Abstract

   Oxaliplatin, a key chemotherapeutic agent, often induces resistance in
   colorectal cancer (CRC) treatment, highlighting the urgent need for
   reliable biomarkers to predict treatment efficacy. In this study, we
   aimed to identify key genes associated with oxaliplatin resistance in
   CRC and to evaluate their potential as prognostic biomarkers. Using CRC
   patient data from the TCGA dataset, we categorized patients into
   oxaliplatin-resistant and -sensitive groups and conducted differential
   expression analysis. Key feature genes were identified through
   univariate Cox analysis, LASSO regression, and stepwise multivariate
   Cox regression. The predictive value of the identified markers was
   validated using logistic regression, weighted gene co-expression
   network analysis (WGCNA), and external validation in GEO cohorts. The
   tumor microenvironment (TME) was assessed using the MCP-counter
   algorithm, and CRC cell experiments were performed to evaluate changes
   in drug sensitivity following oxaliplatin exposure. Based on TCGA CRC
   data, we constructed a prognostic index derived from a three-gene
   signature associated with oxaliplatin resistance. This index was
   significantly correlated with progression-free survival (PFS) in
   oxaliplatin-resistant CRC patients and showed robust prognostic
   performance, with AUCs of 0.848 and 0.861 in gastric cancer and
   pancreatic adenocarcinoma cohorts, respectively. Notably, TNFAIP2
   knockout significantly reduced clonogenic ability in CRC cells
   following oxaliplatin treatment. Our results identify TLE4, TNFAIP2,
   and ARGLU1 as key contributors to oxaliplatin resistance in CRC. The
   oxaliplatin resistance–related gene signature (ORGSig) serves as a
   promising tool for predicting treatment response and prognosis in CRC
   patients receiving oxaliplatin-based chemotherapy. This signature also
   offers potential for guiding personalized therapy and overcoming drug
   resistance in clinical practice.

Supplementary Information

   The online version contains supplementary material available at
   10.1038/s41598-025-05556-2.

   Keywords: Chemoresistance, Oxaliplatin, CRC, Prognostic marker, TNFAIP2

   Subject terms: Cancer, Cell biology, Gastroenterology

Introduction

   Colorectal cancer (CRC) is among the most prevalent malignancies of the
   digestive system, with over one million new cases diagnosed
   annually^[32]1. Current treatment strategies include surgical
   resection, radiotherapy, chemotherapy, targeted therapy, and
   immunotherapy^[33]2–[34]4. Oxaliplatin, a key chemotherapeutic agent,
   is widely used in combination with 5-fluorouracil and folinic acid for
   CRC treatment^[35]5. Although oxaliplatin has improved clinical
   outcomes, approximately 40–50% of patients with stage II or III CRC
   develop resistance, resulting in poor prognosis^[36]6. Therefore,
   identifying biomarkers associated with oxaliplatin resistance and
   prognosis is crucial for addressing CRC’s molecular heterogeneity and
   improving therapeutic outcomes.

   Oxaliplatin, a third-generation platinum-based chemotherapeutic agent,
   inhibits tumor progression by disrupting DNA synthesis and impairing
   cell division in cancer cells^[37]7–[38]9. Its role in CRC treatment
   was first established in 1996^[39]10. Although oxaliplatin can be used
   as monotherapy, it is commonly administered in combination with other
   agents to enhance efficacy. In advanced and metastatic CRC, it is
   typically combined with 5-fluorouracil and folinic acid. Beyond CRC,
   oxaliplatin also exhibits antitumor activity in breast, bladder,
   prostate, brain, and lung cancers, as well as lymphomas^[40]11–[41]15.
   Despite its clinical benefits, resistance frequently emerges,
   especially in aggressive tumors. Mechanisms contributing to
   chemoresistance include enhanced DNA repair, inhibition of apoptosis,
   altered drug transport, detoxification, and epigenetic
   modifications^[42]16. Recent studies have shown that UPF1 upregulation
   promotes oxaliplatin resistance by regulating TOP2A activity and
   maintaining cancer stemness in CRC^[43]17. Additionally, long
   non-coding RNAs (lncRNAs) and feedback loops contribute significantly
   to the regulation of chemotherapeutic responses in tumor cells^[44]18.
   These findings underscore the complexity of oxaliplatin resistance
   mechanisms. However, the molecular heterogeneity of CRC hinders a clear
   understanding of these mechanisms, highlighting the need to further
   explore the genes involved. Developing a gene panel based on
   oxaliplatin resistance–related genes is therefore essential for
   enhancing CRC treatment and prognosis.

