Abstract
Background:
Primary non-response to anti-tumor necrosis factor (anti-TNF) therapy remains a major clinical challenge in ulcerative colitis (UC); however, public-data transcriptomic studies often yield unstable single-gene biomarkers and limited replication.
Methods:
We analyzed three independent pretreatment UC mucosal transcriptome cohorts from infliximab-treated patients (total n = 79). The induction response was assessed at 4–8 weeks according to the original definition in each study. To prioritize cross-cohort biology over single-probe overlap, we integrated recurrent effect direction, pathway convergence, and sample-level scoring. We defined a 68-gene consensus signature for interpretation, evaluated generalization with leave-one-dataset-out (LODO) scoring, tested attenuation after adjusting for baseline inflammatory/remodeling proxies, and examined cellular localizations in public epithelial (GSE116222) and rectal immune (GSE125527) single-cell references.
Results:
The 68-gene consensus signature was higher in non-responders across all three cohorts. In the LODO analysis, the held-out non-response axis remained elevated in the held-out non-responders, with a pooled random-effects size of Hedges’ g = 1.36 (95% confidence interval: 0.83‐1.89; p = 5.65 × 10−7; I2 = 0%). The pathway analyses consistently implicated inflammatory responses, TNF-α/NF-κB, IL-6–JAK–STAT3, interferon signaling, hypoxia, and epithelial–mesenchymal transition (EMT), whereas oxidative phosphorylation was enriched in the responders. Adjustments for inflammatory proxies with and without EMT attenuated the effects, but the associations remained directionally positive in all cohorts and retained nominal significance in two cohorts under the strictest model. In public single-cell references, the signature activities were enriched in cycling/stem-like and stress-inflammatory epithelial states and were highest in the rectal myeloid/dendritic cells (M/DCs).
Conclusion:
In the infliximab-treated UC cohorts, induction-phase non-response is associated with reproducible pretreatment mucosal-inflammatory-remodeling states that remain evident in the held-out cross-cohort analyses while localizing most strongly to rectal M/DCs and stress-associated epithelial states in public single-cell references. The signal overlaps substantially with the baseline inflammatory/remodeling burden, but its direction is consistent across the cohorts. These findings support a biologically coherent disease-state framework rather than a deployable clinical predictor.
Introduction
Ulcerative colitis (UC) is a chronic inflammatory disorder of the colonic mucosa and is marked by relapsing tissue injury, persistent immune activation, and substantial therapeutic heterogeneity. Although anti-tumor necrosis factor (anti-TNF) antibodies remain an important treatment option for moderate-to-severe UC, a substantial fraction of patients show primary non-response despite clinical assessment before treatment. This clinical problem prolongs exposure to ineffective therapy, delays alternative treatment, and may reflect a distinct pretreatment biological state rather than random therapeutic failure (Rutgeerts et al., 2005; Raine et al., 2022; ; Vieujean et al., 2025). Modern management approaches for UC increasingly emphasize treat-to-target strategies and endoscopic/histologic endpoints, but a therapeutic ceiling persists even with advanced therapies (Turner et al., 2021; ; West et al., 2023; ). Therefore, primary non-response remains common and continues to drive biomarker discovery (; ). The main goal of these efforts is to define pretreatment biological states that help identify patients who are less likely to benefit from TNF neutralization alone (; ; ; ).
Many transcriptomic studies have attempted to identify the pretreatment biomarkers of anti-TNF response in UC. Early bulk mucosal studies proposed multigene predictors, but their cross-cohort stability was variable (; ; Toedter et al., 2011). Oncostatin M (OSM)-centered inflammatory signaling subsequently emerged as a prominent correlate of anti-TNF non-response in inflammatory bowel disease (IBD), supporting a biologically plausible myeloid/stromal inflammatory model of therapeutic failure (West et al., 2017). Similarly, later studies implicated immune activation and trafficking programs (; Yang et al., 2023; Peña-Cearra et al., 2023; Pu et al., 2024). More recently, omics-driven and network or machine-learning approaches have not only expanded the candidate space but also underscored biomarker instability when the sample size is modest relative to the feature dimensionality (; Wang et al., 2025a; Shanthamallu et al., 2024; Varma and Simon, 2006).
Public-data transcriptomic studies are limited by practical challenges, such as recovering biologically coherent signals across cohorts without turning cohort reuse into circular validation (Varma and Simon, 2006). We addressed this by emphasizing the recurrent effect direction, pathway convergence, and held-out effect-size estimation. We therefore considered three linked questions: can pretreatment primary anti-TNF non-response in UC be captured as a coherent mucosal program across independent cohorts; does this program localize to plausible cellular compartments in public single-cell references; and does the association persist after accounting for baseline inflammatory/remodeling burden? These questions are most relevant in a disease-biology framework, where the aim is to characterize the pretreatment tissue state. To address these questions, we performed a hypothesis-driven analysis of three pretreatment UC mucosal transcriptome cohorts of infliximab-treated patients. Additionally, we defined a compact 68-gene consensus signature for interpretation, evaluated generalization with leave-one-dataset-out (LODO) scoring, examined attenuation after progressively stricter inflammatory/remodeling adjustments, and used public epithelial and immune single-cell references to provide the cellular context. Thus, we sought to describe the pretreatment biology of anti-TNF non-response in UC such that it is reproducible and biologically coherent.
Methods
Study design
We conducted a multicohort public-data transcriptomic study focused on pretreatment primary anti-TNF non-response in UC. Here, three independent infliximab-treated UC mucosal transcriptome cohorts were analyzed. We used a fixed 68-gene consensus signature for interpretation and a separate LODO scoring scheme for held-out effect-size estimation. Public single-cell references were used only to provide the cellular context and were not treated as discovery cohorts.
