Multi-omics profiling unveils biological and clinical insights into pulmonary sarcomatoid carcinoma Authors: Liangdong Sun, Jue Wang, Yilv Yan, Shengnan Duan, Di Wang, Jiameng Hu, Huansha Yu, Ningrui Wang, Min Zhang, Run Zhang, Likun Hou, Junjie Hu, Xinsheng Zhu, Dongliang Bian, Shiqi Hu, Tao Ge, Yan Chen, Lu Han, Jinyi Song, Ming Ding, Peng Zhang, Lele Zhang, Haiyang Hu Published: 2026 ABSTRACT SummaryPulmonary sarcomatoid carcinoma (PSC) is a rare yet highly invasive type of lung cancer that remains poorly characterized. Here, we conduct multi-omics analyses of PSC by integrating whole-exome sequencing, transcriptomic, proteomic, and phosphoproteomic data from 86 PSC patients, alongside single-cell RNA sequencing (scRNA-seq) data from 3 patients receiving neoadjuvant chemoimmunotherapy. We find an elevated ferroptosis suppression signature in PSC tumors and demonstrate their vulnerability to ferroptosis by targeting FTL and SLC3A2. Immune landscape characterization reveals an association between cold tumors, pleomorphic carcinoma, and PTEN mutations. Proteomic profiling identifies three distinct patient subtypes, each exhibiting unique features regarding microenvironment dysregulation, genetic alterations, and potential therapeutic targets. scRNA-seq analysis suggests roles of epithelial-mesenchymal transition in the progression from carcinoma cells to sarcomatoid cells and in immunotherapy resistance in PSC. Altogether, this multi-omics characterization of PSC serves as a rich resource for investigating mechanistic insights and identifying potential therapeutic targets.Keywords: pulmonary sarcomatoid carcinoma, proteogenomic, scRNA-seq, epithelial-mesenchymal transition, immunotherapy Introduction Pulmonary sarcomatoid carcinoma (PSC) is a rare histological subtype of non-small cell lung cancer (NSCLC) with a poor prognosis, accounting for ∼0.4% of all lung cancers.1 The World Health Organization classifies PSC into five histological subtypes, including pleomorphic carcinoma (PC), spindle cell carcinoma (SCC), giant cell carcinoma (GCC), carcinosarcoma, and pulmonary blastoma, and the first three subtypes constitute over 90% of PSC cases.2,3,4 The treatment for PSC has followed the standard first-line approach for NSCLC, yet the outcomes have proven unsatisfactory, with PSC exhibiting conventional treatment-refractory behavior,5,6 thereby raising expectations for additional therapeutic targets and strategies. Immunotherapy using immune checkpoint blockade (ICB) and targeted therapies has significantly progressed the treatment of conventional NSCLC, enhancing patient survival rates7,8 and potentially opening avenues for therapeutic approaches to PSC. PSC exhibits pathologically intratumor heterogeneity, characterized by a sarcomatoid element alongside an epithelial component.9 Recent progress in genomic10,11,12,13,14 and transcriptomic12 sequencing studies has provided initial insights into the molecular landscape of PSC. Laser capture microdissection-based isolation of epithelial and sarcomatoid components and subsequent sequencing analysis uncovered a common origin of these two components and revealed a notable role of epithelial-mesenchymal transition (EMT) in the carcinogenesis of PSC.12,15 Nevertheless, our understanding of the molecular landscape and heterogeneity of PSC remains incomplete. By combining proteomic and phosphoproteomic profiling with genomic analysis, the proteogenomic approach has revealed critical biological pathways, distinct disease subtypes, and potential therapeutic targets in the lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), the two dominant NSCLC subtypes.16,17,18,19 However, no systematic proteomic study of PSC has been reported thus far, even though proteins are essential in determining the tumor phenotype. In this study, we performed a comprehensive proteogenomic characterization of tumors and paired normal adjacent tissues (NATs) from 86 PSC patients by leveraging quantitative proteomics and phosphoproteomics with genomic and transcriptomic sequencing. In addition, single-cell RNA sequencing (scRNA-seq) of tumors from 3 PSC patients receiving neoadjuvant ICB treatment was employed to characterize cellular and molecular features of immunotherapy responses. Our integrative proteogenomic and single-cell transcriptomic analyses provided critical insights into genomic aberrations, tumor microenvironment (TME), disease subgroups, critical biological pathways of PSC, and cellular and molecular characteristics of the PSC patients receiving ICB treatment, which together provide a better understanding of patient stratification with personalized treatment strategies. The resulting dataset also serves as a valuable resource for further research in the field of PSC. Results The proteogenomic landscape of PSC To characterize the proteogenomic landscape of PSC in China, we collected tumor and paired NAT samples from 86 surgically resected PSC patients (Tongji University [TU]-PSC cohort). We employed four molecular profiling strategies to characterize the TU-PSC cohort, including whole-exome sequencing (WES), bulk RNA sequencing (RNA-seq), and isobaric tandem mass tag (TMT) labeling-based global proteomics and phosphoproteomics (Figures 1A and S1A; Table S1). A total of 9,969 proteins and 26,002 reliable phosphorylation sites (phosphosites) from 5,514 phosphoproteins were identified using TMT-based proteomic and phosphoproteomic data, respectively. Robust reproducibility and high data quality were maintained across the entire dataset (Figures S1B–S1G; Table S1). Sample-wise correlation analysis showed higher expression concordances between mRNA and protein in tumor samples compared with NAT samples (Figure S1H). Multi-omics landscape of PSC (A) Workflow of PSC multi-omics profiling. (B) Genetic profiles in the TU-PSC cohort. The significantly mutated genes with q value <0.1 are indicated in red. (C) Mutation frequency of somatic mutations across four PSC cohorts. (D) Kaplan-Meier curves comparing overall survival for KRAS-mutant and KRAS wild-type (WT) patients (log rank test). (E) Kaplan-Meier curves comparing overall survival for high TMB and low TMB patients (log rank test). (F) Trinucleotide motif frequency plots and enriched mutational signatures identified in the TU-PSC cohort. Based on WES data, we assessed the genomic alterations in the TU-PSC cohort (Table S1, STAR Methods). The genomic landscape largely reaffirmed previously reported somatic alterations in PSC12,13,14 (Figures 1B and 1C). We identified TP53 (65%), KRAS (16%), SMARCA4 (12%), PIK3CA (10%), and PTEN (7%) as significant recurrent somatic mutations (q < 0.1; STAR Methods) (Figure 1B). The mutation distribution of the TU-PSC cohort was roughly similar to that of the other three PSC cohorts.10,12,13 Of note, the PTEN mutation rate in the TU-PSC cohort was comparable to that in the Yang et al.12 and Zhou et al.10 cohorts but relatively higher than that in the Nagano et al.13 cohort (Figure 1C), suggesting a possible geographical difference. The KRAS mutation rate in the TU-PSC cohort was relatively lower than that observed in the Nagano et al. and Zhou et al. cohorts, yet it was comparable to that in the Yang et al. cohort (Figure 1C). Similar to the previous observation in PSC,20 KRAS mutation was associated with worse survival in the TU-PSC cohort (Figure 1D). We found longer overall survival (OS) in the patients with a higher tumor mutational burden (TMB) (Figure 1E), consistent with the previous lung cancer studies.21,22 Significant mutations, such as TP53 and SMARCA4, were associated with higher TMB in the TU-PSC cohort (Figure S1I). Through mutational signature analysis, we identified three dominant mutational signatures in PSC, namely Sig1 (Cosmic 4, smoking), Sig2 (Cosmic 13, APOBEC), and Sig3 (Cosmic 1, 5-methylcytosine deamination) (Figure 1F). Further analysis showed that patients with SMARCA4 and KRAS mutations exhibited higher Sig1 activity and SMARCA4 mutations were also associated with lower Sig2 activity (Figures S1J and S1K). Survival analysis revealed the prognostic value of these three mutational signatures, indicating their potential involvement in the progression of PSC (Figure S1L). Effects of genomic alterations in PSC We first assessed the regulatory effects of somatic copy-number alterations (CNAs) on mRNA, protein, and phosphoprotein expression in the TU-PSC cohort. The WES analysis identified 27,080 somatic CNAs, which may affect the abundances of mRNA, protein, and phosphoprotein in either “cis” or “trans” modes, corresponding to the diagonal and vertical patterns, respectively (Figure S2A). We further focused on 593 cancer-associated genes (CAGs)23 to nominate critical genes within CNA regions, of which 590 CAGs were detected in the TU-PSC cohort (Figure 2A). We identified 16 CAGs associated with significant cis-effects across all three omics levels (Figure 2A; Table S2), which were enriched in the ErbB signaling pathway, growth hormone synthesis, secretion and action, bacterial invasion of epithelial cells, insulin signaling pathway, and yersinia infection (Figure 2B). Effects of genetic alterations on proteogenomic profiles in PSC (A) The Venn diagram showing the significant cis-events restricted to cancer-associated genes (CAGs) across multiple data types. (B) Enriched pathway for 16 cis CAG-related significant cis-events shared at mRNA, protein, and phosphoprotein levels. (C) Genome-wide arm amplification and deletions. (D) Genome-wide focal amplification and deletions. Chromosomal locations of peaks of significantly recurring focal amplifications (red) and deletions (blue) were filtered by FDRs. (E) Boxplot showing the KDM5D mRNA expression differences between Yq amplification and Yq non-amplification tumors (Wilcoxon ranked-sum test). (F) Boxplot showing the PTEN protein differences between PTEN-mutant and PTEN wild-type (WT) tumors (Wilcoxon ranked-sum test). (G) Boxplots showing the SMARCA4 expression differences between SMARCA4-mutant and SMARCA4 WT tumors (Wilcoxon ranked-sum test). (H) Heatmap of SMARCA4 phosphosites between SMARCA4-mutant and SMARCA4 WT tumors. (I) Significantly upregulated pathways in SMARCA4-mutant tumors compared with SMARCA4 WT tumors via gene set enrichment analysis. (J) Boxplots showing the PRC2 complex protein differences between SMARCA4-mutant and SMARCA4 WT tumors (Wilcoxon ranked-sum test). CNA analysis identified the most frequent gains in chromosomes 1q, 5p, 7p, 7q, and 8q and losses in chromosome 8p in the TU-PSC cohort (Figure 2C). In addition, we found amplifications in driver oncogenes, including EGFR (7p11.2, 24 cases), MALAT1 (11q13.1, 31 cases), MYC (8q24, 27 cases), VEGFA (6p21.1, 15 cases), and CDK6 (7q21.2, 22 cases), as well as deletions in crucial tumor suppressors, such as CDKN2A (9p21.3, 29 cases) (Figure 2D). Survival analysis of arm-level somatic CNA events revealed significant associations of amplifications of 21p, 13q, 3p, 16p, 9q, Yp, and Yq with PSC prognosis (Figure S2B). KDM5D, located on chromosome Yq, has been found to promote the development and metastasis of cancer cells.24 We found significantly higher expression of KDM5D in PSC tumors with Yq amplification (Figure 2E). Next, we examined the genomic impacts of crucial mutations in PSC on their cognate gene products and identified five genes (TP53, KRAS, PTEN, MET, and SMARCA4) with significant cis-effects (Figures 2F, 2G, and S2C–S2H). KRAS displayed increased mRNA and protein expression in KRAS-mutant tumors (Figure S2E). Meanwhile, protein and mRNA expression of KRAS were significantly upregulated in tumors with KRAS amplification compared with non-amplification tumors (Figure S2F). In addition, PTEN displayed a decreased protein expression in PTEN-mutant tumors (Figure 2F). Tumors with MET amplification showed higher MET expression at both protein and mRNA levels, while MET mutation showed no association with its protein or mRNA expression (Figures S2G and S2H). SMARCA4, an ATP-dependent catalytic subunit of SWI/SNF chromatin-remodeling complexes,25 displayed decreased mRNA and protein expression in SMARCA4-mutant tumors (Figure 2G). Moreover, our phosphoproteomic data demonstrated a decrease in six phosphosites on SMARCA4 in SMARCA4-mutant tumors (Figure 2H; Wilcoxon rank-sum test, all p < 0.05). Proteome-based gene set enrichment analysis (GSEA) linked SMARCA4 mutation to the upregulation of MYC targets, DNA replication, and oxidative phosphorylation pathways (Figure 2I) and to the downregulation of regulation of the actin cytoskeleton, EMT, and inflammatory response pathways26 (Figure S2I). Notably, SMARCA4-mutant tumors showed an elevated protein abundance of key components of PRC2 complexes, including