   In this study, we focused on three key genes—TLE4, TNFAIP2, and
   ARGLU1—which were significantly upregulated in oxaliplatin-resistant
   CRC samples. These genes were selected for their established roles in
   cell proliferation, tumor immunity, and chemoresistance. TLE4 is
   associated with tumor progression and poor prognosis in CRC; TNFAIP2 is
   involved in inflammatory signaling and drug resistance; and ARGLU1
   plays a central role in DNA damage repair across multiple cancer types.
   Together, these genes form a predictive panel that may help stratify
   CRC patients based on their susceptibility to oxaliplatin resistance.

   Our objective was to improve personalized prognosis prediction and
   treatment selection for CRC patients receiving oxaliplatin-based
   chemotherapy. To this end, we developed an oxaliplatin
   resistance–related gene signature (ORGSig) that demonstrates strong
   predictive power for both prognosis and treatment response. We also
   experimentally validated the functional role of TNFAIP2 in CRC cells,
   providing further insight into the mechanisms driving tumor progression
   and chemoresistance in CRC.

Results

DEGs between oxaliplatin-resistant and oxaliplatin-sensitive patients in TCGA
dataset

   To identify key genes associated with oxaliplatin resistance in CRC, we
   selected patients from the TCGA-CRC cohort who had received
   oxaliplatin-based chemotherapy. Among the 64 patients with complete
   chemotherapy data, 50 were classified as sensitive and 14 as resistant.
   Compared to the sensitive group, the resistant group showed
   upregulation of 397 genes and downregulation of 69 genes (Fig. [45]1A,
   B; Table [46]S1). To explore the biological functions of these
   differentially expressed genes (DEGs), we performed Gene Ontology (GO)
   and pathway enrichment analyses using Metascape. The upregulated genes
   were significantly enriched in: (1) immune response and cell activation
   pathways (e.g., inflammatory response, leukocyte migration, cell
   activation, organismal homeostasis); (2) biological regulation
   processes (e.g., epithelial cell proliferation, population-level
   proliferation, phosphorus metabolism regulation, cell adhesion); and
   (3) signaling and interaction pathways (e.g., integrin-mediated cell
   surface interactions) (Fig. [47]1C). These 397 upregulated genes were
   considered candidate components for constructing the ORGSig.

Fig. 1.

   [48]Fig. 1
   [49]Open in a new tab

   DEGs between oxaliplatin-resistant and oxaliplatin-sensitive patients
   in TCGA dataset. (A, B) Volcano plot and heatmap illustrating the DEGs
   between the chemosensitive group and chemoresistant group in the TCGA
   cohort. (C) A total of 397 upregulated genes were subjected to pathway
   enrichment analysis using the Metascape platform.

Construction of ORGSig for CRC

   To identify key genes with the greatest prognostic impact in CRC
   patients treated with oxaliplatin, we performed univariate Cox
   regression analysis on the 397 upregulated genes. This analysis
   identified 32 genes significantly associated with poor prognosis (Fig.
   [50]2A). LASSO regression further narrowed the candidates to four
   genes: TLE4, TNFAIP2, CRABP2, and ARGLU1 (Fig. [51]2B, C). Subsequent
   stepwise multivariate Cox regression analysis refined the selection to
   three critical genes—TLE4, TNFAIP2, and ARGLU1—for constructing the
   prognostic model.

Fig. 2.

   [52]Fig. 2
   [53]Open in a new tab

   Construction of ORGSig for CRC. (A) 32 ORGSig showing an adverse
   prognosis correlation in the TCGA cohort. (B) The tuning parameter (λ)
   for LASSO regression was determined via 10-fold cross-validation. (C)
   Changes in LASSO coefficients and alternative ORGSig across different
   values of the tuning parameter (λ). (D) ROC curves demonstrating the
   performance of the three-gene panel in CRC samples from the training
   set.

Verification of ORGSig for CRC

   To evaluate the predictive performance of the ORGSig model for
   chemotherapy response, we conducted ROC curve analysis on 114 CRC
   patients with available chemotherapy records, yielding an AUC of 0.788
   (Fig. [54]2D). To further validate the robustness of ORGSig, we tested
   the model in both the TCGA training cohort and an external validation
   cohort ([55]GSE87211). In the TCGA cohort, stratification based on risk
   scores demonstrated significantly improved progression-free survival
   (PFS) in the low-risk group compared to the high-risk group, thereby
   confirming the prognostic utility of the model (Fig. [56]3A). ROC
   analysis showed AUCs of 0.791, 0.767, and 0.937 for 1-, 3-, and 5-year
   PFS predictions, respectively (Fig. [57]3B), indicating strong temporal
   predictive accuracy. For external validation, we selected
   oxaliplatin-treated samples from [58]GSE87211. Patients in the
   high-risk group exhibited significantly poorer survival (P < 0.05; Fig.
   [59]3C). The corresponding ROC analysis yielded AUCs of 0.669, 0.579,
   and 0.666 for 1-, 3-, and 7-year survival predictions, respectively
   (Fig. [60]3D). Collectively, these results support the prognostic
   reliability and predictive utility of ORGSig in oxaliplatin-treated CRC
   patients.