Public bulk transcriptome cohorts
Three cohorts were used from the Gene Expression Omnibus (GEO) database for the core analysis: GSE12251, GSE16879, and GSE23597. Of these, GSE12251 served as the discovery cohort for pathway prioritization, while GSE16879 was used for directional replication, and GSE23597 was considered as an additional independent cohort. For the LODO evaluation, each cohort was treated as the held-out dataset in turn. For the primary analysis, we retained only the pretreatment UC mucosal biopsies and harmonized the original study outcomes to a binary responder/non-responder framework while preserving the endpoint definition and assessment timepoint of each cohort.
In GSE12251, all analyzed samples were pretreatment colonic biopsies obtained at week 0; the study annotation reflected week-8 endoscopic and histologic healing, so the primary non-response was defined as the absence of this combined healing endpoint. In GSE16879, the biopsies were obtained before treatment and again at 4–6 weeks after the first infliximab infusion, with the responses being classified from the endoscopic and histologic findings at the 4–6-week assessment timepoint; we retained the pretreatment UC biopsies and used the responder/non-responder classification of this study for the binary outcome. In GSE23597 (ACT 1 sub-study), the biopsies were collected at baseline and during follow-up; we retained the pretreatment biopsies from the infliximab-treated arms and used the study’s week-8 response classification to define induction response versus non-response.
Because the outcome definitions were not identical across the three cohorts (4–6 weeks vs. 8 weeks, with one cohort using an endoscopic-plus-histologic healing endpoint), we treated the harmonized outcome as a pragmatic induction (4–8-week) non-response definition. The biologic-naive status and exact biopsy locations were not consistently extractable from the deposited sample annotations across all three cohorts, so these factors were not used for stratified modeling. Therefore, the main cross-cohort evidence is summarized using held-out LODO evaluation and random-effects pooling rather than any single-cohort estimate.
Bulk expression preprocessing
Normalized expression matrices were obtained from the GEO series-matrix files and organized into phenotype tables and expression matrices restricted to the selected core samples (). Each cohort was processed independently to preserve the cohort-native normalization while avoiding naive cross-platform pooling. The probe-level matrices were analyzed first and then mapped to gene symbols for cross-cohort directionality summaries. The replication standards emphasized effect direction and pathway/signature consistency rather than pooled expression magnitude.
Differential expression analysis
Within each bulk cohort, the responder versus non-responder comparisons were performed using limma (Ritchie et al., 2015). The linear models used response status as the main term, and the moderated t-statistic from the responder versus non-responder contrast was retained for downstream ranking. The probe-level differential results were mapped to gene symbols using platform annotation rules for the Affymetrix GPL570 platform; when multiple probes were mapped to the same gene symbol, a single representative gene-level record was retained for the cross-cohort direction summary and scoring.
Pathway analysis
Hallmark pathway analysis was performed on the gene-level limma moderated t-statistics using a preranked gene-set enrichment approach (Subramanian et al., 2005; ; ). The ranking statistic was the responder versus non-responder moderated t-value; therefore, negative normalized enrichment scores (NESs) in the raw enrichment outputs corresponded to pathways elevated in the non-responders, whereas positive NESs corresponded to pathways elevated in the responders. For figure readability, the signed NES values were reoriented so that the positive values indicate non-responder enrichment. Hallmark gene sets were obtained from the MsigDB database, and enrichment was performed using the fgsea package. The pathways were summarized per cohort and compared across cohorts using direction consistency, false discovery rate (FDR), and nominal significance counts across the datasets. The analyses were performed using R software version 4.4.3 with limma 3.62.2, fgsea 1.32.0, msigdbr 25.1.1, and Python 3.13.7.
Consensus signature and LODO evaluation
We used two related signatures for different purposes. The fixed 68-gene consensus signature was not selected from prioritized pathways; it was defined strictly from the bulk differential expression results as genes that were directionally elevated in the non-responders and nominally significant at p < 0.01 in all three cohorts. This fixed signature was used for biological interpretation, attenuation analyses, and single-cell localization. The generalization was assessed separately with LODO scoring, in which each cohort was evaluated with a gene set derived only from the other two cohorts (directionally elevated in non-responders and nominal p < 0.05 in both training cohorts). Therefore, the held-out LODO gene sets differed from the fixed consensus signature by design and were used for held-out effect-size estimation rather than definition of the interpretable 68-gene set. For interpretability, we also annotated the 68 consensus genes into broad descriptive groups based on known gene functions and annotations: immune/myeloid activation and antigen-presentation markers, cytokine/inflammatory and stress-signaling mediators, tissue-remodeling/stromal/endothelial or epithelial-stress genes, and intracellular regulatory/stress-response genes. These groups were used only to summarize the signature composition shown in Supplementary Table S1; furthermore, we did not treated as evidence that any subset is independent of inflammation.
Effect-size and replication analyses
Module scores were calculated within each cohort from the genes available on the corresponding microarray platforms. For each signature gene, one representative probe was retained to standardize the expression across the pretreatment samples within that cohort. The sample-level module score was the unweighted mean of these within-cohort gene z-scores, with all included genes being oriented so that a higher expression corresponded to the non-responder-enriched direction. Thus, the scores are relative within-cohort quantities rather than absolute expression measurements. Standardized effect sizes were calculated for the non-responders versus responders using Hedges’ g with confidence intervals (CIs). The primary replication evidence was summarized using the LODO axis (held-out evaluation) with fixed-effects and random-effects meta-style summaries, and the heterogeneity was summarized using I2 (). The consensus-signature effect sizes were treated as descriptive in-sample upper bounds.