EZH2, SUZ12, and EED (Figure 2J). SMARCA4 mutation predicts sensitivity to EZH2 inhibitors in NSCLC,27 suggesting that EZH2 inhibitors may also serve as a treatment option for SMARCA4-mutant PSC. Proteomic alterations associated with tumorigenesis and prognosis in PSC We then investigated the proteomic alterations in PSC using proteomic and phosphoproteomic profiles of both PSC tumors and paired NATs. The principal-component analysis (PCA) showed a clear distinction between tumors and NATs at both protein and phosphoprotein levels (Figure S1G). We identified 1,681 differentially expressed proteins (Wilcoxon rank-sum test, |log2 FC| > 1, false discovery rate [FDR] < 0.05), with 847 proteins upregulated and 834 proteins downregulated in PSC tumors compared with NATs (Figure 3A). Gene Ontology (GO) enrichment analysis showed that cytoplasmic translation, ribosome biogenesis, and rRNA processing terms were upregulated in tumors, while wound healing, blood coagulation, and hemostasis terms were downregulated (Figure 3B). Application of more stringent criteria for differential expression identified 529 upregulated proteins in tumors, which we termed PSC-associated proteins (FDR < 0.01 and concordant differential expression in ≥90% of all 74 tumor-NAT pairs; Figure 3C; Table S3). The potential clinical utility of these proteins was annotated using the Human Protein Atlas,28 including 230 plasma proteins, 124 enzymes, 74 known cancer-related genes, and 44 approved drug targets (Table S3). Among the identified PSC-associated proteins, we noticed that SRSF5, a serine/arginine (SR)-rich family of pre-mRNA splicing factors, showed increased protein expression but decreased mRNA expression in PSC tumors (Figure 3D). The opposite differential expression pattern of SRSF5 at mRNA and protein levels indicated the importance of proteomic profiling of the TU-PSC cohort for guiding the experimental design of subsequent functional studies on SRSF5 in PSC. Proteomic and phosphoproteomic alterations associated with tumorigenesis and prognosis (A) The differential protein expression measured using FDR and log2 fold-change (FC) between tumors and NATs (Wilcoxon rank-sum test, proteins quantified in all samples). (B) Representative enriched biological processes of increased (left) and decreased (right) differentially expressed proteins in tumors compared with NATs. (C) Boxplot showing protein expression differences measured in log2 FC between tumors and paired NATs for representative 50 PSC-associated proteins according to the median log2 FC in paired samples, the potential clinical utilities of which were annotated based on the Human Protein Atlas. CRG, cancer-related gene. (D) Boxplots showing the SRSF5 mRNA (left) and protein (right) expression differences between tumors and NATs (Wilcoxon rank-sum test). (E) The differential expression at phosphorylation site level measured using FDR and log2 FC between tumors and NATs (Wilcoxon rank-sum test, phosphorylation sites quantified in half samples). (F) Comparison of abundance changes (log2 FC) in tumors between protein level and phosphorylation site level (quantified in half samples). (G) Kinase-substrate enrichment analysis (KSEA) results between tumors and NATs. (H) The abundance differences of protein and phosphosites of SRPK1 between tumors and NATs (Wilcoxon rank-sum test). (I) Kaplan-Meier curves comparing overall survival based on median protein abundance of SRPK1 (log rank test). The phosphoproteomics data analysis identified 3,221 differentially expressed phosphosites (Wilcoxon rank-sum test, |log2 FC| > 1, FDR < 0.05), including 1,593 phosphosites upregulated and 1,628 phosphosites downregulated in PSC tumors compared with paired NATs (Figure 3E). The differential expression pattern was largely concordant between phosphosites and their cognate proteins (R = 0.62, p < 0.001; Figure 3F). Notably, 1,307 phosphosites exhibited a much more substantial increase (>2-fold) than the corresponding proteins in tumors, particularly among several phosphosites on VIM of the EMT pathway (Figure 3F). We identified 26 kinases with significantly increased activity in PSC tumors using kinase-substrate enrichment analysis (KSEA) (Figure 3G), including SRPK1, which exhibited elevated levels of both protein and phosphosites in tumors (Figure 3H). We observed worse survival in patients with higher SRPK1 protein expression (Figure 3I), suggesting an oncogenic role of SRPK1 in PSC. Suppressed ferroptosis prompted the cell proliferation of PSC Having delineated the global proteomic changes between PSC tumors and paired NATs, we then tried to identify potential therapeutic targets arising from the dysregulated pathways in PSC. EMT is a hallmark of PSC,12,15 and the mesenchymal phenotype exhibits higher susceptibility to ferroptotic cell death.29,30 Ferroptosis has emerged as a pivotal regulator in tumorigenesis.31 Intriguingly, functional enrichment analysis of 847 significantly upregulated proteins in PSC tumors revealed enrichment of the ferroptosis signaling pathway (Fisher’s exact test, p < 0.05, odds ratio: 2.98; Table S3). Notably, except for CYBB, the other five upregulated ferroptosis-related genes (FTL, FTH1, PCBP1, PCBP2, and SLC3A2) were all annotated as ferroptosis suppressors (Figures S3A and S3B). Among these five ferroptosis suppressors, SLC3A2 and FTL were associated with worse survival (Figures S3C and S3D). Consistently, by quantifying a ferroptosis negative regulation (FNR) signature based on ferroptosis suppressors annotated in the FerrDb database,32 we found a significantly elevated FNR signature in PSC tumors, which was associated with a lower survival rate in PSC patients (Figures S3E and S3F). In contrast, no such results were observed for the annotated ferroptosis drivers in the FerrDb database (Figures S3G and S3H). These observations indicated a repressed state of ferroptosis in PSC. We then investigated the potential role of two selected ferroptosis suppressors, SLC3A2 and FTL, in PSC. Both SLC3A2 and FTL were strongly upregulated in PSC tumors and correlated with poorer survival in the TU-PSC cohort (Figures S3A, S3C, and S3D). After individually knocking down SLC3A2 and FTL, we assessed their impact on cell proliferation in PLA-801D cells, a human cell line of PSC33 (Figures 4A and 4B). The results indicated a significant decrease in cell proliferation following the knockdown of SLC3A2 and FTL, respectively (Figure 4B). Consistent results were also observed in colony formation assays (Figure 4C). Given the established role of SLC3A2 as a subunit of the glutamate/cystine antiporter, which imports cystine in exchange for intracellular glutamate,34 we found a marked decrease in the extracellular glutathione (GSH) concentration in cell culture supernatants following SLC3A2 ablation (Figure 4D). FTL functions as the light-chain subunit of ferritin that is essential for intracellular iron storage.35 We noted a significant increase in Fe2+ concentration after FTL knockdown, as measured using FerroOrange fluorescent probes (Figure 4E). Additionally, we observed a notable downregulation of GPX4 in shFTL cells (Figure 4F), consistent with previous findings.36 Furthermore, we treated different groups of PLA-801D cells with DMSO or ferrostatin-1, a widely used ferroptosis inhibitor.37 After incubation with ferrostatin-1, the extracellular GSH level in shSLC3A2 cells rebounded, along with upregulated expression of GPX4 at the protein level (Figures 4G and 4H). Additionally, the cell proliferation of both shSLC3A2 and shFTL cells partially recovered following treatment with ferrostatin-1, while the proliferation of wild-type (WT) cells remained unaffected (Figure 4I). Ferroptosis is commonly linked to the generation of reactive oxygen species (ROS) and lipid peroxidation,38 leading us to evaluate ROS levels and malondialdehyde (MDA) concentrations (Figures 4J and 4K). Silencing SLC3A2 or FTL significantly increased ROS levels and MDA concentrations, both of which were subsequently reduced following ferroptosis inhibition. In addition to MDA measurement, we also evaluated lipid peroxidation changes using C11 BODIPY in shSLC3A2 and shFTL cells. Upon enhancement of ferroptosis, the butadiene moiety of C11 BODIPY undergoes an oxidation-reduction reaction with ROS, shifting its emission wavelength from 591 to 510 nm.39 We therefore used flow cytometry to detect this emission shift by measuring the fluorescence intensity ratio between the fluorescein isothiocyanate (FITC) and phycoerythrin (PE) channels. After knockdown of SLC3A2 or FTL, the fluorescence intensity ratio in most PLA-801D cells increased significantly, indicating elevated lipid peroxidation and ferroptosis. This ratio was restored upon inhibition of ferroptosis (Figures S3I and S3J). Together, these results indicated that SLC3A2- and FTL-mediated inhibition of ferroptosis may contribute to the progression of PSC, highlighting ferroptosis as a potential therapeutic target in PSC (Figure 4L). Derepression of ferroptosis decreases cell proliferation in PSC (A) Western blot showing the knockdown efficiency of SLC3A2 and FTL in PLA-801D cells. (B) The cell proliferation of PLA-801D cells measured using CCK-8 assays following the knockdown of SLC3A2 (left) and FTL (right), respectively. (C) The cell proliferation of PLA-801D cells measured using colony formation assays following the knockdown of SLC3A2 (upper) and FTL (bottom), respectively. Following the knockdown of SLC3A2 or FTL, PLA-801D cells were plated and cultured for 14 days. All colonies were subsequently stained with crystal violet and quantified using ImageJ software. (D) The extracellular GSH concentration of PLA-801D cells was tested following the knockdown of SLC3A2. (E) Following the knockdown of FTL, PLA-801D cells were stained by FerroOrange and quantified using the microplate reader, respectively. (F) The lysates of PLA-801D cells were immunoblotted with the indicated antibodies following the knockdown of FTL. (G) Following the knockdown of SLC3A2, PLA-801D cells were treated with 10 μM ferrostatin-1 for 24 h. Afterward, the extracellular GSH concentration was tested. (H) Following the knockdown of FTL, PLA-801D cells were treated with 10 μM ferrostatin-1 for 24 h. Then the lysates of PLA-801D cells were immunoblotted with the indicated antibodies. (I) Following the knockdown of SLC3A2 (left) or FTL (right), PLA-801D cells were treated with DMSO or 10 μM ferrostatin-1 for 24 h. Then the cell proliferation of PLA-801D cells was detected by CCK-8 assays, respectively. (J and K) Following the knockdown of SLC3A2 (left) or FTL (right), PLA-801D cells were treated with DMSO or 10 μM ferrostatin-1 for 24 h. Then the fluorescent probes were added into the culture medium to test the relative level of ROS, respectively (J); the cell lysates were collected for bicinchoninic acid assay and the detection of lipid peroxidation MDA (K). (L) The graphic model of the repressed ferroptosis in PSC. NC, negative control; ns, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001. The immune landscape of PSC We next investigated the immune landscape of PSC. Unsupervised consensus clustering using the xCell-derived cell type enrichment scores40 for both tumors and NATs identified two sets of tumors representing immunologically hot and cold clusters, as well as an NAT-enriched cluster (Figures 5A, S4A, and S4B; Table S4). We observed increased immune cells (B cells, CD4+ T cells, CD8+ T cells, and M1 macrophages), expression of immune checkpoints, and immune score and decreased tumor immune dysfunction and exclusion (TIDE) score41 in the hot tumor cluster compared with the cold tumor cluster (Figures 5A–5C), indicating a potential benefit of ICB for the hot tumor cluster. GSEA results revealed significant enrichment of antigen processing and presentation (APP), interferon alpha, interferon gamma, and T cell receptor (TCR) pathways (Figure 5D). In line with the APP pathway enrichment in the hot tumor cluster, higher APP protein activity was also observed in this tumor cluster (Figure S4C). The immune repertoire analysis further showed higher levels of unique TCR complementarity-determining region 3 (CDR3) and TCR reads42 in the hot tumor cluster (Figure S4D), which was consistent with the enrichment of the TCR pathway in this cluster. The immune landscape of PSC (A) Heatmaps showing compositions of cell types, mRNA expression of representative immune-related markers, and activities of enriched pathways across three immune clusters: hot tumor cluster, cold tumor cluster, and NAT-enriched cluster (successive heatmaps illustrate xCell immune signatures, mRNA expression of key immune-related markers, and single-sample GSEA (ssGSEA) pathway scores based on mRNA expression for biological pathways that were differentially regulated in immune cluster). (B and C) Comparison of immune score (B) and TIDE score (C) between hot tumor cluster and cold tumor cluster (Wilcoxon rank-sum test). (D) Significantly activated pathways at protein level in hot tumors compared with cold tumors via gene set enrichment analysis. (E) Comparison of PSC histological subtype distribution between hot tumor cluster and cold tumor cluster (Fisher’s exact test). (F) Comparison of PTEN mutation rate between hot tumor cluster and cold tumor cluster (Fisher’s exact test). (G) Comparison of immune score between PTEN-mutant and PTEN wild-type (WT) tumors (Wilcoxon rank-sum test). (H) Comparison of xCell score of immune cells between PTEN-mutant and PTEN WT tumors (Wilcoxon rank-sum test). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Next, we sought to explore tumor-intrinsic determinants of low immune infiltration in the immune-cold tumors of PSC. When associating histological types with immunotypes, we observed a higher percentage of the PC subtype in the cold tumor cluster (Fisher’s exact test, p < 0.05; Figure 5E). Furthermore, interferon alpha and interferon gamma pathways were significantly downregulated in the PC subtype (Figure S4E). Lower infiltration of CD4+ memory T cells, dendritic cells (DCs), and endothelial cells was observed in the PC subtype (Figure S4F). Additionally, the PC subtype showed a relatively lower immune score than the SCC subtype (Figure S4G). We then explored histological subtype-specific mutations in PSC and found a relatively higher PTEN mutation rate in the PC subtype (Fisher’s exact test, p = 0.08; Figure S4H). Meanwhile, all PTEN-mutant tumors were cold tumors (Fisher’s exact test, p < 0.05; Figure 5F). Furthermore, PTEN-mutant tumors presented with a lower immune score (Figure 5G) and reduced infiltration of multiple immune cells, such as CD8+ T cells, CD4+ Tem cells, and DCs (Figure 5H). GSEA also revealed the downregulation of interferon alpha and APP pathways in PTEN-mutant tumors (Figure S4I), indicating a potential role of PTEN mutation on the loss of immunogenicity in PSC. Proteomic subtypes and molecular features of PSC We next examined the proteomic subtypes in the TU-PSC cohort. The non-negative matrix factorization (NMF)-based unsupervised clustering identified three distinct proteomic subtypes (S1, n = 36; S2, n = 22; S3, n = 16) exhibiting unique features regarding genomic, immunologic, and potential therapeutic targets (Figures 6A, S5A, and S5B; Table S5). These three proteomic subtypes exhibited distinct molecular features: the S1 subtype was associated with the proteasome pathway and higher APP activity, the S2 subtype was associated with the EMT pathway and the most abundant EMT-related phosphosites, and the S3 subtype was characterized by the enrichment of DNA replication pathway and a higher PTEN mutation rate (Figures 6 and S5–S7). Proteomic subtypes and molecular characteristics of the TU-PSC cohort (A) Heatmap showing clinicopathologic features, representative enriched biological processes, and signature proteins of three proteomic subtypes identified using NMF-based unsupervised clustering. (B) Heatmap showing significantly enriched pathways (MSigDB Hallmark) in three proteomic subtypes via gene set enrichment analysis. ∗p < 0.05, ∗∗p < 0.01. (C) Stacked bar plot showing the distribution of PSC immunotype (hot tumor and cold tumor) in each proteomic subtype (chi-squared test). (D) Comparison of PD-L1 mRNA (left) and protein (right) expression among three proteomic subtypes (one-way ANOVA following Tukey’s test). (E) Significant enrichment of proteasome and proteasome complex pathways in the S1 subtype via gene set enrichment analysis. (F) Comparison of protein expression of PSMB8 among three proteomic subtypes (Wilcoxon rank-sum test). (G) Comparison of protein expression of PSMB9 among three proteomic subtypes (one-way ANOVA following Tukey’s test). (H) Significant enrichment of antigen processing and presentation pathway in protein level in the S1 subtype via gene set enrichment analysis. (I) Comparison of antigen processing and presentation pathway protein activity among three proteomic subtypes (one-way ANOVA following Tukey’s test). (J) Comparison of tumor purity among three proteomic subtypes (one-way ANOVA following Tukey’s test). (K) Comparison of stromal score among three proteomic subtypes (one-way ANOVA following Tukey’s test). (L and M) Comparison of the mRNA and protein EMscore among three proteomic subtypes (one-way ANOVA following Tukey’s test). (N) Heatmap of representative EMT-related phosphosites in the S2 subtype. The comparison between proteomic subtypes and immunotypes showed a predominance of hot tumors in the S1 and S2 subtypes, while cold tumors were dominant in the S3 subtype (chi-squared test, p < 0.01; Figure 6C). Consistently, the S1 and S2 subtypes exhibited higher PD-L1 expression (Figure 6D), higher immune score, and lower TIDE score (Figures S5C and S5D) compared with the S3 subtype. GSEA using Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets revealed enrichment of genes associated with the proteasome and proteasome complex in the S1 subtype (Figure 6E). Furthermore, the S1 subtype exhibited higher expression levels of multiple proteasome-related proteins (Figure S5E). PSMB8 and PSMB9, two subunits of the immunoproteasome associated with higher antigen presentation43 and better response to ICB in NSCLC44 and melanoma,45 were specifically upregulated in the S1 subtype (Figures 6F and 6G) and were associated with improved survival in PSC (Figures S5F and S5G). In line with the important role of immunoproteasome in antigen presentation,43 the APP pathway was also strongly induced in the S1 subtype (Figures 6H and 6I). The S2 subtype presented with relatively lower tumor purity and a higher stromal score (Figures 6J and 6K), indicating less carcinoma but more sarcomatoid characteristics in this subtype. Notably, the EMT pathway was specifically activated in the S2 subtype (Figure 6B), consistent with higher mesenchymal characteristics observed in the sarcomatous component of PSC based on RNA-seq data.12 Consistently, key components of the EMT pathway also showed higher expression in the S2 subtype (Figures S5H and S5I). To further confirm this result, we derived an EMscore (higher values indicate more sarcomatoid feature, and lower values indicate more carcinoma feature)12 and found a higher EMscore in the S2 subtype at both mRNA and protein levels (Figures 6L and 6M; STAR Methods), again suggesting more evident sarcomatoid characteristics for the S2 subtype. TGF-β is a well-established and potent inducer of EMT,46,47 and the S2 subtype exhibited significantly activated TGF-β signaling (Figures S5J–S5L). Intriguingly, several phosphosites on VIM and ZEB2, two markers of the mesenchymal phenotype, displayed more pronounced induction than the corresponding proteins in the S2 subtype (Figures 6N, S5M, and S5N). The mesenchymal phenotype has been shown to exhibit increased susceptibility to ferroptotic cell death.29 Notably, we found that the S2 subtype was enriched for iron- and ROS-related biological functions (Figures S6A and S6B) and displayed elevated protein expression of FTL (Figure S6C). Importantly, FTL has been reported to promote EMT through AKT/GSK3β/β-catenin signaling in glioma.48 By knocking down FTL in PLA-801D cells, we observed the reduction of EMscore, coagulation activity, and extracellular matrix activity, along with the downregulation of several key EMT markers (Figures S6D–S6J; Table S5), indicating that targeting ferroptosis may inhibit EMT activity in PSC. DNA replication and cell cycle were hallmarks of the S3 subtype (Figure 6B). We observed higher tumor purity, lower stromal score, and lower EMscore in the S3 subtype relative to the other two subtypes (Figures 6J–6M), suggesting more evident carcinoma characteristics to some extent in the S3 subtype. The S3 subtype showed higher expression of classical LUSC markers (SOX2 and TP63) and lower expression of classical LUAD markers (NAPSA and KRT7) (Figures S7A and S7B). When associating somatic mutations with proteomic subtypes, we found a higher mutation rate of PTEN (Figure S7C) and downregulation of PTEN protein expression in the S3 subtype (Figure S7D), which may contribute to the lower immune activity of the S3 subtype49 (Figures 6C and S5C). Querying compounds that could reverse the S3 subtype signature based on the Library of Integrated Network-Based Cellular Signatures (LINCS)50 revealed an enrichment for TOP inhibitors (TOPi) in the S3 subtype (Figure S7E). Notably, multiple TOP family members, TOP1, TOP2A, and TOP2B, were most abundant at the protein level in the S3 subtype (Figure S7F). We further confirmed that the S3 subtype presented with a significantly higher sensitivity score to topotecan than other tumor subtypes (Figure S7G). In addition to the aforementioned features, the S3 subtype also exhibited higher mRNA expression of MYCN and a higher stemness score (Figures S7H and S7I). The higher protein abundance of certain epigenetic targets in the S3 subtype, such as KDM1A, HDAC1, and HDAC2 (Figure S7J), suggested a potential application of epigenetic therapy for this subtype. Based on phosphoproteomic data, we compared the kinase activities across proteomic subtypes and identified subtype-specific kinases (Figure S7K). We also conducted unsupervised clustering analysis using phosphoprotein data and identified three phosphoproteomic subtypes (Pho1, n = 32; Pho2, n = 19; Pho3, n = 21), which were largely congruent with the proteomic subtypes (Figure S7L; Table S5). Consistently, GSEA revealed similar pathway enrichment between the phosphoproteomic subtypes and proteomic subtypes (Figures S7M and 6B). Molecular features associated with ICB immunotherapy resistance of PSC Clinically, PSC patients are typically diagnosed at an advanced stage, and the application of neoadjuvant immunotherapy improves the surgical resectability for NSCLC, possibly also for PSC. Our immune landscape and proteomic subtype analyses indicated that a relatively large proportion of PSC with hot tumor behaviors may benefit from ICB immunotherapy (Figures 5A and 6A). To explore molecular features associated with ICB immunotherapy efficacy in PSC, we prospectively collected four fresh PSC tumor samples for scRNA-seq analysis from three PSC patients receiving neoadjuvant chemoimmunotherapy treatment (neoadjuvant ICB-PSC cohort, Figure 1A; Table S6). This cohort includes two samples collected by biopsy before treatment from one patient with pathological complete response (pCR) (tissue 1) and one patient with non-major pathologic response (NMPR) (tissue 4) (baseline, n = 2), as well as two samples taken from surgical resections after chemoimmunotherapy (tissue 2, pCR and tissue 3, MPR) (post-treatment, n = 2) (Figure S8A). The neoadjuvant ICB-PSC cohort also included paired samples of baseline and post-treatment from a pCR patient (tissue 1 and tissue 2 from patient 1) (Figure S8A). After doublet removal and stringent quality control, we obtained 60,932 cells and categorized these cells into 22 clusters without apparent batch effect (Figures S8A and S8B), which mainly comprised immune cells, epithelium cells, and stromal cells (Figure S8C). Cell clustering analysis identified 12 major cell types: B cell, DC, endothelial cell, epithelium cell, macrophage, mast cell, monocyte, neutrophil, natural killer (NK) cell, plasma cell, stromal cell, and T cell (Figures 7A, S8D, and S8E), which was confirmed by manual annotation with the expression of canonical marker genes (Figure 7B). scRNA-seq profiles for PSC with ICB immunotherapy (A) UMAP plot of all scRNA-seq data showing 12 cell types. (B) Heatmap showing the expression of top marker genes of 12 cell types. (C) Expression comparison of EMT-related genes (VIM, COL1A1, COL6A1, and TIMP1) between carcinoma and sarcomatoid cells (Wilcoxon rank-sum test). (D) Expression comparison of EMT-related transcription factors (TFs) (ZEB2 and SNAI2) between carcinoma and sarcomatoid cells (Wilcoxon rank-sum test). (E) Expression comparison of epithelial genes (KRT7 and EPCAM) between carcinoma and sarcomatoid cells (Wilcoxon rank-sum test). (F) Expression comparison of MHC-II genes (HLA-DMA, HLA-DRA, HLA-DRB1, and HLA-DPB1) between carcinoma and sarcomatoid cells (Wilcoxon rank-sum test). (G) Significantly activated pathways in NMPR tumors compared with pCR tumors via gene set enrichment analysis. (H) Expression comparison of EMT-related genes (VIM, COL1A1, COL4A1, and COL6A1) between NMPR and pCR tumors in baseline samples (Wilcoxon rank-sum