An Integrative Machine Learning Approach Identifies a Cancer- Associated Fibroblast–Driven Predictive Model for Early-Stage Lung Squamous Cell Carcinoma

An Integrative Machine Learning Approach Identifies a Cancer- Associated Fibroblast–Driven Predictive Model for Early-Stage Lung Squamous Cell Carcinoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article An Integrative Machine Learning Approach Identifies a Cancer- Associated Fibroblast–Driven Predictive Model for Early-Stage Lung Squamous Cell Carcinoma Yuan Li, Fanping Zhang, Hongxia Duan, Wenhao Weng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8948144/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Lung squamous cell carcinoma (LUSC) remains associated with unfavorable clinical outcomes, even in early-stage disease. Increasing evidence indicates that the tumor microenvironment (TME), particularly cancer-associated fibroblasts (CAFs), plays a crucial role in tumor progression and therapeutic resistance. However, the prognostic and therapeutic implications of CAF-related molecular signatures in early-stage LUSC remain insufficiently defined. Methods Transcriptomic profiles and corresponding clinical data of patients with early-stage LUSC were retrieved from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) cohorts. Stromal and CAF infiltration levels were quantified, and key CAF-associated modules were identified using weighted gene co-expression network analysis (WGCNA). A CAF-related prognostic signature was subsequently constructed using multivariate Cox regression analysis. Functional enrichment analysis was performed to explore underlying biological mechanisms. The predictive value of the signature for chemotherapy and immune checkpoint blockade response was evaluated. In addition, FSTL3, a core gene within the model, was selected for in vitro validation to assess its effects on cell proliferation, colony formation, and apoptosis in LUSC cells. Results We established a four-gene CAF-related prognostic signature that effectively stratified patients into high- and low-risk groups with significantly different overall survival outcomes. Multivariate analysis confirmed that the CAF-based risk score served as an independent prognostic factor in early-stage LUSC. Enrichment analysis demonstrated that epithelial–mesenchymal transition (EMT) was prominently activated in the high-risk group. Furthermore, risk stratification based on the CAF signature was associated with differential predicted responses to chemotherapy and immune checkpoint inhibitors. Functional experiments revealed that FSTL3 knockdown significantly inhibited cell proliferation and colony formation while promoting apoptosis, supporting the biological relevance of the identified signature. Conclusions We established a robust CAF-derived four-gene prognostic model that effectively predicted patient survival and therapeutic response in early stage LUSC. Experimental validation highlighted the biological and clinical relevance of the CAF-associated signature and identified FSTL3 as a potential therapeutic target for early stage LUSC. cancer-associated fibroblasts lung squamous cell carcinoma prognostic biomarker tumor microenvironment checkpoint inhibitor FSTL3 Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 INTRODUCTION Lung cancer is the leading cause of cancer-related mortality worldwide [ 1 ]. Non-small cell lung cancer (NSCLC) accounts for approximately 85–90% of all cases, with lung squamous cell carcinoma (LUSC) representing one of its major histological subtypes [ 2 ]. Although advances in surgery, chemotherapy, and immunotherapy have improved outcomes in selected patient populations, therapeutic progress in LUSC has been comparatively limited, particularly due to the low frequency of actionable driver mutations and the absence of well-defined molecular subtypes amenable to targeted intervention Despite advances in surgery, chemotherapy, radiotherapy, and immunotherapy, therapeutic progress in LUSC has lagged behind that in lung adenocarcinoma (LUAD), largely due to the relatively low frequency of actionable driver mutations and the absence of clearly defined molecular subtypes suitable for targeted intervention [ 3 – 5 ]. For instance, a recent study in a Chinese cohort demonstrated that nearly 90% of LUAD patients carried mutations in just four key genes [ 5 ], underscoring the molecular tractability of LUAD compared with LUSC. Even among patients with early-stage NSCLC, approximately 25% experience postoperative recurrence, and most recurrent tumors are no longer amenable to curative resection[ 5 , 6 ]. Furthermore, the reduced prevalence of targetable genomic alterations in LUSC may contribute to limited benefit from precision therapies and immune checkpoint blockade [ 7 ]. These challenges highlight an urgent need to identify alternative biological determinants that drive disease progression and influence therapeutic responsiveness in LUSC. Tumor progression is increasingly recognized as a dynamic process shaped not only by malignant cells but also by the tumor microenvironment (TME) [ 8 ]. Reciprocal interactions between tumor cells and stromal components critically regulate invasion, metastasis, immune evasion, and treatment resistance [ 9 ]. Among stromal cell populations, cancer-associated fibroblasts (CAFs) represent a dominant and highly heterogeneous component of the TME [ 10 ]. CAFs encompass functionally diverse subtypes with both tumor-promoting and tumor-restraining properties, and their phenotypic plasticity allows them to profoundly influence tumor behavior [ 10 , 11 ]. Consequently, CAFs have emerged as promising prognostic biomarkers and potential therapeutic targets across multiple malignancies targets [ 12 ]. Recent single-cell transcriptomic studies have further demonstrated that specific CAF subpopulations are closely associated with patient survival and responsiveness to immunotherapy, including immune checkpoint blockade [ 10 ]. These findings suggest that CAF-related gene signatures may serve as clinically actionable tools to stratify patients, guide treatment decisions, and minimize unnecessary toxicity in non-responders. However, the prognostic and therapeutic relevance of CAF-associated molecular markers in early-stage LUSC remains insufficiently characterized. In the present study, we integrated transcriptomic data and CAF infiltration profiles from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) databases. Focusing specifically on stage I–II LUSC to enhance clinical applicability, we employed weighted gene co-expression network analysis (WGCNA) to identify CAF-related hub genes and constructed a CAF-based risk score incorporating both CAF infiltration levels and gene signature components. The prognostic and therapeutic implications of this model were systematically evaluated using Cox regression and functional enrichment analyses, and further validated in an independent cohort. Importantly, in vitro experiments confirmed the biological relevance of the identified signature, and functional assays identified FSTL3 as a potential therapeutic target in early-stage LUSC. MATERIALS AND METHODS 2.1 2.1. Data collection and preparation Gene profiles and clinical information (level 3 data) of The Cancer Genome Atlas Lung Squamous Cell Carcinoma (TCGA-LUSC) samples were downloaded from UCSC Xena data hubs ( https://xenabrowser.net/hub/ ). Gene expression data from TCGA were experimentally measured using the Illumina HiSeq 2000 RNA platform at TCGA genome characterization center, University of North Carolina. Genes were mapped onto the human genome coordinates using the UCSC Xena HUGO ProbeMap. Stages I-IIB were also selected. The gene-level transcription estimates were provided as log 2 (x + 1) transformed RSEM-normalized counts. The normalized gene expression profiles and clinical information of the GSE157010 [ 13 ] dataset were downloaded from the GEO database using the GEO2R package. The mean abundance was determined if one gene had multiple probes. Genes that did not have expression values in at least half of the samples were excluded from analysis. 2.2. Stromal score and CAF infiltration estimation ESTIMATE software was used to estimate stromal cell infiltration in tumor samples [ 14 ]. To estimate the CAF abundance, MCP_Counter (Microenvironment Cell Populations-counter) and xCell (cell-type enrichment analysis of bulk transcriptomes) methods were additionally followed using the R package ‘immunedeconv’ [ 15 , 16 ]. 2.3. Weighted gene co-expression network analysis The top 5,000 genes with the highest median absolute deviations (MAD) were selected from the TCGA-LUSC dataset. Gene expression and traits (xCell CAF score and ESTIMATE stromal score) of TCGA-LUSC samples were used to construct a co-expression network using the R package WGSCNA. Co-expression modules that included highly correlated genes were identified. The eigengene of each module was evaluated using the first principal component of module expression. Hub genes were identified by measuring gene significance (GS) for traits and module membership (MM). 2.4. CAF-related prognostic model construction and validation Univariate COX regression analysis of the hub genes was performed using the“coxphw” function of R package“survival.” Genes with significant p-values in both TCGA-LUSC and GSE157010 datasets were selected for subsequent LASSO (Least Absolute Shrinkage and Selection Operator (LASSO). LASSO Cox regression analysis was used to select candidate genes that were highly associated with OS. A Cox proportional hazards model was constructed using TCGA-LUSC dataset. This model was also used to calculate the risk score of the GSE157010 dataset. Survival analysis between patients with high- and low-risk scores was performed using R package survminer. 2.5. Enrichment analysis Gene ontology gene set overrepresented analysis (GSOA) was performed by Metascape online service [ 17 ]. All genes in the genome were used as the background. Gene sets with a p-value < 0.01 were grouped into clusters based on their membership similarities. The most statistically significant term within a cluster was selected to represent that cluster. Differential gene expression analyses were performed using the LIMMA package. Gene set enrichment (GSEA) was performed using the clusterProfiler package, whereas single-sample GSEA (ssGSEA) was performed using the GSVA method. 2.6 Prediction of response to chemo- and immuno- therapy The R package “pRRophetic” was used to predict drug sensitivity against Cisplatin, Paclitaxel, Docetaxel, Gemcitabine, Vinorelbine and Etoposide, which are listed in https://www.cancer.org/cancer/lung-cancer/treating-non-smallcell/chemotherapy.html . IC 50 values were transformed using Ln. Finally, the TIDE ( http://tide.dfci.harvard.edu/ ) online service was used to predict responses blockade therapy. 2.7 Cell lines and cell culture Human LUSC cell lines (H226 and H2170) and normal human bronchial epithelial cell line (BEAS-2B) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). H226 and H2170 cells were cultured in RPMI-1640 medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS; HyClone, GE Healthcare Life Sciences, Logan, UT, USA), 100 U/mL penicillin, and 0.1 mg/mL streptomycin (Gibco). BEAS-2B cells were maintained in DMEM/F12 medium (1:1), supplemented with 10% FBS and 1% penicillin-streptomycin. All the cells were incubated at 37°C in a humidified atmosphere containing 5% CO₂. Adherent cells were detached using 0.05% trypsin-EDTA (Invitrogen, Carlsbad, CA, USA) for passaging or further experiments. 2.8. Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) assays Total RNA was isolated from LUSC cell lines using the miRNeasy Serum/Plasma Kit (Qiagen) according to the manufacturer’s instructions. RNA quality and purity were evaluated using the NanoDrop system (Thermo Fisher Scientific). The RNA samples were immediately stored at -80°C for future use. cDNA was synthesized using the PrimeScript RT Reagent Kit (Takara, Shiga, Japan). RNA expression levels were quantified using the SensiFAST SYBR LO-ROX Kit (Bioline, London, UK) on a QuantStudio 6 Flex RT-PCR System (Applied Biosystems, Foster City, CA, USA). ACTB was used as the internal control for mRNA quantification. The relative RNA expression levels were determined using the 2-ΔΔ CT method. Primers used in this study are listed in Suppl. Table. 1 2.9 Transfection of Small Interfering RNA H226 and H2170 cells were seeded in 6-well plates at a density of 2.5 × 10 cells per well. Two distinct double-stranded siRNAs targeting separate coding regions of FSTL3 (or a Silencer negative control siRNA (Thermo Fisher Scientific] ) were transfected using Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA, USA) following the manufacturer’s protocol. After 48 h, the transfected cells were harvested for total RNA and protein extraction, as well as for downstream functional assays. 2.10 Colony formation assay For the colony formation assay, the cells were seeded at a density of 800 cells/well in 6-well plates. Colony formation was allowed to proceed for 7–10 days, with the culture medium refreshed every three days. After incubation, the cell colonies were fixed with 4% paraformaldehyde in PBS (Thermo Fisher Scientific, Fair Lawn, NJ, USA) for 30 min and stained overnight with 0.1% crystal violet solution (Thermo Fisher Scientific). The colonies were quantified using ImageJ software, which allowed the assessment of the relative changes in colony formation. 