Specificity analysis against baseline inflammatory burden
We next evaluated whether the consensus signature largely reflected the baseline inflammatory and remodeling burden. First, we created a simple inflammatory proxy using the within-cohort z-scored expressions of seven recurrent mucosal inflammatory genes observed in the non-responders (IL1B, TLR2, TNFAIP6, NAMPT, PTGS2, CXCR2, and TREM1) using a single representative probe per gene when available. These genes were selected because they were repeatedly elevated in the non-responders in the present cross-cohort analyses and are consistent with the mucosal inflammatory biology discussed in prior anti-TNF and IBD biomarker studies (; ; ; Toedter et al., 2011; West et al., 2017). We then constructed stricter Hallmark-based inflammatory proxies with and without epithelial–mesenchymal transition (EMT) from the available genes in the corresponding MSigDB gene sets. Because these proxies overlap biologically with the consensus signature, the stricter models were interpreted as sensitivity analyses rather than formal causal disentanglement. Within each cohort, the linear models estimated the associations between the consensus-signature score and response status before and after adjustment for these proxies.
Class-level anti-TNF sensitivity analysis (golimumab cohort)
As a sensitivity analysis beyond infliximab, we evaluated the fixed 68-gene consensus signature in a golimumab-treated UC cohort (GSE92415) derived from the PURSUIT-SC induction study (Sandborn et al., 2014; Telesco et al., 2018). Here, we retained the pretreatment mucosal biopsies and used the study’s week-6 response classification to define induction response versus non-response. This analysis was treated as supportive class-level evidence and was not incorporated into the core three-cohort effect-size pooling.
Single-cell localization support (processed public references)
The single-cell analyses involved processed public data and were intended to provide cellular context rather than discovery evidence.
Epithelial reference
The GSE116222 dataset served as the epithelial reference and corresponds to the single-cell atlas of human colon in UC (Smillie et al., 2019). Using the processed expression matrix and accompanying metadata, we assigned broad epithelial states using a curated marker panel and a simple maximum marker-score rule. The 68-gene consensus signature was then scored at the single-cell level using the genes available in the reference matrix (Smillie et al., 2019; ). Inter-state score differences were assessed using the Kruskal–Wallis test and one-versus-rest Wilcoxon rank-sum tests with Benjamini–Hochberg adjustment.
Immune rectal reference
The GSE125527 dataset served as the immune reference. We used the processed expression matrix and metadata tables, restricted the analysis to rectal tissue cells, and summarized the signature activity across the annotated immune cell classes using genes present in the reference matrix (). Inter-cell-class differences were assessed separately for the diseased and healthy rectal cells using the Kruskal–Wallis test and one-versus-rest Wilcoxon rank-sum tests with Benjamini–Hochberg adjustment. These tests were used as descriptive localization support at the cell level but not as donor-level inferential tests. The reported single-cell summary values represent within-reference relative activities and should be interpreted as enrichment patterns rather than direct evidence of cell-of-origin.
Results
Cohort assembly and analysis design
The three pretreatment UC mucosal transcriptome cohorts from infliximab-treated patients were assembled in a discovery-replication framework. The analysis set included 23 patients from GSE12251 (12 responders, 11 non-responders), 24 from GSE16879 (8 responders, 16 non-responders), and 32 from GSE23597 (25 responders, 7 non-responders), for a total of 79 patients, as described in Table 1.
TABLE 1
| Cohort | Role | Therapy | Platform | Endpoint/timepoint (induction) | Biopsy/source notes from deposited records | n | Responders | Non-responders |
|---|---|---|---|---|---|---|---|---|
| GSE12251 | Discovery | Infliximab | GPL570 | Week-8 endoscopic-plus-histologic healing; non-response = absence of combined healing | Pretreatment colonic biopsies at week 0 | 23 | 12 | 11 |
| GSE16879 | Replication | Infliximab | GPL570 | 4–6-week endoscopic and histologic response after first infusion | Pretreatment mucosal biopsies from actively inflamed UC mucosa | 24 | 8 | 16 |
| GSE23597 | Cohort 3 | Infliximab | GPL570 | Week-8 response classification in the ACT 1 gene-expression sub-study | Pretreatment colonic biopsy samples from infliximab-treated arms | 32 | 25 | 7 |
Summary of the core cohorts used for the primary claim.
Endpoint definitions and biopsy/source notes are reported from the deposited GEO records and associated study descriptions. Biologic-naive status and granular anatomic biopsy locations were not consistently available across all three cohorts and were therefore not modeled.
The response assessment definitions and timepoints were cohort-native: GSE12251 used week-8 endoscopic-plus-histologic healing; GSE16879 used endoscopic and histologic responses at 4–6 weeks after first infliximab infusion; GSE23597 used week-8 response classifications from an ACT 1 sub-study. The overall study overview, evidence framework, and working model are summarized in Figure 1. Because single-probe replication was uneven across the cohorts, the downstream inference focused on recurring effect direction, convergent pathways, and sample-level scoring rather than ranking each cohort’s candidate biomarkers.
FIGURE 1
Recurrent pretreatment inflammatory-remodeling program is associated with anti-TNF non-response
Across the cohorts, the non-responders shared a recurring inflammatory and tissue-remodeling program that included TLR2, IL1B, IL1RN, PTGS2, TNFAIP6, TREM1, CXCR2, NAMPT, CLEC4E, and CD93. Several of these genes are consistent with prior reports of mucosal inflammation and anti-TNF response biology, placing the present findings within an established biological context rather than proposing a wholly new pathway (; ; ; Toedter et al., 2011; West et al., 2017). The pathway analyses converged on an inflammatory-remodeling program in pretreatment non-responders, including inflammatory responses, TNF-α/NF-κB signaling, IL-6–JAK–STAT3 signaling, interferon signaling, hypoxia, and EMT. In contrast, oxidative phosphorylation was reproducibly enriched in the responders. These pathway results are summarized in Figure 2.
