The gene expression profiles and corresponding clinical data were attained from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/). A total of 232 autophagy-related genes were extracted from the Human Autophagy Database (HADb, http://www.autophagy.lu/index.html). The platform information corresponding to all data sets in the present study was shown in Table 1. Data analysis was performed with the R software (version 4.0.2). The “sva” R package was used to remove batch effects between different data sets, and then the merged data were normalized. The principal component analysis (PCA) was used to verify the results of batch correction.

The relative abundances of 22 immune cells in the intestinal mucous tissue of active UC and healthy control were inferred by the “CIBERSORT” algorithm (Newman et al., 2015), which applies a matrix of reference gene expression values (LM22 file) to represent the minimum values of different cell types. The deconvolution algorithm was performed by analyzing 1,000 permutations and the LM22 file, which was downloaded from the CIBERSORT web portal (https://cibersort.stanford.edu/). The screening condition for qualified samples was p < 0.05. The results of the immune infiltration patterns were visualized using the “barplot” package, “corrplot” package, and “pheatmap” package in R.

Based on the original data, we utilized the “limma” R package to filter the differentially expressed genes (DEGs) between active UC samples and healthy samples, with adjusted p-value (adj. p) < 0.05 and |log2 fold change (FC)| > 1 as thresholds. The R packages “ggplot2” and “pheatmap” were used to visualize the results of differential expression analysis.

WGCNA, a systematic biological method that uses gene expression data to construct scale-free networks, can search gene modules with cooperative expression and identify correlations between modules and phenotypes (Langfelder and Horvath 2008). We constructed a co-expression network using the “WGCNA” R package. At first, we established a weighted adjacency matrix and the soft threshold power (β) was 13. Then the adjacency matrix was converted to a topological overlap matrix (TOM) to prepare for hierarchical clustering analysis (Ravasz et al., 2002). The gene modules were identified using a dynamic tree cut algorithm with a minimum size of 30, and gene modules were clustered according to the cutoff value of 0.4. Genes within an identified module had similar expression patterns and were strongly related to the occurrence of active UC.

To further understand the underlying pathomechanism of active UC, the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were performed using the “clusterProfiler” package based on the DEGs between healthy control and active UC.

The data in GSE114527, GSE92415, and GSE38713 were statistically analyzed with Graphpad Prism 8 (Graphpad Software Inc., United States). All data were represented as the mean ± SE for quantitative variables. One-way ANOVA was used for more than two comparison groups. All other statistical analysis was implemented by R (v4.1.0, http://www.R-project.org), and p < 0.05 was considered statistically significant.

The analysis flow chart of this study was shown in Figure 1. The expression profile data of GSE53306, GSE87466, and GSE134025 downloaded from the GEO database were preprocessed and evaluated, and a total of 36 healthy controls and 46 active UC samples were obtained for subsequent analysis. The results of PCA revealed that the batch effect between three data sets was eliminated (Figures 2A,B).

Various microbes, damaged organelles, and aggregates are some of the sources of inflammatory signals (Deretic et al., 2013; Deretic and Levine 2018). The cytoplasmic clearance of autophagy is crucial to the normative development, differentiation, and function of immune cells (Morishita and Mizushima 2019). UC is a chronic inflammatory intestinal barrier disease and mucosal immune dysregulation may be related to the pathogenesis of UC (Kobayashi et al., 2020). Therefore, to clarify the pathogenesis of active UC, we compared the difference of 22 immune cells infiltrating in intestinal mucosal of healthy controls and active UC patients through the CIBERSORT algorithm (Figures 2C–F). Filter out samples with p > 0.05, and finally obtain 43 active UC and 21 healthy controls from the total set. The histogram visualized the composition of the immune cell populations in each sample of active UC and healthy controls (Figure 2C). Figures 2D,F showed the abundance difference in immune cell infiltration of intestinal mucosal between active UC and healthy control. The immune cells with higher infiltration levels in the intestinal mucosa of patients with active UC than that of healthy controls include activated mast cells (p < 0.001), CD4⁺ naive T cell (p = 0.003), activated CD4⁺ memory T cells (p = 0.001), follicular helper T cells (p < 0.001), M0 macrophages (p = 0.002), M1 macrophages (p = 0.029), activated dendritic cells (p = 0.02), and neutrophils (p < 0.001). However, healthy controls were in possession of higher levels of infiltration of CD4⁺ memory resting T cells (p < 0.001), resting dendritic cells (p = 0.03), resting mast cells (p < 0.001), and M2 macrophages (p < 0.001) in intestinal mucosa. Besides, there was a weak correlation between the components of different immune cell infiltration (Figure 2E). Activated mast cells were slightly closely related to other immune cells. It had a weak negative correlation with M2 macrophages (r = −0.61) and CD4⁺ memory resting T cells (r = −0.52) and a weak positive correlation with neutrophils (r = 0.56).

