentially expressed according to prognosis after RT. Considering the prognosis of CCRT, genes such as TJP3, ANO8, EVPL, MAP3K2, RAB2A, SLC35G5, GARS1 and WDR59 were mutated and while CAPN8, PPME1, DPYSL5, JPH3 and EMID1 were differentially expressed. The Hallmark pathways including MYC targets v1, oxidative phosphorylation, coagulation, MTORC1 signaling, mitotic spindle, G2M checkpoint, and E2F targets were found to be significant. Notably, among those 28 genes, mutation of TJP3 were simultaneously observed in relation to prognosis after both RT and CCRT.

Several studies have investigated biomarkers and molecular mechanisms associated with radioresistance in OSCC. Chen et al. compared the prognostic values of genes in patients from TCGA-OSCC cohort with (n = 162) and without RT (n = 118), identifying 60 S ribosomal protein L36a (RPL36A) as a promising biomarkers associated with radioresistance in a Taiwanese population¹¹. Overexpression NEDD8 gene was observed extensively in primary tumors compared to normal tissues and was correlated with poor responsiveness to RT in TCGA OSCC cohort. Furthermore, NEDD8 knockdown in HSC4 cells sensitizes them to RT¹². An in vitro analysis initially compared mRNA expression levels between parental OSCC cells and radioresistant cells. Subsequently, using TCGA OSCC database, six candidate genes (KCNJ2, CLEC18C, P3H3, PIK3R3, SERPINE1 and TMC8) were identified as being closely associated with prognosis and radioresistance¹³. In vitro and in vivo studies demonstrated that IGFBP3, a potential biomarker for radiosensitivity, promotes ionizing radiation-mediated OSCC cell death by increasing reactive oxygen species production through NF- κB activation and cytokine production¹⁴.

However, these studies using patient data didn’t differentiate between RT alone and CCRT, which may involve somewhat different mechanisms. In our study analyzing mutation and differentially expressed gene (DEG)s associated with OS or PFS, we observed limited overlap in genes between RT and CCRT groups. Only TJP3 was mutated in both RT resistant and CCRT resistant tissues. As main anti-cancer treatments, ionizing radiation and chemotherapeutic agents trigger cell death by causing DNA damage. Disruption of the DNA damage response can lead to increased sensitivity or resistance of cancer cells to these genotoxic agents¹⁵. Few studies have directly compared the mechanisms underlying resistance between chemotherapy and radiotherapy. Hao et al. reported that cisplatin sensitivity is closely associated with radiation therapy efficacy, as they found that Solute Carrier Family 1 Member 6 (SLC1A6) contributed to reduced cisplatin sensitivity in radioresistant nasopharyngeal cells by altering drug metabolism profiles and genes¹⁶. In contrast, our results showed limited overlap in pathways and genes between RT and CCRT resistance, suggesting potentially different mechanisms for these two types of treatment resistance. Still, no studies to date have specifically suggested different mechanisms between RT and CCRT resistance. Further studies with tailored approaches are needed to elucidate these difference.

We performed WES on a 56-year-old male diagnosed with RIS involving the right buccal mucosa. This sarcoma developed three years after CCRT for OSCC. The mean latency interval for RIS is reported to be 11.1 years, with an average radiation dose of 62 Gy used in head and neck region. Osteosarcoma and fibrosarcoma was identified as the most prevalent histological type among RIS¹⁷. Elevated genomic instability and higher immune cell infiltrations were found in RIS than in primary sarcomas^18,19. Our study provides insights into the temporal evolution of RIS at the genomic level by longitudinally analyzing somatic mutations within a single patient. Notably, the genes associated with the prognosis of RT were mainly distributed in the primary OSCC tissue, whereas the genes associated with the prognosis of CCRT were evenly distributed from the primary tissue to the inflamed tissue and the osteosarcoma. RT primarily affects the tumor cells directly, which could explain why prognostic genes are mainly found in the primary OSCC tissue, while CCRT potentially affects a broader range of tissues and cellular processes, leading to a more even distribution. Also, CCRT may have a more significant impact on the tumor microenvironment and the immune system, explaining the presence of prognostic genes in inflamed tissues¹⁸. The distribution of prognostic genes may reflect the complex interplay between treatment modalities, tumor biology and the host response to therapy and further study would be necessary.