FIGURE 2
Consensus 68-gene signature and held-out LODO evaluation support cross-cohort association
The 68-gene consensus signature defined from the genes upregulated in the non-responders and nominally significant at p < 0.01 in all three cohorts was consistently higher in the pretreatment non-responders in each dataset. Because this fixed signature was derived using information from all cohorts, the within-cohort effect sizes are descriptive in-sample upper bounds (Hedges’ g = 1.85, 1.48, and 2.03 for GSE12251, GSE16879, and GSE23597, respectively), with a pooled random-effects association of Hedges’ g = 1.78 (95% CI: 1.22–2.34; p = 6.0 × 10−10; I2 = 0%). The signature was compositionally mixed rather than being restricted to a single canonical inflammatory marker set. Descriptive annotations of the 68 genes showed immune/myeloid and antigen-presentation components (including CD86, CD93, FCGR2A, HLA-DQA1, LILR-family genes, PTPRC, TLR1, and TLR8), inflammatory or stress-signaling mediators (including IL-6, CSF3, PTGS2, SOCS3, IFIT3, and SOD2), and remodeling or epithelial/stromal stress-associated genes (including CHI3L1, COL8A1, HGF, IGFBP5, IL11, INHBA, SELE, SERPINE1, TFPI2, and ZEB1). The full descriptive groupings are provided in Supplementary Table S1.
We then investigated whether the same pattern persisted when each cohort was scored only using the genes derived from the other two cohorts. In these LODO analyses, the non-response axis remained higher in the non-responders, with effect sizes of Hedges’ g = 1.52, 0.90, and 1.72 for held-out GSE12251, GSE16879, and GSE23597, respectively. The random-effects pooling yielded Hedges’ g = 1.36 (95% CI: 0.83–1.89; p = 5.65 × 10−7; I2 = 0%). The LODO-derived gene sets were intentionally broader than the fixed consensus signature: 632, 1,007, and 524 genes were used when holding out GSE12251, GSE16879, and GSE23597, respectively. The 68 fixed consensus genes were contained within each LODO gene set, indicating biological consistency between the compact interpretation signature and broader training-derived held-out axes while preserving their different analytical roles. This comparison is provided in Supplementary Table S7. The LODO score distributions and pooled effect-size summaries are shown in Figure 3.
FIGURE 3
Public single-cell references support localization to both myeloid and epithelial compartments
Two processed public single-cell references were used to place the bulk signal within the cellular context. In the epithelial reference GSE116222, the consensus-signature activity differed across marker-defined epithelial states (Kruskal–Wallis p = 3.91 × 10−219) and was relatively higher in the cycling, stem/transit-amplifying, and stress-inflammatory epithelial states while being lower in the absorptive states (Smillie et al., 2019; ). The one-versus-rest tests supported higher signature activities in the cycling, stem/transit-amplifying, and stress-inflammatory epithelial states after Benjamini–Hochberg adjustment (Supplementary Table S7). In the rectal immune reference GSE125527, the consensus-signature activity differed across immune cell classes in both diseased and healthy rectal cells (Kruskal–Wallis p < 1 × 10−300 for each subset). The signal was highest in the myeloid/dendritic cells (M/DCs) in diseased rectal tissue (mean score of 0.583), exceeding those of the NK (0.241), T (0.210), and B (0.170) cell compartments. The M/DCs remained higher than all other diseased rectal immune cells in the one-versus-rest tests after Benjamini–Hochberg adjustment (adjusted p = 1.05 × 10−156). The same rank order was observed in the healthy rectum, where the M/DCs again showed the highest average score (0.526) (). Some representative consensus gene-expression patterns across the epithelial and immune cell populations are shown in Supplementary Figure S3. Together, these two references point to a mixed pattern with a prominent myeloid/inflammatory component alongside epithelial stress/proliferative activity. The single-cell localization support is summarized in Figure 4.
FIGURE 4
Baseline inflammatory burden explains part of the non-response axis
Adjustment for a simple inflammatory proxy attenuated the non-response association, but the positive residual coefficients remained across all three cohorts. With stricter hallmark inflammatory and inflammatory-plus-EMT adjustments, the response-status coefficient remained positive in all cohorts and retained nominal significance in the GSE16879 (p = 0.030) and GSE23597 (p = 0.0077) datasets under the strictest joint model. In contrast, in the discovery cohort GSE12251, the adjusted coefficient was essentially null (beta = 0.016, p = 0.88), indicating that the signature-response association in this small cohort is statistically collinear with the baseline mucosal inflammatory and tissue-remodeling burden. These analyses place the non-response axis close to baseline inflammatory severity while showing residual response-associated signals in the two replication cohorts after the same proxy adjustments. The specificity benchmarks are summarized in Figure 5.
FIGURE 5
As an external class-level anti-TNF sensitivity analysis, the 68-gene consensus signature was also directionally higher in week-6 non-responders in the golimumab-treated PURSUIT-SC cohort (GSE92415; pretreatment biopsies; n = 87), with a modest but significant standardized association (Hedges’ g = 0.50, 95% CI: 0.07–0.92; p = 0.022; Supplementary Table S6; Supplementary Figure S1) (Sandborn et al., 2014; Telesco et al., 2018).
Discussion
Main interpretation
This study identifies a reproducible pretreatment mucosal state associated with induction-phase non-response in infliximab-treated UC cohorts. The data support a broader inflammatory-remodeling axis that recurs across three infliximab-treated cohorts, remains evident in the held-out LODO analyses, and is most active in rectal M/DCs together with stress-associated epithelial states. The main contribution of this work is a replicated biological interpretation of pretreatment heterogeneity and not a standalone predictive test.