2.11 Invasion assay Cell invasion was assessed using BioCoat Matrigel Invasion Chambers with 8.0 µm pore polyester membranes (BD Biosciences, Franklin Lakes, NJ, USA). Cells (5 × 10 cells/well) were seeded into inserts containing serum-free medium and placed in 24-well plates containing medium supplemented with 10% FBS. After 48 h, cells that had migrated to the lower surface of the membrane were fixed and stained using a Diff-Quik kit (Thermo Fisher Scientific). 2.12 Cellular apoptosis assay Cell apoptosis was assessed using the Muse™ Annexin V & Dead Cell Kit (Luminex Corp, Austin, TX, USA) according to the manufacturer’s instructions. Briefly, 5 × 10⁵ cells were collected and 100 µL of the suspension was mixed with 100 µL of Muse™ Annexin V & Dead Cell Reagent. The percentage of apoptotic cells was measured using a Muse™ Cell Analyzer (Millipore Corp., Billerica, MA, USA). RESULTS 3.1 Patient demographics and characteristics The present study included data from two publicly available datasets of patients with early stage LUSC (stages I-II), including the TCGA dataset (TCGA-LUSC; n = 406) and GEO dataset (GSE157010; n = 235) (Table 1 ). The median ages of the patients in TCGA and GEO datasets were 69 and 68 years, respectively. OS and gene expression data were available for both cohorts of patients and used for the analyses. Table 1 Demographic and clinical information of the cohorts included in this study. TCGA (n = 406) GSE157010 (n = 235) Sex Male 297 (73.1%) 153 (60.5%) Female 109 (26.9%) 82 (39.5%) Age Median 69 68 Range 39–90 46–89 Pathologic stage Stage I-IB 244 (60.0%) not available Stage II-IIB 162 (40.0%) not available T stage T1 not available 63 (26.8%) T2 not available 139 (59.1%) T3 not available 31 (13.2%) NA not available 2 (0.9%) Follow up Alive 241 (59.4%) 119 (50.6%) Death 165 (46.6%) 116 (49.4%) 5-years survival rate 45/406 97/235 Mean follow up time 33 46 3.2. High stromal score and CAF infiltration are associated with poor OS Subjective scoring of the patterns of stromal tissues within the TME is an interesting and simplified way of defining cancer-associated alterations of the stromal tissue by combining a variety of stromal tissue components, including CAFs and their abundance, into a single stromal score, which has been shown to be of significant prognostic value in different studies both as an independent prognosticator of clinical outcome and in combination with conventional staging systems [ 18 ]. Interestingly, the tumor stromal score has been shown to significantly correlate with the CAF score in various solid tumors such as colorectal cancer [ 18 ]. Accordingly, we investigated whether stromal infiltration into the TME could be a prognostic factor in patients with LUSC. To do this, we first calculated an “Estimate stromal score” and studied its association with OS of the LUSC patients. Our results indicated a significant association between higher stromal scores and poor OS of patients in both TCGA and GEO cohorts ( p = 0.0081 and p = 0.013, Fig. 1 A and 1 B, respectively). In addition to OS, patients with higher stromal scores in the TCGA cohort had lower disease-specific survival (DSS) ( Supp. Figure 1 ). However, owing to the unavailability of DSS data, we were unable to perform a similar analysis in the GEO cohort. To more specifically address CAF abundance within the TME, CAF infiltration was calculated for both cohorts of patients using MCP_Counter and xCell CAF methods. patients with LUSC with higher CAF infiltration, as indicated by both methods, had shorter OS ( p ≤ 0.01). Kaplan–Meier survival plots for patients with high vs. low stromal scores and CAF infiltration are presented in Fig. 1 C-F. 3.3. WGCNA identifies hub CAF-associated genes Given the associations of stroma score and CAF infiltration with OS in the studied LUSC patients, we sought to identify hub CAF-associated genes using WGCNA analysis. To construct the scale-free topology co-expression network, a soft threshold power (β) of 3 was estimated (scale-free R2 = 0.936) in the TCGA LUSC dataset of early-stage patients ( Supp. Figure 2 A). Hierarchical clustering revealed 8 co-expression modules ( Supp. Figure 2 B), of which the blue and turquoise modules had the strongest positive correlation with stroma score and CAF infiltration (Fig. 3 A). Therefore, we continued to identify the hub genes in these two modules. Using a set of minimal criteria including an MM score of over 0.5, GS score of over 0.4 and the significance p value of less than 0.05 in the cox regression analysis, a total of 271 genes from the blue and turquoise modules were identified and selected as the hub CAF-associated genes. Using these hub genes, we performed GO biological process enrichment analysis to identify the significant biological gene clusters. Our analysis revealed 20 clusters with significant biological pathways (Fig. 2 B. Among these clusters, CAF-associated genes were highly associated with the inflammatory response, response to external stimuli, immune response, and regulation of the MAPK cascade, indicating the potential function(s) of CAF-associated genes. 3.4. Construction of the stromal CAF-based prognostic risk model Following the identification of hub CAF-associated genes, we investigated whether these genes could act as independent prognosticators of OS using univariate Cox regression analysis. Among the 271 studied genes, 19 were significantly correlated with OS in both cohorts (Fig. 3 A). Lasso Cox regression analysis was performed to develop a prognostic CAF-associated gene signature ( Supp. Figure 3 ). A final set of four genes was selected to construct the Cox model (formula: 0.0629× PLXNA4 + 0.0775 × C1QTNF1 + 0.0564 × GFPT2 + 0.0908 × FSTL3). The LUSC patients in TCGA cohort were then assigned to two high-risk and low-risk groups using our 4-gene CAF signature, and survival analysis was performed. Our results showed that patients with LUSC in the high-risk group had significantly shorter OS than those in the low-risk group ( p = 8.0e-3) (Fig. 3 B). We validated the developed gene signature in a GEO cohort using the same model. The results were consistent with those of TCGA cohort, indicating that patients with lower risk scores had shorter OS ( p = 5.3e-4). These findings show that the 4-gene-based risk score is significantly associated with patient prognosis in early stage LUSC. 3.5. Functional analysis reveals the hallmark inflammatory response pathway To investigate the potential functional differences between high- and low-risk patients, differential expression analysis and hallmark gene set enrichment analysis were performed for the GEO cohort. The analysis identified 10 hallmark pathways with the “epithelial mesenchymal transition” (EMT) being the most relevant pathway. The ridge plot demonstrating the results of the GSEA analysis is shown in Fig. 4 A. To further validate the EMT association, a similar analysis was performed based on the gene ontology resource. These results were in accordance with those obtained from the GSEA analysis, indicating that EMT-associated genes were the most upregulated genes in the high-risk LUSC patients of the studied cohort (Fig. 4 B). Finally, single-sample GSEA, in which an EMT hallmark enrichment score was assigned to each sample, was performed. We found that our CAF risk score was highly associated with the EMT hallmark enrichment scores of patients in both the TCGA and GEO datasets (r = 0.84, p ≤ 0.0001 and r = 0.78, p ≤ 0.0001, respectively) (Fig. 4 C-D). The significant differences in EMT hallmark genes between the high- and low-risk patients may at least partially explain the association of the higher risk score with poor OS. 3.6 CAF-based signature is associated with chemo- and immuno-therapy response To delineate the possibility that a lower clinical response to therapy is responsible for poor prognosis in high-risk LUSC patients, we decided to study whether our CAF-associated gene signature was associated with the clinical response to therapy. To this end, the sensitivity of patients in different risk groups to various chemotherapeutic agents, including cisplatin, paclitaxel, docetaxel, gemcitabine, vinorelbine, and etoposide, was predicted. Ln-transformed IC 50 values were used to differentiate patients with varying degrees of sensitivity to chemotherapeutic drugs. Interestingly, we found that the sensitivity to three drugs (Paclitaxel, Gemcitabine and Vinorelbine) was significantly different between the high- and low-risk CAF groups of patients in both datasets (Fig. 5 A). Next, we attempted to predict the response to immunotherapy by using the CAF-based risk score. Similarly, our 4-gene CAF signature was able to differentiate responders from non-responders to immunotherapy in both TCGA and GEO datasets, indicating that patients with higher risk scores were less likely to respond to immunotherapy (Fig. 5 B-C). 3.6 CAF-based signature is associated with chemo- and immuno-therapy response To delineate the possibility that a lower clinical response to therapy is responsible for poor prognosis in high-risk LUSC patients, we decided to study whether our CAF-associated gene signature was associated with the clinical response to therapy. To this end, the sensitivity of patients in different risk groups to various chemotherapeutic agents, including cisplatin, paclitaxel, docetaxel, gemcitabine, vinorelbine, and etoposide, was predicted. Ln-transformed IC 50 values were used to differentiate patients with varying degrees of sensitivity to chemotherapeutic drugs. Interestingly, we found that the sensitivity to three drugs (Paclitaxel, Gemcitabine and Vinorelbine) was significantly different between the high- and low-risk CAF groups of patients in both datasets (Fig. 5 A). Next, we attempted to predict the response to immunotherapy by using the CAF-based risk score. Similarly, our 4-gene CAF signature was able to differentiate responders from non-responders to immunotherapy in both TCGA and GEO datasets, indicating that patients with higher risk scores were less likely to respond to immunotherapy (Fig. 5 B-C). 3.7 Identification of the biological role of a four-gene CAF signature in LUSC To further confirm the specific biological role of the four-gene signature, we compared the mRNA expression levels of the four genes between a normal human bronchial epithelial cell line and the two LUSC cell lines. qRT-PCR analysis revealed that, compared with normal bronchial epithelial cells, FSTL3 consistently showed elevated expression in both LUSC cell lines, whereas the other three genes did not display consistent changes (Fig. 6 A). To explore the functional significance of FSTL3 in LUSC, siRNA-mediated knockdown was performed in H226 and H2170 cells. qRT-PCR confirmed that transfection with FSTL3-specific siRNA efficiently reduced FSTL3 mRNA expression 48 h after transfection (Fig. 6 B). indicated successful gene silencing. Functionally, colony formation assays demonstrated that FSTL3 knockdown significantly inhibited clonogenic growth in both the cell lines ( p < 0.05; Fig. 7 A). This suggests that FSTL3 contributes to the proliferative capacity of LUSC. We examined the efficacy of FSTL3 in suppressing cellular invasion, which contributes to cancer progression and metastasis. The transwell assay. Knockdown of FSTL3 inhibited cell invasion significantly more effectively than individual treatments in both the LUSC cell lines ( p < 0.05; Fig. 7 B). Finally, to determine whether FSTL3 affected apoptotic regulation, an Annexin V-FITC/PI binding assay was performed. The results revealed a significant increase in apoptotic cell populations following FSTL3 knockdown in both LUSC cell lines ( p < 0.05; Fig. 7 C), indicating that FSTL3 functions as an oncogenic factor in LUSC, promoting cell proliferation and invasion and inhibiting apoptosis. These findings highlight FSTL3 as a potential therapeutic target and a critical downstream effector of the CAF-related gene signature in LUSC. DISCUSSION Given the unparalleled prevalence and mortality of NSCLS as well as its poor prognosis and unresponsiveness to therapy, the search for relevant prognostic and predictive markers has been the center of attention in recent years. Nonetheless, such attempts to find clinically reliable prognostic markers have not been very successful, calling for further research, herein [ 19 ]. A great deal of Evidence suggests that in addition to the genetic characteristics of the primary tumor, the fingerprint of the reciprocal interactions between tumor cells and their surrounding stroma is a critical determinant of tumor fate and progression, which can be exploited to predict tumor behavior [ 20 , 21 ]. Fibroblasts within the tumor tissue (CAFs) are among the major cell types of the tumor stroma that affect tumor progression through their interactions with other stromal cells as well as tumor cells [ 22 , 23 ]. Given the heterogeneity of CAFs within different tumors, profiling can provide useful information regarding the future behavior of the tumor and its functional assessment [ 22 ]. Supporting this notion, stroma-associated markers have been shown to be better prognostic tools than tumor-intrinsic parameters in prognostic tools [ 20 ]. In line with this, stromal score has been previously used by others as a prognostic marker associated with survival and is calculated based on a weighted sum of different types of stromal tissues that are above a certain threshold [ 18 ]. Regarding LUSC, certain CAF populations have been reported to be associated with higher cancer cell proliferation [ 22 ], accumulation of single-nucleotide variants (SNPs), and tumor growth [ 24 ]. However, to date, there