Keratin 1 and keratin 18 as immunohistochemical markers for distinguishing pulmonary metastases of oral squamous cell carcinoma from primary lung squamous cell carcinoma Authors: Yoshiko Watanabe, Hisami Kato, Rei Ishikawa, Kimio Yoshimura, Hidetaka Yamada, Ryosuke Miyazaki, Sayuri Takeuchi, S M Jahurul Haque, Katsuhiro Yoshimura, Yasuhiro Sakai, Miki Sugita, Yusuke Inoue, Yusuke Takanashi, Keigo Sekihara, Kazuhito Funai, Kazuma Masumoto, Kazuya Shinmura Published: 2026 ABSTRACT AbstractAccurate distinction between pulmonary metastases of oral squamous cell carcinoma (OSCC) and primary lung squamous cell carcinoma (LSCC) is critical for clinical management, but practical immunohistochemical markers remain limited in routine practice. We screened three public mRNA expression datasets, GSE84846 for OSCC and The Cancer Genome Atlas and Clinical Proteomic Tumor Analysis Consortium for LSCC, and selected six candidate markers for immunohistochemical evaluation in 39 OSCC and 92 LSCC specimens. Among them, keratin 1 (K1; encoded by KRT1) and keratin 18 (K18; encoded by KRT18) showed the most significant differential expression between OSCC and LSCC (both p < 0.0001). K1 was predominantly expressed in OSCC and rarely in LSCC, whereas K18 showed the opposite pattern. K1 showed high specificity (0.989) and an odds ratio of 117.8, while K18 showed a specificity of 0.974 and an odds ratio of 59.1. Receiver operating characteristic analysis confirmed strong discriminatory performance for K1 (AUC, 0.865) and K18 (AUC, 0.872). In four whole-exome sequencing-confirmed pairs of primary OSCC and lung metastases, staining patterns for both markers were concordant between the primary oral tumors and pulmonary lesions. These findings support K1 and K18 as practical markers for distinguishing pulmonary metastases of OSCC from primary LSCC.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-53003-7.Keywords: Immunohistochemistry, K1, K18, Lung squamous cell carcinoma, Oral squamous cell carcinoma, Pulmonary metastasisSubject terms: Biomarkers, Cancer, Oncology Introduction Oral squamous cell carcinoma (OSCC) is a common malignant epithelial neoplasm in the head and neck region and a major histological subtype of oral cancer1–3. Despite advances in surgery, radiotherapy, and systemic therapy that have improved locoregional control, distant metastasis remains a determinant of prognosis. In OSCC patients treated with curative intent, 5%–15% eventually develop distant metastases4,5, most frequently to the lung4–7. Once pulmonary metastasis occurs, therapeutic options become limited, and overall survival is reduced8. Treatment strategies differ between pulmonary metastasis of OSCC and primary lung squamous cell carcinoma (LSCC)9–11. In general, pulmonary metastasis of OSCC is regarded as a distant metastatic disease and is commonly managed with systemic therapy, although selected patients with solitary or limited pulmonary metastases may consider pulmonary metastasectomy. Conversely, primary LSCC is staged and treated as non-small cell lung cancer, and potentially curative anatomic lung resection with lymph node evaluation is considered in resectable cases, followed by stage-adapted adjuvant therapy when indicated. Furthermore, these differences are associated with distinct prognostic implications. Because OSCC and lung cancer share etiologic risk factors, particularly tobacco exposure, OSCC patients not uncommonly present with pulmonary nodules during initial staging or follow-up12,13. Thus, pulmonary lesions may represent either metastatic OSCC or a second primary LSCC. Accurate discrimination is essential, as management strategies and prognostic implications differ substantially. The detection of squamous cell carcinoma in the lung of OSCC patients poses a diagnostic dilemma: distinguishing pulmonary metastasis from second primary LSCC. This distinction has consequences for tumor–node–metastasis staging, therapeutic decision-making (e.g., systemic versus local treatment), and prognostic assessment. Radiological and clinical features alone are insufficient to reliably differentiate these entities; therefore, pathological evaluation is central to the diagnostic process. However, pathological distinction is challenging because pulmonary metastases of OSCC and primary LSCC share histomorphological features and exhibit similar expression of commonly used immunohistochemical (IHC) markers, including p40, p63, and high-molecular-weight cytokeratins14. Ancillary markers, such as CK7, CK20, and thyroid transcription factor-1, may assist in diagnostic workflows, but they are not specifically tailored to this differential diagnosis and frequently yield inconclusive results15. Therefore, diagnostic uncertainty persists in a subset of cases and may lead to suboptimal therapeutic decision-making. Molecular approaches, including microsatellite analysis and next-generation sequencing–based clonal comparison, can provide definitive evidence of metastatic relatedness16–18; however, these techniques are technically complex, resource-intensive, and not practical for routine diagnosis. Recently, keratin 13 (K13; encoded by KRT13) and UPK1B have been proposed as candidate IHC markers to assist in this