Fig. 3.

   [61]Fig. 3
   [62]Open in a new tab

   Verification of ORGSig for CRC. (A) The Kaplan–Meier survival curves
   showing notable variations in PFS based on different risk scores in the
   TCGA cohort. (B) Time-dependent ROC curve analysis for 1-, 3-, and
   7-year survival in the TCGA cohort. (C) The Kaplan–Meier survival
   curves showing notable variations in PFS based on different risk scores
   in the GEO cohort. (D) Time-dependent ROC curve analysis for 1-, 3-,
   and 7-year survival in the GEO cohort.

The application of the prognostic nomogram

   To determine whether ORGSig serves as an independent prognostic factor,
   we performed univariate and multivariate Cox regression analyses. In
   the univariate analysis, M stage, clinical stage, and risk score were
   significantly associated with prognosis (Fig. [63]4A). Multivariate
   analysis further confirmed that the risk score remained an independent
   predictor, with a hazard ratio (HR) of 1.460 (95% CI: 1.240–1.700, P <
   0.001) (Fig. [64]4B). To enhance the clinical utility of ORGSig, we
   constructed a nomogram integrating the risk score with key
   clinicopathological variables to predict 1-, 2-, and 3-year PFS (Fig.
   [65]4C). As expected, higher risk scores were associated with poorer
   outcomes. Calibration plots showed strong concordance between predicted
   and observed survival probabilities, indicating good model reliability
   (Fig. [66]4D). In addition, ROC curve analyses demonstrated that the
   risk score outperformed traditional clinical parameters (T, N, and M
   stages), with AUC values of 0.763, 0.713, and 0.741 at different time
   points (Fig. [67]4E). Collectively, these findings support ORGSig as a
   robust and clinically valuable prognostic tool for CRC patients.

Fig. 4.

   [68]Fig. 4
   [69]Open in a new tab

   The application of the prognostic nomogram. (A, B) Univariate and
   multivariate Cox regression analyses evaluating association between
   clinicopathological factors and ORGSig with PFS.(C) Nomogram for
   predicting PFS. (D) The calibration curves for 1-,3- and 5- year PFS.
   (E) The ROC curve of risk score and clinical characteristics was
   performed based on 1-, 3-, and 5-year PFS.

Three genes as the best combination in predicting tumor progression

   To further validate the reliability of ORGSig, we performed
   multivariable logistic regression to assess its ability to predict
   tumor progression after chemoresistance. This analysis provided deeper
   insight into the association between the gene panel and disease
   progression. We constructed multiple regression models using individual
   genes and combined gene sets to compare their AUC values (Fig. [70]5A).
   Among all combinations, the model incorporating all three core
   biomarkers achieved the highest AUC of 0.797, indicating superior
   predictive performance (Fig. [71]5B). While the AUCs of other
   combinations were slightly lower, they still demonstrated considerable
   predictive potential (Fig. [72]5C). These findings support the
   hypothesis that the three key genes contribute to chemoresistance and
   are associated with differences in PFS in CRC. To evaluate the
   generalizability of ORGSig, we extended the analysis to other
   oxaliplatin-treated cancers in the TCGA database, including stomach
   adenocarcinoma (STAD, n = 22) and pancreatic adenocarcinoma (PAAD, n =
   20). Multivariable logistic regression was applied to each cancer type,
   and AUC values were calculated for different gene combinations. In both
   STAD and PAAD cohorts, the three-gene model showed the highest
   predictive accuracy, with AUCs of 0.848 and 0.861, respectively (Fig.
   [73]5D–I).

Fig. 5.

   [74]Fig. 5
   [75]Open in a new tab

   Three genes as the best combination in predicting tumor progression.
   (A, D, G) Evaluation of all gene combinations and their corresponding
   AUC values calculated by a multivariate logistic regression model in
   patients with COREAD, STAD, and PAAD. (B, C, E, F, H, I) ROC curves
   depicting the performance of the top two combinations in predicting
   clinical outcomes for patients with COREAD, STAD, and PAAD.