Biological context
The enrichment pattern in the rectal M/DC compartments, together with the TNF-α/NF-κB, IL-6–JAK–STAT3, interferon, and hypoxia programs, is consistent with a cytokine-rich myeloid inflammatory milieu coupled to epithelial stress and remodeling. Prior works have likewise implicated myeloid/stromal inflammatory states, including OSM-driven biology, in anti-TNF non-response in IBD (West et al., 2017; Thomas et al., 2024; ). To understand how the 68-gene consensus signature relates to these established frameworks, we compared its components with the classical OSM-driven axis. While the OSM axis is primarily stromal and macrophage-centric, our 68-gene signature captures the downstream transcriptional endpoints of myeloid activation (such as CD86, LILRB genes, and CSF2RB) alongside direct downstream targets of IL-6/OSM signaling (such as SOCS3 and BCL6). This places the consensus signature as a broader downstream readout of the myeloid-remodeling axis.
The single-cell results add an epithelial layer to this interpretation. The consensus-signature activity was concentrated in both cycling/stem-like epithelial cells and M/DCs: a pattern compatible with lamina propria myeloid activation occurring alongside epithelial regeneration and stress remodeling. A plausible tissue-level model for this is that inflammatory cytokines, such as IL-6, TNF-α, and OSM, act on the epithelial progenitor and reparative compartments during active mucosal injury, producing a proliferative and stress-associated state marked by hypoxia and EMT-related programs. In the pretreatment biopsies, high activities of this myeloid-epithelial remodeling axis may therefore indicate a refractory inflammatory tissue state. Because the primary cohorts were sampled before treatment, this interpretation remains broader than anti-TNF-specific resistance alone and includes the possibility of general inflammatory refractoriness.
Specificity and confounding
An important interpretive challenge is separating treatment resistance from the baseline inflammatory burden. In our data, adjustment for the inflammatory and EMT-related proxies reduced the response-status coefficient substantially but did not remove the association across cohorts. The near-null adjusted coefficient in GSE12251 (p = 0.88) suggests that, in this small discovery cohort, the signature-response association is statistically inseparable from the overall baseline inflammatory and tissue-remodeling burden. This may reflect a subset in which the refractory disease is dominated by inflammatory scaling or limited power (n = 23) to estimate the residual effects in the presence of strong collinearity. In contrast, the persistence of significant adjusted associations in the replication cohorts (GSE16879, p = 0.030; GSE23597, p = 0.0077) supports cross-cohort directional stability and suggests that the 68-gene signature captures the response-associated biology, including myeloid-stromal activation and epithelial stem-like stress responses, beyond simple scaling of mucosal inflammation.
Limitations
The present study has clear limitations: it is retrospective, is based on public transcriptomic data, and cannot support causal inference. The outcome definitions were harmonized across the studies rather than being prospectively standardized, and the biologic-naive status, exact biopsy locations, medication-regimen details, and baseline endoscopic or histologic severity data were not uniformly available in the deposited annotations. The 68-gene consensus signature was selected through a stringent cross-cohort nominal threshold and directional agreement rule, but this rule was designed for reproducible biological interpretation rather than formal family-wise error control across all possible intersections. Moreover, we did not exhaustively evaluate alternative threshold or intersection/union rules for signature construction. All primary bulk cohorts used microarray platforms, which have more limited dynamic ranges and reduced sensitivities for low-abundance transcripts than next-generation sequencing. The specificity analyses used inflammatory and EMT proxy scores but did not include direct deconvolution or cell-type abundance covariates; thus, residual confounding by the cellular composition remains possible. The single-cell datasets provide cellular context rather than treatment-response validation, and the per-cell tests do not account for donor-level clustering. In addition, the immune reference was restricted to rectal tissue, whereas the bulk biopsy locations were not uniformly documented as rectal; hence, location-dependent cellular composition or gene-expression differences could influence the interpretation. The immune-reference analysis also used only the subset of signature genes represented in that matrix, and the golimumab cohort should be interpreted as supportive class-level evidence rather than broad anti-TNF generalization.
Moving toward clinically actionable stratification will require targeted validation rather than additional retrospective reanalyses alone. Therefore, the priority next steps include testing the 68-gene signature in prospective pretreatment biopsy cohorts with standardized response endpoints, comparing the signature with established inflammatory-response frameworks such as the OSM axis (West et al., 2017), and evaluating whether a smaller quantitative polymerase chain reaction or targeted RNA panel can reproduce the signal. Future models should also include endoscopic severity, histology, C-reactive protein, fecal calprotectin, drug exposure, and immunogenicity variables because therapeutic drug monitoring () and anti-drug antibody or HLA-associated immunogenicity (; Zhang et al., 2025; ) can affect apparent treatment failure independently of the baseline tissue state. Mechanistic validation should focus on the main biological components supported by the present findings: myeloid/stromal inflammatory programs (; ; Wang et al., 2025b), epithelial stress and remodeling (; ; ), hypoxia/stress pathways (; ), pH-sensing inflammatory pathways (), proteomic disease-activity signatures (Veyssière et al., 2025), microbial interaction networks (), and candidate inflammatory targets such as FPR2 (Yang et al., 2026). Overall, induction-phase non-response in the infliximab-treated UC cohorts is associated with a reproducible pretreatment mucosal inflammatory-remodeling state that maps most strongly to the rectal M/DCs and epithelial stress/proliferative states in public single-cell references. The association is attenuated but not extinguished after accounting for the baseline inflammatory/remodeling burden.
Statements
Data availability statement
Publicly available datasets were analyzed in this study. All datasets analyzed in this study are publicly available from the NCBI Gene Expression Omnibus (GEO) database: GSE12251 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE12251), GSE16879 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE16879), GSE23597 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE23597), GSE92415 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE92415), GSE116222 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE116222), and GSE125527 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE125527). Source code and the public reproducibility package are available at GitHub (https://github.com/Zhenghongwei11/Pretreatment-nonresponse-axis-in-UC; release v1.2.1) and are archived on Zenodo (version-specific DOI: 10.5281/zenodo.21496340; concept DOI: 10.5281/zenodo.19028074).