distinction19; however, their diagnostic performance requires validation in larger and independent cohorts. Based on these limitations, there is a need for practical biomarkers capable of distinguishing pulmonary metastases of OSCC from primary LSCC in routine diagnostic pathology. In the present study, we integrated transcriptomic screening with IHC and genomic analyses. We identified candidate genes differentially expressed between OSCC and LSCC, evaluated their expression in clinical specimens, and validated their diagnostic utility using genetically confirmed metastatic pairs. Through this comprehensive approach, we aimed to establish practical IHC biomarkers, particularly keratin 1 (K1; encoded by KRT1) and keratin 18 (K18; encoded by KRT18), that are applicable to routine pathological practice. This study also provides a practical framework for integrating public transcriptomic screening, institutional immunohistochemical validation, and genomic confirmation in biomarker development for differential diagnosis. Results Differentially expressed genes between OSCC and LSCC To identify candidate IHC markers that distinguish pulmonary metastases of OSCC from primary LSCC, we analyzed three mRNA expression datasets: GSE84846 (99 OSCC cases)20, The Cancer Genome Atlas (TCGA) (501 LSCC cases)21, and Clinical Proteomic Tumor Analysis Consortium (CPTAC) (108 LSCC cases)22. Candidate genes were prioritized using predefined cross-platform expression filters designed to identify transcripts showing consistent directional differences between OSCC and LSCC. Using the predefined criteria described in the Methods, we identified 12 transcripts upregulated and 11 transcripts downregulated in OSCC compared with LSCC (Tables 1 and 2). Table 1Upregulated genes selected through cross-platform mRNA expression screening in oral squamous cell carcinoma (OSCC) compared with lung squamous cell carcinoma (LSCC).GeneMedian expression value aSymbolDescriptionOSCC (GSE84846; >4)LSCC (TCGA; <20)LSCC (CPTAC; <−0.15)ACTC1Actin Alpha Cardiac Muscle 14.2251.218−0.601CTSGCathepsin G4.0607.597−1.370HBDHemoglobin Subunit Delta5.8270.515−0.415HSPB3Heat Shock Protein Family B (Small) Member 34.8908.835−1.246KRT1Keratin 14.23917.626−0.156LGALS2Galectin 24.96913.346−0.353MYBPHMyosin Binding Protein H4.6241.980−0.152MYH2Myosin Heavy Chain 24.8630.000−2.728SLNSarcolipin4.2052.247−1.412TCAPTitin-Cap4.4433.515−0.709TNNC2Troponin C2, Fast Skeletal Type4.2673.364−0.837TNNI2Troponin I2, Fast Skeletal Type4.00111.351−1.133a Values in parentheses in the column headings denote the predefined cutoff values used for gene selection. Expression values from the three datasets may not be directly comparable due to variations in data processing and normalization methods. Upregulated genes selected through cross-platform mRNA expression screening in oral squamous cell carcinoma (OSCC) compared with lung squamous cell carcinoma (LSCC). a Values in parentheses in the column headings denote the predefined cutoff values used for gene selection. Expression values from the three datasets may not be directly comparable due to variations in data processing and normalization methods. Table 2Downregulated genes selected through cross-platform mRNA expression screening in oral squamous cell carcinoma (OSCC) compared with lung squamous cell carcinoma (LSCC).GeneMedian expression value aSymbolDescriptionOSCC (GSE84846; <−2)LSCC (TCGA; >3000)LSCC (CPTAC; >0)APOL1Apolipoprotein L1−2.3194644.7510.732CAPN1Calpain 1−2.3007818.8860.633EPCAMEpithelial Cell Adhesion Molecule−2.8354229.7350.708IGFBP3Insulin Like Growth Factor Binding Protein 3−2.9037928.4831.590KRT15Keratin 15−2.0549519.4443.197KRT18Keratin 18−2.2065580.0280.049MYOFMyoferlin−2.1713640.0330.005RPL27ARibosomal Protein L27a−2.11911872.6990.115SLC7A5Solute Carrier Family 7 Member 5−2.2085843.7471.033STIP1Stress Induced Phosphoprotein 1−2.5924497.1090.428TOP2ADNA Topoisomerase II Alpha−2.1693813.2141.439a Values in parentheses in the column headings denote the predefined cutoff values used for gene selection. Expression values from the three datasets may not be directly comparable due to variations in data processing and normalization methods. Downregulated genes selected through cross-platform mRNA expression screening in oral squamous cell carcinoma (OSCC) compared with lung squamous cell carcinoma (LSCC). a Values in parentheses in the column headings denote the predefined cutoff values used for gene selection. Expression values from the three datasets may not be directly comparable due to variations in data processing and normalization methods. Based on consistent differential expression across platforms and feasibility for IHC evaluation in formalin-fixed, paraffin-embedded (FFPE) tissue, four genes were selected for validation: KRT1 (upregulated in OSCC) and KRT18, MYOF, and RPL27A (downregulated in OSCC). In addition, KRT13 and UPK1B, which were recently proposed as potential markers for distinguishing pulmonary metastases of OSCC from primary LSCC19, were included for comparison, resulting in a total of six candidate markers. K1 and K18 outperform other candidates in differentiating primary OSCC from LSCC IHC staining was performed on 39 primary OSCC and 92 primary LSCC specimens. Staining