is no information regarding the prognostic role of CAF infiltration in patients with early stage LUSC. To examine this possibility, stromal scores and the extent of CAF infiltration were calculated in two cohorts of LUSC patients from TCGA and GEO databases. Survival analysis revealed that a higher stromal score and CAF infiltration, calculated using two different methods, were associated with OS and DSS in patients with LUSC (Fig. 1 and Suppl. Figure 1 ). These findings are in line with the results of other studies indicating the role of CAFs and tumor stroma in the progression and prognosis of various solid tumors including lung cancer [ 19 , 25 , 26 ]. Although pathological slides from tumor specimens stained with hematoxylin–eosin (HE) are often available for many solid cancer patients [ 18 ], objective analysis of the slides to obtain reliable stroma-related prognostic markers is not common, is cumbersome and is largely affected by the pathologist/user variation and thus a mRNA-based gene expression score representing the same prognostic value is more valuable [ 14 ]. Given that surgery in combination with chemotherapy and radiotherapy is the treatment of choice for patients with stage I-II LUSC patients [ 27 ], tumor specimens from resected tumor tissues are always available. Therefore, the development of tissue-based mRNA expression markers is a clinically feasible strategy to identify prognostic markers. Accordingly, using tumor tissue mRNA expression data from both cohorts, we performed WGCNA analysis to first identify the hub genes associated with CAF and stromal score and then check whether these CAF-associated genes can independently predict OS. Our analysis identified 271 hub genes, of which a final set of 4 were used to construct a prognostic model for patients with LUSC. Our 4-gene CAF panel was able to independently predict OS in patients with LUSC in TCGA cohort. The performance of the gene signature was subsequently confirmed in GEO patients, which were used as the validation cohort (Fig. 3 C ) . These findings suggest that our 4-gene CAF signature can potentially be used as a prognostic marker in patients with stage I-II LUSC who have undergone surgery. Although the association between stromal score and OS has been previously shown in LUSC patients, the scores were calculated based on a deep analysis of pathological slides for late-stage patients, whereas our CAF panel can predict OS in early stage patients using only gene expression data [ 18 ]. Furthermore, in contrast to our study, most previous reports on the prognostic role of tumor stroma and CAF-related markers have only used single markers; however, combinational panels of markers have been shown to be superior to single-gene markers in terms of prognostic accuracy [ 28 ]. Nevertheless, the number of patients included in the present study was limited, and these findings need to be further validated in larger cohorts of patients in prospectively designed studies. Biological pathway enrichment analysis of the identified CAF-associated hub genes revealed their role in EMT. EMT has been linked to various malignancies, including lung cancer, and has well-established roles in tumor metastasis, progression, and recurrence [ 29 – 31 ], which may at least partially explain the mechanism responsible for worse prognosis in patients with higher CAF and stromal risk scores. Among the genes used to construct the CAF-associated prognostic model, glutamine-fructose-6-phosphate transaminase 2 ( GFPT2) plays a critical role as a regulator of metabolomic reprogramming in lung cancer [ 32 ]. Additionally, upregulation of follistatin-like 3 ( FSTL3 ) has been shown to promote the proliferation and migration of NSCLC cells through regulation of the lncRNA DSCAM-AS1/miR-122-5p Axis [ 33 ]. These findings emphasize the importance of the identified gene signature in LUSC pathogenesis and justify their combined use in prognostic panels for early prediction of OS. Importantly, among patients with stage I and II LUSC who underwent surgical resection, recurrence occurred in up to 35% of the cases [ 13 ], which could be at least partially due to the ineffectiveness of therapy. Therefore, the prediction of response to therapy in patients with stage I and II LUSC can identify patients who may benefit the most from the selected treatment options and spare those in whom the therapy is unlikely to be effective. In line with this goal, we also tried to determine whether our CAF panel could predict the response to chemotherapy and immunotherapy. Our results showed that patients with lower risk scores were more likely to respond to certain chemotherapy drugs, including Paclitaxel, Gemcitabine and Vinorelbine, as well as immunotherapy. Although immunotherapy has revolutionized the treatment of lung cancer, the majority of patients (60–80%) do not benefit from it [ 34 ], making the identification of these patients paramount. This is especially important since selection of best treatment options tailored based on the patients` CAF status at early stages of adjuvant therapy may decrease the chances of recurrence and spare those patients who may benefit the least from certain treatments from the toxicities associated with therapy. In support of our findings, it has been previously shown that the presence of CD90 + stromal cells within the TME can functionally inactivate tumor-infiltrating lymphocytes [ 35 ]. Furthermore, among our 4-gene CAF panel, Plexin-A4 (Plxna4) is shown to be upregulated in lung cancer cells and linked to the impaired homing capacity of cytotoxic T lymphocytes to the tumor site [ 33 , 36 ]. These findings may explain the series of immunomodulatory events mediated by CAFs and stromal cells, which could potentially reduce the efficacy of immunotherapy in patients less effective Taken together, our findings highlight the significance of CAF-related gene expression profiling for determining the prognosis and response to therapy in patients with LUSC. Our 4-gene CAF signature can independently predict OS and response to chemotherapy and immunotherapy in patients with early stage LUSC undergoing therapeutic resection. The prognostic and predictive values of this CAF-related panel need to be further validated in large-scale cohorts of patients alone and in combination with conventional clinical parameters. CONCLUSIONS In summary, our study demonstrated that CAFs and the extent of their infiltration into the tumor microenvironment are strongly associated with survival and therapeutic response in patients with LUSC. A CAF-related four-gene signature serves as an independent prognostic factor and predictor of the response to chemotherapy and immunotherapy. Among these genes, FSTL3 has emerged as a key effector that promotes proliferation and invasion, and inhibits apoptosis in LUSC cells. These findings highlight the critical role of the tumor stroma and CAFs in shaping tumor behavior and underscore the potential of CAF-based gene signatures, particularly FSTL3, as prognostic markers and therapeutic targets in lung squamous cell carcinoma. Abbreviations LUSC: Lung squamous cell carcinoma NSCLC: non-small cell lung cancer OS: overall survival TME: tumor microenvironment CAFs: cancer-associated fibroblasts WGCNA: weighted gene co-expression network analysis EMT: epithelial–mesenchymal transition FBS: fetal bovine serum CCK-8: Cell Counting Kit-8 TCGA: The Cancer Genome Atlas GEO: Gene Expression Omnibus qRT-PCR: Quantitative reverse transcription PCR RT: room temperature WB: Western blotting Declarations Acknowledgments: We acknowledge and appreciate our colleagues for their valuable suggestions and technical assistance in this study. Data Availability Statement: All datasets used in this study are publicly available. The TCGA-LUSC cohort was obtained from The Cancer Genome Atlas (TCGA) through the Genomic Data Commons (GDC) data portal (https://portal.gdc.cancer.gov/) under the accession TCGA-LUSC. The bulk RNA-seq dataset GSE157010 was downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE157010. Tumor immune dysfunction and exclusion (TIDE) scores were obtained from the TIDE web server (http://tide.dfci.harvard.edu/). All the codes used for data processing and analysis are available from the corresponding author upon request. Ethical approval This study did not require ethical approval as all data analyzed were obtained from publicly accessible databases and previously published studies. Consent to participate Not applicable. This research did not involve any new human participants. Consent to publish All authors have read and approved the final version of the manuscript and consented to its submission for publication. Author Contribution Yuan Li, Fanping Zhang, Hongxia Duan, and Wenhao Weng conceived of and designed the experiments. Yuan Li and Fanping Zhang analyzed the data. Yuan Li conducted all cell experiments. Yuan Li, Fanping Zhang, Hongxia Duan and Wenhao Weng prepared and revised the manuscript. All authors have read and approved the final version of the manuscript. References R.L. Siegel, K.D. Miller, H.E. Fuchs, A. Jemal, Cancer statistics, 2022, CA: a cancer journal for clinicians, (2022). Z. Cheng, C. Yu, S. Cui, H. Wang, H. Jin, C. Wang, B. Li, M. Qin, C. Yang, J. He, circTP63 functions as a ceRNA to promote lung squamous cell carcinoma progression by upregulating FOXM1, Nature communications, 10 (2019) 1–13. A.A. Schegoleva, A.A. Khozyainova, A.A. Fedorov, T.S. Gerashchenko, E.O. Rodionov, E.B. Topolnitsky, N.A. Shefer, O.V. Pankova, A.A. Durova, M.V. Zavyalova, Prognosis of different types of non-small cell lung cancer progression: current state and perspectives, Cell Physiol Biochem, 55 (2021) 29–48. N.A. Rizvi, M.D. Hellmann, J.R. Brahmer, R.A. Juergens, H. Borghaei, S. Gettinger, L.Q. Chow, D.E. Gerber, S.A. Laurie, J.W. Goldman, Nivolumab in combination with platinum‐based doublet chemotherapy for first-line treatment of advanced non–small-cell lung cancer, Journal of Clinical Oncology, 34 (2016) 2969. Y. Zhang, Y. Sun, Y. Pan, C. Li, L. Shen, Y. Li, X. Luo, T. Ye, R. Wang, H. Hu, Frequency of Driver Mutations in Lung Adenocarcinoma from Female Never-Smokers Varies with Histologic Subtypes and Age at DiagnosisDriver Mutations in Lung Adenocarcinoma, Clinical cancer research, 18 (2012) 1947–1953. P. Kumari, G.K. Vishwakarma, A. Bhattacharjee, Thresholding of prominent biomarkers of breast cancer on overall survival using classification and regression tree, Cancer Biomarkers, (2021) 1–10. A. Addeo, A. Passaro, U. Malapelle, G.L. Banna, V. Subbiah, A. Friedlaender, Immunotherapy in non-small cell lung cancer harbouring driver mutations, Cancer treatment reviews, 96 (2021). D.F. Quail, J.A. Joyce, Microenvironmental regulation of tumor progression and metastasis, Nature medicine, 19 (2013) 1423–1437. J.H. Zhang, C.Y. Song, Y. Tian, X.Y. Yang, Single-Cell RNA Sequencing in Lung Cancer: Revealing Phenotype Shaping of Stromal Cells in the Microenvironment, Front Immunol, 12 (2022). P.M. Galbo, X.X. Zang, D.Y. Zheng, Molecular Features of Cancer-associated Fibroblast Subtypes and their Implication on Cancer Pathogenesis, Prognosis, and Immunotherapy Resistance, Clinical Cancer Research, 27 (2021) 2636–2647. A.F. Irvine, S. Waise, E.W. Green, B. Stuart, G.J. Thomas, Characterising cancer-associated fibroblast heterogeneity in non-small cell lung cancer: a systematic review and meta-analysis, Sci Rep, 11 (2021) 3727. S. Togo, U.M. Polanska, Y. Horimoto, A. Orimo, Carcinoma-associated fibroblasts are a promising therapeutic target, Cancers, 5 (2013) 149–169. R. Bueno, W.G. Richards, D.H. Harpole, K.V. Ballman, M.-S. Tsao, Z. Chen, X. Wang, G. Chen, L.R. Chirieac, M.H. Chui, Multi-institutional prospective validation of prognostic mRNA signatures in early stage squamous lung cancer (Alliance), Journal of Thoracic Oncology, 15 (2020) 1748–1757. K. Yoshihara, M. Shahmoradgoli, E. Martínez, R. Vegesna, H. Kim, W. Torres-Garcia, V. Treviño, H. Shen, P.W. Laird, D.A. Levine, Inferring tumour purity and stromal and immune cell admixture from expression data, Nature communications, 4 (2013) 1–11. E. Becht, N.A. Giraldo, L. Lacroix, B. Buttard, N. Elarouci, F. Petitprez, J. Selves, P. Laurent-Puig, C. Sautès-Fridman, W.H. Fridman, Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression, Genome biology, 17 (2016) 1–20. D. Aran, Cell-type enrichment analysis of bulk transcriptomes using xCell, Bioinformatics for Cancer Immunotherapy, Springer2020, pp. 263–276. Y. Zhou, B. Zhou, L. Pache, M. Chang, A.H. Khodabakhshi, O. Tanaseichuk, C. Benner, S.K. Chanda, Metascape provides a biologist-oriented resource for the analysis of systems-level datasets, Nature communications, 10 (2019) 1–10. J.N. Kather, J. Krisam, P. Charoentong, T. Luedde, E. Herpel, C.-A. Weis, T. Gaiser, A. Marx, N.A. Valous, D. Ferber, Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study, PLoS medicine, 16 (2019) e1002730. R.M. Bremnes, T. Dønnem, S. Al-Saad, K. Al-Shibli, S. Andersen, R. Sirera, C. Camps, I. Marinez, L.