Differential expression analysis was performed based on the expression profile of 36 healthy controls and 46 active UC patients from three different data sets (GSE53306, GSE87466, and GSE134025). We obtained a total of 423 DEGs, including 256 upregulated DEGs and 167 downregulated DEGs. The volcano plot and hierarchical clustering heatmap plot showed the distribution of DEGs between healthy controls and active UC (Figures 3A,B). The heatmap saw the expression pattern of the top 50 DEGs between active UC patients and healthy controls (Figure 3B).

Then, GO and KEGG enrichment analyses were performed on the 423 DEGs to get a better understanding of the potential molecular mechanisms and functions of active UC. Most of the GO terms of biological process (BP) were focused on the migration and chemotaxis of immune cells, such as neutrophils, myeloid leukocytes, and granulocytes (Figure 3C). Autophagy-mediated lipolysis was critical for neutrophil self-renewal by supplying free fatty acids to the mitochondrial respiration pathway necessary for neutrophil differentiation (Riffelmacher et al., 2017). Other significantly enriched GO terms were also compactly related to immune cell adhesion and immune response. Moreover, the enriched KEGG pathway showed in Figure 3D. A great majority of inflammation-related pathways were activated at the active UC group, including cytokine-cytokine receptor interaction, IL-17 signaling pathway, and TNF signaling pathway. More importantly, several pathways associated with autophagy were enriched, such as the PI3K-Akt signaling pathway and NF-kappa B signaling pathway. It was worth mentioning that the aberrant activation of inflammasome caused by autophagy-related protein deficiency was pivotal for the progression of inflammatory disease (Takahama et al., 2018).

The WGCNA network could divide genes into different modules with similar biological functions (Bryan et al., 2018). We identified four modules in the module classification through WGCNA (Figure 3E). Based on the module eigengene (ME), and the turquoise (r = −0.79, P = 8e − 19) and blue (r = 0.74, P = 2e − 15) modules with higher connectivity were selected as featured modules (Figure 3F). This also indicated that the turquoise module was highly negatively correlated with the occurrence of active UC, while the blue module was strongly positively correlated with active UC. However, the other two modules had a correlation of less than 0.5 with active UC.

To further narrow the range of autophagy-related hub genes, we used a Venn diagram to obtain the common genes of autophagy-related genes and DEGs with the genes in blue and turquoise modules, respectively (Figures 4A,B). Three hub genes were obtained from both two Venn analyses. Based on these results, we concluded that CASP1, SERPINA1, and CCL2 were autophagy-related hub genes with a remarkable positive correlation with active UC (Figure 4A), while IL24, NAMPT, and CASP4 were autophagy-related hub genes with a prominent negative correlation with active UC (Figure 4B).

Subsequently, we verified the differential expression of these six hub genes by another two data sets downloaded from GEO. GSE107499 contained an expression profile from inflamed and non-inflamed colon tissue from patients with active UC. Figure 4C revealed that CASP1 (p = 2.9e − 11), SERPINA1 (p = 3.5e − 14), CCL2 (p = 4.4e − 15), IL24 (p = 1.8e − 10), NAMPT (p = 1.5e − 08), and CASP4 (p = 2.22e − 16) were all upregulated in the lesion colon tissue of active UC patients. Furthermore, the expression of these six hub genes of active UC patients was higher than inactive UC patients and healthy controls in GSE59071 (Figure 4D). It was worth noting that, except for SERPINA1 (p < 0.01), there was no statistically significant difference in the expression of the other five hub genes in the intestinal mucosal tissues of healthy controls and inactive UC patients. Given the analysis results of WGCNA, the genes in the turquoise module were negatively correlated with active UC, and the genes in the blue module were positively correlated with active UC. Therefore, CASP1, SERPINA1, and CCL2 in the blue module were upregulated in the active UC samples, which was logical and could be used for further analysis.

According to the median expression levels of CASP1, SERPINA1, and CCL2, 46 active UC patients in the comprehensive data set were divided into high and low expression groups. The heat map showed the top 20 genes with significant differences between the high and low expression groups based on the screening threshold as |log2FC| > 1 and p < 0.05. We could conclude that CASP1 (Figure 5A), SERPINA1 (Figure 5B), and CCL2 (Figure 5C) could divide active UC patients into two groups with significantly different expressions.