test). (I) Comparison of C1 program score between NMPR and pCR tumors in baseline samples (Wilcoxon rank-sum test). (J) The significant enriched KEGG pathways for the C1 program. (K) Comparison of C1 program score at mRNA, protein, and phosphoprotein levels among three proteomic subtypes in the TU-PSC cohort (Wilcoxon rank-sum test). Our proteomic subtype characterization analysis revealed a higher stromal score and more evident sarcomatoid characteristics in the S2 subtype, which exhibited a specific enrichment of the EMT pathway (Figure 6). Notably, an important role of the EMT pathway in the carcinogenesis of PSC has been reported.12,15 We therefore first investigated the involvement of the EMT pathway in the transition process from pulmonary carcinoma cells to sarcomatoid cells in PSC based on scRNA-seq. Combined with the results from inferCNV (Figure S8F), we derived malignant cells with carcinoma cells from epithelium cells and sarcomatoid cells from stromal and endothelial cells. We further performed cell trajectory analysis in all malignant cells of scRNA data. The pseudotime trajectory axis suggested a relocation of EMT from the carcinoma cells at the starting area to the sarcomatoid cells at the terminal area (Figures S8G and S8H), suggesting a possible involvement of the EMT pathway in inducing transcriptional remodeling from carcinoma cells to sarcomatoid cells in PSC. Pseudotemporal expression dynamics analysis further highlighted the top pseudotime-dependent genes exhibiting a gradual upregulation from pseudotime 0 to 100, including COL1A1, COL6A1, TIMP1, and VIM, which are involved in EMT (Figure S8I). Consistently, sarcomatoid cells showed higher expression of EMT-related genes and transcription factors compared with carcinoma cells (Figures 7C and 7D). In addition to higher EPCAM and KRT7 expression, multiple major histocompatibility complex (MHC)-II members were highly expressed in carcinoma cells (Figures 7E and 7F). We then explored the interaction between malignant cells and immune cells, and results revealed that ligand-receptor pairs involved in the MHC-II pathway (HLA-DMA_CD4, HLA-DMB_CD4, HLA-DQA1_CD4, and HLA-DQB1_CD4) were more frequent between carcinoma cells and DCs and macrophages. However, the interactions between ligand-receptor pairs involved in collagen-related pathways (COL6A1_CD44, COL6A2_CD44, and COL6A3_CD44) were found to be more frequent between sarcomatoid cells and DCs and macrophages (Figures S9A and S9B). Based on baseline tumor cells, GSEA revealed the enrichment of EMT-related pathways in the NMPR patient (Figure 7G). The EMT-related genes, including VIM, COL1A1, COL4A1, and COL6A, were highly expressed in baseline tumors from the NMPR patient compared with the pCR patient (Figure 7H). To identify shared cell states across tumors, despite predominant patient-specific differences in malignant cells,51 we used the Harmony algorithm52 to integrate malignant cells into 6 subsets shared across tumors (Figures S9C and S9D). Intriguingly, the activity of the top 1 C1 program was higher in tumor cells from the NMPR patient compared with the pCR patient, while the other five programs exhibited the opposite pattern of activity (Figures 7I and S9E). KEGG and GO enrichment analysis revealed an enrichment of the EMT-related processes in the C1 program (Figures 7J and S9F). These results suggested that the activation of the EMT pathway may be associated with ICB resistance in PSC. Notably, when associating the C1 program with proteomic subtypes in the TU-PSC cohort, we found a higher C1 program score in the S2 subtype (Figure 7K), suggesting that the S2 subtype exhibiting the higher EMT features might be resistant to ICB immunotherapy, despite both S1 and S2 proteomic subtypes exhibiting higher PD-L1 expression in the TU-PSC cohort (Figure 6D). Discussion In this study, we present an integrated molecular portrait of PSC by leveraging large-scale genomics, transcriptomics, proteomics, phosphoproteomics, and scRNA-seq analysis for Chinese PSC patients. This approach provides an up-to-date, comprehensive multi-omics landscape of this rare lung cancer. Comprehensive multi-omics data provide a deeper exposition of PSC, such as distinct proteomic subtypes, immunotype, potential therapeutic vulnerabilities, immune regulators, and indicators of immunotherapy resistance. Overall, this study provides a rich resource for the exploration of mechanistic insights and therapeutic targets in PSC. Single-omic data are not sufficient for comprehensively characterizing PSC, which is under complex regulation, and our study provides additional layers of characterization from proteomics and phosphoproteomics data. We identified a suppressed ferroptosis signature in PSC tumors based on proteomic profiling. Ferroptosis plays important roles in both lung cancer development and therapy.53,54 We found an aberrant expression of the ferroptosis pathway in PSC tumors, with an elevated FNR signature significantly associated with poor survival in PSC patients. Notably, SLC3A2 and FTL, two critical ferroptosis repressors with distinct mechanisms of action,55 were both upregulated in PSC tumors. SLC3A2 is known to play an essential role in system Xc–, which mediates the uptake of cystine.55 High expression of SLC3A2 was positively correlated with ferroptosis resistance in several cancer types.56,57,58 FTL is responsible for intracellular iron storage.35 Dysregulation of redox homeostasis and iron metabolism barriers also play important roles in the mechanism of ferroptosis.59 By performing in vitro experiments, we showed that blocking either the expression of SLC3A2 or FTL remarkably reduced tumor cell proliferation by inducing ferroptosis in the PSC cell line through reduced GSH levels. These results demonstrated that ferroptosis significantly affects the proliferation of PSC and thus may represent a potential target for treating PSC. With the proteogenomic characterization of PSC, we were able to determine how genes associated with disease phenotypes are regulated. PTEN is a frequently epigenetically downregulated or somatically mutated tumor suppressor in many types of cancer.60 We identified PTEN as a significantly mutated gene in the TU-PSC cohort, particularly in the PC histological subtype of PSC. Notably, the PTEN mutation was much more pronounced in the cold tumor cluster of PSC, indicating its involvement in the loss of immunogenicity. PTEN deficiency promotes immune suppression and therapeutic resistance to immunotherapy in various cancer types.61,62 Since restoring the function of PTEN is currently unfeasible, a more practical strategy to mitigate the effects of PTEN loss is to inhibit PI3K signaling.63 Preclinical studies have shown that combining PI3Kβ inhibition with ICB leads to positive responses in PTEN-deficient mouse models of melanoma and breast cancer,64,65 providing clues for treating PTEN-mutant PSC. Regarding hot tumors of PSC, Seth et al. reported an immune-high subtype associated with high immune score and infiltration of immune cells in PSC,66 showing similar biological features to the hot tumor cluster in our study. PSC consists of five histological subtypes, emphasizing substantial intertumoral heterogeneity. Therefore, molecular subtyping based on omics data is essential for deepening our understanding of intertumoral heterogeneity and advancing personalized medicine for PSC. The analysis of transcriptome and DNA methylation data provided the initial insights into the molecular classification of PSC.12 Compared with RNA and DNA methylation, proteins are the end products of gene expression, directly reflecting the functional state of cells and tissues. Therefore, protein profiling aligns more closely with actual biological processes, whereas RNA or DNA methylation indicates potential genetic activity or epigenetic modifications, which may not always correlate with functional outcomes. Based on large-scale proteomics data from 74 PSC patients, we provided an initial PSC proteomic stratification, revealing three PSC subtypes with unique molecular features: the S1 subtype showed higher antigen presentation activity and proteasome subunit expression; the S2 subtype showed EMT pathway enrichment and the most abundant EMT-related phosphorylated sites; and the S3 subtype was characterized by cell proliferation and may benefit from TOPi. Furthermore, several proteomic subtype-enriched kinases were identified, while their therapeutic utility requires further investigation. These findings highlight the potential of these molecular subtypes to guide personalized treatment selection for PSC. One notable feature of PSC is its high ranking in TMB and leukocyte fraction, indicating a T cell-inflamed TME in PSC.12 PD-1/PD-L1 ICB appears to be highly effective for PSC,67 potentially due to the relatively higher percentage of hot tumors in PSC (57.3%) compared with LUAD (26.8%)16 and LUSC (25.7%).17 Our proteomic analysis identified that two (S1 and S2) of the three PSC proteomic subtypes exhibited hot tumor features. However, the S2 subtype was distinguished from the S1 subtype by its higher EMT pathway activity. Yang et al. revealed the important roles of EMT in the carcinogenesis of PSC.12 By performing scRNA-seq analysis of samples from PSC patients receiving neoadjuvant chemoimmunotherapy, we further found evidence suggesting a potential role of EMT in the progression from carcinoma cells to sarcomatoid cells and in immunotherapy resistance in PSC. EMT is associated with increased cancer cell plasticity, invasiveness, and resistance to immunotherapy in multiple cancers.68,69,70 In addition, EMT may predict an unfavorable outcome after ICB therapy in PD-L1-high but not PD-L1-low NSCLC.71 Although both S1 and S2 subtypes exhibited higher PD-L1 expression, the S2 subtype, characterized by stronger EMT features, may be resistant to ICB immunotherapy. Alongside the higher PD-L1 expression, TGF-β signaling in the S2 subtype may contribute to EMT and immune evasion in cancer, thereby compromising the responsiveness to ICI therapy.71 This mechanistic insight also underlines the potential utility of bispecific antibodies targeting TGF-β and PD-L1 for the S2 subtype as a therapeutic strategy in PSC.72 Importantly, we found higher FTL protein expression in the S2 subtype and demonstrated that derepressing ferroptosis by knocking down FTL may inhibit EMT activity in the PLA-801D cell line, suggesting that ferroptosis may serve as a potential therapeutic vulnerability in PSC, particularly within the EMT-enriched S2 proteomic subtype. Notably, during the peer-review process of our study, pharmacologically inducing ferroptosis through the inhibition of FSP1 showed significant therapeutic benefits in preclinical LUAD mouse models.73 Combined with our findings, these results underscore the potential of targeting ferroptosis as a therapeutic strategy for patients with PSC and other forms of LUAD. In conclusion, our integrative multi-omics portrait of PSC provides a critical resource for PSC biology and sheds light on the biological underpinnings, disease vulnerability, and treatment selection of PSC. Our study underscores the important value of proteome-based subtyping and how it can contribute to the personalized therapies to a certain extent, thereby bringing us closer to the goal of proteomics-driven precision medicine in PSC. Fundamentally, our work holds important clinical implications in PSC, and we look forward to further developing potential therapeutic strategies for this rare lung cancer. Limitations of the study This study has several limitations. While our study included the largest cohort to date for investigating the multi-omics features of PSC, the sample size, particularly in the scRNA-seq analysis, remained limited. Although baseline NMPR tumors showed enhanced EMT features and EMT is a hallmark for PSC, specific criteria for distinguishing ICB NMPR patients based on the EMT activity require further exploration. The retrospective and single-center design of this study may also lead to several inherent biases, such as selection bias. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. Haiyang Hu (haiyanghu@cpu.edu.cn). Materials availability This study did not generate new unique reagents. Data and code availability •The WES and RNA-seq raw data have been deposited in the Genome Sequence Archive in the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA008505 that are publicly accessible at http://bigd.big.ac.cn/gsa-human. The proteomic data have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix, OMIX007422). QC and enrichment metrics data have been deposited in Zenodo (https://zenodo.org/records/17595641). This paper does not report original code. The exact analysis scripts for the two-component normalization and KSEA have been deposited in Zenodo (https://zenodo.org/records/17696161). Sample annotation, processed, and normalized data files are provided in Table S1. The software and code used in this study are referenced in their corresponding STAR Methods sections and in the key resources table. The WES and RNA-seq raw data have been deposited in the Genome Sequence Archive in the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA008505 that are publicly accessible at http://bigd.big.ac.cn/gsa-human. The proteomic data have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix, OMIX007422). QC and enrichment metrics data have been deposited in Zenodo (https://zenodo.org/records/17595641). This paper does not report original code. The exact analysis scripts for the two-component normalization and KSEA have been deposited in Zenodo (https://zenodo.org/records/17696161). Sample annotation, processed, and normalized data files are provided in Table S1. The software and code used in this study are referenced in their corresponding STAR Methods sections and in the key resources table. •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact, Haiyang Hu (haiyanghu@cpu.edu.cn), upon request. Any additional information required to reanalyze the data reported in this work paper is available from the lead contact, Haiyang Hu (haiyanghu@cpu.edu.cn), upon request. Acknowledgments This study was supported by the National Natural Science Foundation of China (grant no. 82430053, 82125001, 82471347, 82201948, and 82173393), the Shanghai Science and Technology Committee (grant no. 24SF1904500), Innovation Program of Shanghai Municipal Education Commission (grant no. 2023ZKZD33), Tongji University Medicine-X Interdisciplinary Research Initiative (grant no. 2025-0554-ZD-03, 2026-0554-ZD-06, and 2026-0554-YB-19), and foundation of Shanghai Pulmonary Hospital (grant no. TJ-FK-YXJC003, LYRC202402, FKLY20004, and FKYC2604). We thank Prof. Fan Bai and Dr. Yang Liu (Biomedical Pioneering Innovation Center, Peking University, Beijing) for expert guidance and insightful suggestions to our study. The graphical abstract and Figure 4L were generated with BioRender. Author contributions Conceptualization, H.H., L.Z., P.Z., and M.D.; methodology, L.S., J.W., L.Z., Y.Y., and S.D.; validation, J.W., D.W., Jiameng Hu, H.Y., and N.W.; formal analysis, L.S., J.W., and Y.Y; investigation, L.S., J.W., Y.Y., and S.D.; resources, M.D., P.Z., L.Z., and H.H.; data curation, L.S., J.W., and Y.Y.; writing – original draft, L.S., J.W., Y.Y., and S.D.; writing – review and editing, L.S., J.W., Y.Y., M.Z., R.Z., L. Hou, Junjie Hu, D.B., S.H., T.G., Y.C., L. Han, J.S., M.D., P.Z., L.Z., and H.H.; visualization, L.S., J.W., Y.Y., and L.Z.; supervision, H.H., L.Z., P.Z., and M.D.; funding acquisition, P.Z., H.H., L.Z., and M.D. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesRabbit monoclonal anti-CD98/SLC3A2AbclonalCat#A24735; RRID:AB_3677543Rabbit monoclonal anti-FTLAbclonalCat#A11241; RRID:AB_2861523Rabbit polyclonal-GPX4AbclonalCat#A1933; RRID:AB_2763960Mouse monoclonal anti-β-ActinAbclonalCat#AC004; RRID:AB_2737399Biological samplesPaired tumor, normal adjacent lung tissues from a cohort of 86 PSC patientsShanghai Pulmonary Hospital, Tongji UniversityThis paperFresh tumor lung tissues from a cohort of 3 PSC patients receiving neoadjuvant immunotherapyShanghai Pulmonary Hospital, Tongji UniversityThis paperCritical commercial assaysQIAamp Fast DNA tissue kitQIAGENCat#51404SureSelect Human All ExonV6 KitAgilent TechnologiesCat#5190-8865NEBNext Ultra DNA Library Prep KitNEBCat#E7530LTRIzol ReagentInvitrogenCat#15596026NEBNext® UltraTM RNA Library Prep KitNEBCat#E7530LDeposited dataProteomics data of TU-PSC cohortThis paperOMIX database: OMIX007422; https://ngdc.cncb.ac.cn/omixWES and RNA-seq data of TU-PSC cohortThis paperGSA database: HRA008505; http://bigd.big.ac.cn/gsa-humanRaw WB resultsThis paperMendeley Data; https://doi.org/10.17632/7wgpmf85r4.1GENCODE (version 22)GENCODEhttps://www.gencodegenes.orgGRCh38/hg38NCBIhttps://www.ncbi.nlm.nih.gov/assembly/GCF_000001405.26/Human Protein AtlasUhlén et al.28https://www.proteinatlas.orgMSigDB v7.1 gene setsLiberzon et al.74https://www.gseamsigdb.orgPhosphoSitePlusHornbeck et al.75https://www.phosphosite.orgExperimental models: Cell linesHuman: PLA-801DATCCN/AOligonucleotidessiRNA targeting sequence: human shControl, 5′-ATGGTAGCGTACACTTATGAT-3′This paperN/AsiRNA targeting sequence: human shSLC3A2#1, 5′-CTAGCTCATACCTGTCTGATT-3′This paperN/AsiRNA targeting sequence: human shSLC3A2#2, 5′-GCTGGGTCCAATTCACAAGAA-3′This paperN/AsiRNA targeting sequence: human shFTL#1, 5′-GCTCACTCTCAAGCACGACTA-3′This paperN/AsiRNA targeting sequence: human shFTL#2, 5′-CTGGAGACTCACTTCCTAGAT-3′This paperN/ASoftware and algorithmsGATK (version 4.1.7.0)Broad Institutehttps://software.broadinstitute.org/gatk/Picard (version 2.22.8)GitHubhttp://broadinstitute.github.io/picard/MutSigCV (version 1.4)Lawrence et al.76https://www.genepattern.org/modules/docs/MutSigCVMutect2 (version 4.1.7.0)Cibulskis et al.77https://github.com/broadinstitute/gatk/mutect2Strelka2 (version 2.9.10)Kim et al.78https://github.com/Illumina/strelkaCNVkit (version 0.9.6)Talevich et al.79https://github.com/etal/cnvkitGISTIC (version 2.0.23)Mermel et al.80http://broadinstitute.github.io/gistic2/STAR (version 2.7.6a)Dobin et al.81https://github.com/alexdobin/STARSalmon (version 1.3.0)Patro et al.82https://combine-lab.github.io/salmon/FlowJo software (version 10.9.0)Waters Corporationhttps://www.flowjo.com/flowjo10/overviewImageJ software (version 2.14.0)National Institutes of Healthhttps://imagej.net/ij/MaxQuant software (version 2.0.2.0)Tyanova et al.83http://www.maxquant.orgXcalibur software (version 4.1)Thermo Fisher Scientifichttps://www.thermofisher.cn/cn/zh/home/industrial/mass-spectrometry/liquid-chromatography-mass-spectrometry-lc-ms/lc-ms-software/lc-ms-data-acquisition-software/xcalibur-data-acquisition-interpretation-software.htmlConsensusClusterPlus (version 1.64.0)Wilkerson et al.84https://bioconductor.org/packages/ConsensusClusterPlus/ESTIMATE (version 1.0.13)Yoshihara et al.85https://bioinformatics.mdanderson.org/public-software/estimate/xCell (version 1.1.0)Aran et al.40https://xcell.ucsf.edu/NMF R-package (version 0.26)Gaujoux and Seoighe.86https://cran.r-project.org/web/packages/NMF/index.htmlsurvminer (version 0.4.9)GitHubhttps://github.com/kassambara/survminersurvival (version 3.5–5)GitHubhttps://github.com/therneau/survivalR software (version 4.3.1)R Development Core Teamhttps://www.r-project.org/DreamAIPei Wang Lab.87https://github.com/WangLab-MSSM/DreamAIKSEAapp (version 0.99.0)Wiredja et al.88github.com/casecpb/KSEA/Seurat version (version 4.2.0)Hao et al.89https://satijalab.org/seurat/ Experimental model and study participant details Clinical specimens All PSC tumor and NAT samples for this study were collected from Shanghai Pulmonary Hospital (Tongji University, Shanghai, China) between December 2009 to April 2021. This study was approved by the Institutional Review Board of Shanghai Pulmonary Hospital (K23-309) and written informed consents of all participating patients were obtained. Paired histologically-NATs were collected from the same patient at tumor resection, which were procured from tumor-adjacent regions confirmed to be free of tumor infiltration by pathological assessment. Tissues were obtained and stored in liquid nitrogen less than 30 min after resection. Each sample underwent only one freeze-thaw cycle prior to proteogenomic analyses. Histological sections from each case were reviewed by two board-certified pathologists to confirm the assigned pathology. The clinicopathological parameters were collected and summarized in Table S1, including gender, age, smoking status, tumor site, histologic type, tumor size, date of surgical resection, TNM staging (AJCC cancer staging system 8th edition), and so on. This cohort included 73 males and 13 females with a median age at diagnosis of 63 (range 37–79). In total, 27, 33, 25, and 1 patient was classified as TNM stages I, II III, and IV respectively. Overall survival (OS) was defined as the interval between surgery and death or last date of follow-up. Cell lines The PLA-801D cell line (human pulmonary giant cell carcinoma cell, male) was obtained from the Institute of Biochemistry and Cell Biology at the Chinese Academy of Science. Method details Specimen acquisition and preparation For accurate tumor cellularity analysis, hematoxylin and eosin (H&E) staining of the middle section of each tissue was done. Pathological features of all tumor samples were confirmed independently by two certified and experienced pathologists. Acceptable PSC tumor tissue segments were determined by two board-certified pathologists based on total cellularity (>50%) and necrosis (<20%). Finally, high-quality tumors and paired NATs from 86 patients were chosen for WES, RNA-seq, proteomic, and phosphoproteomic profiling). PSC tumor specimens and NATs were cryo-pulverized using the Covaris CryoPREP instrument, and then divided into three parts. The first part was snap-frozen in the liquid nitrogen and stored at −80°C for DNA extraction and WES; the second part was stored in the 1 mL RNAlater Reagent in −80°C for RNA extraction and bulk RNA-seq; the remaining samples were treated with SDS lysis buffer and kept at −80°C for the subsequent proteomic and phosphoproteomic analyses. Genomic and transcriptomic profiling experiments DNA extraction and whole exome sequencing (WES) Genomic DNA was extracted from tumors and NATs using QIAamp Fast DNA tissue kit (QIAGEN) according to the manufacturer’s protocol. Total DNA was quantified by the Qubit 2.0 Flurometer (Life Technologies) and NanoDrop 2000 (Thermo Fisher Scientific) and the integrity was assessed by TapeStation (Agilent Technologies). For Illumina sequencing library construction, the genomic DNA was fragmented to an average size of 180–280 bp using a Covaris focused-ultrasonicator. Then, WES libraries were prepared and captured using the Agilent SureSelect Human All Exon V6 kit (Agilent Technologies) following the manufacturer’s instructions. The DNA library with 150 bp paired-end reads was sequenced on the Illumina Novaseq 6000 platform. RNA extraction and RNA sequencing Total RNA was extracted and purified from fresh frozen tissues using the TRIzol reagent (Invitrogen). All RNA analytes were assayed for RNA concentration and purity using Qubit® RNA Assay Kit in Qubit® 2.0 Flurometer (Life Technologies) and NanoPhotometer® spectrophotometer (IMPLEN). RNA integrity was assessed using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies). Samples with RNA integrity number (RIN) > 6.0 were considered high quality and used to prepare the transcriptome library. Total RNA-seq library construction was performed from the RNA samples using NEBNext® UltraTM RNA Library Prep Kit for Illumina® (NEB) following manufacturer’s recommendations and index codes were added to attribute sequences to each sample. Quality control was performed at every step, and the libraries were quantified using the Agilent Bioanalyzer 2100 system (Agilent Technologies). Then, paired-end libraries were sequenced on an Illumina Novaseq 6000 platform and 150 bp paired-end reads were generated. Peptides preparation for MS analysis Protein extraction and digestion For the TMT 11-plex proteomic and phosphoproteomic experiments, cryopulverized human PSC and corresponding NAT samples were homogenized in a lysis buffer containing 1% SDS, 10 mM HEPES (pH 7.0), 2 mM MgCl2, Benzonase (1:1 000 dilution), and Roche PhosSTOP (1:50 dilution). After homogenization, samples were centrifuged at 12 000 rpm for 15 min to collect the supernatant. Protein concentrations were determined using the BCA assay. For each sample, 4.5mg proteins were reduced with 2 mM DTT (Sigma, D9163) for 20 min on an end-to-end rotator and then alkylated with 20 mM iodoacetamide (Sigma, I1149) in the dark at room temperature for another 20 min. Protein precipitation was carried out using the methanol-chloroform method (1 part lysate: 4 parts methanol: 1 part chloroform: 3 parts H2O). The protein pellet was washed with 1 mL of methanol, then reconstituted in 8 M urea containing 20 mM Tris HCl and 10 mM EDTA (pH 7.5). Protein digestion was performed in 8 M urea buffer using Lys-C enzyme (Promega, #VA1170) at a 1:200 enzyme-to-protein ratio for 3 h. Then three volumes of 20mM Tris-Cl, pH 7.5 was used to dilute the 8 M Urea to 2M. Trypsin (Promega, #V5280) was then added at a 1:100 enzyme-to-protein ratio for an overnight digestion at 37°C in a 2 M urea diluted solution. The digested peptides were acidified with trifluoroacetic