Ethics statement
Ethical approval was not required for the study involving humans in accordance with the local legislation and institutional requirements. Written informed consent to participate in this study was not required from the participants or their legal guardians/next of kin in accordance with the national legislation and institutional requirements.
Author contributions
HZ: Conceptualization, Visualization, Software, Data curation, Formal Analysis, Writing – original draft, Methodology. XZ: Writing – review and editing, Validation, Methodology. WC: Conceptualization, Writing – review and editing, Project administration, Supervision.
Funding
The author(s) declared that financial support was not received for this work and/or its publication.
Acknowledgments
The authors thank the investigators who generated and deposited the GEO datasets used in this study.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was used in the creation of this manuscript. The authors used ChatGPT (OpenAI) for language editing. The authors reviewed and approved the final text.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2026.1845510/full#supplementary-material
Abbreviations
UC, ulcerative colitis; anti-TNF, anti-tumor necrosis factor; EMT, epithelial–mesenchymal transition.
References
1
AnjieS. I.HulshoffM. S.D’HaensG. (2023). Efficacious dosing regimens for anti-TNF therapies in inflammatory bowel disease: where do we stand?Expert Opin. Biol. Ther.23 (4), 341–351. 10.1080/14712598.2023.2198086
2
ArijsI.LiK.ToedterG.QuintensR.Van LommelL.Van SteenK.et al (2009a). Mucosal gene signatures to predict response to infliximab in patients with ulcerative colitis. Gut58 (12), 1612–1619. 10.1136/gut.2009.178665
3
ArijsI.De HertoghG.LemaireK.QuintensR.Van LommelL.Van SteenK.et al (2009b). Mucosal gene expression of antimicrobial peptides in inflammatory bowel disease before and after first infliximab treatment. PLoS One4 (11), e7984. 10.1371/journal.pone.0007984
4
BelarifL.DangerR.KermarrecL.Nerrière-DaguinV.PengamS.DurandT.et al (2019). IL-7 receptor influences anti-TNF responsiveness and T cell gut homing in inflammatory bowel disease. J. Clin. Invest.129 (5), 1910–1925. 10.1172/JCI121668
5
BolandB. S.HeZ.TsaiM. S.OlveraJ. G.OmilusikK. D.DuongH. G.et al (2020). Heterogeneity and clonal relationships of adaptive immune cells in ulcerative colitis revealed by single-cell analyses. Sci. Immunol.5 (50), eabb4432. 10.1126/sciimmunol.abb4432
6
BrunM. K.GehinJ. E.BjørlykkeK. H.WarrenD. J.KlaasenR. A.SextonJ.et al (2024). Clinical consequences of infliximab immunogenicity and the effect of proactive therapeutic drug monitoring: exploratory analyses of the randomised, controlled NOR-DRUM trials. Lancet Rheumatol.6 (4), e226–e236. 10.1016/S2665-9913(23)00341-7
7
ChenL.ZhangC.NiuR.XiongS.HeJ.WangY.et al (2025). Multi-omics biomarkers for predicting efficacy of biologic and small-molecule therapies in adults with inflammatory bowel disease: a systematic review. United Eur. Gastroenterol. J.13 (4), 517–530. 10.1002/ueg2.12720
8
CuiG.FanQ.LiZ.GollR.FlorholmenJ. (2021). Evaluation of anti-TNF therapeutic response in patients with inflammatory bowel disease: current and novel biomarkers. EBioMedicine66, 103329. 10.1016/j.ebiom.2021.103329
9
CuiG.FlorholmenJ.GollR. (2022). Could mucosal TNF transcript as a biomarker candidate help optimize Anti-TNF biological therapy in patients with ulcerative colitis?Front. Immunol.13, 881112. 10.3389/fimmu.2022.881112
10
Culmsee-HolmF. B.BuhlE.KraaerM. T.SteenholdtC.AinsworthM. A. (2025). Identifying genetic factors influencing the development of anti-drug antibodies in inflammatory bowel disease: a scoping review. J. Inflamm. Res.18, 17589–17610. 10.2147/JIR.S546252
11
Di MattiaM.SalleseM.NeriM.LopetusoL. R. (2024). Hypoxic functional regulation pathways in the GI tract: focus on the HIF-1α and microbiota’s crosstalk. Inflamm. Bowel Dis.30 (8), 1406–1418. 10.1093/ibd/izae046
12
DignassA.RathS.KleindienstT.StallmachA. (2023). Review article: translating STRIDE-II into clinical reality - opportunities and challenges. Aliment. Pharmacol. Ther.58 (5), 492–502. 10.1111/apt.17622
13
EdgarR.DomrachevM.LashA. E. (2002). Gene expression omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res.30 (1), 207–210. 10.1093/nar/30.1.207
14
GisbertJ. P.ChaparroM. (2020). Predictors of primary response to biologic treatment [anti-TNF, vedolizumab, and ustekinumab] in patients with inflammatory bowel disease: from basic science to clinical practice. J. Crohns Colitis14 (5), 694–709. 10.1093/ecco-jcc/jjz195
15
GudiñoV.Bartolomé-CasadoR.SalasA. (2025). Single-cell omics in inflammatory bowel disease: recent insights and future clinical applications. Gut74 (8), 1335–1345. 10.1136/gutjnl-2024-334165
