Transcriptomic discovery of prognostic biomarker candidates

We performed a comparative transcriptomic analysis using the TCGA COADREAD dataset to identify prognostic biomarkers associated with obesity in colorectal cancer (CRC). Samples were classified into three groups—adjacent normal tissue (AN), tumor tissue from healthy-weight CRC patients (HT), and tumor tissue from overweight/obese CRC patients (OT). A total of 139 genes demonstrated a consistent monotonic trend in expression across the AN–HT–OT axis, with 61 showing a downward pattern and 78 showing an upward pattern (Fig. 1A). Of these, six representative genes—NNT, PANK3, PPARGC1B, RTL6 (downward), and FAM220A, TMEM9 (upward)—were selected based on consistent trends and survival relevance. Their expression profiles showed progressive changes across AN, HT, and OT groups, as visualized in bar plots (Fig. 1B), reflecting obesity-associated transcriptional shifts.

Fig. 1

Discovery of obesity-associated prognostic biomarker candidates in colorectal cancer (CRC). A Schematic overview of gene expression trend analysis using TCGA COADREAD data. Samples were stratified into adjacent normal (AN), tumor tissue from healthy-weight CRC patients (HT), and tumor tissue from overweight/obese CRC patients (OT). A total of 139 genes showed monotonic expression changes across the AN–HT–OT axis, with 61 genes displaying a downward trend and 78 genes showing an upward trend. B Bar plots showing expression patterns of six candidate genes (NNT, PANK3, PPARGC1B, RTL6, FAM220A, and TMEM9) across the three groups. Four genes showed a downward trend (higher in AN, lower in OT), and two genes showed an upward trend (lower in AN, higher in OT). C 5-year overall survival analysis in the OT group revealed six genes significantly associated with prognosis. High expression of downward-trend genes (NNT, PANK3, PPARGC1B, RTL6) and low expression of upward-trend genes (FAM220A, TMEM9) were associated with favorable survival outcomes. OT, tumor tissue samples from overweight and obese patients with CRC (BMI ≥ 25 kg/m2); HT, tumor tissue samples from healthy weight patients with CRC (body mass index [BMI] < 25 kg/m2); AN, adjacent normal tissue samples from group HT

These six candidates also demonstrated statistically significant associations with overall survival in the OT group: NNT, PANK3, PPARGC1B, and RTL6 (downward trend) and FAM220A and TMEM9 (upward trend) (Fig. 1C). In the OT group, higher expression of downward-trend genes and lower expression of upward-trend genes correlated with improved survival outcomes. This pattern was not observed in the HT group, suggesting a potential obesity-specific prognostic effect. Accordingly, these genes were selected for further validation by immunohistochemistry (IHC) in an independent CRC patient cohort.

Clinical and pathological characteristics by NNT expression

Clinical validation was conducted in an independent cohort of 448 CRC patients from Gachon University Gil Medical Center. Of these, 127 (28.3%) were classified as having obesity-associated CRC (BMI ≥ 25 kg/m2), while 321 (71.7%) were healthy-weight CRC patients (Table 1). The mean patient age was 65.1 ± 11.4 years, with 60.0% male and 40.0% female. BMI values differed markedly between the groups, with healthy-weight patients showing a mean BMI of 21.7 ± 2.1 kg/m2 and overweight/obese patients showing a mean BMI of 27.3 ± 2.1 kg/m2. (t test, p < 0.0001). Most tumors were located in the sigmoid colon or rectum and were moderately differentiated. Based on TNM staging, 15.1% were stage I, 32.6% stage II, 33.0% stage III, and 11.8% stage IV.