acid (TFA) to a pH of 2–3. Desalting was performed using HLB cartridges (60 mg, Waters Oasis HLB, WAT094226). A solution of 0.1% formic acid and 40% acetonitrile were used to elute the peptides from HLB resin. Ten percent of each eluate was dried using a speedvac, and peptide concentrations for global proteomic analysis were measured using a BCA assay. For phospho-peptides enrichment, the remaining eluate was supplemented with 2 M lactic acid for phosphopeptide enrichment. Preparation of reference samples The proteomic and phosphoproteomic analyses of PSC samples were structured as TMT 11-plex experiments. The reference samples for TMT-11 experiments were prepared by pooling all 74 pairs of PSC tumor and adjacent non-tumor samples, ensuring equal protein amounts. Aliquots (4.5 mg) of this reference mixture were frozen and consistently used in each multiplex experiment from the protein extraction and digestion stage. Phosphopeptide enrichment Phosphopeptides were enriched using titanium dioxide (TiO2) beads. The beads were washed twice with a binding buffer containing 50% acetonitrile, 0.1% TFA, and 2 M lactic acid, then re-suspended in the same buffer. Digested samples were incubated overnight with TiO2 beads at a 1:14 peptides-to-beads ratio with continuous rotation at room temperature. After incubation, the supernatant was removed by centrifugation, and the beads were washed three times with the binding buffer. Phosphopeptides were eluted twice using 600 μL of 100 mM K2HPO4 (pH 9.2). The combined eluates were acidified immediately to a pH of 2–3 using TFA and desalted using 10 mg C18 SepPak cartridges (Waters Oasis HLB, 186000383). The eluates were then dried using a vacuum centrifuge and the phospho-peptides powder were ready for further TMT label. TMT 11-plex labeling TMT 11-plex labeling was performed on 74 unique PSC tumor samples and their paired NAT samples, distributed across 15 TMT 11-plex experiments. Each of the 11-plex channels was used to label individual samples, with one channel reserved for a reference sample and the remaining ten for individual sample labeling, as described previously.90 For proteome analysis, desalted peptides were reconstituted in 200 mM HEPES (pH 8.0) to a final concentration of 1 mg/mL. A 50 μg aliquot from each sample was labeled with 150 μg of TMT 11-plex reagent. For phosphoproteome analysis, dried phosphopeptides were reconstituted in 200 mM HEPES (pH 8.0) to a final concentration of 0.5 mg/mL, and 30 μg of phosphopeptides were labeled with 120 μg of TMT 11-plex reagent, with a final acetonitrile concentration of 40%. After a 1-h incubation at room temperature, the reaction was quenched with hydroxylamine for 30 min. The labeled peptides were pooled, acidified with 20 μL of 20% TFA, lyophilized, and desalted using 10 mg C18 SepPak columns (Waters Oasis HLB, 186000383). Offline HPLC fractionation To enhance protein identification depth and reduce sample complexity, peptide fractionation was performed using basic high pH reverse-phase high-pressure liquid chromatography (RP-HPLC). The mobile phases used were Solvent A (10 mM ammonium formate, pH 9.9) and Solvent B (10 mM ammonium formic acid in 80% acetonitrile, pH 9.9). TMT-labeled peptides were solubilized in Solvent A and centrifuged at 13 000 rpm for 10 min to remove any insoluble peptides. The soluble peptides were then loaded onto a ZORBAX 300 Extend-C18 column (2.1 mm × 150 mm, 3.5 μm particle size, Agilent). Peptide separation was achieved using a Thermo Scientific Ultimate 3000 HPLC system with basic reversed-phase chromatography. For proteome analysis, a 150-min gradient at a flow rate of 0.15 mL/min was used, with the following settings: (min: %B) 0:4; 101:65; 106:100; 124:100; 126:0. Eluents were collected into 96 fractions over 5 to 100 min. To simplify downstream analysis, these 96 fractions were concatenated into 24 pooled fractions, which were subsequently acidified with TFA. For phosphoproteome analysis, a 100-min gradient was applied: (min: %B) 0:4; 75.5:65; 80:100; 98:100; 100:0. Eluents were collected and combined into 12 fractions over 5 to 75.5 min. Similar to the proteome analysis, fractions were acidified with TFA. The peptides and phosphopeptides were dried using a speedvac, desalted with stage tip columns, and re-suspended in 15 μL of 4% ACN/0.1% formic acid for nanoLC-MS/MS analysis, with 4 μL injected per run. LC-MS/MS for proteomic and phosphoproteomic analysis LC-MS/MS analysis for both proteomic and phosphoproteomic samples was conducted using an EASY-nLC 1200 UHPLC system (Thermo Fisher Scientific). Fractionated peptides were loaded onto an in-house packed capillary nanospray column (15 cm × 75 μm, 1.9 μm particle size, 10 μm tip opening). The mobile phases were 0.1% formic acid (Mobile Phase A) and 90% acetonitrile with 0.1% formic acid (Mobile Phase B), at a flow rate of 250 nL/min. For proteome and phosphoproteome analysis, a 110-min LC-MS/MS method was employed, consisting of a 10-min column equilibration step, a 20-min sample loading step, and the gradient profile: (min: %B) 0:8; 95:28; 100:100; 110:100. The same LC system and column configuration were used for both analyses. Both global proteome and phosphoproteome analyses were carried out on an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific) using an MS2-based TMT method in data-dependent acquisition mode with Xcalibur software (version 4.1). The spray voltage was set to 2.2 kV, and the ion transfer tube temperature was maintained at 275°C. MS1 data were acquired in the Orbitrap analyzer with a resolution of 120 000, an AGC target of 4e5, and an m/z range from 375 to 1 500. The maximum ion injection time was set to 50 ms. Charge states from 2 to 6 were selected, and a dynamic exclusion of 60 s with isotopes excluded was applied. MS2 scans were performed at a resolution of 30 000 using a quadrupole isolation window of 0.7. Genomic and transcriptomic data analysis Somatic mutation detection and filtering After excluding low-quality reads, the exome sequencing data was first aligned to GRCh38/hg38 sequence reference using BWA MEM (version 0.7.10) (http://bio-bwa.sourceforge.net/). The resulting BAM files were further processed including marking duplicates with Picard (version 2.22.8) (https://broadinstitute.github.io/picard/) and base quality recalibration with BaseRecalibrator from GATK (version 4.1.7.0) (https://gatk.broadinstitute.org/hc/en-us). To ensure robust variant calling, we employed a consensus approach utilizing two independent algorithms: Mutect277 (GATK version 4.1.7.0) and Strelka278 (version 2.9.10), both run with default parameters on paired PSC samples. The definitive set of single nucleotide variants (SNVs) and insertions-deletions (INDELs) was defined as the intersection of the outputs from both callers. This high-confidence set was then subjected to filtering against short tandem repeat regions (annotations sourced from the UCSC Table Browser, https://genome.ucsc.edu/cgi-bin/hgTables), followed by functional annotation using Funcotator in GATK (version 4.1.7.0). Deciphering of mutational signatures in PSC The r package “Sigminer” (version 2.2.2)91 was used to extract mutational signatures from the WES data. Bayesian variant of NMF algorithm was used to decipher mutational signatures in cancer somatic mutations stratified by 96 base substitutions in trinucleotide sequence contexts, and the optimal factorization of k value is selected when the magnitude of the cophenetic correlation coefficient begins to drop significantly. The gradient boosting machine learning technique yields an individual score for each mutational signature, combining likelihood with cosine similarity and exposure of signatures using the NNLS algorithm. After decomposing into four signatures, signatures were compared against validated single base substitution mutational catalogs retrieved from COSMIC (https://cancer.sanger.ac.uk/signatures/), and cosine similarity was calculated to identify the best match. Analysis of significantly mutated genes The significance of mutated genes was evaluated in tumors by MutSigCV (version 1.4)76 in GenePattern, and mutated genes were deemed significant above the background mutation rate if false discovery rate (FDR) for the genes was less than 0.1. Comparison of mutation frequency in PSC with previous studies Combining previously reported whole-exome sequencing or targeted sequencing cohorts (two Chinese cohorts10,12 and one Japanese cohort13), we compared the distribution of the somatic mutations across these four cohorts. Tumor mutational burden (TMB) and cancer-associated driver mutations TMB was calculated via R package “maftools” (version 2.16.0). Cancer-associated driver mutations were acquired from DriverDBv3.92 Copy number variation calling Somatic copy number variation (CNV) with discrete copy number segments was determined using the copy number calling pipeline of CNVkit (version 0.9.6).79 These were further examined using GISTIC (version 2.0.23)80 to identify gained or lost focal-level and arm-level events as well as CNV at the gene level. Q value less than 0.25 was considered significant. To exclude more false positives, a log2 ratio greater than 0.4 or less than −0.4 was deemed as the threshold for gain or loss, respectively. Default value was used for other parameters. RNA quantification After removal of adaptors and low-quality reads, RNA-seq cleaned sequence data were aligned to human reference sequence hg38 using STAR (version 2.7.6a).81 The resulting BAM files were analyzed for gene expression using Salmon (version 1.3.0)82 against a transcriptomic reference GENCODE (version 22, https://www.gencodegenes.org). Then the resulting gene-level counts for protein-coding genes were normalized and transformed into transcripts per million (TPM) values, followed by applying log2 transformation. Data rows of redundant gene symbols were aggregated by calculating the average expression values. Batch effects were adjusted using the “combat” function in R package “sva” (version 3.48.0). Proteomic and phosphoproteomic data analysis MS database searching The MS raw files generated by LC-MS/MS were analyzed using MaxQuant software (version 2.0.2.0)83 against the UniProt human proteome database (version 2019-03-07, containing 20,404 sequences). For quantification, TMT 11-plex MS2 reporter ion quantification was applied, with a reporter mass tolerance set at 0.003 Da. To reduce precursor co-fragmentation interference, the precursor intensity fraction (PIF) filter was set to 0.5. Carbamidomethylation of cysteine (C) was designated as a fixed modification. In the proteomic analysis, oxidized methionine (M), protein N-terminal acetylation, and deamidation of asparagine and glutamine (NQ) were set as variable modifications. In phosphoproteomic analysis, oxidized methionine (M), protein N-terminal acetylation, asparagine and glutamine deamidation (NQ), and phosphorylation of serine, threonine, or tyrosine (S/T/Y) were also treated as variable modifications. A target-decoy strategy was used to control the false discovery rate (FDR) at 1% for peptides, proteins, and phosphorylation sites. A maximum of five modifications per peptide was permitted, and trypsin was used as the enzyme with a maximum of two missed cleavage sites allowed. The peptide tolerance for the first and main searches was set at 20 ppm and 4.5 ppm, respectively, with a minimum peptide length of seven amino acids. For each TMT labeling dataset, channel purities were corrected based on the kit LOT number. Sites with computed localization probability greater than 0.75 were considered as confidently localized. The FDR filtering was set to 0.01 on PSM, PTM site and protein level. As for the rules of handling missing-value, only proteins and phosphorylation sites quantified in at least half samples were retained and imputed for the missing values using “RegImpute” algorithm. The imputation method was implemented using DreamAI tool, an ensemble algorithm developed during the NCI-CPTAC Dream Proteomics Imputation Challenge (version 0.1.0).87 Two-component normalization of TMT ratios It was assumed that for every sample there would be a set of unregulated proteins and phosphosites that have abundance comparable to the common reference (CR) sample. In the normalized sample, these proteins and phosphosites should have a log TMT ratio centered at zero. In addition, there were proteins and phosphosites that were either up or downregulated compared with the CR. A normalization scheme was employed that attempted to identify the unregulated proteins or phosphosites, and centered the distribution of these log-ratios around zero in order to nullify the effect of differential protein loading and/or systematic