intensity and extent were quantified using the H-score method (range: 0–300). To capture both binary diagnostic classification and the full distribution of staining intensity, we evaluated marker performance using both dichotomized H-scores and continuous H-score-based receiver operating characteristic (ROC) analysis. A uniform cutoff value of 50 was applied to dichotomize cases as positive (≥ 50) or negative (< 50), which enabled calculation of Fisher’s exact test, sensitivity, specificity, and ORs (Table 3; Fig. 1, Supplementary Fig. S1). Table 3Immunohistochemical performance of six candidate markers for differentiating oral squamous cell carcinoma (OSCC) from lung squamous cell carcinoma (LSCC).AntibodyFrequency in OSCCFrequency in LSCCP valueSensitivityaSpecificityaOR (95% CI) a, bK122/39 (56.4%)1/92 (1.1%)< 0.00010.5640.989117.765(14.863–933.119)K1310/39 (25.6%)23/92 (25.0%)1.0000.2500.7440.967(0.409–2.284)K181/39 (2.6%)56/92 (60.9%)< 0.00010.6090.97459.111(7.769–449.740)MYOF27/39 (69.2%)47/92 (51.1%)0.0820.5110.3080.464(0.210–1.026)RPL27A2/39 (5.1%)6/92 (6.5%)1.0000.0650.9491.291(0.249–6.694)UPK1B1/39 (2.6%)13/92 (14.1%)0.0640.1410.9746.253(0.789–49.576)a Sensitivity, specificity, and ORs were calculated for OSCC versus LSCC for K1 and for LSCC versus OSCC for K13, K18, MYOF, RPL27A, and UPK1B, according to the expected direction of marker expression. b OR, odds ratio; CI, confidence interval. Immunohistochemical performance of six candidate markers for differentiating oral squamous cell carcinoma (OSCC) from lung squamous cell carcinoma (LSCC). 117.765 (14.863–933.119) 0.967 (0.409–2.284) 59.111 (7.769–449.740) 0.464 (0.210–1.026) 1.291 (0.249–6.694) 6.253 (0.789–49.576) a Sensitivity, specificity, and ORs were calculated for OSCC versus LSCC for K1 and for LSCC versus OSCC for K13, K18, MYOF, RPL27A, and UPK1B, according to the expected direction of marker expression. b OR, odds ratio; CI, confidence interval. Representative immunohistochemical staining of six candidate proteins—keratin 1 (K1), K13, K18, myoferlin (MYOF), ribosomal protein L27a (RPL27A), and uroplakin 1B (UPK1B)—identified in this study as differentially expressed between oral squamous cell carcinoma (OSCC) and lung squamous cell carcinoma (LSCC), or selected based on recent literature19. For each marker, a representative image from either a positive or negative case is shown, depending on which staining pattern is predominant in the corresponding tumor group. Representative images of positive staining for all six markers are provided in Supplementary Fig. S1 (scale bars = 50 μm). Of these markers, only K1 and K18 exhibited significant differential expression between OSCC and LSCC (p < 0.0001 by Fisher’s exact test). K1 was positive in 22 of 39 OSCC cases (56.4%) but only in 1 of 92 LSCC cases (1.1%), corresponding to a specificity of 0.989 and an odds ratio (OR) of 117.8. In contrast, K18 was positive in 56 of 92 LSCC cases (60.9%) and only in 1 of 39 OSCC cases (2.6%), yielding a specificity of 0.974 and an OR of 59.1. The remaining four markers (K13, MYOF, RPL27A, and UPK1B) did not show significant discriminatory ability between the tumor types. K13 exhibited nearly identical positivity rates in OSCC and LSCC (25.6% vs. 25.0%, p = 1.000), whereas MYOF showed insufficient specificity. RPL27A and UPK1B demonstrated low sensitivity. Collectively, these markers had insignificant p-values and wide confidence intervals (CIs), indicating limited diagnostic utility. To evaluate discriminatory performance across the full range of staining intensities, ROC curve analysis was conducted using continuous H-score values, and areas under the curves (AUCs) were calculated (Table 4), with ROC curves for K1 and K18 shown in Fig. 2. Consistent with the dichotomized analysis, K18 (AUC = 0.872, 95% CI: 0.799–0.933) and K1 (AUC = 0.865, 95% CI: 0.784–0.936) showed the highest diagnostic accuracy. UPK1B exhibited modest performance (AUC = 0.601), whereas K13, MYOF, and RPL27A were largely uninformative, with AUCs of 0.407, 0.331, and 0.466, respectively. Taken together, binary and continuous analyses consistently identified K1 and K18 as the most diagnostically informative IHC markers for distinguishing primary OSCC from LSCC. Table 4Area under the curve (AUC) values for six candidate immunohistochemical markers in differentiating oral squamous cell carcinoma (OSCC) from lung squamous cell carcinoma (LSCC).Primary antibodyAUC (95% CI) a, bK10.865 (0.784–0.936)K130.407 (0.308–0.506)K180.872 (0.799–0.933)MYOF0.331 (0.241–0.425)RPL27A0.466 (0.376–0.553)UPK1B0.601 (0.517–0.679)a AUCs were calculated for OSCC versus LSCC for K1 and for LSCC versus OSCC for K13, K18, MYOF, RPL27A, and UPK1B, according to the expected direction of marker expression. b AUC, area under the curve; CI, confidence interval. Area under the curve (AUC) values for six candidate immunohistochemical markers in differentiating oral squamous cell carcinoma (OSCC) from lung squamous cell carcinoma (LSCC). a AUCs were calculated for OSCC versus LSCC for K1 and for LSCC versus OSCC for K13, K18, MYOF, RPL27A, and UPK1B, according to the expected direction of marker expression. b AUC, area under the curve; CI, confidence interval. Receiver operating