The immune-infiltrating landscape in healthy control, UC_inactive and UC_active groupsFigure 1 illustrates the study process in the form of a flowchart.Figure 1Flowchart of the the research (by Biorender).Initially, we gathered data from three GEO cohorts (GSE48958, GSE53306, and GSE75214), which included bulk RNA sequencing information from healthy controls, inactive ulcerative colitis (UC), and active UC samples. To ensure accuracy and reliability, we merged the three GEO cohorts after correcting for batch effects, resulting in a comprehensive cohort of 31 healthy controls, 41 inactive UC, and 97 active UC samples (Fig. 2A,B). PCA indicated that the active UC samples were notably distinct from the other two groups (Fig. 2C). We then employed the CIBERSORT algorithm to compare the differences in immune cells among the three groups. Samples with p > 0.05 were considered reliable, and we obtained the relative percentages of 22 immune cells in 3 healthy controls, 10 inactive UC, and 68 active UC samples (Fig. 2D). Figure 2E depicted the correlations between the 22 immune cells, revealing that M2 macrophages were negatively correlated with activated mast cells (R = − 0.6), while follicular helper T cells were positively correlated with naïve B cells (R = 0.59) and resting dendritic cells (R = 0.65). Furthermore, CD8 + T cells were positively correlated with regulatory T cells (Tregs) (R = 0.61) and activated natural killer (NK) cells (R = 0.51). Finally, we compared the immune cell infiltrations and immune functions between the groups, as shown in Fig. 2F,G. As expected, the active UC group exhibited the highest level of immune cell infiltration and activation of several immune functions.Figure 2The immune-infiltrating landscape in healthy control, UC_inactive and UC_active groups. (A,B). PCA analysis showing the distribution of samples in GSE48958, GSE53306 and GSE75214 before (A) and after (B) batch effect correction. (C) PCA analysis showing the distribution of samples in healthy control, UC_inactive and UC_active groups. (D) The relative percent of 22 immune cells in each sample in the total set. (E) The correlation analysis between 22 immune cells based on the total set. Red and blue indicates a positive or negative correlation, respectively. The darker the color, the stronger the correlation. (F) Comparison of immune cell infiltrations in healthy control, UC_inactive and UC_active groups. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (G) Immune function analysis between healthy control and UC_active groups (up), UC_inactive and UC_active groups (down). (R software (version 4.2.1) (https://mirrors.tuna.tsinghua.edu.cn/CRAN/)).Functional enrichment analysis and polyamine metabolic reprogramming in healthy control, UC_inactive and UC_active groupsWe employed a filtering approach, selecting differentially expressed genes (DEGs) with an adjusted p-value < 0.05 and |log2 FC|> 1 in both cohort A (consisting of healthy controls and active UC samples) (Fig. 3A,B) and cohort B (comprising inactive UC and active UC samples) (Figure S2). GO analysis revealed that immune-related pathways, such as cytokine-mediated signaling pathway, leukocyte migration, secretory granule membrane, secretory granule lumen, immune receptor activity, and cytokine activity, were enriched in the UC_active group (Fig. 3C,E). Similarly, KEGG analysis demonstrated enrichment of cytokine-cytokine receptor interaction, PI3K-Akt signaling pathway, and chemokine signaling pathway in the UC_active group in both cohort A (Fig. 3D) and cohort B (Fig. 3F).Figure 3Differential expression analysis and functional enrichment analysis between UC_active and healthy control group. (A,B). Volcano plots showing the DEGs between UC_active and healthy control (A) or UC_inactive group (B). (C,D) GO (C) and KEGG (D) analysis between healthy control and UC_active group. (E, F) GO (E) and KEGG (F) analysis between UC_inactive and UC_active group.Apart from immune-related factors, nutrients and metabolites are also significant regulators of immune cell function25. When immune cells are activated, they undergo metabolic changes to meet the elevated energy requirements and sustain their immune functions26. The interplay between immune cells and metabolism orchestrates both innate and adaptive immunity26,27. Therefore, we analyzed the metabolic pathway profile between the UC_active and healthy control groups (Fig. 4A) or UC_inactive groups (Fig. 4B). We observed changes in numerous metabolic pathways in the UC_active group, including classical energy metabolism pathways, such as citric acid cycle, fatty acid degradation, and glycogen degradation. Interestingly, we also noticed that the polyamine biosynthesis pathway was active in the UC_active group. Recent studies have highlighted the significant role of polyamines in mammalian immune systems, particularly inflammation, although the underlying mechanism remains controversial 9. Hence, we selected polyamine metabolism for further investigation.Figure 4The difference on metabolic molecular signatures between UC_active and healthy control (A) or UC_inactive group (B). (R software (version 4.2.1) (https://mirrors.tuna.tsinghua.edu.cn/CRAN/)).WGCNA to identify hub modules and hub genes associated with active UCWe performed WGCNA to identify hub modules and hub