GSE38713 was a data set containing whole-genome transcriptional analysis data of colonic biopsies from patients with active and inactive UC treated with 5-aminosalicylic acid (5-ASA) and azathioprine. It contained a total of 43 biopsies: 13 healthy controls, eight inactive UC, seven non-involved active UC, and 15 involved active UC. The inclusion and exclusion criteria for active and inactive UC patients were strict. We compared the expression of autophagy-related hub genes in the above samples. We found no difference in the expression of CASP1, SERPINA1, and CCL2 in involved intestinal mucosa of active UC patients after treatment with 5-ASA or azathioprine (p > 0.05, Figure 5D), which was still significantly higher than that of healthy controls. And the expression of CASP1, SERPINA1, and CCL2 showed no difference with or without drug treatment (p > 0.05, Figure 5D).

UC is a chronic relapsing gastrointestinal disease, and the alternation of remission and relapse is an important feature of it (Halmos and Gibson 2015). When treated with 5-ASA (Figure 5E) or azathioprine (Figure 5F), the expression of CASP1, SERPINA1, and CCL2 in the non-involved intestinal mucosa of active UC patients were not different from that of patients in remission, but significant differences were found between the involved and non-involved intestinal mucosa of active UC patients. In addition, after 5-ASA or azathioprine treatment, the expression of CASP1 and SERPINA1 in involved intestinal mucosa of active UC was significantly higher than that of UC patients in remission, but the expression of CCL2 was not different between the two groups (p > 0.05).

According to these results, we could speculate that the intestinal mucosal invasion of active UC patients may lead to the defect in autophagy, but not related to the treatment of 5-ASA or azathioprine. What’s more, no or less impaired autophagy may occur in the non-involved intestinal mucosa of active UC patients and remission patients.

Golimumab is a fully human IgG1 Kappa monoclonal antibody against tumor necrosis factor (TNF)-α (Sandborn et al., 2014). Subcutaneous injections of golimumab can induce a clinical response, remission, and mucosal healing, and improve quality of life in patients with active UC. It has been used in clinical practice for many years in adult patients with UC. In GSE92415, the expression of CASP1, SERPINA1, and CCL2 in the intestinal mucosa of active UC patients in the golimumab response and non-response groups was not different compared to the placebo group before golimumab treatment, which was higher than that in the healthy control group (Figure 6A). However, after 6 weeks of golimumab treatment, the expressions of CASP1 were still significantly higher than those of healthy controls (p < 0.0001, Figure 6B). It is worthwhile mentioning that the expression of SERPINA1 in the golimumab response group was significantly reduced (p < 0.05). The expression of CCL2 was dramatically reduced, and there was no statistical difference with the healthy control group (p > 0.05). While the placebo group showed the same changes. Therefore, after the treatment of golimumab, whether CCL2 was involved in the autophagy repair of impaired intestinal mucosal in active UC patients remained to be further studied. Besides, the expression of CASP1 and CCL2 in the intestinal mucosa of active UC patients with or without golimumab treatment for 6 weeks in the golimumab response group did not change. And the same results were also observed in the non-responsive groups. It is important to highlight that the expression of SERPINA1 was significantly down-regulated in the golimumab response group after receiving golimumab treatment for 6 weeks (Figure 6C).

GC can prevent TNF-induced increase intestinal permeability and fatal shock, and also plays an essential role in the treatment of active UC (Ballegeer et al., 2018). In GSE114527, there was no difference in the expression of CASP1 in the intestinal mucosa of active UC patients in the GC-responsive and non-responsive groups (Figure 6C). After 3 days of GC treatment, there was little change in CASP1 expression. For SERPINA1, after the treatment of GC for 3 days, there was no difference between the GC response group and the healthy control. In the GC non-responsive group, SERPINA1 expression in the intestinal mucosa of active UC patients was not different before and after GC treatment for 3 days, which was always higher than that of healthy controls (p < 0.05). These results indicated that the response to GC therapy could improve the autophagy damage of intestinal mucosa of active UC patients through SERPINA1 involved pathways. It must also be mentioned that CCL2 did not differ across all groups (p > 0.05).

Based on our findings, it can be concluded that among GC or golimumab responders, repair of impaired autophagy of intestinal mucosal in patients with active UC depends on SERPINA1.