Applying WCGNA to ORGSig and clinical characteristics

   To further validate the reliability of ORGSig in predicting tumor
   progression after CRC chemotherapy, we applied Weighted Gene
   Co-expression Network Analysis (WGCNA). This approach enabled us to
   investigate associations among ORGSig, PFS, risk score, and gene
   modules with co-expression patterns similar to those of the key genes.
   We selected a soft-thresholding power of β = 8 to construct a
   scale-free network (Fig. [76]6A). By merging modules with similar
   expression profiles using dynamic tree cutting, we obtained a
   consolidated network comprising seven modules (Fig. [77]6B). The
   relationships between each module and clinical outcomes, including PFS
   and risk score, are shown in Fig. [78]6C. Notably, PFS showed a
   negative correlation with risk score, highlighting the model’s
   effectiveness in predicting chemoresistance and tumor recurrence. The
   three core genes were located in distinct modules: TLE4 in the brown
   module (risk score correlation: 0.5; PFS correlation: − 0.02), TNFAIP2
   in the yellow module (risk score: 0.43; PFS: − 0.04), and ARGLU1 in the
   turquoise module (risk score: 0.36; PFS: 0.01). Gene enrichment
   analysis further revealed that these modules were enriched in pathways
   related to post-chemotherapy cell proliferation, metabolism, and immune
   responses (Fig. [79]6D–F).

Fig. 6.

   [80]Fig. 6
   [81]Open in a new tab

   Applying WCGNA to ORGSig and clinical characteristics. (A)
   Determination of the soft threshold for WGCNA (power = 8) based on
   scale independence and mean connectivity. (B) Construction of a cluster
   tree using all genes from CRC patients treated with oxaliplatin, with
   similar genes grouped into color-coded modules, and merged dynamically.
   (C) Heatmap matrix displaying the relationship between gene modules and
   clinical traits. (D-F) Dot chart illustrating pathways enriched through
   GO analysis. TLE4, TNFAIP2 and ARGLU1 were classified into brown,
   yellow and turquoise modules respectively.

Associations of ORGSig model with immune cell infiltration

   Based on the enrichment analysis of DEGs, which revealed activation of
   immune-related pathways, we further explored immunological differences
   between high- and low-risk score groups. Using the ESTIMATE algorithm,
   we calculated immune, stromal, and ESTIMATE scores in the TCGA cohort.
   The high-risk group exhibited significantly higher immune infiltration
   and stromal content, as reflected by elevated immune, stromal, and
   ESTIMATE scores (Fig. [82]7A). MCP-counter analysis revealed increased
   infiltration of multiple immune cell types—including cytotoxic T cells,
   NK cells, B cells, monocytes, myeloid dendritic cells, endothelial
   cells, and cancer-associated fibroblasts—in the high-risk group (Fig.
   [83]7B). To investigate potential immunotherapeutic targets, we
   examined immune checkpoint gene expression and found that HHLA2 was
   more highly expressed in the low-risk group, whereas TNFRSF9, TIGIT,
   CD28, CD80, HAVCR2, CD200R1, BTLA, CD276, NRP1, CTCN1, ICOS, IDO1,
   CTLA4, LAG3, and TNFRSF8 were significantly upregulated in the
   high-risk group (Fig. [84]7C). Additionally, we assessed treatment
   sensitivity and observed that high-risk patients showed better
   responses to anti-CTLA4 therapy and chemotherapy (P < 0.05, Fig.
   [85]7D). Collectively, these findings suggest that the high-risk score
   group exhibits enhanced immune activity and may benefit from
   immunotherapy, providing important insights for personalized treatment
   strategies.

Fig. 7.

   [86]Fig. 7
   [87]Open in a new tab

   Associations of ORGSig model with immune cell infiltration. (A)
   Histogram depicting immune cell infiltration analysis.(B) Comparison of
   immune cell proportions based on MCP counter between the low-risk and
   high-risk groups in the TCGA cohort. (C) Expression distribution of
   common immune checkpoints in high- and low-risk score patients from the
   TCGA cohort. (D) Sensitivity analysis for anti-PD-1 and anti-CTLA4
   treatments in high- and low-risk groups.