-T. Busund, The role of tumor stroma in cancer progression and prognosis: emphasis on carcinoma-associated fibroblasts and non-small cell lung cancer, Journal of thoracic oncology, 6 (2011) 209–217. I.O. Bello, M. Vered, D. Dayan, A. Dobriyan, R. Yahalom, K. Alanen, P. Nieminen, S. Kantola, E. Läärä, T. Salo, Cancer-associated fibroblasts, a parameter of the tumor microenvironment, overcomes carcinoma-associated parameters in the prognosis of patients with mobile tongue cancer, Oral oncology, 47 (2011) 33–38. R. Kalluri, M. Zeisberg, Fibroblasts in cancer, Nature Reviews Cancer, 6 (2006) 392–401. H. Nakamura, M. Sugano, T. Miyashita, H. Hashimoto, A. Ochiai, K. Suzuki, M. Tsuboi, G. Ishii, Organoid culture containing cancer cells and stromal cells reveals that podoplanin-positive cancer-associated fibroblasts enhance proliferation of lung cancer cells, Lung Cancer, 134 (2019) 100–107. S. Neri, H. Hashimoto, H. Kii, H. Watanabe, K. Masutomi, T. Kuwata, M. Tsuboi, K. Goto, A. Ochiai, G. Ishii, Cancer cell invasion driven by extracellular matrix remodeling is dependent on the properties of cancer-associated fibroblasts, Journal of cancer research and clinical oncology, 142 (2016) 437–446. Y. Zhu, L. Zhang, H. Zha, F. Yang, C. Hu, L. Chen, B. Guo, B. Zhu, Stroma-derived fibrinogen-like protein 2 activates cancer-associated fibroblasts to promote tumor growth in lung cancer, Int. J. Biol. Sci., 13 (2017) 804–814. M.L. Wikberg, S. Edin, I.V. Lundberg, B. Van Guelpen, A.M. Dahlin, J. Rutegård, R. Stenling, Å. Öberg, R. Palmqvist, High intratumoral expression of fibroblast activation protein (FAP) in colon cancer is associated with poorer patient prognosis, Tumor Biology, 34 (2013) 1013–1020. S. Miura, N. Mitsuhashi, H. Shimizu, F. Kimura, H. Yoshidome, M. Otsuka, A. Kato, T. Shida, D. Okamura, M. Miyazaki, Fibroblast growth factor 19 expression correlates with tumor progression and poorer prognosis of hepatocellular carcinoma, BMC cancer, 12 (2012) 1–15. D. Anusewicz, M. Orzechowska, A.K. Bednarek, Lung squamous cell carcinoma and lung adenocarcinoma differential gene expression regulation through pathways of Notch, Hedgehog, Wnt, and ErbB signalling, Scientific reports, 10 (2020) 1–15. G. Sebastiani, A. Alberti, Non invasive fibrosis biomarkers reduce but not substitute the need for liver biopsy, World journal of gastroenterology: WJG, 12 (2006) 3682. Y. Otsuki, H. Saya, Y. Arima, Prospects for new lung cancer treatments that target EMT signaling, Developmental Dynamics, 247 (2018) 462–472. D. Greaves, Y. Calle, Epithelial Mesenchymal Transition (EMT) and Associated Invasive Adhesions in Solid and Haematological Tumours, Cells, 11 (2022) 649. J.-Y. Shih, P.-C. Yang, The EMT regulator slug and lung carcinogenesis, Carcinogenesis, 32 (2011) 1299–1304. W. Zhang, G. Bouchard, A. Yu, M. Shafiq, M. Jamali, J.B. Shrager, K. Ayers, S. Bakr, A.J. Gentles, M. Diehn, GFPT2-expressing cancer-associated fibroblasts mediate metabolic reprogramming in human lung adenocarcinoma, Cancer research, 78 (2018) 3445–3457. L. Gao, X. Chen, Y. Wang, J. Zhang, Up-regulation of FSTL3, regulated by lncRNA DSCAM-AS1/miR-122-5p Axis, promotes proliferation and migration of non-small cell lung Cancer cells, OncoTargets and therapy, 13 (2020) 2725. A. Ribas, J.D. Wolchok, Cancer immunotherapy using checkpoint blockade, Science, 359 (2018) 1350–1355. H. Yang, S. Berezowska, P. Dorn, P. Zens, P. Chen, R.-W. Peng, T.M. Marti, G.J. Kocher, R.A. Schmid, S.R. Hall, Tumor-infiltrating lymphocytes are functionally inactivated by CD90+ stromal cells and reactivated by combined Ibrutinib and Rapamycin in human pleural mesothelioma, Theranostics, 12 (2022) 167. W. Celus, A.I. Oliveira, S. Rivis, H.H. Van Acker, E. Landeloos, J. Serneels, S.T. Cafarello, Y. Van Herck, R. Mastrantonio, A. Köhler, Plexin-A4 Mediates Cytotoxic T-cell Trafficking and Exclusion in Cancer, Cancer immunology research, 10 (2022) 126–141. Additional Declarations No competing interests reported. Supplementary Files Tables.docx SupplementalMaterials.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8948144","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":607936517,"identity":"ae1f5dd9-80a4-4db7-a8f8-fd2d34b27d36","order_by":0,"name":"Yuan Li","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Li","suffix":""},{"id":607936518,"identity":"9359b384-1866-49ec-ab84-4fb3b9cbe15d","order_by":1,"name":"Fanping Zhang","email":"","orcid":"","institution":"Tongji University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Fanping","middleName":"","lastName":"Zhang","suffix":""},{"id":607936519,"identity":"04db8cf8-2e7e-4ea5-8a8d-d8992b8f09cf","order_by":2,"name":"Hongxia Duan","email":"","orcid":"","institution":"Nantong University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hongxia","middleName":"","lastName":"Duan","suffix":""},{"id":607936520,"identity":"912cc923-3a37-4a44-a19b-77af06d75862","order_by":3,"name":"Wenhao Weng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABE0lEQVRIie3QMUvEMBTA8SeBdHnerZXKfQOhEIiDcPdVLgidWjlwuaFD4cBbdO+kX0GXzoVAu0TnDAcqhTrooHRU5FLRLZ66OeQ/BV5+PBIAl+sfNwUgJQEocdCff0no9IPQPxAMewI/krA+k484Xx0Nl6ddN0tXu9RL7psZjEd7GWnvbETdRAeo2mNfXRdBXrVI8YGxHA4ZL+l+aCFcx5wlJ1JkOinIdiaR+hENEEpRmHf5NnL7ZMi7FBc6brpP4r1uJBpZk2RSXOoYgq8tZBOZqJiTt0qKK1XxACtDsCUBhuYtknIb2Vkq9pKnUpzXi6bDVE6GXrTV4Xw84vWitRGTdXv/VcR+v588fztyuVwuV98a8EljugRoB+MAAAAASUVORK5CYII=","orcid":"","institution":"Shanghai Children's Hospital, Shanghai Jiao Tong University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Wenhao","middleName":"","lastName":"Weng","suffix":""}],"badges":[],"createdAt":"2026-02-23 14:24:58","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8948144/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8948144/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105037825,"identity":"63cf1695-0a8b-49f1-8b26-a53325f50041","added_by":"auto","created_at":"2026-03-20 07:40:43","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":89146,"visible":true,"origin":"","legend":"\u003cp\u003eCAF infiltration and stromal score are associated with OS in stage I and II LUSC patients\u003c/p\u003e\n\u003cp\u003eData were extracted from the TCGA-LUSC (n=406) and GSE157010 (n=235) cohorts. \u003cstrong\u003eA\u003c/strong\u003e, Survival analysis based on stromal scores calculated using ESTIMATE software; \u003cstrong\u003eB\u003c/strong\u003e and \u003cstrong\u003eC\u003c/strong\u003e, Survival analysis based on the extent of CAF infiltration calculated using two different methods (MCP_Counter and xCell, respectively).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/d362bb33acb74238e27cc1d2.png"},{"id":105037799,"identity":"cedf95e9-5e84-45e2-9a7f-68f681296836","added_by":"auto","created_at":"2026-03-20 07:40:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":104293,"visible":true,"origin":"","legend":"\u003cp\u003eWeighted gene co-expression network analysis identifies hub genes associated with CAF and stromal scores.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA\u003c/strong\u003e Represents the identified co-expression modules based on hierarchical clustering. \u003cstrong\u003eB\u003c/strong\u003e Illustratedrepresentation of the identified gene clusters associated with different biological pathways.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/3ea6f2c0adc8539e583cfa0c.png"},{"id":105037797,"identity":"930835c7-9ef4-4a65-85cf-7d6d9e4ded4e","added_by":"auto","created_at":"2026-03-20 07:40:35","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":79731,"visible":true,"origin":"","legend":"\u003cp\u003eCAF-Based Prognostic Risk Model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA\u003c/strong\u003e, Hazard ratios (95% CI) for individual hub genes in two cohorts of LUSC patients: TCGA-LUSC (n=406) and GSE157010 (n=235). B and \u003cstrong\u003eC\u003c/strong\u003e, Kaplan-Meier plots representing patient survival based on the calculated risk scores in TCGA and GEO cohorts, respectively; \u003cstrong\u003eD\u003c/strong\u003e and \u003cstrong\u003eE\u003c/strong\u003e, represents individual patients` live/dead status based on the calculated risk scores at different time points.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/9a25e35d93c5a353a9221f2b.png"},{"id":105037801,"identity":"b913e7a0-17d7-4f97-9770-cd8842122114","added_by":"auto","created_at":"2026-03-20 07:40:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":131811,"visible":true,"origin":"","legend":"\u003cp\u003eThe 4-gene CAF signature is associated with EMT, Represents the ridgeplot for gene set enrichment analysis; \u003cstrong\u003eB\u003c/strong\u003e, EMT-associated genes are the most upregulated genes in the high-risk LUSC patients; \u003cstrong\u003eC\u003c/strong\u003e and \u003cstrong\u003eD\u003c/strong\u003e, Scatterplots of single sample GSEA enrichment analysis in which each dot represents an individual patient. The CAF risk score is highly associated with the EMT hallmark enrichment scores of the patients in both TCGA and GEO datasets (r=0.84, \u003cem\u003ep \u003c/em\u003e≤0.0001 and r=0.78, \u003cem\u003ep\u003c/em\u003e ≤0.0001, respectively).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/bf74471c909a4aa3ec6b0cb3.png"},{"id":105037877,"identity":"5194559b-4fa0-4814-b9fc-e23dc9301f0d","added_by":"auto","created_at":"2026-03-20 07:40:51","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":95162,"visible":true,"origin":"","legend":"\u003cp\u003eCAF-associated gene signature predicts response to chemo- and immuno- therapy. \u003cstrong\u003eA\u003c/strong\u003erepresents the differential sensitivity to Paclitaxel, Gemcitabine, and Vinorelbine in LUSC patients with high vs. low CAF risk scores; data extracted from TCGA-LUSC (n=406) and GSE157010 (n=235) cohorts. \u003cstrong\u003eB\u003c/strong\u003e and \u003cstrong\u003eC\u003c/strong\u003e, represent the differential CAF risk scores between responders and non-responders to immunotherapy.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/a24c94fb525be3a198376386.png"},{"id":105037868,"identity":"ec92e88e-27f5-4c21-844a-7cda6e4a8033","added_by":"auto","created_at":"2026-03-20 07:40:48","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":46124,"visible":true,"origin":"","legend":"\u003cp\u003eExpression of PLXNA4, C1QTNF1, GFPT2, and FSTL3 in human LUSC cells\u003cstrong\u003eA \u003c/strong\u003emRNA expression levels of PLXNA4, C1QTNF1, GFPT2, and FSTL3 in human LUSC cell lines were measured by qRT-PCR.\u003cstrong\u003eB \u003c/strong\u003eKnockdown efficiency of two siRNAs targeting FSTL3 in two LUSC cell lines. Data are presented as mean ± SD. *, p \u0026lt; 0.05; **, p \u0026lt; 0.01. LUSC, lung squamous cell carcinoma.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/17bac066af534ccb5b0d4738.png"},{"id":105037628,"identity":"835431af-2b48-4fa9-b7cd-8a50c8435b04","added_by":"auto","created_at":"2026-03-20 07:39:58","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":269252,"visible":true,"origin":"","legend":"\u003cp\u003eThe anticancer effect of FSTL3 on colony formation, invasion and apoptosis in LUSC cell lines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA\u003c/strong\u003e Colony formation assay was performed to assess the clonogenic potential of H226 and H2170 cells following 48-hour transfection with two siRNAs targeting FSTL3. Representative images of the colonies were captured, and colony numbers were quantified. \u003cstrong\u003eB\u003c/strong\u003eInvasion assay was conducted to evaluate the invasive capacity of H226 and H2170 cells after 48-hour transfection with FSTL3 siRNAs. Cell invasion was quantified in three randomly selected fields per membrane. \u003cstrong\u003eC\u003c/strong\u003e Annexin V assay was performed to determine the percentage of apoptotic cells in H226 and H2170 cell lines following 48-hour FSTL3 siRNA transfection. Data are presented as mean ± SD. *, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.05; **, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0.01.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/fde72d904f6555e5c22eae67.png"},{"id":109172169,"identity":"40e77836-793f-4da6-a780-c8c6f1f972a7","added_by":"auto","created_at":"2026-05-13 09:03:18","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":966345,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/9d723db0-1714-4163-98c6-24a3fe4d5c51.pdf"},{"id":105037629,"identity":"0d78a28d-d30d-43e1-bdab-cb8c89b3dee4","added_by":"auto","created_at":"2026-03-20 07:39:58","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":18090,"visible":true,"origin":"","legend":"","description":"","filename":"Tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/b18f605fec352b4442a8f631.docx"},{"id":105037800,"identity":"f9bdc38f-8328-4b27-8cf7-c03fb3a993df","added_by":"auto","created_at":"2026-03-20 07:40:36","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":264285,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalMaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-8948144/v1/d147803182adf930b7b8353c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"An Integrative Machine Learning Approach Identifies a Cancer- Associated Fibroblast–Driven Predictive Model for Early-Stage Lung Squamous Cell Carcinoma","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eLung cancer is the leading cause of cancer-related mortality worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Non-small cell lung cancer (NSCLC) accounts for approximately 85\u0026ndash;90% of all cases, with lung squamous cell carcinoma (LUSC) representing one of its major histological subtypes [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Although advances in surgery, chemotherapy, and immunotherapy have improved outcomes in selected patient populations, therapeutic progress in LUSC has been comparatively limited, particularly due to the low frequency of actionable driver mutations and the absence of well-defined molecular subtypes amenable to targeted intervention\u003c/p\u003e \u003cp\u003eDespite advances in surgery, chemotherapy, radiotherapy, and immunotherapy, therapeutic progress in LUSC has lagged behind that in lung adenocarcinoma (LUAD), largely due to the relatively low frequency of actionable driver mutations and the absence of clearly defined molecular subtypes suitable for targeted intervention [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. For instance, a recent study in a Chinese cohort demonstrated that nearly 90% of LUAD patients carried mutations in just four key genes [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], underscoring the molecular tractability of LUAD compared with LUSC.