MS variation. A 2-component Gaussian mixture model-based normalization algorithm was used to achieve this effect. The two Gaussians N(mi1, si1) and N(mi2, si2) for a sample i were fitted and used in the normalization process as follows: the mode mi of the log-ratio distribution was determined for each sample using kernel density estimation with a Gaussian kernel and Shafer-Jones bandwidth. A two-component Gaussian mixture model was then fit with the mean of both Gaussians constrained to be mi, i.e., mi1 = mi2 = mi. The Gaussian with the smaller estimated standard deviation si = min (si1, si2) was assumed to represent the unregulated component of proteins/phosphosites, and was used to normalize the sample. The sample was standardized using (mi,si) by subtracting the mean mi from each protein/phosphosite and dividing by the standard deviation si. Proteogenomic analysis quality control and batch effect assessment We used principal component analysis to assess the presence of a batch effect in the sample/CR ratio data based on TMT-plex. There was no overall apparent batch effect in the global proteome or phosphoproteome datasets. To assess data quality, we assessed the Pearson correlations of log TMT reporter ion intensities between all pairs of TMT plexes, separately for sample and CR channels. For each given pair of TMT plexes, correlations between CRs were consistently higher than correlations of respective sample channels enabling precise relative quantitation of proteins and phosphorylation sites using the CR channels as common denominator. Integrated analysis Defining cancer-associated genes Cancer-associated genes (CAG) were compiled using the Census website https://cancer.sanger.ac.uk/census. CNA-driven cis and trans effects Correlations between copy number alterations (CNA) and mRNA, proteome, and phosphoproteome were determined using Pearson correlation. In addition, p values (corrected for multiple testing using Benjamini-Hochberg method) for assessing the statistical significance of the correlation values were also calculated. CNA trans-effects for a given gene were determined by identifying genes with statistically significant (FDR <0.05) positive or negative correlations. Tumor-NAT differential expression analysis Differential expression analysis was performed for tumors and paired NATs using the Wilcoxon signed-rank test. p values were adjusted using the Benjamini-Hochberg method. Each feature was required to be non-missing values. Proteins (collapsed into gene-level) differentially expressed between tumors and NATs (FDR <0.05, log2FC > 1 or < −1) were further used for KEGG enrichment analysis. PSC associated-proteins identification We filtered proteins with at least 2-fold upregulated in more than 90% tumor-NAT pairs to identify PSC-associated proteins. Then, proteins of specific type or function including cancer related genes, enzymes, plasma proteins, and drug targets (FDA approved drug targets and potential drug targets) were annotated by the Human Protein Atlas (HPA, https://www.proteinatlas.org).28 Kinase Substrate Enrichment Analysis The kinase-substrate enrichment analysis (KSEA) was conducted by using the KSEAapp R package (version 0.99.0).88 Substrates of every kinase were collected from the PhosphoSitePlus database and NetworKIN.75,93,94 A kinase score was given for each kinase based exclusively on the collective phosphorylation status of its substrates and transformed into Z score. For the kinase enrichment analysis, the threshold for significantly enriched kinases was p value <0.05. Gene set enrichment analysis For gene set enrichment analysis (GSEA), results for all proteins were ranked by log2-fold change and evaluated with the “GSEA” algorithm.95 “Hallmark” and “KEGG” gene sets were acquired from MSigDb (https://www.gsea-msigdb.org/gsea/msigdb).74 We filtered GSEA results based on the criterion of p value <0.05 and visualized the candidate pathways based on the normalized enrichment score. Ferroptosis associated genes Ferroptosis negative regulation genes and drivers were acquired from FerrDb V2 (http://www.zhounan.org/ferrdb/current/). Immune subtype identification and downstream analysis The abundances of 64 different cell types in PSC tumors and NAT samples were computed via xCell (version 1.1.0)40 (https://xcell.ucsf.edu/) using the RNA-seq data. Consensus clustering was performed using the R package “ConsensusClusterPlus” (version 1.64.0).84 Specifically, 80% of the original samples were randomly subsampled without replacement and partitioned into 3 major clusters using the K-Means algorithm, which was repeated 200 times. Furthermore, ESTIMATE algorithm (version 1.0.13)85 was used to derive the overall immune score and stromal score, and tumor purity for each sample based on RNA-seq data. Immune repertoire and Tumor Immune Dysfunction and Exclusion score estimation We used a computational algorithm named TRUST4 to evaluate the immune repertoire and to extract T and B cell receptors (TCRs and BCRs, respectively) complementarity-determining region 3 (CDR3) sequences. Then we used the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm41 to estimate a tide score and the predicted response to immune checkpoint blockade (ICB). Unsupervised clustering using NMF using proteomic and phosphoproteomic data We conducted NMF-based clustering using proteomic data and phosphoproteomic data in PSC tumors, separately. Briefly, given a factorization rank k (where k is the number of clusters), NMF decomposes a p × n data matrix V into two matrices W and H such that multiplication of W and H approximates V. Matrix H is a k × n matrix whose entries represent weights for each sample (1 to N) to contribute to each cluster (1 to k), whereas matrix W is a p × k matrix representing weights for each feature (1 to p) to contribute to each cluster (1 to k). Matrix H was used to assign samples to clusters by choosing the k with maximum score in each column of H. For each sample, we calculated a cluster membership score as the maximal fractional score of the corresponding column in matrix H. To remove uninformative features from the dataset before NMF clustering we kept proteins with the top standard deviation (top 20th percentile) across all tumor samples. As for clustering using phosphoproteins, we kept feature with top standard deviation (top 95th percentile) across all tumor samples. Each column in the data matrix was further scaled and standardized such that all features from different data types were represented as z-scores. The resulting data matrix of z-scores was converted to a non-negative input matrix required by NMF as follows. 1.Create one data matrix with all negative numbers zeroed. Create one data matrix with all negative numbers zeroed. 2.Create another data matrix with all positive numbers zeroed and the signs of all negative numbers removed. Create another data matrix with all positive numbers zeroed and the signs of all negative numbers removed. 3.Concatenate both matrices resulting in a data matrix twice as large as the original, but containing only positive values and zeros and hence appropriate for NMF. Concatenate both matrices resulting in a data matrix twice as large as the original, but containing only positive values and zeros and hence appropriate for NMF. The resulting matrix was then subjected to NMF analysis leveraging the NMF R-package (version 0.26)86 and using the factorization method described in Brunet’s research.96 To determine the optimal factorization rank k (number of clusters) we calculated cophenetic correlation coefficient measuring how well the intrinsic structure of the data was recapitulated after clustering across 50 iterations. The optimal k was defined as the maximum of cophenetic score for cluster numbers between k = 2 and 8. Having determined the optimal factorization rank k, in order to achieve robust factorization of the proteomic data matrix V, we repeated the NMF analysis using 200 iterations with random initializations of W and H and performed the partitioning of samples into clusters as described above. EMscore derivation We acquired the RNAseq data of microdissected epithelial and sarcomatoid components in PSC,12 and performed differentially expressed gene analysis. EMscore was calculated by median expression of sarcomatoid genes subtracts median expression of epithelial genes. TGF-β signaling activity TGF-β signaling activity was generated using PROGENy97 based on proteome expression data. CMAP analysis Differentially expressed genes among the PSC proteomic subtypes were identified using the Wilcoxon rank-sum test. The identified differentially expressed genes and proteins were then filtered for gene symbols measured in the L1000 assay (978 landmark genes). These NMF-specific signatures were used as input to calculate connectivity scores against the LINCS L1000 perturbation-response signatures using CLUE (https://clue.io). The topotecan response signatures was acquired from iLINCS (http://www.ilincs.org/ilincs/). Single-cell RNA sequencing analysis of PSC with ICB Sample collection Sample collection informed consent was obtained for each patient. We obtained the primary tumor tissue by percutaneous pulmonary biopsy, bronchoscopy biopsy, or endobronchial ultrasound (EBUS) biopsy before drug administration. After the last cycle of neoadjuvant therapy, the fresh tumor tissues were collected immediately by surgical resection. We collected 4 tumor samples (2 baseline and 2 post-treatment) from 3 patients for scRNA-seq. The clinicopathological parameters were collected and summarized in Table S6. Single-cell RNA library preparation and sequencing A 40-micron sterile strainer (Corning) was used to separate single cells after digestion. The cells were then centrifuged at 300 × g 4°C for 5 min and cell pellets were resuspended in 1mL PBS (HyClone). Cell suspensions were counted with TC20 automated cell counter (Bio-Rad) to determine cell concentration and viability. The concentration of single-cell suspension was adjusted to 1 × 105 cells/mL in PBS. Single-cell suspension was then loaded onto a microfluidic chip (GEXSCOPE Single Cell RNA-seq Kit, Singleron Biotechnologies) and scRNA-seq libraries were constructed according to the manufacturer’s instructions (Singleron Biotechnologies). The resulting scRNA-seq libraries were sequenced on an Illumina HiSeq ×10 instrument with 150 bp paired end reads. scRNA-seq data processing The matrix of read counts per sample was further analyzed by the Seurat package (version 4.2.0)89 in R software (version 4.3.1). For each cell, we used four quality control measures. Cells falling into any of the following four categories were excluded: (1) nFeature_RNA <300 or >6 000, (2) > 20% UMIs of mitochondria genes, (3) nCount_RNA <300 or >50 000, and (4) housekeeping score (defined as the sum of the UMIs of three canonical housekeeping genes: ACTB, GAPDH and MALAT1) < 1. To exclude data from droplets containing more than one cell, doublet detection and removal were performed using DoubletFinder (version 2.0.3).98 DecontX, a Bayesian method, was used to estimate and remove contamination in individual cells.99 An expected doublet rate parameter of 0.025 was used, and doublet score thresholds were chosen manually to divide putative singlet and neotypic doublet modes in the score distribution. Predicted doublets were then removed from gene-by-barcode matrices. Subsequently, we performed data normalization, identification of highly variable genes, and PCA with Seurat’s classic workflow. Then, Harmony algorithm was used to correct the potential batches among sample resources. Uniform Manifold Approximation and Projection (UMAP) was used for visualization. The Seurat “FindNeighbors” and “FindClusters” functions were applied to detect communities and find cell clusters. And then the “FindAllMarkers” function with default settings was used to find markers for each identified cluster. To annotate the identity and lineage of each cell type, the differentially expressed genes from each cluster were analyzed to a previously reported set of cell-type markers. To identify malignant cells from non-immune cells, inferCNV was used to estimate the copy number variations100 with Macrophage, Monocyte, Mast cell and DC as normal reference, and the parameters were default. To estimate the signatures of gene module and specific pathways, we calculated the enrichment scores using the AddModuleScore function in Seurat. The cell-cell communications and signaling pathways were analyzed using “CellChat” R package (version 1.1.3). Survival analysis Overall survival was measured from the date of surgery to the date of last follow-up or death. We used Kaplan-Meier analysis to explore survival differences