TNFAIP2 confers OXA resistance in CRC cells in vitro

   The role of TNFAIP2 in CRC initiation, progression, and chemoresistance
   remains incompletely understood. In this study, we focused on its
   potential involvement in oxaliplatin resistance. Data from the Human
   Protein Atlas revealed elevated TNFAIP2 expression in CRC tissues (Fig.
   [88]8A). Immunohistochemistry (IHC) analysis further confirmed
   significantly higher TNFAIP2 levels in the oxaliplatin-resistant group
   compared to the sensitive group (Fig. [89]8B, C). To evaluate the
   prognostic significance of TNFAIP2 in CRC, we utilized the Kaplan–Meier
   Plotter online tool. High TNFAIP2 expression was significantly
   associated with poor overall survival (OS) (P = 0.01, Fig. [90]8D).
   Subgroup analysis indicated a trend toward reduced survival in patients
   undergoing surgery alone (P = 0.14, Fig. [91]8E) and a significant
   association with poor prognosis in patients receiving postoperative
   chemotherapy (P = 0.019, Fig. [92]8F). To validate the functional role
   of TNFAIP2 in oxaliplatin resistance, we established TNFAIP2 knockdown
   SW480 cell lines using shRNA, with silencing efficiency confirmed by
   qRT-PCR (Fig. [93]8G). Colony formation assays showed a significant
   reduction in clonogenic potential following TNFAIP2 silencing under
   baseline conditions. Upon oxaliplatin treatment, knockdown cells
   exhibited a near-complete loss of colony-forming ability (Fig. [94]8H,
   I). Additionally, TNFAIP2 knockdown markedly enhanced
   oxaliplatin-induced apoptosis compared to control cells (Fig. [95]8J,
   K). Collectively, these results suggest that TNFAIP2 plays a key role
   in mediating oxaliplatin resistance in CRC and may serve as a promising
   therapeutic target to overcome chemoresistance.

Fig. 8.

   [96]Fig. 8
   [97]Open in a new tab

   TNFAIP2 confers OXA resistance in CRC cells in vitro. (A)
   Immunohistochemical data of TNFAIP2 in HPA database. Scale bar = 100
   μm. (B, C) Representative IHC images of TNFAIP2 in 8 CRC tissues with
   different groups. Scale bar = 100 μm. (D, E, F) Kaplan–Meier survival
   curves of CRC patients with TNFAIP2 expression in the Kaplan–Meier
   Plotter database. (G) qRT–PCR was employed to validate the knockdown
   efficiency in various cell lines. (H, I) Analysis of the clonogenic
   assay. (J, K) Annexin V-FITC and PI staining showing apoptosis in the
   indicated CRC cells treated with OXA (30 μM) for 48 h.

Discussion

   Oxaliplatin is widely used in the treatment of CRC, but resistance to
   this agent remains a significant clinical challenge^[98]19. Tumor cells
   that acquire resistance can evade oxaliplatin-induced cytotoxicity and
   contribute to tumor recurrence, particularly in relapsed and metastatic
   settings^[99]20–[100]22. Previous studies have demonstrated a strong
   association between dysregulated gene expression and chemotherapy
   resistance^[101]23. Consequently, there is increasing interest in
   identifying reliable prognostic biomarkers to predict response to
   oxaliplatin-based chemotherapy and guide treatment strategies.

   Compared with previous studies, our work was distinguished by a
   rigorous, multi-step strategy to identify candidate genes associated
   with oxaliplatin resistance. For example, Yin et al. developed a gene
   signature based on resistance-related pathways and validated it across
   multiple cohorts. In contrast, we employed a comprehensive pipeline
   that incorporated differential expression analysis, univariate Cox
   regression, LASSO regression, and multivariate Cox regression, ensuring
   both statistical robustness and clinical relevance^[102]24. Unlike
   approaches that rely on single datasets or specific pathways, we
   utilized TCGA data for gene identification and further validated our
   findings using the GEO database and other independent tumor cohorts,
   thereby enhancing generalizability. While Zhou et al. focused on
   lncRNAs linked to oxaliplatin sensitivity, our study targets
   protein-coding genes directly implicated in resistance mechanisms,
   offering clearer therapeutic implications^[103]25. Furthermore,
   although Lin et al. proposed a resistance-related signature based on
   large-scale colon cancer profiling, our model integrates gene
   expression data with survival outcomes and external validation,
   yielding a robust and clinically applicable prognostic tool^[104]26.
   This integrative approach improves predictive accuracy and supports the
   development of personalized treatment strategies for CRC patients
   receiving oxaliplatin.