\u003c/p\u003e \u003cp\u003eEven among patients with early-stage NSCLC, approximately 25% experience postoperative recurrence, and most recurrent tumors are no longer amenable to curative resection[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Furthermore, the reduced prevalence of targetable genomic alterations in LUSC may contribute to limited benefit from precision therapies and immune checkpoint blockade [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. These challenges highlight an urgent need to identify alternative biological determinants that drive disease progression and influence therapeutic responsiveness in LUSC.\u003c/p\u003e \u003cp\u003eTumor progression is increasingly recognized as a dynamic process shaped not only by malignant cells but also by the tumor microenvironment (TME) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Reciprocal interactions between tumor cells and stromal components critically regulate invasion, metastasis, immune evasion, and treatment resistance [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Among stromal cell populations, cancer-associated fibroblasts (CAFs) represent a dominant and highly heterogeneous component of the TME [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. CAFs encompass functionally diverse subtypes with both tumor-promoting and tumor-restraining properties, and their phenotypic plasticity allows them to profoundly influence tumor behavior [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Consequently, CAFs have emerged as promising prognostic biomarkers and potential therapeutic targets across multiple malignancies targets [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRecent single-cell transcriptomic studies have further demonstrated that specific CAF subpopulations are closely associated with patient survival and responsiveness to immunotherapy, including immune checkpoint blockade [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. These findings suggest that CAF-related gene signatures may serve as clinically actionable tools to stratify patients, guide treatment decisions, and minimize unnecessary toxicity in non-responders. However, the prognostic and therapeutic relevance of CAF-associated molecular markers in early-stage LUSC remains insufficiently characterized.\u003c/p\u003e \u003cp\u003eIn the present study, we integrated transcriptomic data and CAF infiltration profiles from The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) databases. Focusing specifically on stage I\u0026ndash;II LUSC to enhance clinical applicability, we employed weighted gene co-expression network analysis (WGCNA) to identify CAF-related hub genes and constructed a CAF-based risk score incorporating both CAF infiltration levels and gene signature components. The prognostic and therapeutic implications of this model were systematically evaluated using Cox regression and functional enrichment analyses, and further validated in an independent cohort. Importantly, in vitro experiments confirmed the biological relevance of the identified signature, and functional assays identified FSTL3 as a potential therapeutic target in early-stage LUSC.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 2.1. Data collection and preparation\u003c/h2\u003e \u003cp\u003eGene profiles and clinical information (level 3 data) of The Cancer Genome Atlas Lung Squamous Cell Carcinoma (TCGA-LUSC) samples were downloaded from UCSC Xena data hubs (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://xenabrowser.net/hub/\u003c/span\u003e\u003cspan address=\"https://xenabrowser.net/hub/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Gene expression data from TCGA were experimentally measured using the Illumina HiSeq 2000 RNA platform at TCGA genome characterization center, University of North Carolina. Genes were mapped onto the human genome coordinates using the UCSC Xena HUGO ProbeMap. Stages I-IIB were also selected. The gene-level transcription estimates were provided as log\u003csub\u003e2\u003c/sub\u003e(x\u0026thinsp;+\u0026thinsp;1) transformed RSEM-normalized counts. The normalized gene expression profiles and clinical information of the GSE157010 [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] dataset were downloaded from the GEO database using the GEO2R package. The mean abundance was determined if one gene had multiple probes. Genes that did not have expression values in at least half of the samples were excluded from analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Stromal score and CAF infiltration estimation\u003c/h2\u003e \u003cp\u003eESTIMATE software was used to estimate stromal cell infiltration in tumor samples [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. To estimate the CAF abundance, MCP_Counter (Microenvironment Cell Populations-counter) and xCell (cell-type enrichment analysis of bulk transcriptomes) methods were additionally followed using the R package \u0026lsquo;immunedeconv\u0026rsquo; [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Weighted gene co-expression network analysis\u003c/h2\u003e \u003cp\u003eThe top 5,000 genes with the highest median absolute deviations (MAD) were selected from the TCGA-LUSC dataset. Gene expression and traits (xCell CAF score and ESTIMATE stromal score) of TCGA-LUSC samples were used to construct a co-expression network using the R package WGSCNA. Co-expression modules that included highly correlated genes were identified. The eigengene of each module was evaluated using the first principal component of module expression. Hub genes were identified by measuring gene significance (GS) for traits and module membership (MM).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. CAF-related prognostic model construction and validation\u003c/h2\u003e \u003cp\u003eUnivariate COX regression analysis of the hub genes was performed using the\u0026ldquo;coxphw\u0026rdquo; function of R package\u0026ldquo;survival.\u0026rdquo; Genes with significant p-values in both TCGA-LUSC and GSE157010 datasets were selected for subsequent LASSO (Least Absolute Shrinkage and Selection Operator (LASSO). LASSO Cox regression analysis was used to select candidate genes that were highly associated with OS. A Cox proportional hazards model was constructed using TCGA-LUSC dataset. This model was also used to calculate the risk score of the GSE157010 dataset. Survival analysis between patients with high- and low-risk scores was performed using R package survminer.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Enrichment analysis\u003c/h2\u003e \u003cp\u003eGene ontology gene set overrepresented analysis (GSOA) was performed by Metascape online service [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. All genes in the genome were used as the background. Gene sets with a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.01 were grouped into clusters based on their membership similarities. The most statistically significant term within a cluster was selected to represent that cluster. Differential gene expression analyses were performed using the LIMMA package. Gene set enrichment (GSEA) was performed using the clusterProfiler package, whereas single-sample GSEA (ssGSEA) was performed using the GSVA method.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Prediction of response to chemo- and immuno- therapy\u003c/h2\u003e \u003cp\u003eThe R package \u0026ldquo;pRRophetic\u0026rdquo; was used to predict drug sensitivity against Cisplatin, Paclitaxel, Docetaxel, Gemcitabine, Vinorelbine and Etoposide, which are listed in \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cancer.org/cancer/lung-cancer/treating-non-smallcell/chemotherapy.html\u003c/span\u003e\u003cspan address=\"https://www.cancer.org/cancer/lung-cancer/treating-non-smallcell/chemotherapy.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. IC\u003csub\u003e50\u003c/sub\u003e values were transformed using Ln. Finally, the TIDE (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://tide.dfci.harvard.edu/\u003c/span\u003e\u003cspan address=\"http://tide.dfci.harvard.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) online service was used to predict responses blockade therapy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Cell lines and cell culture\u003c/h2\u003e \u003cp\u003eHuman LUSC cell lines (H226 and H2170) and normal human bronchial epithelial cell line (BEAS-2B) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). H226 and H2170 cells were cultured in RPMI-1640 medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (FBS; HyClone, GE Healthcare Life Sciences, Logan, UT, USA), 100 U/mL penicillin, and 0.1 mg/mL streptomycin (Gibco). BEAS-2B cells were maintained in DMEM/F12 medium (1:1), supplemented with 10% FBS and 1% penicillin-streptomycin. All the cells were incubated at 37\u0026deg;C in a humidified atmosphere containing 5% CO₂. Adherent cells were detached using 0.05% trypsin-EDTA (Invitrogen, Carlsbad, CA, USA) for passaging or further experiments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8. Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) assays\u003c/h2\u003e \u003cp\u003eTotal RNA was isolated from LUSC cell lines using the miRNeasy Serum/Plasma Kit (Qiagen) according to the manufacturer\u0026rsquo;s instructions. RNA quality and purity were evaluated using the NanoDrop system (Thermo Fisher Scientific). The RNA samples were immediately stored at -80\u0026deg;C for future use. cDNA was synthesized using the PrimeScript RT Reagent Kit (Takara, Shiga, Japan). RNA expression levels were quantified using the SensiFAST SYBR LO-ROX Kit (Bioline, London, UK) on a QuantStudio 6 Flex RT-PCR System (Applied Biosystems, Foster City, CA, USA). ACTB was used as the internal control for mRNA quantification. The relative RNA expression levels were determined using the 2-ΔΔ CT method. Primers used in this study are listed in \u003cb\u003eSuppl. Table. 1\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.9 Transfection of Small Interfering RNA\u003c/h2\u003e \u003cp\u003eH226 and H2170 cells were seeded in 6-well plates at a density of 2.5 \u0026times; 10 cells per well. Two distinct double-stranded siRNAs targeting separate coding regions of FSTL3 (or a Silencer negative control siRNA (Thermo Fisher Scientific] ) were transfected using Lipofectamine RNAiMAX (Invitrogen, Carlsbad, CA, USA) following the manufacturer\u0026rsquo;s protocol. After 48 h, the transfected cells were harvested for total RNA and protein extraction, as well as for downstream functional assays.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.10 Colony formation assay\u003c/h2\u003e \u003cp\u003eFor the colony formation assay, the cells were seeded at a density of 800 cells/well in 6-well plates. Colony formation was allowed to proceed for 7\u0026ndash;10 days, with the culture medium refreshed every three days. After incubation, the cell colonies were fixed with 4% paraformaldehyde in PBS (Thermo Fisher Scientific, Fair Lawn, NJ, USA) for 30 min and stained overnight with 0.1% crystal violet solution (Thermo Fisher Scientific). The colonies were quantified using ImageJ software, which allowed the assessment of the relative changes in colony formation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.11 Invasion assay\u003c/h2\u003e \u003cp\u003eCell invasion was assessed using BioCoat Matrigel Invasion Chambers with 8.0 \u0026micro;m pore polyester membranes (BD Biosciences, Franklin Lakes, NJ, USA). Cells (5 \u0026times; 10 cells/well) were seeded into inserts containing serum-free medium and placed in 24-well plates containing medium supplemented with 10% FBS. After 48 h, cells that had migrated to the lower surface of the membrane were fixed and stained using a Diff-Quik kit (Thermo Fisher Scientific).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e2.12 Cellular apoptosis assay\u003c/h2\u003e \u003cp\u003eCell apoptosis was assessed using the Muse\u0026trade; Annexin V \u0026amp; Dead Cell Kit (Luminex Corp, Austin, TX, USA) according to the manufacturer\u0026rsquo;s instructions. Briefly, 5 \u0026times; 10⁵ cells were collected and 100 \u0026micro;L of the suspension was mixed with 100 \u0026micro;L of Muse\u0026trade; Annexin V \u0026amp; Dead Cell Reagent. The percentage of apoptotic cells was measured using a Muse\u0026trade; Cell Analyzer (Millipore Corp., Billerica, MA, USA).\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Patient demographics and characteristics\u003c/h2\u003e \u003cp\u003eThe present study included data from two publicly available datasets of patients with early stage LUSC (stages I-II), including the TCGA dataset (TCGA-LUSC; n\u0026thinsp;=\u0026thinsp;406) and GEO dataset (GSE157010; n\u0026thinsp;=\u0026thinsp;235) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The median ages of the patients in TCGA and GEO datasets were 69 and 68 years, respectively. OS and gene expression data were available for both cohorts of patients and used for the analyses.