characteristic (ROC) curves for keratin 1 (K1) and K18 in differentiating primary oral squamous cell carcinoma (OSCC) from primary lung squamous cell carcinoma (LSCC). ROC curves were generated using continuous H-scores. The area under the curve (AUC) values are presented in the figure and correspond to those summarized in Table 4. Utility of K1 and K18 in the differential diagnosis of pulmonary metastases of OSCC Five paired specimens consisting of a primary OSCC and a clinically suspected metastatic squamous cell carcinoma in the lung [POL (pair of OSCC and lung tumor) #1–POL#5] (Supplementary Table S1) were analyzed. These paired cases were analyzed as a proof-of-concept set to assess whether marker expression remained concordant between primary OSCCs and their corresponding pulmonary lesions. In the primary OSCC components, conventional IHC markers (p40, CK7, CK20, and ARL13B) exhibited staining patterns consistent with previously reported profiles23–26. In the corresponding pulmonary lesions, these markers showed concordant staining patterns between the oral and lung components for all five pairs (Table 5, Supplementary Fig. S2). Table 5Immunohistochemical profiles and somatic mutation concordance in paired cases of oral squamous cell carcinoma (OSCC) and lung tumors (POL cases).CaseImmunohistochemical profileaSomatic mutation status (No. 1)bSomatic mutation status (No. 2)bAntibodyOSCCLTGene mutation[RefSeq ID]OSCCLTGene mutation[RefSeq ID]OSCCLTPOL#1p40CK7CK20ARL13B++−−++−−SPATA31D1-c.G3240T (p.R1080R)[NM_001001670]++SORBS2- c.G781A (p.G261S)[NM_001145674]++POL#2p40CK7CK20ARL13B+−−−+−−−FSIP2-c.G16564A (p.E5522K)[NM_173651]++DIAPH3-c.G2679C (p.K893N)[NM_001042517]++POL#3p40CK7CK20ARL13B+−−−+−−−NOTCH1-c.A1331G (p.Y444C)[NM_017617]++HRAS-c.G34A (p.G12S)[NM_005343]++POL#4p40CK7CK20ARL13B++−−++−−TP53-c.G644T (p.S215I)[NM_000546]++NF2-c.C1021T (p.R341X)[NM_000268]++POL#5p40CK7CK20ARL13B+−−−+−−−PIK3CA-c.G3019C (p.G1007R)[NM_006218]+−VMA21-c.A184G (p.R62G)[NM_001017980]+−a Immunohistochemical expression is indicated as positive (+) or negative (−) for OSCC and lung tumors (LT). Cutoff values for positivity were set at 50% for p40, 10% for CK7 and CK20, and 15% for ARL13B. b In the OSCC column, “+” indicates that the somatic variant was detected in OSCC by WES. In the LT column, “+” indicates that the same somatic variant identified in OSCC was also detected in the corresponding LT by PCR and Sanger sequencing. Conversely, “−” indicates that the variant was not detected in the LT. Immunohistochemical profiles and somatic mutation concordance in paired cases of oral squamous cell carcinoma (OSCC) and lung tumors (POL cases). p40 CK7 CK20 ARL13B + + − − + + − − SPATA31D1-c.G3240T (p.R1080R) [NM_001001670] SORBS2- c.G781A (p.G261S) [NM_001145674] p40 CK7 CK20 ARL13B + − − − + − − − FSIP2-c.G16564A (p.E5522K) [NM_173651] DIAPH3-c.G2679C (p.K893N) [NM_001042517] p40 CK7 CK20 ARL13B + − − − + − − − NOTCH1-c.A1331G (p.Y444C) [NM_017617] HRAS-c.G34A (p.G12S) [NM_005343] p40 CK7 CK20 ARL13B + + − − + + − − TP53-c.G644T (p.S215I) [NM_000546] NF2-c.C1021T (p.R341X) [NM_000268] p40 CK7 CK20 ARL13B + − − − + − − − PIK3CA-c.G3019C (p.G1007R) [NM_006218] VMA21-c.A184G (p.R62G) [NM_001017980] a Immunohistochemical expression is indicated as positive (+) or negative (−) for OSCC and lung tumors (LT). Cutoff values for positivity were set at 50% for p40, 10% for CK7 and CK20, and 15% for ARL13B. b In the OSCC column, “+” indicates that the somatic variant was detected in OSCC by WES. In the LT column, “+” indicates that the same somatic variant identified in OSCC was also detected in the corresponding LT by PCR and Sanger sequencing. Conversely, “−” indicates that the variant was not detected in the LT. Based on this concordance, we next determined whether the lung tumors represented true metastases from the corresponding OSCCs. Clonal relatedness was assessed by comparing somatic mutations identified in the primary OSCCs with those detected in the matched lung lesions. To evaluate clonal relatedness, whole-exome sequencing (WES) of the primary OSCCs detected between 5 and 192 somatic mutations per tumor. All protein-altering variants are listed in Supplementary Tables S2–S6. To contextualize the mutational profiles, we determined whether the variants involved genes frequently mutated in Japanese OSCC, as reported by Osawa et al27. Variants affecting these genes are listed in Supplementary Table S7. This comparison was performed to provide biological context and was not used to select variants for downstream validation. Two representative somatic mutations from each case were evaluated in the matched lung tumors by polymerase chain reaction (PCR) amplification followed by Sanger sequencing. Identical mutations were confirmed in the lung lesions of POL#1–POL#4, but not in POL#5, indicating that POL#1–POL#4 represent true metastatic pairs, whereas POL#5 likely represents a second primary tumor (Table 5, Supplementary Fig. S3). MYOF and RPL27A demonstrated limited discriminatory performance in the primary OSCC–LSCC analysis; thus, IHC evaluation in the POL cases focused on K1 and K18, illustrating opposite expression patterns and strong discriminatory performance between primary OSCC and