To evaluate the prognosis following RT and CCRT, we divided patients based on a 24-month OS. The choice of 24 months was based on Struckmeier et al.’s report, which indicated that 82.86% of recurrences manifested within the first two years, with the poorest outcomes observed in patients who experience recurrence within 12 months²⁰. Similarly, other studies showed that up to 76% of recurrences occur within the first two years²¹. Two-year OS also differs markedly based on clinical risk factors, ranging from 54% in stage 1 to 3% in stage IV of OSCC²²and from 43.9% in patients without extracapsular extension to 19.6% in those with it²³. Based on this 24-month OS stratification, we performed univariate Cox regression analysis to further identify the genes associated with both prognosis and post-treatment tumor aggressiveness following RT or CCRT.

Both DNA mutation and RNA expression levels of CLTCL1, RGL3, H3C12, SLC35G5, GARS1 and JPH3 demonstrated significant prognostic value in multivariate Cox analyses. Specifically, CLTCL1, RGL3 and H3C12 were associated with RT and SLC35G5, GARS1 and JPH3 with CCRT. And TJP3(tight junction protein 3) was simultaneously mutated in relation to prognosis after both RT and CCRT and this mutation was observed in the primary OSCC tissue of the RIS patient. TJP3 plays a cell scaffolding role and is involved in epithelial differentiation. It also promotes T cell immunity escape and chemoresistance in breast cancer²⁴ and has been shown to positively correlated with chemoresistance in bladder cancer cell lines²⁵. However, there is no significant evidence indicating that TJP3 is related to radiation resistance. Among six genes (CLTCL1, RGL3, H3C12, SLC35G5, GARS1 and JPH3) which were significant in multivariate Cox analysis, SLC35G5 and GARS1 were reported to be related with chemotherapy resistant. SLC35G5 were only 2 mutated genes(SLC35G5, TUBA3D) shared between platinum resistant cases pre- and post- neoadjuvant chemotherapy in high grade serous ovarian carcinoma²⁶. GARS1 is a potential prognostic marker and therapeutic target for pan-cancer immunotherapy and could effectively predict the response to anti-PD-L1²⁷.

This study has several limitations. We defined RT or CCRT resistance based on progression free survival within 24 months, as there is no effective method to evaluate immediate resistance. Our analysis included open datasets, such as TCGA and GEO, in conjunction with a single patient case from our institution. Despite this, the longitudinal data from such a rare case holds its own significance. Additionally, the cohort size is relatively small because only 95 patients received PO RT and 83 received PO CCRT. To address this limitation, we analyzed in vitro data to further validate our findings, focusing on the role of TJP3.

Conclusion

Through a comparative analysis of DNA mutation and mRNA expression data from patients who received RT or CCRT for OSCC treatment, we identified 28 significant prognostic genes associated with OS and PFS and these genes. And these genes were validated in WES data from a patient who developed RIS, functioning as indicators of RT and CCRT resistance. DNA mutations and RNA expression levels of CLTCL1, RGL3 and H3C12 showed significant prognostic value in multivariate Cox analyses for RT, while SLC35G5, GARS1 and JPH3 were identified as prognostic genes for CCRT. Limited overlap between RT and CCRT was observed, with only TJP3 being common to both. In longitudinal WES data from a radiation induced sarcoma patient, RT prognostic genes were predominantly found in the primary tumor, whereas CCRT prognostic genes were distributed more evenly across primary tumor tissue, inflamed granulation tissue, inflammatory myofibroblastic proliferation and RIS tissue.

These findings could provide insights into the molecular mechanisms underlying RT and CCRT resistance in OSCC, paving the way for advancements in precision medicine. .