16
HausmannM.SeuwenK.de VallièreC.BuschM.RuizP. A.RoglerG. (2024). Role of pH-sensing receptors in colitis. Pflugers Arch.476 (4), 611–622. 10.1007/s00424-024-02943-y
17
HigginsJ. P.ThompsonS. G. (2002). Quantifying heterogeneity in a meta-analysis. Stat. Med.21 (11), 1539–1558. 10.1002/sim.1186
18
HodgesP.RobertsC.ParkesM.KennedyN. A.GoodhandJ.AhmadT.et al (2026). Clinical utility of pre-treatment four-digit-resolution HLA genotyping to guide anti-tumor necrosis factor choice and concomitant immunomodulator use. J. Crohns Colitis20 (2), jjaf233. 10.1093/ecco-jcc/jjaf233
19
HouF.BianX.JingD.GaoH.ZhuF. (2024). Hypoxia, hypoxia-inducible factors and inflammatory bowel diseases. Gastroenterol. Rep. (Oxf).12, goae030. 10.1093/gastro/goae030
20
JohnsonJ. L.SargsyanD.NeimanE. M.HartA.StojmirovicA.KosoyR.et al (2024). Gene coexpression networks reveal a broad role for lncRNAs in inflammatory bowel disease. JCI Insight9 (3), e168988. 10.1172/jci.insight.168988
21
KorotkevichG.SukhovV.BudinN.ShpakB.ArtyomovM. N.SergushichevA. (2016). Fast gene set enrichment analysis. bioRxiv. 10.1101/060012
22
KorsunskyI.WeiK.PohinM.KimE. Y.BaroneF.MajorT.et al (2022). Cross-tissue, single-cell stromal atlas identifies shared pathological fibroblast phenotypes in four chronic inflammatory diseases. Med3 (7), 481–518.e14. 10.1016/j.medj.2022.05.002
23
LeoneG. M.ManganoK.PetraliaM. C.NicolettiF.FagoneP. (2023). Past, present and (foreseeable) future of biological anti-TNF alpha therapy. J. Clin. Med.12 (4), 1630. 10.3390/jcm12041630
24
LiberzonA.BirgerC.ThorvaldsdóttirH.GhandiM.MesirovJ. P.TamayoP. (2015). The molecular signatures database (MSigDB) hallmark gene set collection. Cell Syst.1 (6), 417–425. 10.1016/j.cels.2015.12.004
25
MaC.JairathV.FeaganB. G.Peyrin-BirouletL.DaneseS.SandsB. E.et al (2024). Interpreting modern randomized controlled trials of medical therapy in inflammatory bowel disease. Nat. Rev. Gastroenterol. Hepatol.21 (11), 792–808. 10.1038/s41575-024-00989-y
26
MelogranaF.SudhakarP.YousefiB.CaenepeelC.FalonyG.Vieira-SilvaS.et al (2024). Individual-network based predictions of microbial interaction signatures for response to biological therapies in IBD patients. Front. Mol. Biosci.11, 1490533. 10.3389/fmolb.2024.1490533
27
Melón-ArdanazE.VenyM.CorralizaA. M.GudiñoV.Garrido-TrigoA.Sanzo-MachucaÁ.et al (2025). Differential effects of tofacitinib on macrophage activation contribute to lack of response in ulcerative colitis patients. J. Crohns Colitis19 (6), jjaf076. 10.1093/ecco-jcc/jjaf076
28
MoranG. W.GordonM.SinopolouV.RadfordS. J.DarieA. M.VuyyuruS. K.et al (2025). British society of gastroenterology guidelines on inflammatory bowel disease in adults: 2025. Gut74 (Suppl. 2), s1–s101. 10.1136/gutjnl-2024-334395
29
OuJ.LiL.ZhiF.HuangB.ZhaoX. (2025). Developing models for the diagnosing of ulcerative colitis and prognosis of anti-TNF-α non-response based on neutrophil extracellular trap-associated genes. Front. Immunol.16, 1530508. 10.3389/fimmu.2025.1530508
30
ParikhK.AntanaviciuteA.Fawkner-CorbettD.JagielowiczM.AulicinoA.LagerholmC.et al (2019). Colonic epithelial cell diversity in health and inflammatory bowel disease. Nature567 (7746), 49–55. 10.1038/s41586-019-0992-y
31
Peña-CearraA.CasteloJ.LavínJ. L.Gonzalez-LopezM.Pascual-ItoizM. A.FuertesM.et al (2023). Mitochondrial dysfunction-associated microbiota establishes a transmissible refractory response to anti-TNF therapy during ulcerative colitis. Gut Microbes15 (2), 2266626. 10.1080/19490976.2023.2266626
32
PuD.WangP.WangX.TianY.GongH.MaX.et al (2024). Focusing on non-responders to infliximab with ulcerative colitis, what can we do first and next?Int. Immunopharmacol.141, 112943. 10.1016/j.intimp.2024.112943
33
RaineT.BonovasS.BurischJ.KucharzikT.AdaminaM.AnneseV.et al (2022). ECCO guidelines on therapeutics in ulcerative colitis: medical treatment. J. Crohns Colitis16 (1), 2–17. 10.1093/ecco-jcc/jjab178
34
RitchieM. E.PhipsonB.WuD.HuY.LawC. W.ShiW.et al (2015). Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res.43 (7), e47. 10.1093/nar/gkv007
35
RutgeertsP.SandbornW. J.FeaganB. G.ReinischW.OlsonA.JohannsJ.et al (2005). Infliximab for induction and maintenance therapy for ulcerative colitis. N. Engl. J. Med.353 (23), 2462–2476. 10.1056/NEJMoa050516
36
SandbornW. J.FeaganB. G.MaranoC.ZhangH.StraussR.JohannsJ.et al (2014). Subcutaneous golimumab induces clinical response and remission in patients with moderate-to-severe ulcerative colitis. Gastroenterology146 (1), 85–95. 10.1053/j.gastro.2013.05.048
37
ShanthamalluU. S.KilpatrickC.JonesA.RubinJ.SalehA.BarabásiA. L.et al (2024). A network-based framework to discover treatment-response-predicting biomarkers for complex diseases. J. Mol. Diagn.26 (10), 917–930. 10.1016/j.jmoldx.2024.06.008