genes associated with active UC. After clustering analysis, no outlier sample was detected in the total set (Figure S3A,B). We selected 13 as the soft threshold power (β) in both cohort A and cohort B based on the scale-free topology model fit and mean connectivity (Figs. 5A and S4A). In cohort A, we identified 6 modules (Fig. 5B), and the red module had the strongest negative correlation with active UC (R = − 0.79, p < 0.001) (Fig. 5C). Therefore, we designated the red module as the hub module, and filtered out hub genes based on MM > 0.8 and GS > 0.5 (Fig. 5D, Table S4). In cohort B, we also identified 5 modules (Figure S4B), and the green module had the strongest positive correlation with active UC (R = 0.66, p < 0.001) (Figure S4C). Similarly, we defined the green module as the hub module in cohort B, and identified hub genes in the right upper quadrant of the module eigengene heatmap (Figure S4D, Table S5).Figure 5WGCNA to identify hub modules and hub genes based on healthy control and active UC. (A) Selection of the soft-thresholding powers. The left panel shows the relationship between soft-thresholding powers and scale free topology model fit, while the right shows the relationship between soft-thresholding powers and mean connectivity. (B) Module clustering dendrogram based on a dissimilarity measure (1-TOM). Different modules are colored differently. (C) Heatmap of the correlation between each module and trait. Red and blue indicates a positive or negative correlation, respectively. The darker the color, the stronger the correlation. (D) Scatter plot of the relationship between module membership (MM) and gene significance (GS) in red module. Genes in right upper quadrant are identified as hub genes in red module.Identification and verification of the key PRGs associated with active UCTo identify key PRGs associated with active UC, we intersected DEGs, hub genes, and PRGs, resulting in 5 key genes in cohort A (Fig. 6A) and 2 key genes in cohort B (Figure S4E). We subsequently verified the expression of these key genes among healthy control, UC_inactive, and UC_active groups in the total set (Fig. 6B). The expressions of TGM2 and TRIM22 sequentially increased in healthy control, UC_inactive, and UC_active groups, while PPARG sequentially decreased. Additionally, the expressions of NNMT and PTGS2 in UC_active group were significantly higher than those in healthy control and UC_inactive groups. ROC curves showed that each key PRG had a significant clinical diagnostic value for active UC (Figs. 6C, S4F). We further confirmed these results with external validation sets GSE87473 and GSE107499, where differential expression analysis and ROC curves supported the key genes as biomarkers indicating active UC (Figure S5). We also investigated the relationships between key genes and immune infiltrating cells based on the total set. As shown in the heatmap (Fig. 6D), NNMT, PTGS2, TGM2, and TRIM22 had a positive association with most immune infiltrating cells, while PPARG had a negative association. Among the 5 key genes, NNMT and PTGS2 had much stronger correlation with immune cells, particularly type I Tregs, natural killer T cells, neutrophils, mast cells, and immature dendritic cells. These findings suggest that the identified 5 key genes could serve as potential biomarkers for active UC and are closely associated with immune infiltrating cells.Figure 6Identification and verification of the key PRGs in healthy control and active UC. (A) Venn diagram of the overlap among DEGs (between healthy control and UC_active group), hub genes in red module and PRGs. Genes in the center overlapping region are identified as key PRGs associated with active UC. (B) Differential expression analysis of key PRGs in healthy control, UC_inactive and UC_active groups. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (C) ROC curves of each key PRG for diagnosis of active UC. (D) Heatmap of the correlation between each key PRG and immune cells. Red and blue indicates a positive or negative correlation, respectively. The darker the color, the stronger the correlation.The role of key PRGs in therapeutic response of UC patientsAlthough significant advances have been made in the treatment of UC, there remains a group of patients who are unresponsive to existing therapies28. Therefore, we investigated the role of key genes in the therapeutic response of UC patients and evaluated their potential as biomarkers to improve drug sensitivity. Glucocorticoids (GCs) are essential drugs for combatting inflammatory diseases, and high doses of GCs are commonly administered to UC patients with severe symptoms or experiencing an acute relapse29,30. We compared the expressions of key PRGs in UC patients at different stages of treatment and found no significant difference in the intestinal mucosa of GC responders or non-responders before or after GC treatment (Figure S6A), suggesting that GCs may not affect these 5 key genes. With the advancements in therapeutic strategies for UC, we focused on emerging targeted biological agents31. Infliximab (IFX), the first TNF antagonist approved for treating UC, can induce and maintain remission and mucosal healing32,33. Golimumab, the third anti-TNF therapy for the treatment of moderate-to-severe UC, was also examined34. Vedolizumab (VDZ), a first-line