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDemographic and clinical information of the cohorts included in this study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTCGA (n\u0026thinsp;=\u0026thinsp;406)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eGSE157010 (n\u0026thinsp;=\u0026thinsp;235)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e297 (73.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e153 (60.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e109 (26.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e82 (39.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e68\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e39\u0026ndash;90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e46\u0026ndash;89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003ePathologic stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eStage I-IB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e244 (60.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eStage II-IIB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e162 (40.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eT stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eT1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e63 (26.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eT2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e139 (59.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eT3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e31 (13.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003enot available\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2 (0.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eFollow up\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eAlive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e241 (59.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e119 (50.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eDeath\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e165 (46.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e116 (49.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e5-years survival rate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e45/406\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e97/235\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMean follow up time\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e46\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.2. High stromal score and CAF infiltration are associated with poor OS\u003c/h2\u003e \u003cp\u003eSubjective scoring of the patterns of stromal tissues within the TME is an interesting and simplified way of defining cancer-associated alterations of the stromal tissue by combining a variety of stromal tissue components, including CAFs and their abundance, into a single stromal score, which has been shown to be of significant prognostic value in different studies both as an independent prognosticator of clinical outcome and in combination with conventional staging systems [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Interestingly, the tumor stromal score has been shown to significantly correlate with the CAF score in various solid tumors such as colorectal cancer [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Accordingly, we investigated whether stromal infiltration into the TME could be a prognostic factor in patients with LUSC. To do this, we first calculated an \u0026ldquo;Estimate stromal score\u0026rdquo; and studied its association with OS of the LUSC patients. Our results indicated a significant association between higher stromal scores and poor OS of patients in both TCGA and GEO cohorts (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.0081 and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB, respectively). In addition to OS, patients with higher stromal scores in the TCGA cohort had lower disease-specific survival (DSS) (\u003cb\u003eSupp.\u003c/b\u003e Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). However, owing to the unavailability of DSS data, we were unable to perform a similar analysis in the GEO cohort.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo more specifically address CAF abundance within the TME, CAF infiltration was calculated for both cohorts of patients using MCP_Counter and xCell CAF methods. patients with LUSC with higher CAF infiltration, as indicated by both methods, had shorter OS (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026le;\u0026thinsp;0.01). Kaplan\u0026ndash;Meier survival plots for patients with high vs. low stromal scores and CAF infiltration are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC-F.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.3. WGCNA identifies hub CAF-associated genes\u003c/h2\u003e \u003cp\u003eGiven the associations of stroma score and CAF infiltration with OS in the studied LUSC patients, we sought to identify hub CAF-associated genes using WGCNA analysis. To construct the scale-free topology co-expression network, a soft threshold power (β) of 3 was estimated (scale-free R2\u0026thinsp;=\u0026thinsp;0.936) in the TCGA LUSC dataset of early-stage patients (\u003cb\u003eSupp.\u003c/b\u003e Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). Hierarchical clustering revealed 8 co-expression modules (\u003cb\u003eSupp.\u003c/b\u003e Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB), of which the blue and turquoise modules had the strongest positive correlation with stroma score and CAF infiltration (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Therefore, we continued to identify the hub genes in these two modules. Using a set of minimal criteria including an MM score of over 0.5, GS score of over 0.4 and the significance \u003cem\u003ep\u003c/em\u003e value of less than 0.05 in the cox regression analysis, a total of 271 genes from the blue and turquoise modules were identified and selected as the hub CAF-associated genes. Using these hub genes, we performed GO biological process enrichment analysis to identify the significant biological gene clusters. Our analysis revealed 20 clusters with significant biological pathways (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB. Among these clusters, CAF-associated genes were highly associated with the inflammatory response, response to external stimuli, immune response, and regulation of the MAPK cascade, indicating the potential function(s) of CAF-associated genes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Construction of the stromal CAF-based prognostic risk model\u003c/h2\u003e \u003cp\u003eFollowing the identification of hub CAF-associated genes, we investigated whether these genes could act as independent prognosticators of OS using univariate Cox regression analysis. Among the 271 studied genes, 19 were significantly correlated with OS in both cohorts (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Lasso Cox regression analysis was performed to develop a prognostic CAF-associated gene signature (\u003cb\u003eSupp.\u003c/b\u003e Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). A final set of four genes was selected to construct the Cox model (formula: 0.0629\u0026times; PLXNA4\u0026thinsp;+\u0026thinsp;0.0775 \u0026times; C1QTNF1\u0026thinsp;+\u0026thinsp;0.0564 \u0026times; GFPT2\u0026thinsp;+\u0026thinsp;0.0908 \u0026times; FSTL3). The LUSC patients in TCGA cohort were then assigned to two high-risk and low-risk groups using our 4-gene CAF signature, and survival analysis was performed. Our results showed that patients with LUSC in the high-risk group had significantly shorter OS than those in the low-risk group (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.0e-3) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). We validated the developed gene signature in a GEO cohort using the same model. The results were consistent with those of TCGA cohort, indicating that patients with lower risk scores had shorter OS (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5.3e-4). These findings show that the 4-gene-based risk score is significantly associated with patient prognosis in early stage LUSC.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Functional analysis reveals the hallmark inflammatory response pathway\u003c/h2\u003e \u003cp\u003eTo investigate the potential functional differences between high- and low-risk patients, differential expression analysis and hallmark gene set enrichment analysis were performed for the GEO cohort. The analysis identified 10 hallmark pathways with the \u0026ldquo;epithelial mesenchymal transition\u0026rdquo; (EMT) being the most relevant pathway. The ridge plot demonstrating the results of the GSEA analysis is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA. To further validate the EMT association, a similar analysis was performed based on the gene ontology resource. These results were in accordance with those obtained from the GSEA analysis, indicating that EMT-associated genes were the most upregulated genes in the high-risk LUSC patients of the studied cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). Finally, single-sample GSEA, in which an EMT hallmark enrichment score was assigned to each sample, was performed. We found that our CAF risk score was highly associated with the EMT hallmark enrichment scores of patients in both the TCGA and GEO datasets (r\u0026thinsp;=\u0026thinsp;0.84, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026le;\u0026thinsp;0.0001 and r\u0026thinsp;=\u0026thinsp;0.78, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026le;\u0026thinsp;0.0001, respectively) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC-D). The significant differences in EMT hallmark genes between the high- and low-risk patients may at least partially explain the association of the higher risk score with poor OS.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e3.6 CAF-based signature is associated with chemo- and immuno-therapy response\u003c/h2\u003e \u003cp\u003eTo delineate the possibility that a lower clinical response to therapy is responsible for poor prognosis in high-risk LUSC patients, we decided to study whether our CAF-associated gene signature was associated with the clinical response to therapy. To this end, the sensitivity of patients in different risk groups to various chemotherapeutic agents, including cisplatin, paclitaxel, docetaxel, gemcitabine, vinorelbine, and etoposide, was predicted. Ln-transformed IC\u003csub\u003e50\u003c/sub\u003e values were used to differentiate patients with varying degrees of sensitivity to chemotherapeutic drugs. Interestingly, we found that the sensitivity to three drugs (Paclitaxel, Gemcitabine and Vinorelbine) was significantly different between the high- and low-risk CAF groups of patients in both datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). Next, we attempted to predict the response to immunotherapy by using the CAF-based risk score. Similarly, our 4-gene CAF signature was able to differentiate responders from non-responders to immunotherapy in both TCGA and GEO datasets, indicating that patients with higher risk scores were less likely to respond to immunotherapy (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB-C).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e3.6 CAF-based signature is associated with chemo- and immuno-therapy response\u003c/h2\u003e \u003cp\u003eTo delineate the possibility that a lower clinical response to therapy is responsible for poor prognosis in high-risk LUSC patients, we decided to study whether our CAF-associated gene signature was associated with the clinical response to therapy. To this end, the sensitivity of patients in different risk groups to various chemotherapeutic agents, including cisplatin, paclitaxel, docetaxel, gemcitabine, vinorelbine, and etoposide, was predicted. Ln-transformed IC\u003csub\u003e50\u003c/sub\u003e values were used to differentiate patients with varying degrees of sensitivity to chemotherapeutic drugs. Interestingly, we found that the sensitivity to three drugs (Paclitaxel, Gemcitabine and Vinorelbine) was significantly different between the high- and low-risk CAF groups of patients in both datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). Next, we attempted to predict the response to immunotherapy by using the CAF-based risk score. Similarly, our 4-gene CAF signature was able to differentiate responders from non-responders to immunotherapy in both TCGA and GEO datasets, indicating that patients with higher risk scores were less likely to respond to immunotherapy (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB-C).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Identification of the biological role of a four-gene CAF signature in LUSC\u003c/h2\u003e \u003cp\u003eTo further confirm the specific biological role of the four-gene signature, we compared the mRNA expression levels of the four genes between a normal human bronchial epithelial cell line and the two LUSC cell lines. qRT-PCR analysis revealed that, compared with normal bronchial epithelial cells, FSTL3 consistently showed elevated expression in both LUSC cell lines, whereas the other three genes did not display consistent changes (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo explore the functional significance of FSTL3 in LUSC, siRNA-mediated knockdown was performed in H226 and H2170 cells. qRT-PCR confirmed that transfection with FSTL3-specific siRNA efficiently reduced FSTL3 mRNA expression 48 h after transfection (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB). indicated successful gene silencing. Functionally, colony formation assays demonstrated that FSTL3 knockdown significantly inhibited clonogenic growth in both the cell lines (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). This suggests that FSTL3 contributes to the proliferative capacity of LUSC. We examined the efficacy of FSTL3 in suppressing cellular invasion, which contributes to cancer progression and metastasis. The transwell assay. Knockdown of FSTL3 inhibited cell invasion significantly more effectively than individual treatments in both the LUSC cell lines (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB). Finally, to determine whether FSTL3 affected apoptotic regulation, an Annexin V-FITC/PI binding assay was performed. The results revealed a significant increase in apoptotic cell populations following FSTL3 knockdown in both LUSC cell lines (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC), indicating that FSTL3 functions as an oncogenic factor in LUSC, promoting cell proliferation and invasion and inhibiting apoptosis. These findings highlight FSTL3 as a potential therapeutic target and a critical downstream effector of the CAF-related gene signature in LUSC.