Patient and methods

Patient

A 56-year-old male patient was diagnosed with OSCC at the buccal mucosa in September 2020, classified as cT4aN1M0. The patient underwent radical CCRT from October to December 2020, including radiotherapy up to 67.5 Gy over 30 fractions and weekly cisplatin chemotherapy for six weeks. There were no subsequent recurrences, but exophytic lesions persisted at the site of radiation in oral cavity. Therefore, subsequent biopsies revealed inflamed granulation tissue in December 2021, inflammatory myofibroblastic proliferation in March 2022, and radiation-induced osteosarcoma in February 2024, which led to a surgery. However, the sarcoma recurred after surgery and the patient expired. With the patient’s consent, tissues diagnosed in 2020 (OSCC), 2021 (inflamed granulation tissue), 2022 (inflammatory myofibroblastic proliferation), and 2024 (radiation-induced sarcoma) were analyzed using next-generation sequencing (NGS). This study was conducted with approval of the Institutional Review Board of Seoul National University Dental Hospital (IRB No: CRI24004).

Validation population

This study analyzed data from the TCGA portal, excluding cases with tumors of the pharynx, hypopharynx, larynx, and oropharynx. After removing 15 patients with a prior malignancy, 305 OSCC patients were included, with whole-exome and mRNA sequencing data available for 297 and 299 patients, respectively (Table S1). Clinical and survival data were obtained from the Genomic Data Commons (GDC)²⁸with poor prognosis defined as overall survival within 24 months. For external validation, 96 HPV-negative OSCC patients from the GSE41613 dataset were analyzed, excluding one case with unknown treatment status. Additionally, RNA-seq data from GSE217645, consisting of radiosensitive and isogenic radioresistant HNSCC cell lines, were used to validate genes associated with therapy response, following outlier removal via hierarchical clustering. All analyses were conducted in compliance with the GDC Data Use Agreement and the dbGaP Data Use Certification.

Somatic mutation and mRNA sequencing data

FastQ files generated from the patient’s NGS were processed alongside normal blood samples using a pipeline that included FastQC, Trimmomatic, BWA-MEM, DeDuplicate, GATK BaseRecalibrator, GATK Mutect2 for somatic calling, GATK CalculateContamination, SnpEff for annotation, and GATK Funcotator. This process yielded variant call format (VCF) and mutation annotation format (MAF) files, documenting DNA mutations. Whole exome somatic mutation and mRNA sequencing (RNA-Seq) data, released on March 29, 2022, were downloaded from the TCGA data portal. Somatic variants were extracted in MAF format, with whole exome sequencing conducted on the Illumina Genome Analyzer IIX platform. Loss-of-function mutations were defined as nonsense mutations, frameshift indels, and splice-site mutations. RNA-Seq by Expectation Maximization (RSEM) data were generated using the Illumina HiSeq 2000 RNA-Seq platform (version 2), and normalized RSEM data were used to estimate mRNA expression levels.

Statistical analysis

Fisher’s exact tests were applied to identify genes with significant mutation differences with MAFtools²⁹. Differentially expressed genes (DEGs) were identified with DESeq2 based on unstranded read counts, using a negative binomial distribution model. Cox proportional hazards regression analyses—both univariate and multivariate—were conducted using DNA mutation and RNA transcripts per million (TPM) data to evaluate overall survival, adjusting for relevant clinical covariates. The proportional hazards assumptions were tested using Schoenfeld and dfbeta residuals. Multivariate logistic regression models were constructed through stepwise selection based on the Akaike Information Criterion (AIC), and model performance was assessed using the area under the curve (AUC). Visualizations were generated using Python’s matplotlib and R’s ggplot2. All statistical analyses were performed in R (v4.3.2, Vienna, Austria; http://www.R-project.org/).

Biological pathways analysis

GSEA was performed on the DEGs. GSEA calculates an enrichment score (ES) by examining the ranked gene list, increasing a running-sum statistic for each gene present in the gene set and decreasing it for genes not in the set. Additionally, pathway enrichment analyses for GO and KEGG were conducted based on DNA mutations in OSCC patients^30–32. GO enrichment analysis was carried out using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) from the National Institutes of Health, Bethesda, MD.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

S.K. contributed to the conception, design, data acquisition, analysis, interpretation, draft and critically revise the manuscript. H.Y., J.L., Y.P. and J.L. were engaged in sample collection and preparation. I.K. and K.M.P. contributed to the conception, design, draft and critically revise the manuscript. All authors approved the final manuscript.

Data availability

The data that support the findings of this study are available from the Cancer Genome Atlas [http://gdc.cancer.gov/, accession number: phs000178.v11.p8] and GSE41613 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE41613]. All the other data associated with this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.