LSCC, together with K13 and UPK1B, which were previously proposed as differential diagnostic markers. In the WES-confirmed metastatic pairs (POL#1–POL#4), the staining patterns for all four markers were concordant between the oral and pulmonary components. In contrast, POL#5 exhibited discordant expression patterns across all four markers (Fig. 3). Immunohistochemical (IHC) profiles of keratin 1 (K1), K13, K18, and uroplakin 1B (UPK1B) in paired oral squamous cell carcinoma–lung tumor (POL) cases. (A) Summary of IHC positivity (+) and negativity (–) for the four markers in primary oral squamous cell carcinomas (OSCCs) and their corresponding lung tumors. POL#1–POL#4 represent true pairs of primary OSCCs and their pulmonary metastases, whereas POL#5 represents a non-matching pair, in which the lung tumor corresponds to a primary lung squamous cell carcinoma (LSCC) rather than a metastasis. The H-score cutoff for positivity was set at 50. (B) Representative IHC staining of the four markers in POL#4. (C) Representative IHC staining in POL#5. K18 is physiologically localized in the cytoplasm; therefore, cytoplasmic staining was considered positive, whereas nuclear staining was regarded as nonspecific (scale bars = 50 μm). Although these findings suggest that K13 and UPK1B may also contribute to distinguishing pulmonary metastases of OSCC from primary LSCC, their limited sensitivity for detecting LSCC (Table 3) is a major limitation. Together with our earlier results identifying K1 and K18 as the most diagnostically informative discriminators between primary OSCC and primary LSCC, these data support the practicality and diagnostic potential of K1 and K18 as IHC tools for distinguishing pulmonary metastases of OSCC from primary LSCC. Discussion In this study, we identified K1 and K18 as practical IHC markers with diagnostic potential for distinguishing pulmonary metastases of OSCC from primary LSCC. By integrating transcriptomic data from three independent datasets (GSE84846, TCGA, and CPTAC) with IHC validation of 131 clinical specimens, we demonstrated that K1 and K18 exhibit distinct tumor type–specific expression patterns. K1 was predominantly expressed in OSCC and rarely detected in LSCC, whereas K18 exhibited the opposite pattern. These differences were supported by strong diagnostic performance metrics, including high specificity (0.989 for K1, 0.974 for K18), strong ORs (117.8 and 59.1, respectively), and high AUC values (0.865 and 0.872, respectively). Furthermore, in four WES-confirmed OSCC lung metastases, the expression patterns of K1 and K18 were consistent between primary and metastatic lesions, reinforcing their diagnostic utility in distinguishing pulmonary metastases of OSCC from primary LSCC. Collectively, these results establish K1 and K18 as practical and biologically grounded biomarkers for resolving this clinically significant diagnostic dilemma in surgical pathology. K1 and K18 exhibited excellent diagnostic performance in the present study, as reflected by high specificity, strong ORs, and favorable AUC values, supporting their potential as discriminative biomarkers. The biological plausibility of using K1 and K18 as discriminator markers lies in their distinct expression patterns, which reflect epithelial lineage and tumor differentiation status. K1 is a keratin typically expressed in keratinizing stratified squamous epithelium, particularly within differentiated spinous and granular layer cells28, and is frequently retained in well-differentiated, keratinizing OSCCs29. As a differentiation-associated “keratinizing” keratin, K1 reflects suprabasal/terminal squamous maturation, consistent with keratinizing, well-differentiated OSCC phenotypes. K18 is primarily expressed in simple and glandular epithelia28 and was expressed in a substantial subset of LSCC cases in the present study, thereby potentially reflecting the retention or aberrant activation of simple epithelial keratin programs during bronchial squamous carcinogenesis30. As a prototypical “simple-epithelium” keratin, it reflects bronchial epithelial programs variably retained in LSCC subsets. These expression patterns mirror the distinct histogenetic origins of OSCC and LSCC: OSCC arises from the stratified squamous epithelium of the oral mucosa, whereas LSCC arises from bronchial epithelium undergoing squamous metaplasia and subsequent dysplasia. Differences in epithelial lineage and differentiation likely underlie the contrasting keratin expression profiles observed in the present study, supporting K1 and K18 as discriminator markers rather than merely descriptive indicators of the epithelial phenotype. Overall, K1 reflects keratinizing or terminal squamous differentiation characteristic of stratified squamous epithelium, whereas K18 represents a simple-epithelium program retained or aberrantly activated during bronchial squamous carcinogenesis. Despite these conceptual distinctions, previous studies have not assessed K1 or K18 as diagnostic markers for distinguishing pulmonary metastases of OSCC from primary LSCC, nor validated their performance in clinically or genetically confirmed metastatic cases. Our study bridges this gap by showing that K1 is strongly expressed in OSCC but largely absent