   To explore the mechanisms underlying oxaliplatin resistance, we
   analyzed CRC samples from the TCGA database by comparing
   oxaliplatin-resistant and sensitive groups. Differential expression
   analysis identified a set of upregulated genes, which were subsequently
   subjected to enrichment analysis. These genes showed significant
   associations with tumor immunity, cell signaling, and malignant
   biological processes. While both upregulated and downregulated genes
   can contribute to drug resistance, our study focused on upregulated
   genes for several reasons. First, upregulated genes are often actively
   involved in cellular pathways that mediate resistance, facilitating
   tumor cells’ ability to evade the cytotoxic effects of oxaliplatin.
   Second, their elevated expression may reflect an adaptive response to
   chemotherapy, highlighting their potential as biomarkers or therapeutic
   targets. In contrast, downregulated genes typically represent loss of
   normal cellular functions, which are less directly associated with
   resistance mechanisms. We therefore prioritized upregulated genes to
   identify those most functionally relevant to oxaliplatin resistance.
   Using a stepwise analytical pipeline, we first identified DEGs,
   followed by univariate Cox regression to evaluate associations with
   patient survival. The resulting candidates were refined using LASSO and
   multivariate Cox regression, yielding three key genes: TLE4, TNFAIP2,
   and ARGLU1 (Fig. [105]2). These genes were not only significantly
   upregulated in resistant samples but also correlated with poor
   prognosis, enhancing their clinical relevance. The predictive
   performance of this three-gene signature was subsequently validated
   using independent CRC cohorts from the GEO database, as well as other
   oxaliplatin-treated tumor types from the TCGA database. These results
   confirm the robustness and translational potential of the identified
   gene panel.

   TLE4 is a member of the highly conserved TLE transcriptional
   corepressor family. While TLE proteins do not directly bind DNA, they
   interact with transcription factors such as Hes1 and Runx2 to form
   repressive complexes^[106]27. Previous research suggests TLE4 functions
   as a tumor suppressor in hematologic malignancies and is located on
   chromosome 9q, a region frequently deleted in acute myeloid leukemia
   (AML)^[107]28. Loss of TLE4 expression promotes leukemic cell
   proliferation and activates the Wnt signaling pathway, enhancing
   pro-inflammatory signaling^[108]29. In contrast, elevated TLE4
   expression in CRC correlates with advanced Dukes stage, lymph node
   metastasis, and poor prognosis, likely through activation of the
   JNK–c-Jun pathway, which promotes proliferation and invasion^[109]30.
   Additionally, TLE4 polymorphisms have been linked to shorter PFS
   following gemcitabine treatment in HER2-negative metastatic breast
   cancer, indicating a role in chemoresistance^[110]31. Wei et al. also
   reported TLE4 overexpression in cisplatin-resistant bladder cancer
   xenografts, supporting its broader involvement in drug
   resistance^[111]32. Together, these findings suggest TLE4 may suppress
   tumor development in hematologic cancers but contribute to
   chemoresistance in solid tumors such as CRC and bladder cancer. Further
   studies are warranted to clarify TLE4’s role in oxaliplatin resistance
   and evaluate its potential as a therapeutic target or predictive
   biomarker in CRC.

   ARGLU1 is a well-conserved bifunctional protein that serves as both an
   activator and effector in the nuclear receptor system. Its
   overexpression inhibits the proliferation of gastric and head and neck
   squamous cell carcinoma cells, whereas silencing ARGLU1 suppresses the
   growth and migration of breast cancer cells^[112]33–[113]35.
   Functionally, ARGLU1 plays a key role in transcriptional regulation and
   the DNA damage response. While its overexpression promotes cancer cell
   proliferation, its downregulation produces the opposite effect.
   Notably, ARGLU1 also contributes to chemoresistance by enhancing DNA
   repair capacity^[114]36. These findings are consistent with our data,
   which show that reducing ARGLU1 expression sensitizes CRC cells to
   oxaliplatin.

   TNFAIP2 is a novel oncoprotein implicated in the initiation and
   progression of multiple cancers^[115]37–[116]39. Elevated TNFAIP2
   expression has been identified as a risk factor in cervical cancer,
   contributing to increased susceptibility and incidence^[117]40. In
   gastric cancer, its serum levels are elevated and associated with
   inflammation and tumor angiogenesis^[118]41. TNFAIP2 overexpression is
   also linked to cisplatin resistance by promoting epithelial–mesenchymal
   transition (EMT) in urothelial carcinoma cells^[119]42. In head and
   neck squamous cell carcinoma, high TNFAIP2 expression correlates with
   poor prognosis and chemoresistance. Mechanistically, TNFAIP2 competes
   with KEAP1 to stabilize NRF2, thereby inhibiting cisplatin-induced
   oxidative stress and apoptosis^[120]43. In breast cancer, TNFAIP2
   promotes chemoresistance by activating RAC1 via IQGAP1^[121]44.
   Collectively, these findings support a strong role for TNFAIP2 in
   mediating resistance to platinum-based therapies. Our results further
   confirm its high expression in oxaliplatin-resistant CRC tissues and
   its association with poor prognosis. Silencing TNFAIP2 in CRC cells
   enhances sensitivity to oxaliplatin, underscoring its potential as a
   therapeutic target.