biological targeting α4β7 integrin, was recommended for UC patients in the case of immunosuppressive or TNF antagonist therapy failure35. After treatment with IFX or VDZ, the expressions of all the key genes in responders were significantly different from those in non-responders, and the levels of expressions in responders after treatment were closer to those in healthy controls (Figs. 7, S6B). Specifically, compared with non-responders, NNMT, PTGS2, TGM2, and TRIM22 were remarkably decreased in the intestinal mucosa of responders receiving treatment, while PPARG was increased. However, despite being a TNF antagonist, golimumab had different effects on these 5 key genes compared to IFX. Differential expression analysis indicated that in golimumab-responders, the expressions of NNMT, TGM2, and PPARG were changed before and after golimumab treatment, while PTGS2 and TRIM22 did not differ across all groups (Figure S6C). In summary, the key genes remained unchanged with GCs treatment but showed variations with novel biological agents. This hints at a potential link between these genes and the effectiveness of biological agents, especially IFX and VDZ, in drug response. However, further experimental validation is needed to understand the underlying mechanisms.Figure 7Biological agents repaired intestinal mucosa of UC patients by regulating the key PRGs. (A,B) Differential expression analysis of key PRGs in different treatment groups with IFX (A) and VDZ (B) in GSE73661. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).Expression of key PRGs in colonic lamina propria immune cells of active UC by scRNA analysisTo further elucidate the associations between 5 key genes and active UC, we analyzed a single-cell RNA sequencing (scRNA-seq) database (GSE162335) consisting of information from colonic lamina propria immune cells from 11 active UC patients to identify the cell populations expressing key genes in active UC. Quality control indicated that all samples met the criteria, with more than 50 detected genes (nFeature) and less than 5% mitochondrial genes (Figure S7A). After preprocessing, a dataset comprising 18,179 high-quality cell samples was obtained, and a strong positive correlation was observed between nFeature and sequencing depth (total number of UMIs, nCount) (R = 0.86) (Figure S7B). Based on selected thresholds, 1500 variable genes were identified (Figure S7C), and PCA was utilized to decrease the dimensionality of the dataset by focusing on these variable genes (Figure S7D). After selecting the top 20 principal components to further reduce dimensionality, t-SNE method was used to cluster the cell samples, resulting in 20 subgroups (Figs. 8A, S8A). We calculated the contribution value of genes and defined marker genes for each subgroup, and the top 10 marker genes were visualized in the heatmap (Figure S8B).Figure 8The cell annotation and the expression levels of key PRGs in colonic lamina propria immune cell types in active UC. (A,B) Visualisation of clustering (A) and annotation (B) in the tSNE plot. (C) The tSNE plots showing distribution of key PRGs in different immune cell types. (D,E) The violin plots (D) and bubble chart (E) describing expression levels and expression proportion of key PRGs in different immune cell types, respectively.We then annotated and categorized cell samples into six cell types, including B cells, T cells, monocytes, common myeloid precursor cells (CMP), tissue stem cells, and macrophages (Fig. 8B). We found that T cells, B cells, and monocytes were the main immune infiltrating cells in the colonic lamina propria of active UC patients. We subsequently investigated the expression of key genes in different cell types (Fig. 8C). TRIM22 had the most extensive distribution, including B cells, T cells, monocytes, and tissue stem cells, while PPARG expression was low in all cell types and sporadically distributed in a subset of T cells (Fig. 8D,E). Significantly, PTGS2 was markedly enriched in CMP and monocytes, and NNMT and TGM2 were specifically expressed in tissue stem cells (Fig. 8D,E). These results highlight the close relationships between CMP, monocytes, tissue stem cells, and PRGs, and their crucial roles in colonic immunity, even though these cell types were present in small numbers.Verification of diagnostic efficiency of the key PRGs in colitis-associated CRC (CAC)Persistent inflammation is one of major causes of cancer development, which can lead to the formation of dysplastic precursor lesions in multiple regions of the colon via a phenomenon called field cancerization36. Given the increased risk of CAC development in patients with ulcerative colitis (UC), we evaluated the diagnostic potential of key genes for CAC.We first merged two Gene Expression Omnibus (GEO) cohorts containing healthy controls and CAC patients after batch effect correction, resulting in a comprehensive cohort comprising 13 healthy controls and 26 CAC samples (Fig. 9A,B). PCA showed marked differences between CAC samples and healthy controls (Fig. 9C). Metabolism analysis indicated that both CAC and UC shared multiple changes in comparison to healthy controls, particularly in the active polyamine biosynthesis pathway (Fig. 9D). Notably, the expressions of NNMT, PTGS2, TGM2, and TRIM22 were significantly upregulated in CAC tumor tissue, while