in LSCC, whereas K18 exhibits the opposite pattern. Moreover, these expression profiles are maintained in WES-confirmed OSCC primary–lung metastasis pairs. These results indicate that K1 and K18 not only reflect epithelial origin but also capture tumor type–specific biological signatures that persist after metastatic progression, supporting their potential utility as diagnostically informative biomarkers for differentiating pulmonary metastases of OSCC from primary LSCC. K13 and UPK1B expression have been proposed as strategies for distinguishing LSCC from pulmonary metastases of OSCC19, as significantly higher expression of both markers was observed in LSCC; however, this study was based on a relatively small cohort (pulmonary metastasis of OSCC, n = 5; LSCC, n = 12). In the present study, we evaluated a larger cohort (OSCC, n = 39; LSCC, n = 92). In addition, the initial screening in the previous study19analyzed only three OSCC lung metastases and three LSCC samples, which may have increased sampling variability. In our cohort, neither K13 nor UPK1B achieved statistically significant discrimination between the two tumor types. Although UPK1B exhibited high specificity and modest diagnostic performance (AUC = 0.601), its low sensitivity limits its use as a standalone marker. Consistent with our findings, Lennartz et al31. reported comparable K13 positivity rates in OSCC (55.9%) and LSCC (63.2%), further suggesting limited discriminatory value. Our study and that of Lennartz et al. analyzed primary OSCC and primary LSCC, whereas Goda et al19. compared pulmonary metastases of OSCC with primary LSCC, a difference in study design that may partly account for the discrepant findings. Taken together, these observations suggest that the diagnostic performance of K13 and UPK1B may depend on tumor origin and clinical context, and their utility for distinguishing metastatic OSCC from primary LSCC should be validated in larger, well-characterized cohorts specifically designed to address this clinical question. Beyond its diagnostic importance, the accurate distinction between pulmonary metastases of OSCC and primary LSCC has direct clinical implications. The management of recurrent or metastatic OSCC, including pulmonary metastases, typically prioritizes systemic therapy, such as platinum-based chemotherapy and immune checkpoint inhibitors targeting the PD-1/PD-L1 pathway, including pembrolizumab and nivolumab, particularly in cases of disseminated or unresectable disease. Solitary and resectable lesions may be considered for pulmonary metastasectomy under certain conditions, often using parenchyma-sparing sublobar resection9,11,32. In contrast, primary LSCC is managed according to standard lung cancer guidelines, which usually involve anatomic lung resection with nodal evaluation. Postoperative platinum-based adjuvant therapy may be indicated depending on tumor stage and biomarkers (e.g., PD-L1 expression and actionable driver alterations)10,33. Prognosis and long-term follow-up strategies also differ fundamentally between pulmonary metastasis of OSCC and primary LSCC, emphasizing the need for precise diagnostic classification. In daily practice, however, diagnostic tools remain suboptimal, particularly when histologic features and conventional IHC markers overlap. Our results provide a practical pathology-based solution by introducing K1 and K18 as complementary biomarkers that may be applied to routine FFPE specimens and readily interpreted by pathologists. Incorporating these markers into diagnostic workflows may improve clinical decision-making, ensure appropriate treatment selection, and ultimately optimize patient outcomes. Several limitations should be acknowledged regarding the generalizability of our results. The IHC performance of K1 and K18 was evaluated exclusively in our institutional cohort, and these markers were not tested in an independent external validation series. In addition, the WES-confirmed paired cases of OSCC and lung tumors (POL cases) were intended as a small proof-of-concept set to support the biological consistency of K1/K18 expression between primary and metastatic lesions, rather than to provide external validation. Thus, the diagnostic performance estimates should be interpreted as hypothesis-generating. Future multi-center studies incorporating independent external cohorts are needed to confirm the reproducibility and broader applicability of K1 and K18 for distinguishing pulmonary metastases of OSCC from primary LSCC. In conclusion, the integration of transcriptomic screening, IHC validation, and genomic confirmation underscores the practical diagnostic value of K1 and K18 as diagnostically informative and biologically plausible biomarkers for distinguishing pulmonary metastases of OSCC from primary LSCC in routine pathology. Methods Tissue samples FFPE tissue blocks were obtained from 39 patients with OSCC (70 ± 13 years; male/female = 25/14; pT stage: pT1/pT2–4, 10/29; pN stage: pN0/pN1–2/pNx = 9/2/28) and 92 patients with LSCC (67 ± 9 years; male/female = 83/9; pT1/pT2–4, 12/80; pN0/pN1–2/pNx = 54/28/10) treated at Hamamatsu University Hospital, Japan. In addition, FFPE tissue blocks from five paired cases consisting of a primary OSCC and