   Furthermore, this study primarily focuses on patients receiving
   oxaliplatin-based chemotherapy. Future research should explore the
   potential interactions between oxaliplatin and other therapeutic
   modalities, such as immune checkpoint inhibitors, to gain deeper
   insight into the mechanisms and clinical implications of oxaliplatin
   resistance in CRC. Comprehensive investigation into combination
   treatment strategies is essential to better understand their effects on
   resistance development and to refine predictive models for improved
   clinical applicability.

Conclusions

   In conclusion, this study identified three key genes—TLE4, TNFAIP2, and
   ARGLU1—associated with oxaliplatin resistance in CRC patients from the
   TCGA-CRC dataset. The resulting three-gene signature demonstrated
   strong prognostic value and clinical relevance, offering a novel and
   effective tool for predicting oxaliplatin resistance and guiding
   personalized treatment strategies. Furthermore, the model exhibited
   potential applicability across other cancer types, underscoring its
   broader translational significance. Future studies should aim to
   validate this gene panel in larger, prospective clinical cohorts and
   work toward developing standardized protocols to facilitate its
   integration into clinical practice.

Methods and material

Data collection

   Clinical and transcriptomic data for the TCGA CRC cohort were obtained
   from the TCGA repository in TPM-normalized format for downstream
   analyses. After excluding samples lacking prognostic information, 114
   cases were included in the prognostic evaluation and designated as the
   training cohort, among which 64 had documented chemotherapy response
   data. For external validation of the prognostic model, the
   [122]GSE87211 dataset from the GEO database was used as an independent
   validation cohort. A total of 83 cases receiving oxaliplatin-based
   chemotherapy were selected from 243 available samples based on
   treatment records.

Identification of differentially expressed ORGSig

   To identify pivotal genes associated with chemotherapy, a cohort of 64
   cases with comprehensive chemotherapy records from the TCGA dataset was
   chosen. Chemotherapy responses in all 64 cases were characterized into
   four categories: complete response (CR), partial response (PR), stable
   disease (SD), and clinical progressive disease (PD). For analytical
   clarity, patients exhibiting SD and PD were grouped as chemoresistant,
   while those displaying CR and PR were categorized as chemosensitive in
   terms of chemotherapy response. Differential expression analysis of
   ORGSig between these two groups was conducted using the “limma”
   package, employing criteria of P < 0.05 and |log2(Foldchange[FC])| >
   mean (|Log2(FC)|) + 2 SD (|Log2(FC)|)^[123]45.

Development and validation of prognostic models in both the training and
validation cohorts

   A prognostic model was established using the TCGA dataset as the
   training cohort. Initially, univariate cox proportional hazard
   regression analysis was implemented to identify DEGs with significant
   prognostic implications (P < 0.01). Subsequently, the LASSO technique
   combined with cox proportional hazards regression was employed to
   reduce variables and prevent overfitting. Lastly, multivariate stepwise
   cox regression analysis was performed to select the risk model
   comprising the ORGSig, defining the resultant model as the ORGSig. Risk
   score for each sample was calculated as follows: Risk Score
   = 0.7047583×TLE4[exp] + 0.4238049×TNFAIP2[exp] + 0.4774622×ARGLU1[exp].
   For validation of the prognostic model, patients in the training cohort
   were stratified into low- and high-risk score groups based on the
   median risk score. The divergence in PFS between these groups was
   evaluated employing Kaplan–Meier survival analysis. Time-dependent ROC
   curve was constructed to determine the AUC for 1-year, 3-year, and
   5-year PFS in CRC patients. This analysis was replicated in the
   external validation dataset [124]GSE87211.

Association between risk score and immune characteristics

   The MCP-counter algorithm was utilized to determine the enriched
   presence of immune cells, endothelial cells, and fibroblasts in
   individual samples^[125]46. Prediction of response to immune checkpoint
   blockade therapy relied on the TIDE model^[126]47. Additionally, the
   expression levels of conventional immune checkpoints were compared in
   different groups.

Development and evaluation of the nomogram in the TCGA dataset

   To delve deeper into identifying independent prognostic factors among
   various potential variables, univariate and multivariate analyses using
   the cox regression model were performed. Subsequently, a nomogram
   integrating the ORGSig and independent prognostic clinical factors was
   designed for practical applicability.

Testing the applicability of ORGSig

   In addition to TCGA-CRC, we extended our analysis to include patients
   who underwent oxaliplatin treatment from various TCGA projects to
   validate the effectiveness of the Oxaliplatin resistance gene panel.
   Logistic regression models were constructed using the R package
   “glmnet” showcasing various gene combinations and their corresponding
   models. The top-performing models and their respective AUC values for
   distinct cancer types were presented accordingly.