PPARG was downregulated (Fig. 9E), mirroring the results obtained for UC. ROC curves were used to evaluate the diagnostic significance of key genes in distinguishing between healthy controls and CAC (Fig. 9F). All five key genes showed AUC values above 0.70, with NNMT, PTGS2, and TGM2 exceeding 0.85, suggesting their potential role in CAC diagnosis.Figure 9Verification of diagnostic efficiency of the key PRGs in CAC. (A,B) PCA analysis showing the distribution of samples in GSE4183 and GSE37283 before (A) and after (B) batch effect correction. (C) PCA analysis showing the distribution of samples in healthy control and CAC groups. (D) The differences on metabolic reprogramming between healthy control and CAC groups. (R software (version 4.2.1) (https://mirrors.tuna.tsinghua.edu.cn/CRAN/)). (E) Differential expression analysis of key PRGs between healthy control and CAC groups. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (F) ROC curves of each key PRG for CAC diagnosis.Validation of the role of polyamines and NNMT in inflammation and CAC in vitro and vivo modelsIn order to explore the relationship between polyamines and intestinal inflammation, we conducted experiments in both in vitro and in vivo models. We detected changes in mRNA and protein levels of polyamine metabolism enzymes in HT29 cells treated with LPS. As shown in Fig. 10A,B, the expression of polyamine metabolism enzymes changed after LPS stimulation, suggesting that inflammation can lead to polyamine remodeling. Inflammatory conditions upregulated the polyamine transporter ATP13A2 and the polyamine synthesis enzymes ADM1 and ODC1, while downregulating to some extent the polyamine degradation enzymes SMOX and PAOX. These findings suggest that cells promote polyamine synthesis and inhibit polyamine degradation in order to increase intracellular polyamine levels and alleviate damage caused by inflammation. To further investigate the impact of polyamines on UC, we supplemented exogenous spermine and spermidine to HT29 cells after LPS stimulation and evaluated changes in inflammatory cytokines (Fig. 10C,D). The results demonstrated that polyamine supplementation significantly reduced the levels of inflammatory cytokines, suggesting that polyamine levels can directly inhibit inflammation. Furthermore, we investigated the expression of five key genes in the cell inflammation model (Fig. 10E), mouse colitis model (Fig. 10F), and mouse CAC model (Fig. 10G). Consistent with previous findings, NNMT, PTGS2, TGM2, and TRIM22 were upregulated to varying degrees in colitis and CAC models, while PPARγ was significantly downregulated. (As mice do not have an ortholog for the human TRIM22 gene, only the mRNA of the other four genes was detected in the mice).Figure 10Verification of the role of polyamines and key PRGs in inflammation or CAC. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (A,B) The mRNA (A) and protein (B) levels of polyamine metabolic enzymes in HT29 treated with LPS. (C,D) The mRNA (C) and protein (D) levels of inflammatory cytokines in HT29 cells supplemented with exogenous spermine and spermidine following LPS stimulation. (E–G) The mRNA expression levels of key PRGs in HT29 treated with LPS (E), DSS-induced colitis model (F), and AOM/DSS-induced CAC model (G).Based on our previous findings (Figs. 6A and S4E), NNMT and PTGS2 have been identified as effective diagnostic markers for both UC and control groups, with good diagnostic performance in both active and non-active UC. Given the well-established role of PTGS2 in inflammation, we focused our investigation on the impact of NNMT on UC and CAC. We verified the knockdown efficacy of three siRNAs targeting NNMT using qPCR and western blot (Fig. 11A,B), and identified siNNMT1#1 and #2 as the most effective for subsequent experiments. In LPS-treated HT29 cells, we observed that knockdown of NNMT significantly reduced cell viability and increased LPS-induced cell death (Fig. 11C). Furthermore, we examined the expression levels of inflammatory factors (Fig. 11D) and found that IL6, TGFβ, and TNFα were upregulated following NNMT knockdown, indicating a potential anti-inflammatory role for NNMT. We also detected the expression of NNMT1 in an AOM/DSS-induced CAC model and observed a significant upregulation of NNMT in tumor tissues (Fig. 11E). Given our previous observation of NNMT enrichment in tissue stem cells, we investigated the effect of NNMT on cell stemness. Our analysis of classical stem cell biomarkers (Fig. 11F) showed that knockdown of NNMT resulted in downregulation of multiple stem cell markers, including SOX4, SOX2, Nanog, and CD44, emphasizing the crucial role of NNMT in maintaining cell stemness.Figure 11Verification of the role of NNMT in inflammation and cell stemness. (ns no significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001) (A,B). The mRNA (A) and protein (B) levels of NNMT in HT29 transfected with siNNMT. (C) The effect of NNMT on cell viability was evaluated using the CCK8 assay in HT29 cells treated with LPS. (D) The mRNA levels of inflammatory cytokines in HT29 cells transfected with siNNMT following LPS stimulation. (E) The protein expression levels of NNMT in CAC tissue were detected with IHC. (F) The protein expression levels of stemness-related genes in HT29 cells transfected with siNNMT.