a subsequent lung squamous cell carcinoma arising in the same patient were collected. These cases were designated as POL cases (Supplementary Table S1). All tissue samples were collected and analyzed with the approval of the Clinical Research Review Board of Hamamatsu University School of Medicine (approval number: 24–242). All methods were performed in accordance with the relevant guidelines and regulations, and the study was conducted in accordance with the Declaration of Helsinki. The requirement for informed consent was waived because of the retrospective nature of the study and the use of anonymized samples from pathological archives, with an opt-out procedure implemented through a public information notice. Collection of publicly available mRNA expression data Array-based mRNA expression data for OSCC (n = 99) were obtained from the Gene Expression Omnibus (GEO) database (accession number: GSE84846)20, maintained by the National Center for Biotechnology Information. RNA sequencing–based mRNA expression data for LSCC were retrieved from two independent cohorts: The Cancer Genome Atlas (TCGA) project (project ID: LUSC; n = 501)21 and the Clinical Proteomic Tumor Analysis Consortium (CPTAC; n = 108)22. Both datasets were accessed via the Genomic Data Commons data portal (https://portal.gdc.cancer.gov). Identification of genes differentially expressed between OSCC and LSCC To identify genes specifically expressed in OSCC relative to LSCC, mRNA expression data from GEO, TCGA, and CPTAC datasets were systematically analyzed. Genes upregulated in OSCC were selected using the following criteria: (1) a median expression value > 4 in OSCC samples from GSE84846; (2) a median expression value < 20 in LSCC samples from TCGA; and (3) a median expression value < − 0.15 in LSCC samples from CPTAC. Conversely, genes downregulated in OSCC compared with LSCC were identified using the following criteria: (1) a median expression value < − 2 in OSCC samples from GSE84846; (2) a median expression value > 3000 in LSCC samples from TCGA; and (3) a median expression value > 0 in LSCC samples from CPTAC. These thresholds were applied as stringent, cross-platform filters to prioritize transcripts exhibiting consistent directionality across GEO, TCGA, and CPTAC and clear separation between OSCC and LSCC, while yielding a manageable number of candidates for downstream validation. Immunohistochemical analysis FFPE tissue sections were subjected to IHC staining using the Histofine Simple Stain MAX PO kit (Nichirei, Tokyo, Japan), with 3,3′-diaminobenzidine tetrahydrochloride used as the chromogen, according to previously described protocols34. Details of the primary antibodies and staining conditions are provided in Supplementary Table S8. Positive immunoreactivity was defined as cytoplasmic staining for K1, K13, K18, and RPL27A, and as cytoplasmic and/or membranous staining for MYOF and UPK1B, as previously reported19,28,30,35,36. For these six markers, IHC expression was quantified using the H-score method, calculated by multiplying staining intensity (0 = negative; 1 = weak; 2 = moderate; 3 = strong) by the percentage of tumor cells exhibiting each intensity (range: 0–100%), resulting in a total H-score ranging from 0 to 300. Based on the cutoff proposed by Jeon et al37., H-scores ≥ 50 were considered positive, whereas H-scores < 50 were considered negative. The expression of p40, CK7, CK20, and ARL13B was assessed by IHC staining based on their expected subcellular localizations: nuclear for p40, cytoplasmic for CK7 and CK20, and primary cilia for ARL13B38–41. Unlike the six markers assessed using H-scores, these proteins were evaluated using predefined cutoff values for the proportion of positively stained tumor cells, as described previously14,42,43: 50% for p40, 10% for CK7 and CK20, and 15% for ARL13B. IHC images were acquired using a Leica DMD108 digital microimaging system (Leica Microsystems, Wetzlar, Germany). Whole-exome sequencing analysis Genomic DNA was extracted from OSCC tissues and matched adjacent noncancerous tissues of POL cases using the QIAamp DNA FFPE Advanced UNG Kit (QIAGEN, Limburg, Netherlands). The concentration and quality of double-stranded DNA (dsDNA) were assessed using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA), a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific), and a 4150 TapeStation system (Agilent Technologies, Santa Clara, CA, USA). DNA libraries were prepared using the Twist Library Preparation Enzymatic Fragmentation Kit 2.0 and enriched with the Twist Human Exome 2.0 Panel (Twist Bioscience, South San Francisco, CA, USA). Paired-end sequencing (150 bp) was conducted using an Illumina NovaSeq X Plus platform (Illumina, Inc., San Diego, CA, USA), targeting a minimum mean coverage of ≥ 300× for tumor samples and ≥ 150× for matched normal samples. Raw sequencing reads were processed using Fastp44 for adapter trimming and quality filtering. The cleaned reads were aligned to the GRCh38 primary human genome assembly (Ensembl release 95)45 using BWA-MEM46. Alignment files were processed and compressed using SAMtools47. Somatic variants were identified using three independent variant callers: Mutect2 (GATK4)48, Strelka249, and VarScan250, with