WGCNA analysis

   The R package “WGCNA” was employed to establish a co-expression
   network. The methodology involved selecting TCGA-CRC patients who
   received oxaliplatin, integrating all mRNA into a gene expression
   cluster analysis. Following sample clustering and removal of outliers,
   the soft-thresholding power was determined using an algorithm to create
   a scale-free co-expression network and identify gene modules. Each
   module, distinguished by a distinct color, encompassed genes
   demonstrating analogous expression profiles. Through dynamic tree cut
   analysis, genes were classified, and modules sharing high similarity
   were merged to form cohesive dynamic clusters. Furthermore, the
   relationship between different modules and two clinical attributes, PFS
   and risk score, was examined. Modules containing crucial
   Oxaliplatin-resistant genes were prioritized for GO analysis.

Patient sample

   A total of 16 fresh tissue samples were obtained from patients,
   including 8 cases of oxaliplatin-sensitive and 8 cases of
   oxaliplatin-resistant colorectal cancer patients (stages II-III) who
   received adjuvant chemotherapy based on oxaliplatin after surgery. All
   tissues were embedded in paraffin to investigate the correlation
   between TNFAIP2 and oxaliplatin resistance. Tumor responses were
   evaluated using the Response Evaluation Criteria in Solid Tumors
   (RECIST). Written informed consent for the use of tumor tissues and
   clinical data in this study was obtained from all participants. This
   study was conducted in accordance with the Declaration of Helsinki and
   approved by the Ethics Committee of Yancheng First Hospital.

Cell cultures

   Immortalized human colorectal tumor cell lines SW480 was bought from
   cells bank of the Chinese Academy of Science (Shanghai, China). All
   cells were seeded in Dulbecco’s Modifed Eagle’s Medium (DMEM,
   Invitrogen, USA). 10% Fetal Bovine Serum (Clark bioscience, USA) and
   ampicillin and streptomycin (Gibco, USA) were added to the DMEM. All
   cells were cultured in 5% CO[2] at 37 ℃.

Cellular transfection

   The shRNA transfections were performed by the manufacturer’s protocol.
   shTNFAIP2#1 (CTTCACCAAAGGGAAGAAGAA), shTNFAIP2#2
   (CCACAAACTTCGTGGATCAAA) and shNC (UUUGUACUACACAAAAGUACUG) were
   purchased from GenePharma (Shanghai, China).

IHC

   IHC was performed following a previously described method^[127]48. The
   following antibody was used: anti-TNFAIP2 (Santa Cruz Biotechnology,
   sc-28318; dilution 1:200). The staining intensity was evaluated
   manually by two independent, experienced pathologists using the
   following scoring system: 0 = no staining, 1 = weak staining, 2 =
   moderate staining, and 3 = strong staining. The percentage of
   positively stained cells was assessed using four categories: 1 (0–10%),
   2 (11–50%), 3 (51–80%), and 4 (81–100%). The final IHC score was
   calculated by multiplying the intensity score by the percentage of
   positive cells.

RNA isolation and qRT–PCR

   RNA isolation and qRT–PCR were conducted according to our previously
   described study^[128]48. The gene-specific primers are as follows
   (5’–3’): TNFAIP2-F, CCTGGGCCTGAAAGTTCCTT, R, 5’-
   GGCATCCTTTATACCGGCCA-3’. β-actin—F, ACCCTGAAGTACCCCATCGAG, R,
   AGCACAGCCTGGATAGCAAC.

Colony formation assay

   In the colony formation assay, 1000 cells were plated in 6-well plates
   and allowed to grow for approximately 10 days with or without
   oxaliplatin treatment (2.0 µM)^[129]49. Subsequently, the colonies were
   stained with crystal violet and counted at the conclusion of the
   experiment.

Flow cytometry analysis

   The transfected cells were cultured with 0 or 30 µM oxaliplatin for 48
   h. Flow cytometry was performed using an Annexin V-FITC Apoptosis
   Detection Kit from Vazyme Biotech Co. Ltd. (Nanjing, China).

Statistical analysis

   The data are expressed as means ± standard deviation (SD), with all
   experiments replicated in triplicate. Statistical analyses were
   conducted using GraphPad Prism 8.0 or R software (version 4.2.1). A
   significance threshold of P < 0.05 was utilized. Statistical
   significance is denoted as follows: ns, not significant; *P < 0.05;
   **P < 0.01; ***P < 0.001.

Electronic supplementary material

   Below is the link to the electronic supplementary material.
   [130]Supplementary Material 1^ (38.5KB, xlsx)

Acknowledgements