Table 1 Baseline characteristics

To further explore clinical differences between the obesity-associated and healthy-weight CRC groups, we compared clinicopathological characteristics based on BMI classification (Table 2). No significant differences were observed between the two groups with respect to age, sex, diabetes mellitus, family history, anemia, white blood cell counts, serum CEA levels, tumor location, tumor differentiation, or TNM stage. In contrast, the prevalence of smoking was lower in the overweight/obese CRC group than in the healthy-weight group (9.4% vs 17.8%). Both univariate and multivariate analyses indicated an inverse association between smoking and overweight/obese CRC (univariate p = 0.028; multivariable OR 0.466, 95% CI 0.229–0.949, p = 0.035). This trend is consistent with the well-established inverse relationship between smoking and body weight. [22, 23]. In addition, we assessed the clinicopathological associations of the six candidate genes identified in the TCGA analysis. Univariate and multivariate analyses revealed distinct clinical correlations for several of these genes. Specifically, PANK3 was associated with TNM stage; PPARGC1B with serum CEA levels and tumor differentiation; RTL6 with anemia and tumor differentiation; FAM220A with both tumor differentiation and TNM stage; and TMEM9 with anemia, tumor differentiation, and TNM stage (Online Resources 1–5). NNT expression showed no significant association with conventional clinicopathological variables (BMI < 25 vs ≥ 25: univariate p = 0.293; multivariable OR 0.768, 95% CI 0.491–1.202, p = 0.249), supporting that its prognostic impact is largely independent of baseline features (Table 3).

Table 2 Clinical characteristics of patients with colorectal cancer (CRC)Table 3 Clinicopathologic significance of NNT expressionNNT as a potential independent prognostic marker in obesity-related CRC

To validate our findings, we performed IHC and survival analysis, using the GMC cohort. First, we stratified the 448 patients into OT and HT groups according to BMI (cutoff value of 25 kg/m2), and each group was further classified into high and low subgroups based on IHC expression (Fig. 2A). Thereafter, we performed survival analysis to evaluate the association between six protein expression, BMI status, and 5-year event-free survival (EFS) and disease-specific survivals (DSS). Among the six candidate proteins in the GMC cohort, survival analysis validated NNT as the sole protein with a significant association with patient survival. In the OT group, survival analysis revealed that the rate of 5-year EFS was lower in the lower expression subgroup (NNTLow) compared to that in the higher expression subgroup (NNTHigh), and the HR of NNTLow group versus the NNTHigh group was 2.20 in the multivariate model (CPH model adjusted for age and sex, 95% confidence interval [CI]: 1.01–4.80, p < 0.05). In the HT group and the entire cohort, the 5-year EFS did not differ between the NNTLow and NNTHigh subgroups. In addition, the other five candidates showed no difference in the 5-year EFS and DSS (Fig. 2C and D), regardless of expression and BMI status, suggesting that NNT may be a specific prognostic biomarker for overweight CRC patients.

Fig. 2

Discovery and clinical validation of new biomarker candidate specific for obesity-associated colorectal cancer. A Representative immunohistochemical staining of six candidate proteins—NNT, PANK3, PPARGC1B, RTL6, FAM220A, and TMEM9—in colorectal cancer tissue. Staining intensity and distribution were scored semi-quantitatively as follows: score 0, faint and rare immunostaining; score 1 + , focal and weak staining; score 2 + , moderate and diffuse staining; and score 3 + , strong and diffuse staining. B–C 5-year event-free survival (B) and 5-year disease-specific survival (C) according to the expression of six candidates (NNT, PANK3, PPARGC1B, RTL6, FAM220A, and TMEM9) in the group OT, group HT, and all patients with CRC; FAM220A, family with sequence similarity 220 member A; NNT, Nicotinamide nucleotide transhydrogenase; PANK3, pantothenate kinase 3; PPARGC1B, peroxisome proliferator-activated receptor gamma coactivator 1-beta; RTL6, retrotransposon Gag like 6; TMEM9, transmembrane protein 9