38
SmillieC. S.BitonM.Ordovas-MontanesJ.SullivanK. M.BurginG.GrahamD. B.et al (2019). Intra- and inter-cellular rewiring of the human colon during ulcerative colitis. Cell178 (3), 714–730.e22. 10.1016/j.cell.2019.06.029
39
SubramanianA.TamayoP.MoothaV. K.MukherjeeS.EbertB. L.GilletteM. A.et al (2005). Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U S A.102 (43), 15545–15550. 10.1073/pnas.0506580102
40
TelescoS. E.BrodmerkelC.ZhangH.StraussR.JohannsJ.MazumderB.et al (2018). Gene expression signature for prediction of golimumab response in a phase 2a open-label trial of patients with ulcerative colitis. Gastroenterology155 (4), 1008–1011.e8. 10.1053/j.gastro.2018.06.077
41
ThomasT.FriedrichM.Rich-GriffinC.PohinM.AgarwalD.PakpoorJ.et al (2024). A longitudinal single-cell atlas of anti-tumour necrosis factor treatment in inflammatory bowel disease. Nat. Immunol.25 (11), 2152–2165. 10.1038/s41590-024-01994-8
42
ToedterG.LiK.MaranoC.MaK.SagueS.HuangC. C.et al (2011). Gene expression profiling and response signatures associated with differential responses to infliximab treatment in ulcerative colitis. Am. J. Gastroenterol.106 (7), 1272–1280. 10.1038/ajg.2011.83
43
TurnerD.RicciutoA.LewisA.D’AmicoF.DhaliwalJ.GriffithsA. M.et al (2021). STRIDE-II: an update on the selecting therapeutic targets in inflammatory bowel disease (STRIDE) initiative of the international organization for the study of IBD (IOIBD): determining therapeutic goals for treat-to-target strategies in IBD. Gastroenterology160 (5), 1570–1583. 10.1053/j.gastro.2020.12.031
44
VarmaS.SimonR. (2006). Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics7, 91. 10.1186/1471-2105-7-91
45
VeyssièreM.HammoudiN.Le BourhisL.HassidD.BonnetJ.Tran MinhM. L.et al (2025). Blood proteomic signatures associated with disease activity in inflammatory bowel diseases. J. Crohns Colitis19 (9), jjaf162. 10.1093/ecco-jcc/jjaf162
46
VieujeanS.JairathV.Peyrin-BirouletL.DubinskyM.IacucciM.MagroF.et al (2025). Understanding the therapeutic toolkit for inflammatory bowel disease. Nat. Rev. Gastroenterol. Hepatol.22 (6), 371–394. 10.1038/s41575-024-01035-7
47
WangM.LiangL.TangZ.HanJ.WuL.LiuL.et al (2025a). Multi-omics derivation of a core gene signature for predicting therapeutic response and characterizing immune dysregulation in inflammatory bowel disease. Front. Immunol.16, 1611598. 10.3389/fimmu.2025.1611598
48
WangJ.GuayH.ChangD. (2025b). Crohn’s disease and ulcerative colitis share 2 molecular subtypes with different mechanisms and drug responses. J. Crohns Colitis19 (3), jjae152. 10.1093/ecco-jcc/jjae152
49
WestN. R.HegazyA. N.OwensB. M. J.BullersS. J.LinggiB.BuonocoreS.et al (2017). Oncostatin M drives intestinal inflammation and predicts response to tumor necrosis factor-neutralizing therapy in patients with inflammatory bowel disease. Nat. Med.23 (5), 579–589. 10.1038/nm.4307
50
WestJ.TanK.DeviJ.MacraeF.ChristensenB.SegalJ. P. (2023). Benefits and challenges of treat-to-target in inflammatory bowel disease. J. Clin. Med.12 (19), 6292. 10.3390/jcm12196292
51
YangX.ShiJ.WangG.ChenH.YeY.ZhongJ.et al (2023). Novel mRNA signature for anti-TNF-α therapy primary response in patients with ulcerative colitis. Inflamm. Bowel Dis.29 (9), 1458–1469. 10.1093/ibd/izad060
52
YangW. S.WangX. Z.WuW.LiY.WangG. F.ZhouQ. T.et al (2026). Formyl peptide receptor 2 is a potential biomarker and therapeutic target for inflammatory bowel disease. Acta Pharmacol. Sin.47 (3), 677–688. 10.1038/s41401-025-01695-4
53
ZhangQ.SharipM.RobertsC.ShakwehE.ParkesM.AhmadT. (2025). HLA-DQA1*05:01 and DQA1*05:05 inform choice of anti-tumor necrosis factor and concomitant use of immunomodulators in patients with inflammatory bowel disease. J. Crohns Colitis19 (11), jjaf195. 10.1093/ecco-jcc/jjaf195
Summary
Keywords
anti-tumor necrosis factor, epithelial remodeling, infliximab, mucosal transcriptome, myeloid cells, primary non-response, single-cell RNA-sequencing, ulcerative colitis
Citation
Zheng H, Zhuang X and Chen W (2026) A reproducible pretreatment mucosal-inflammatory-remodeling state is associated with induction-phase anti-tumor necrosis factor non-response in ulcerative colitis. Front. Genet. 17:1845510. doi: 10.3389/fgene.2026.1845510
Received
02 April 2026
Revised
30 June 2026
Accepted
02 July 2026
Published
30 July 2026
Volume
17 - 2026
Edited by
Emidio Scarpellini, KU Leuven, Belgium
Reviewed by
Qingsheng Liu, Hangzhou Hospital of Traditional Chinese Medicine, China
Anna-Teresa D'Hooghe, KU Leuven, Belgium
Updates
Copyright
© 2026 Zheng, Zhuang and Chen.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Wenbiao Chen, 2008001@qzmc.edu.cn
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.