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eGiven the unparalleled prevalence and mortality of NSCLS as well as its poor prognosis and unresponsiveness to therapy, the search for relevant prognostic and predictive markers has been the center of attention in recent years. Nonetheless, such attempts to find clinically reliable prognostic markers have not been very successful, calling for further research, herein [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. A great deal of Evidence suggests that in addition to the genetic characteristics of the primary tumor, the fingerprint of the reciprocal interactions between tumor cells and their surrounding stroma is a critical determinant of tumor fate and progression, which can be exploited to predict tumor behavior [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Fibroblasts within the tumor tissue (CAFs) are among the major cell types of the tumor stroma that affect tumor progression through their interactions with other stromal cells as well as tumor cells [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Given the heterogeneity of CAFs within different tumors, profiling can provide useful information regarding the future behavior of the tumor and its functional assessment [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Supporting this notion, stroma-associated markers have been shown to be better prognostic tools than tumor-intrinsic parameters in prognostic tools [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In line with this, stromal score has been previously used by others as a prognostic marker associated with survival and is calculated based on a weighted sum of different types of stromal tissues that are above a certain threshold [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Regarding LUSC, certain CAF populations have been reported to be associated with higher cancer cell proliferation [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], accumulation of single-nucleotide variants (SNPs), and tumor growth [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. However, to date, there is no information regarding the prognostic role of CAF infiltration in patients with early stage LUSC. To examine this possibility, stromal scores and the extent of CAF infiltration were calculated in two cohorts of LUSC patients from TCGA and GEO databases. Survival analysis revealed that a higher stromal score and CAF infiltration, calculated using two different methods, were associated with OS and DSS in patients with LUSC (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e \u003cb\u003eand Suppl. Figure\u0026nbsp;1\u003c/b\u003e). These findings are in line with the results of other studies indicating the role of CAFs and tumor stroma in the progression and prognosis of various solid tumors including lung cancer [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAlthough pathological slides from tumor specimens stained with hematoxylin\u0026ndash;eosin (HE) are often available for many solid cancer patients [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], objective analysis of the slides to obtain reliable stroma-related prognostic markers is not common, is cumbersome and is largely affected by the pathologist/user variation and thus a mRNA-based gene expression score representing the same prognostic value is more valuable [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Given that surgery in combination with chemotherapy and radiotherapy is the treatment of choice for patients with stage I-II LUSC patients [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], tumor specimens from resected tumor tissues are always available. Therefore, the development of tissue-based mRNA expression markers is a clinically feasible strategy to identify prognostic markers. Accordingly, using tumor tissue mRNA expression data from both cohorts, we performed WGCNA analysis to first identify the hub genes associated with CAF and stromal score and then check whether these CAF-associated genes can independently predict OS. Our analysis identified 271 hub genes, of which a final set of 4 were used to construct a prognostic model for patients with LUSC. Our 4-gene CAF panel was able to independently predict OS in patients with LUSC in TCGA cohort. The performance of the gene signature was subsequently confirmed in GEO patients, which were used as the validation cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC\u003cb\u003e)\u003c/b\u003e. These findings suggest that our 4-gene CAF signature can potentially be used as a prognostic marker in patients with stage I-II LUSC who have undergone surgery. Although the association between stromal score and OS has been previously shown in LUSC patients, the scores were calculated based on a deep analysis of pathological slides for late-stage patients, whereas our CAF panel can predict OS in early stage patients using only gene expression data [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Furthermore, in contrast to our study, most previous reports on the prognostic role of tumor stroma and CAF-related markers have only used single markers; however, combinational panels of markers have been shown to be superior to single-gene markers in terms of prognostic accuracy [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Nevertheless, the number of patients included in the present study was limited, and these findings need to be further validated in larger cohorts of patients in prospectively designed studies.\u003c/p\u003e \u003cp\u003eBiological pathway enrichment analysis of the identified CAF-associated hub genes revealed their role in EMT. EMT has been linked to various malignancies, including lung cancer, and has well-established roles in tumor metastasis, progression, and recurrence [\u003cspan additionalcitationids=\"CR30\" citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], which may at least partially explain the mechanism responsible for worse prognosis in patients with higher CAF and stromal risk scores. Among the genes used to construct the CAF-associated prognostic model, glutamine-fructose-6-phosphate transaminase 2 (\u003cem\u003eGFPT2)\u003c/em\u003e plays a critical role as a regulator of metabolomic reprogramming in lung cancer [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Additionally, upregulation of follistatin-like 3 (\u003cem\u003eFSTL3\u003c/em\u003e) has been shown to promote the proliferation and migration of NSCLC cells through regulation of the lncRNA DSCAM-AS1/miR-122-5p Axis [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. These findings emphasize the importance of the identified gene signature in LUSC pathogenesis and justify their combined use in prognostic panels for early prediction of OS.\u003c/p\u003e \u003cp\u003eImportantly, among patients with stage I and II LUSC who underwent surgical resection, recurrence occurred in up to 35% of the cases [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], which could be at least partially due to the ineffectiveness of therapy. Therefore, the prediction of response to therapy in patients with stage I and II LUSC can identify patients who may benefit the most from the selected treatment options and spare those in whom the therapy is unlikely to be effective. In line with this goal, we also tried to determine whether our CAF panel could predict the response to chemotherapy and immunotherapy. Our results showed that patients with lower risk scores were more likely to respond to certain chemotherapy drugs, including Paclitaxel, Gemcitabine and Vinorelbine, as well as immunotherapy. Although immunotherapy has revolutionized the treatment of lung cancer, the majority of patients (60\u0026ndash;80%) do not benefit from it [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e], making the identification of these patients paramount. This is especially important since selection of best treatment options tailored based on the patients` CAF status at early stages of adjuvant therapy may decrease the chances of recurrence and spare those patients who may benefit the least from certain treatments from the toxicities associated with therapy. In support of our findings, it has been previously shown that the presence of CD90\u003csup\u003e+\u003c/sup\u003e stromal cells within the TME can functionally inactivate tumor-infiltrating lymphocytes [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Furthermore, among our 4-gene CAF panel, Plexin-A4 (Plxna4) is shown to be upregulated in lung cancer cells and linked to the impaired homing capacity of cytotoxic T lymphocytes to the tumor site [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. These findings may explain the series of immunomodulatory events mediated by CAFs and stromal cells, which could potentially reduce the efficacy of immunotherapy in patients less effective\u003c/p\u003e \u003cp\u003eTaken together, our findings highlight the significance of CAF-related gene expression profiling for determining the prognosis and response to therapy in patients with LUSC. Our 4-gene CAF signature can independently predict OS and response to chemotherapy and immunotherapy in patients with early stage LUSC undergoing therapeutic resection. The prognostic and predictive values of this CAF-related panel need to be further validated in large-scale cohorts of patients alone and in combination with conventional clinical parameters.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eIn summary, our study demonstrated that CAFs and the extent of their infiltration into the tumor microenvironment are strongly associated with survival and therapeutic response in patients with LUSC. A CAF-related four-gene signature serves as an independent prognostic factor and predictor of the response to chemotherapy and immunotherapy. Among these genes, FSTL3 has emerged as a key effector that promotes proliferation and invasion, and inhibits apoptosis in LUSC cells. These findings highlight the critical role of the tumor stroma and CAFs in shaping tumor behavior and underscore the potential of CAF-based gene signatures, particularly FSTL3, as prognostic markers and therapeutic targets in lung squamous cell carcinoma.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eLUSC: Lung squamous cell carcinoma\u003c/p\u003e\n\u003cp\u003eNSCLC: non-small cell lung cancer\u003c/p\u003e\n\u003cp\u003eOS: overall survival\u003c/p\u003e\n\u003cp\u003eTME: tumor microenvironment\u003c/p\u003e\n\u003cp\u003eCAFs: cancer-associated fibroblasts\u003c/p\u003e\n\u003cp\u003eWGCNA: weighted gene co-expression network analysis\u003c/p\u003e\n\u003cp\u003eEMT: epithelial–mesenchymal transition\u003c/p\u003e\n\u003cp\u003eFBS: fetal bovine serum\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCCK-8: Cell Counting Kit-8\u003c/p\u003e\n\u003cp\u003eTCGA: The Cancer Genome Atlas\u003c/p\u003e\n\u003cp\u003eGEO: Gene Expression Omnibus\u003c/p\u003e\n\u003cp\u003eqRT-PCR: Quantitative reverse transcription PCR\u003c/p\u003e\n\u003cp\u003eRT: room temperature\u003c/p\u003e\n\u003cp\u003eWB: Western blotting\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u0026nbsp;\u003c/strong\u003eWe acknowledge and appreciate our colleagues for their valuable suggestions and technical assistance in this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement:\u003c/strong\u003e All datasets used in this study are publicly available. The TCGA-LUSC cohort was obtained from The Cancer Genome Atlas (TCGA) through the Genomic Data Commons (GDC) data portal (https://portal.gdc.cancer.gov/) under the accession TCGA-LUSC. The bulk RNA-seq dataset GSE157010 was downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE157010. Tumor immune dysfunction and exclusion (TIDE) scores were obtained from the TIDE web server (http://tide.dfci.harvard.edu/).\u003c/p\u003e\n\u003cp\u003eAll the codes used for data processing and analysis are available from the corresponding author upon request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study did not require ethical approval as all data analyzed were obtained from publicly accessible databases and previously published studies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. This research did not involve any new human participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to publish\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have read and approved the final version of the manuscript and consented to its submission for publication.