Survival impact of NNT expression in obesity-associated CRC

We first evaluated the prognostic relevance of TNM stage in obesity-associated CRC (OT) patients. As expected, patients with late-stage disease (stage III–IV) had significantly poorer 5-year EFS and disease-specific survival DSS compared to those with early-stage disease (stage I–II). In multivariate analysis, the HR for EFS in stage III–IV versus I–II was 12.07 (95% CI: 4.28–34.08, p < 0.001), and the HR for DSS was 8.33 (95% CI: 2.90–23.90, p < 0.001) (Fig. 3A and B). Importantly, the individual prognostic effects of NNT expression and TNM stage are presented separately in Figs. 2 and 3 (for NNT) and Figs. 3A–B (for TNM), both demonstrating significant associations with survival.

Fig. 3

Multivariate test of NNT in the GMC cohort. A–B 5-year event-free survival (EFS) (A) and 5-year disease-specific survival (DSS) (B) considering only tumor node metastasis (TNM) stage in the OT group. Survival analysis with TNM stage alone. C–D GMC cohort multivariate test of 5-year EFS (C) and 5-year DSS (D) combined with NNT and TNM staging system of OT patients. E–F 5-year EFS (E) and 5-year DSS (F) according to NNT expression at the same TNM stage (III-IV). Panels A–B illustrate the individual prognostic impact of TNM stage alone, while panels C–F show the combined analysis integrating NNT expression with TNM staging

Given that NNT expression was independent of conventional clinicopathologic factors, we investigated whether combining NNT expression with TNM staging could improve prognostic stratification in obesity-associated CRC. To this end, patients were categorized into four groups: Group 1 (high NNT expression and early-stage disease [Stage I–II]), Group 2 (low NNT expression and early-stage), Group 3 (high NNT expression and advanced-stage [Stage III–IV]), and Group 4 (low NNT expression and advanced-stage). Multivariate survival analysis revealed a clear and stepwise stratification of survival outcomes across these four groups. Group 1 exhibited the most favorable prognosis, followed by Group 2 and Group 3, while Group 4 demonstrated the worst 5-year event-free survival (EFS) and disease-specific survival (DSS). In particular, Group 4 showed a markedly elevated risk of adverse outcomes compared to Group 1, with an EFS hazard ratio (HR) of 21.20 (95% CI: 2.88–156.08, p < 0.01) and a DSS HR of 14.05 (95% CI: 1.89–104.38, p = 0.01) (Fig. 3C and D). These results demonstrate that integrating NNT expression with TNM stage enables more refined and clinically meaningful risk stratification, allowing for a more accurate prediction of prognosis than either variable alone.

Within the subgroup of patients with advanced-stage disease (TNM stage III–IV), further stratification by NNT expression revealed significant survival differences. Patients with low NNT expression (Group 4) showed markedly lower 5-year event-free survival (EFS) compared to those with high NNT expression (Group 3), with a hazard ratio (HR) of 2.52 (95% CI: 1.09–5.82; p = 0.030). A similar trend was observed for disease-specific survival (DSS) (HR: 2.32, 95% CI: 0.84–5.92), though not statistically significant (Fig. 3E and F). These results indicate that NNT expression stratifies prognosis even within the same TNM stage, suggesting that its combination with staging criteria may enhance prognostic resolution in patients with obesity-associated CRC.

Spatially resolved metabolic and signaling reprogramming associated with NNT expression in obese CRC