associated with protein expression, TMB status, SCNA status. For continuous variables, including mRNA, protein and phosphosites expression, maximally selected rank statistics (maxstat) implemented in survimer R package (version 0.4.9) was performed to determine the optimal cutpoint for selected samples. For the prognostic analysis in TMB, SRPK1, ferroptosis drivers and FNR signatures, tumor samples were stratified based on the median values. Log rank tests and Cox proportional hazards regression for statistical significance were implemented via the survminer (version 0.4.9) and survival (version 3.5–5) packages in R software. Functional experiments Cell culture and reagents The PLA-801D cell line (human pulmonary giant cell carcinoma cell, male) was obtained from the Institute of Biochemistry and Cell Biology at the Chinese Academy of Science and authenticated through Short Tandem Repeats Authentication. All cells were maintained at 37°C with 5% CO2 in DMEM (Gibco) medium supplemented with 10% fetal bovine serum (FBS, Royacel) and penicillin (100 units/mL)/streptomycin (0.1 mg/mL) (Gibco). Cells were cultured in 10 cm dishes or plated in 6 or 96 well plates for the indicated experiments. The chemical compounds used in cell culture experiments included Ferrostatin-1 (Selleck, S7243). The PLA-801D cell line was routinely confirmed as pathogen-free and mycoplasma-free by PCR. Generation of knockdown cell lines The shRNA sequences were cloned into pLKO.1 plasmid (Addgene) as the manufacturer’s instructions described. The shRNA sequences utilized were as follows: human shControl, 5′-ATGGTAGCGTACACTTATGAT-3′; human shSLC3A2#1, 5′-CTAGCTCATACCTGTCTGATT-3′; human shSLC3A2#2, 5′-GCTGGGTCCAATTCACAAGAA-3′; human shFTL#1, 5′-GCTCACTCTCAAGCACGACTA-3′; human shFTL#2, 5′-CTGGAGACTCACTTCCTAGAT-3′; The lentiviral plasmid was co-transfected into 293T cells with packaging plasmids, pMD2.G and psPAX2 (Addgene), using Lipofectamine™ 3000 Transfection Reagent (Invitrogen, L3000001). Lentiviral particles were collected 48 h after transfection and filtered with a 0.45μm filter. After infect the cells with lentivirus for 24 h, 2 μg/mL puromycin (Gibco, A1113803) were used for stable cell line selection. Immunoblotting assay Cells were lysed with 1% Triton X-100 buffer (Beyotime, ST1723) and then boiled in 5× loading buffer for 10 min. Each sample was electrophoresed on 10% PAGE gels, then transferred to PVDF membranes and immunoblotted with the antibodies targeting SLC3A2 (Abclonal, A24375), FTL (Abclonal, A11241), GPX4 (Abclonal, A1933) and β-Actin (Abclonal, AC004). Immunoblots were visualized by a chemiluminescence imaging system (Tanon). Cell proliferation Cell viability and proliferation were measured by using cell counting kit-8 (CCK-8) (Biomake). PLA-801D cells were seeded in 96-well plates at 1×103 per well for cell proliferation measurement following SLC3A2 and FTL knockdown and 2×103 per well for cell viability measurement after Ferrostatin-1 treatment as the experiments designed. Then, 10 μL CCK-8 reagent was added to each well at the time of harvest. The cells were incubated at 37°C for 1.5 h, and absorbance at 450 nm was measured to determine cell viability or proliferation using a microplate reader (Varioskan Flash, Thermo Fisher Scientific). Clone formation PLA-801D cells were plated in 6-well plates at 800 per well. The medium was changed every 3 days. After 14 days of culture, the cells in the plates were fixed with methanol and stained with crystal violet. Pictures of the stained cells were taken with the mobile phone camera.The colony numbers of each well were calculated through ImageJ software. Iron staining and detection 5×105 PLA-801D cells were seeded into 6-well plates. After cell attachment, cells were stained with FerroOrange (Dojingo, F374) in HBSS for exactly 30 min at 37°C and 5% CO2. Then, the plates were observed and pictured by fluorescence microscope (Revolve Generation 2, ECHO). For detection of Fe2+, 2×104 PLA-801D cells were seeded into 96-well plates and then stained with FerroOrange in HBSS. The cells were measured after 30 min using a microplate reader at Ex/Em = 543nm/580nm. Detection of glutathione The cell culture supernatants were also collected to measure extracellular glutathione. GSH and GSSG Assay Kit (Beyotime, S0053) was used as the manufacturer’s instructions described. Add three times the volume of the cell pellet with Protein Removal Reagent solution. Subsequently, subject the sample to two rapid freeze-thaw cycles using liquid nitrogen and a 37°C water bath. Incubate at 4°C or on ice for 5 min. Centrifuge at 4°C, 10,000g for 10 min. The supernatant is collected for total glutathione assay. Add diluted GSH clearance auxiliary solution for GSSG measurement, then vortex immediately to mix thoroughly. Subsequently, add GSH clearance working reagent, vortex again immediately to mix, and incubate at 25°C for 60 min. After adding 150 μL of total glutathione detection working solution into 96-well plates, mix thoroughly and incubate at room temperature for 5 min. Afterward, the plates were measured at 405 nm absorbance at 25°C, and standard curves were plotted to calculated the concentration of GSH. Detection of reactive oxygen species and lipid peroxidation MDA PLA-801D cells were first loaded with fluorescent probes in Reactive Oxygen Species Assay Kit (Beyotime, S0033). After 20 min for incubation, the ROS level of cells was measured by a microplate reader at Ex/Em = 488nm/525nm. For MDA detection, PLA-801D cells were lysed in NP40 lysis buffer (Beyotime, P0013F) for 30 min at 4°C. Next, supernatants were collected after 12000×g, 10 min centrifugation and determined for the concentration of protein through Enhanced BCA Protein Assay Kit (Beyotime, P0010). Afterward, the supernatants remained were reacted as Lipid Peroxidation MDA Assay Kit (Beyotime, S0131) instructed. Add 0.1mL of lysis buffer as an appropriate blank control to a centrifuge tube. Add 0.1 mL of the aforementioned standard solutions at different concentrations to prepare the standard curve. Add 0.1 mL of the sample for analysis. Subsequently, add 0.2 mL of the MDA detection working solution. After thorough mixing, heat at 100°C or in a boiling water bath for 15 min. Allow the water bath to cool to room temperature, then centrifuge at 1000g for 10 min at room temperature. Transfer 200μL of the supernatant to a 96-well plate, then measure the absorbance at 532 nm by a microplate reader. Detection of C11 BODIPY PLA-801D cells were collected from 12-well plates and washed three times with PBS. The cells were then resuspended in 2 μM C11 BODIPY 581/591 (TargetMol, T39652) solution and incubated for 30 min. After staining, the cells were washed three times with PBS and resuspended in FACS buffer (2% FBS in PBS). Flow cytometry was performed on a BD FACSVerse system (BD Biosciences), and data were analyzed using FlowJo software (version 10.9.0). RNA sequencing for control and shFTL PLA-801D cells RNA-seq cleaned sequence data for control and shFTL PLA-801D cells were aligned to human reference sequence hg38 using STAR (version 2.7.6a).81 The resulting BAM files were analyzed for gene expression using Salmon (version 1.3.0)82 against a transcriptomic reference GENCODE (version 22, https://www.gencodegenes.org). Then the resulting gene-level counts for protein-coding genes were normalized and transformed into TPM values, followed by applying log2 transformation. Data rows of redundant gene symbols were aggregated by calculating the average expression values. Quantification and statistical analysis Quantification methods and statistical analysis methods for single-omics and multi-omics analyses were mainly described and referenced in the respective Method Details subsections. Additionally, standard statistical tests were used to analyze the data, including but not limited to Wilcoxon test, One-way ANOVA following Tukey’s test, Fisher’s exact test, Log rank test, and Cox proportional hazards regression model. For categorical variables versus categorical variables, Fisher’s exact test or Chi-square test was used; for categorical variables versus continuous variables, Student’s t test, Wilcoxon test or One-way ANOVA following Tukey’s test were used; and for continuous variables versus continuous variables, Pearson or Spearman’s correlation was used. All statistical tests were two-sided, and statistical significance was considered when p value <0.05 unless otherwise indicated. Multiple comparisons were adjusted by the Benjamini-Hochberg procedure. Kaplan-Meier plots (Log rank test) were used to describe overall survival. Variables associated with overall survival were identified using univariate Cox proportional hazards regression models. Statistical analysis was performed using GraphPad Prism. Student’s t test or Wilcoxon-rank test was used to compare the statistical differences between the experimental groups. For functional experiments, each was repeated at least three times independently, and the results were shown as mean ± standard error of the mean (SEM). The statistical significance of differences was determined by two-way ANOVA or Student’s t test. Details can be found in Results and Figure Legends. Published: June 26, 2026 Footnotes Supplementary data related to this article can be found online at https://doi.org/10.1016/j.xcrm.2026.102887. Contributor Information Ming Ding, Email: mingding@cpu.edu.cn. Peng Zhang, Email: zhangpeng1121@tongji.edu.cn. Lele Zhang, Email: zhanglele@tongji.edu.cn. Haiyang Hu, Email: haiyanghu@cpu.edu.cn. Supplemental information Document S1. Figures S1–S9 Table S1. Clinical Information and multi-omics data in the TU-PSC cohort, related to Figures 1 and S1 Table S2. Effects of genomic alterations in PSC, related to Figure 2 Table S3. Tumor-NAT comparisons reveal tumorigenic changes in PSC, related to Figures 3 and 4 Table S4. Immune landscape of PSC, related to Figures 5 and S4 Table S5. NMF proteomic subtypes with distinct biological features, related to Figures 6 and S5–S7 Table S6. scRNA-seq profiles for PSC with ICB immunotherapy, related to Figures 7, S8, and S9 Document S2. Article plus supplemental information References Associated Data Supplementary Materials Document S1. Figures S1–S9 Table S1. Clinical Information and multi-omics data in the TU-PSC cohort, related to Figures 1 and S1 Table S2. Effects of genomic alterations in PSC, related to Figure 2 Table S3. Tumor-NAT comparisons reveal tumorigenic changes in PSC, related to Figures 3 and 4 Table S4. Immune landscape of PSC, related to Figures 5 and S4 Table S5. NMF proteomic subtypes with distinct biological features, related to Figures 6 and S5–S7 Table S6. scRNA-seq profiles for PSC with ICB immunotherapy, related to Figures 7, S8, and S9 Document S2. Article plus supplemental information Data Availability Statement •The WES and RNA-seq raw data have been deposited in the Genome Sequence Archive in the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA008505 that are publicly accessible at http://bigd.big.ac.cn/gsa-human. The proteomic data have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix, OMIX007422). QC and enrichment metrics data have been deposited in Zenodo (https://zenodo.org/records/17595641). This paper does not report original code. The exact analysis scripts for the two-component normalization and KSEA have been deposited in Zenodo (https://zenodo.org/records/17696161). Sample annotation, processed, and normalized data files are provided in Table S1. The software and code used in this study are referenced in their corresponding STAR Methods sections and in the key resources table. The WES and RNA-seq raw data have been deposited in the Genome Sequence Archive in the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession number HRA008505 that are publicly accessible at http://bigd.big.ac.cn/gsa-human. The proteomic data have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix, OMIX007422). QC and enrichment metrics data have been deposited in Zenodo (https://zenodo.org/records/17595641). This paper does not report original code. The exact analysis scripts for the two-component normalization and KSEA have been deposited in Zenodo (https://zenodo.org/records/17696161). Sample annotation, processed, and normalized data files are provided in Table S1. The software and code used in this study are referenced in their corresponding STAR Methods sections and in the key resources table. •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact, Haiyang Hu (haiyanghu@cpu.edu.cn), upon request. Any additional information required to reanalyze the data reported in this work paper is available from the lead contact, Haiyang Hu (haiyanghu@cpu.edu.cn), upon request.