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYuan Li, Fanping Zhang, Hongxia Duan, and Wenhao Weng conceived of and designed the experiments. Yuan Li and Fanping Zhang analyzed the data. Yuan Li conducted all cell experiments. Yuan Li, Fanping Zhang, Hongxia Duan and Wenhao Weng prepared and revised the manuscript. All authors have read and approved the final version of the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eR.L. Siegel, K.D. Miller, H.E. Fuchs, A. Jemal, Cancer statistics, 2022, CA: a cancer journal for clinicians, (2022).\u003c/li\u003e\n\u003cli\u003eZ. Cheng, C. Yu, S. Cui, H. Wang, H. Jin, C. Wang, B. Li, M. Qin, C. Yang, J. He, circTP63 functions as a ceRNA to promote lung squamous cell carcinoma progression by upregulating FOXM1, Nature communications, 10 (2019) 1\u0026ndash;13.\u003c/li\u003e\n\u003cli\u003eA.A. Schegoleva, A.A. Khozyainova, A.A. Fedorov, T.S. Gerashchenko, E.O. Rodionov, E.B. Topolnitsky, N.A. Shefer, O.V. Pankova, A.A. Durova, M.V. Zavyalova, Prognosis of different types of non-small cell lung cancer progression: current state and perspectives, Cell Physiol Biochem, 55 (2021) 29\u0026ndash;48.\u003c/li\u003e\n\u003cli\u003eN.A. Rizvi, M.D. Hellmann, J.R. Brahmer, R.A. Juergens, H. Borghaei, S. Gettinger, L.Q. Chow, D.E. Gerber, S.A. Laurie, J.W. Goldman, Nivolumab in combination with platinum‐based doublet chemotherapy for first-line treatment of advanced non\u0026ndash;small-cell lung cancer, Journal of Clinical Oncology, 34 (2016) 2969.\u003c/li\u003e\n\u003cli\u003eY. Zhang, Y. Sun, Y. Pan, C. Li, L. Shen, Y. Li, X. Luo, T. Ye, R. Wang, H. Hu, Frequency of Driver Mutations in Lung Adenocarcinoma from Female Never-Smokers Varies with Histologic Subtypes and Age at DiagnosisDriver Mutations in Lung Adenocarcinoma, Clinical cancer research, 18 (2012) 1947\u0026ndash;1953.\u003c/li\u003e\n\u003cli\u003eP. Kumari, G.K. Vishwakarma, A. Bhattacharjee, Thresholding of prominent biomarkers of breast cancer on overall survival using classification and regression tree, Cancer Biomarkers, (2021) 1\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eA. Addeo, A. Passaro, U. Malapelle, G.L. Banna, V. Subbiah, A. Friedlaender, Immunotherapy in non-small cell lung cancer harbouring driver mutations, Cancer treatment reviews, 96 (2021).\u003c/li\u003e\n\u003cli\u003eD.F. Quail, J.A. Joyce, Microenvironmental regulation of tumor progression and metastasis, Nature medicine, 19 (2013) 1423\u0026ndash;1437.\u003c/li\u003e\n\u003cli\u003eJ.H. Zhang, C.Y. Song, Y. Tian, X.Y. Yang, Single-Cell RNA Sequencing in Lung Cancer: Revealing Phenotype Shaping of Stromal Cells in the Microenvironment, Front Immunol, 12 (2022).\u003c/li\u003e\n\u003cli\u003eP.M. Galbo, X.X. Zang, D.Y. Zheng, Molecular Features of Cancer-associated Fibroblast Subtypes and their Implication on Cancer Pathogenesis, Prognosis, and Immunotherapy Resistance, Clinical Cancer Research, 27 (2021) 2636\u0026ndash;2647.\u003c/li\u003e\n\u003cli\u003eA.F. Irvine, S. Waise, E.W. Green, B. Stuart, G.J. Thomas, Characterising cancer-associated fibroblast heterogeneity in non-small cell lung cancer: a systematic review and meta-analysis, Sci Rep, 11 (2021) 3727.\u003c/li\u003e\n\u003cli\u003eS. Togo, U.M. Polanska, Y. Horimoto, A. Orimo, Carcinoma-associated fibroblasts are a promising therapeutic target, Cancers, 5 (2013) 149\u0026ndash;169.\u003c/li\u003e\n\u003cli\u003eR. Bueno, W.G. Richards, D.H. Harpole, K.V. Ballman, M.-S. Tsao, Z. Chen, X. Wang, G. Chen, L.R. Chirieac, M.H. Chui, Multi-institutional prospective validation of prognostic mRNA signatures in early stage squamous lung cancer (Alliance), Journal of Thoracic Oncology, 15 (2020) 1748\u0026ndash;1757.\u003c/li\u003e\n\u003cli\u003eK. Yoshihara, M. Shahmoradgoli, E. Mart\u0026iacute;nez, R. Vegesna, H. Kim, W. Torres-Garcia, V. Trevi\u0026ntilde;o, H. Shen, P.W. Laird, D.A. Levine, Inferring tumour purity and stromal and immune cell admixture from expression data, Nature communications, 4 (2013) 1\u0026ndash;11.\u003c/li\u003e\n\u003cli\u003eE. Becht, N.A. Giraldo, L. Lacroix, B. Buttard, N. Elarouci, F. Petitprez, J. Selves, P. Laurent-Puig, C. Saut\u0026egrave;s-Fridman, W.H. Fridman, Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression, Genome biology, 17 (2016) 1\u0026ndash;20.\u003c/li\u003e\n\u003cli\u003eD. Aran, Cell-type enrichment analysis of bulk transcriptomes using xCell, Bioinformatics for Cancer Immunotherapy, Springer2020, pp. 263\u0026ndash;276.\u003c/li\u003e\n\u003cli\u003eY. Zhou, B. Zhou, L. Pache, M. Chang, A.H. Khodabakhshi, O. Tanaseichuk, C. Benner, S.K. Chanda, Metascape provides a biologist-oriented resource for the analysis of systems-level datasets, Nature communications, 10 (2019) 1\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eJ.N. Kather, J. Krisam, P. Charoentong, T. Luedde, E. Herpel, C.-A. Weis, T. Gaiser, A. Marx, N.A. Valous, D. Ferber, Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study, PLoS medicine, 16 (2019) e1002730.\u003c/li\u003e\n\u003cli\u003eR.M. Bremnes, T. D\u0026oslash;nnem, S. Al-Saad, K. Al-Shibli, S. Andersen, R. Sirera, C. Camps, I. Marinez, L.-T. Busund, The role of tumor stroma in cancer progression and prognosis: emphasis on carcinoma-associated fibroblasts and non-small cell lung cancer, Journal of thoracic oncology, 6 (2011) 209\u0026ndash;217.\u003c/li\u003e\n\u003cli\u003eI.O. Bello, M. Vered, D. Dayan, A. Dobriyan, R. Yahalom, K. Alanen, P. Nieminen, S. Kantola, E. L\u0026auml;\u0026auml;r\u0026auml;, T. Salo, Cancer-associated fibroblasts, a parameter of the tumor microenvironment, overcomes carcinoma-associated parameters in the prognosis of patients with mobile tongue cancer, Oral oncology, 47 (2011) 33\u0026ndash;38.\u003c/li\u003e\n\u003cli\u003eR. Kalluri, M. Zeisberg, Fibroblasts in cancer, Nature Reviews Cancer, 6 (2006) 392\u0026ndash;401.\u003c/li\u003e\n\u003cli\u003eH. Nakamura, M. Sugano, T. Miyashita, H. Hashimoto, A. Ochiai, K. Suzuki, M. Tsuboi, G. Ishii, Organoid culture containing cancer cells and stromal cells reveals that podoplanin-positive cancer-associated fibroblasts enhance proliferation of lung cancer cells, Lung Cancer, 134 (2019) 100\u0026ndash;107.\u003c/li\u003e\n\u003cli\u003eS. Neri, H. Hashimoto, H. Kii, H. Watanabe, K. Masutomi, T. Kuwata, M. Tsuboi, K. Goto, A. Ochiai, G. Ishii, Cancer cell invasion driven by extracellular matrix remodeling is dependent on the properties of cancer-associated fibroblasts, Journal of cancer research and clinical oncology, 142 (2016) 437\u0026ndash;446.\u003c/li\u003e\n\u003cli\u003eY. Zhu, L. Zhang, H. Zha, F. Yang, C. Hu, L. Chen, B. Guo, B. Zhu, Stroma-derived fibrinogen-like protein 2 activates cancer-associated fibroblasts to promote tumor growth in lung cancer, Int. J. Biol. Sci., 13 (2017) 804\u0026ndash;814.\u003c/li\u003e\n\u003cli\u003eM.L. Wikberg, S. Edin, I.V. Lundberg, B. Van Guelpen, A.M. Dahlin, J. Ruteg\u0026aring;rd, R. Stenling, \u0026Aring;. \u0026Ouml;berg, R. Palmqvist, High intratumoral expression of fibroblast activation protein (FAP) in colon cancer is associated with poorer patient prognosis, Tumor Biology, 34 (2013) 1013\u0026ndash;1020.\u003c/li\u003e\n\u003cli\u003eS. Miura, N. Mitsuhashi, H. Shimizu, F. Kimura, H. Yoshidome, M. Otsuka, A. Kato, T. Shida, D. Okamura, M. Miyazaki, Fibroblast growth factor 19 expression correlates with tumor progression and poorer prognosis of hepatocellular carcinoma, BMC cancer, 12 (2012) 1\u0026ndash;15.\u003c/li\u003e\n\u003cli\u003eD. Anusewicz, M. Orzechowska, A.K. Bednarek, Lung squamous cell carcinoma and lung adenocarcinoma differential gene expression regulation through pathways of Notch, Hedgehog, Wnt, and ErbB signalling, Scientific reports, 10 (2020) 1\u0026ndash;15.\u003c/li\u003e\n\u003cli\u003eG. Sebastiani, A. Alberti, Non invasive fibrosis biomarkers reduce but not substitute the need for liver biopsy, World journal of gastroenterology: WJG, 12 (2006) 3682.\u003c/li\u003e\n\u003cli\u003eY. Otsuki, H. Saya, Y. Arima, Prospects for new lung cancer treatments that target EMT signaling, Developmental Dynamics, 247 (2018) 462\u0026ndash;472.\u003c/li\u003e\n\u003cli\u003eD. Greaves, Y. Calle, Epithelial Mesenchymal Transition (EMT) and Associated Invasive Adhesions in Solid and Haematological Tumours, Cells, 11 (2022) 649.\u003c/li\u003e\n\u003cli\u003eJ.-Y. Shih, P.-C. Yang, The EMT regulator slug and lung carcinogenesis, Carcinogenesis, 32 (2011) 1299\u0026ndash;1304.\u003c/li\u003e\n\u003cli\u003eW. Zhang, G. Bouchard, A. Yu, M. Shafiq, M. Jamali, J.B. Shrager, K. Ayers, S. Bakr, A.J. Gentles, M. Diehn, GFPT2-expressing cancer-associated fibroblasts mediate metabolic reprogramming in human lung adenocarcinoma, Cancer research, 78 (2018) 3445\u0026ndash;3457.\u003c/li\u003e\n\u003cli\u003eL. Gao, X. Chen, Y. Wang, J. Zhang, Up-regulation of FSTL3, regulated by lncRNA DSCAM-AS1/miR-122-5p Axis, promotes proliferation and migration of non-small cell lung Cancer cells, OncoTargets and therapy, 13 (2020) 2725.\u003c/li\u003e\n\u003cli\u003eA. Ribas, J.D. Wolchok, Cancer immunotherapy using checkpoint blockade, Science, 359 (2018) 1350\u0026ndash;1355.\u003c/li\u003e\n\u003cli\u003eH. Yang, S. Berezowska, P. Dorn, P. Zens, P. Chen, R.-W. Peng, T.M. Marti, G.J. Kocher, R.A. Schmid, S.R. Hall, Tumor-infiltrating lymphocytes are functionally inactivated by CD90+ stromal cells and reactivated by combined Ibrutinib and Rapamycin in human pleural mesothelioma, Theranostics, 12 (2022) 167.\u003c/li\u003e\n\u003cli\u003eW. Celus, A.I. Oliveira, S. Rivis, H.H. Van Acker, E. Landeloos, J. Serneels, S.T. Cafarello, Y. Van Herck, R. Mastrantonio, A. K\u0026ouml;hler, Plexin-A4 Mediates Cytotoxic T-cell Trafficking and Exclusion in Cancer, Cancer immunology research, 10 (2022) 126\u0026ndash;141.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"cancer-associated fibroblasts, lung squamous cell carcinoma, prognostic biomarker, tumor microenvironment, checkpoint inhibitor, FSTL3","lastPublishedDoi":"10.21203/rs.3.rs-8948144/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8948144/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eBackground\u003c/b\u003e\u003c/p\u003e \u003cp\u003eLung squamous cell carcinoma (LUSC) remains associated with unfavorable clinical outcomes, even in early-stage disease. Increasing evidence indicates that the tumor microenvironment (TME), particularly cancer-associated fibroblasts (CAFs), plays a crucial role in tumor progression and therapeutic resistance. However, the prognostic and therapeutic implications of CAF-related molecular signatures in early-stage LUSC remain insufficiently defined.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e \u003cp\u003eTranscriptomic profiles and corresponding clinical data of patients with early-stage LUSC were retrieved from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) cohorts. Stromal and CAF infiltration levels were quantified, and key CAF-associated modules were identified using weighted gene co-expression network analysis (WGCNA). A CAF-related prognostic signature was subsequently constructed using multivariate Cox regression analysis. Functional enrichment analysis was performed to explore underlying biological mechanisms. The predictive value of the signature for chemotherapy and immune checkpoint blockade response was evaluated. In addition, FSTL3, a core gene within the model, was selected for in vitro validation to assess its effects on cell proliferation, colony formation, and apoptosis in LUSC cells.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e \u003cp\u003eWe established a four-gene CAF-related prognostic signature that effectively stratified patients into high- and low-risk groups with significantly different overall survival outcomes. Multivariate analysis confirmed that the CAF-based risk score served as an independent prognostic factor in early-stage LUSC. Enrichment analysis demonstrated that epithelial\u0026ndash;mesenchymal transition (EMT) was prominently activated in the high-risk group. Furthermore, risk stratification based on the CAF signature was associated with differential predicted responses to chemotherapy and immune checkpoint inhibitors. Functional experiments revealed that FSTL3 knockdown significantly inhibited cell proliferation and colony formation while promoting apoptosis, supporting the biological relevance of the identified signature.\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusions\u003c/b\u003e\u003c/p\u003e \u003cp\u003eWe established a robust CAF-derived four-gene prognostic model that effectively predicted patient survival and therapeutic response in early stage LUSC. Experimental validation highlighted the biological and clinical relevance of the CAF-associated signature and identified FSTL3 as a potential therapeutic target for early stage LUSC.\u003c/p\u003e","manuscriptTitle":"An Integrative Machine Learning Approach Identifies a Cancer- Associated Fibroblast–Driven Predictive Model for Early-Stage Lung Squamous Cell Carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-20 07:02:12","doi":"10.21203/rs.3.rs-8948144/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"40c48aba-983f-40bb-a52c-6b927857554f","owner":[],"postedDate":"March 20th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Rejected","date":"2026-05-13T08:48:33+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-13T09:01:48+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-20 07:02:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8948144","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8948144","identity":"rs-8948144","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}