To elucidate spatial transcriptomic changes associated with NNT, we performed GeoMx DSP on FFPE samples from the validation cohort, profiling both tumor and adjacent normal regions across BMI-defined groups (AN, ON, HT, and OT), and stratifying tumors into NNTHigh and NNTLow where indicated (Fig. 4A, B; Supplementary Fig. S1). To further delineate obesity-related transcriptional differences, NNT expression was compared across four spatial groups (AN, ON, HT, and OT) representing adjacent normal and tumor tissues stratified by BMI. A significant group-level difference was observed (ANOVA p < 0.0001; Supplementary Fig. S1A), showing a progressive decrease from AN → ON → HT → OT. This finding indicates that obesity-related downregulation of NNT emerges already in adjacent normal tissues and becomes more pronounced in tumors. In NNTHigh tumors (Fig. 4C), transcripts involved in metabolic and apoptotic pathways—including PAM, EFNA1, NT5E, MGLL, GULO, ALAD, TXNIP, LGALS3, TGFBR3, and IER3—were significantly upregulated, suggesting increased glycolytic activity, enhanced fatty acid metabolism, and heightened apoptotic signaling. In contrast, genes associated with cellular proliferation and poor prognosis—such as RBM3, NEDD4, KPNB1, ROMO1, E2F1, KHDRBS3, CEBPG, SNRPA, SND1, and DDX18—were markedly downregulated. These transcriptomic shifts indicate that NNTHigh tumors adopt a metabolically active yet less proliferative tumor phenotype.

Fig. 4

Analysis of NNT expression in obese colorectal cancer (CRC) patients using GeoMx spatial transcriptomics technology. A GeoMx analysis of FFPE tissue from obese CRC patients (OT group): Formalin-fixed paraffin-embedded (FFPE) tissue was collected from obese CRC patients, and GeoMx technology was used for spatial transcriptomics. The tumor epithelial regions were isolated based on DNA and PanCK markers to ensure accurate targeting of the tumor cells. B Boxplot of NNT expression groups: The patients were divided into high NNT expression and low NNT expression groups. C Volcano plot of gene expression differences between NNT high and low groups. D GSEA analysis results: The normalized enrichment scores (NES) derived from GSEA analysis are presented. E Heatmap of gene expression related to metabolic pathways, such as glycolysis, fatty acid metabolism, scavengers of reactive oxygen species (ROS), apoptosis, Wnt /β-catenin signaling, and c-Myc signaling. F Boxplots of ACO2, PRDX6, TXNIP, and DDX18 genes

Pathway-level analysis revealed selective enrichment of hallmark programs related to epithelial–mesenchymal transition, hypoxia, fatty acid metabolism, glycolysis, heme metabolism, UV response, and apoptosis in NNTHigh tumors (Fig. 4D). Conversely, oncogenic pathways including Wnt/β-catenin and c-Myc signaling were significantly suppressed. These spatially confined alterations suggest that elevated NNT expression fosters a redox-protective, metabolically reprogrammed microenvironment that may counterbalance tumor-promoting signals. Consistent with this interpretation, spatial mapping of the antioxidant response showed localized upregulation of reactive oxygen species (ROS)-scavenging genes—such as CAT, GPX1–3, SOD1–2, TXNIP, and PRDX6—within NNTHigh tumors (Fig. 4E). Notably, ACO2 (p = 0.0254), PRDX6 (p = 0.0379), and TXNIP (p = 0.0165) were significantly enriched in NNTHigh tumors, implicating enhanced mitochondrial metabolism and redox buffering capacity. In contrast, DDX18 (p = 0.0274) was predominantly expressed in NNTLow tumors, consistent with elevated proliferative potential (Fig. 4F). Within the HT (non-obese tumor) compartment, we further examined four NNT-associated genes—DDX18, PRDX6, TXNIP, and ACO2—to investigate context-specific redox regulation (Supplementary Fig. S1B). DDX18 was higher in NNT-high cases, PRDX6 showed a mild increase in NNT-high tumors, whereas TXNIP and ACO2 were unchanged. These data suggest that NNT selectively modulates antioxidant and metabolic genes, supporting a context-dependent adaptive redox response rather than uniform antioxidant activation.

Together, these spatially resolved transcriptomic insights reveal that high NNT expression reprograms discrete tumor niches toward enhanced energy metabolism and oxidative stress management, while concurrently dampening oncogenic signaling. These findings provide a mechanistic basis for the observed favorable prognostic impact of NNT in obesity-associated CRC.