a minimum variant allele frequency threshold of 0.10. Only variants detected by all three algorithms were considered high-confidence somatic mutations. Variant annotation was performed using ANNOVAR51. Functional impact predictions were obtained using FATHMM-MKL52. Mutation analysis by polymerase chain reaction and Sanger sequencing Genomic DNA was extracted from FFPE tissue sections of lung tumors from POL cases using the QIAamp DNA FFPE Advanced UNG Kit (QIAGEN). PCR was performed on lung tumor DNA to amplify genomic regions harboring somatic mutations identified in OSCC samples by WES. The PCR products were subjected to Sanger sequencing to validate the presence of these mutations in the corresponding lung lesions. The primer sequences are listed in Supplementary Table S9. Statistical analysis Statistical analyses were performed using R software version 4.4.153. For analyses based on dichotomized immunohistochemical H-scores, a predefined cutoff of ≥ 50 was used to classify cases as positive or negative. Associations between marker expression and tumor type were evaluated using Fisher’s exact test, and ORs with 95% CIs were calculated using the epitools package (version 0.5–10.1) in R. Sensitivity and specificity were also determined using these dichotomized H-scores. ROC curve analysis was performed using continuous H-scores to assess diagnostic performance. The AUC and its 95% CI were calculated using the pROC package (version 1.18.5) in R54, with 95% CIs estimated by bootstrap resampling with 200,000 replicates. Electronic Supplementary Material Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We acknowledge Mr. Takaharu Kamo for his technical help. This work was supported in part by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (JSPS) (25K10316). Author contributions Yoshiko Watanabe: Writing – original draft, Resources, Investigation, Writing – review & editing, Hisami Kato: Writing – original draft, Investigation, Formal analysis, Writing – review & editing, Rei Ishikawa: Writing – original draft, Formal analysis, Writing – review & editing, Kimio Yoshimura: Formal analysis, Writing – review & editing, Hidetaka Yamada: Writing – original draft, Investigation, Writing – review & editing, Ryosuke Miyazaki: Investigation, Writing – review & editing, Sayuri Takeuchi: Investigation, Writing – review & editing, S M Jahurul Haque: Investigation, Writing – review & editing, Katsuhiro Yoshimura: Data curation, Writing – review & editing, Yasuhiro Sakai: Data curation, Writing – review & editing, Miki Sugita: Resources, Writing – review & editing, Yusuke Inoue: Resources, Writing – review & editing, Yusuke Takanashi: Resources, Writing – review & editing, Keigo Sekihara: Resources, Writing – review & editing, Kazuhito Funai: Resources, Writing – review & editing, Kazuma Masumoto: Conceptualization, Resources, Writing – review & editing, Kazuya Shinmura: Writing – original draft, Formal analysis, Investigation, Data curation, Project administration, Writing – review & editing. Funding This work was supported in part by a Grant-in-Aid from the Japan Society for the Promotion of Science (25K10316 to K.S.). Data availability The expression datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) under accession number GSE84846 and in The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) through the Genomic Data Commons data portal ([https://portal.gdc.cancer.gov](https:/portal.gdc.cancer.gov)). The uncropped original immunohistochemical images corresponding to Figs. 1 and 3B, C, Supplementary Figure S1, and Supplementary Figure S2 are available in the GitHub repository: [https://github.com/kzshinmu/OSCC-IHC-uncropped-images](https:/github.com/kzshinmu/OSCC-IHC-uncropped-images). All other data generated or analyzed in this study are included in the main article (Results section and Tables 1, 2, 3, 4 and 5) and in the Supplementary Materials (Supplementary Figure S3 and Supplementary Tables S1–S9). Competing interests The authors declare no competing interests. Ethics Approval This study was approved by the Clinical Research Review Board of Hamamatsu University School of Medicine (approval no. 24–242). The requirement for informed consent was waived. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Kazuma Masumoto, Email: kazusan@hama-med.ac.jp. Kazuya Shinmura, Email: kzshinmu@hama-med.ac.jp. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The expression datasets analyzed in this study are publicly available in the Gene Expression Omnibus (GEO) under accession number GSE84846 and in The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) through the Genomic Data Commons data portal ([https://portal.gdc.cancer.gov](https:/portal.gdc.cancer.gov)). The uncropped original immunohistochemical images corresponding to Figs. 1 and 3B, C, Supplementary Figure S1, and Supplementary Figure S2 are available in the GitHub repository: [https://github.com/kzshinmu/OSCC-IHC-uncropped-images](https:/github.com/kzshinmu/OSCC-IHC-uncropped-images). All other data generated or analyzed in this study are included in the main article (Results section and Tables 1, 2, 3, 4 and 5) and in the Supplementary Materials (Supplementary Figure S3 and Supplementary Tables S1–S9).