Crohn’s disease (CD) is a subtype of inflammatory bowel disease (IBD) that has a chronic and relapsing-remitting course, which is influenced by genetic, environmental, and microbial factors (1, 2). The incidence of CD is increasing in emerging industrialized countries in Africa, Asia, and South America (3). Although anti-TNF drugs like infliximab are effective in treating the majority of Crohn’s disease patients (2, 4), up to 30% of patients experience primary non-response to infliximab therapy (5). This increases the risk of adverse drug reactions and imposes a significant economic burden on the patients. However, the reasons behind treatment failure have yet to be fully understood.

Angiogenesis is a characteristic feature of chronic inflammation (6), and microvascular angiogenesis is profoundly influenced by inflammation in IBD (7, 8). This process is closely associated with the pathogenesis of IBD and disease severity (9–11). Furthermore, vascular endothelial growth factor (VEGF), a protein related to angiogenesis, has been reported as a predictive marker for the response to infliximab therapy. Patients with higher baseline levels of VEGF in their blood exhibit a poorer response to infliximab treatment (12). Despite this, there is still a limited understanding of angiogenesis-related genes in the intestinal mucosa of CD patients.

In this study, we aimed to comprehensively investigate the relationship between angiogenesis-related genes and Crohn’s disease (CD). Based on differentially expressed angiogenesis-related genes, we divided CD samples into two clusters (C1 and C2) and analyzed their distinct biological characteristics and response to infliximab treatment. Using machine learning methods, we identified hub genes involved in disease occurrence, development, and angiogenesis. We then constructed predictive scoring systems for disease diagnosis and infliximab treatment response based on these hub genes. We hope that our research will contribute to the advancement of personalized treatment strategies for Crohn’s disease.

Crohn’s disease is a chronic heterogeneous disease with diverse manifestations and progressions, making its diagnosis inherently challenging (28). Moreover, patients often exhibit varied responses to therapeutic drugs, with around 30% of patients showing primary non-response to infliximab treatment (5). This subgroup of patients faces increased economic burden and risks of medication side effects. Therefore, selecting the most effective medications is of paramount importance for these patients.

In this study, we focused on angiogenesis as a starting point to explore the heterogeneity of CD patients and developed models for CD diagnosis and infliximab non-response based on hub genes. The findings of this study may contribute to a better understanding of Crohn’s disease among researchers and provide valuable insights for clinical treatment.

Based on the expression of angiogenesis-related genes, we successfully identified different clusters of Crohn’s disease (C1 and C2). Compared to the C1 cluster, the C2 cluster showed upregulation of multiple metabolic pathways, and poorer response to infliximab. Similar to our findings, Matthew Weiser et al. also identified two subtypes of Crohn’s disease (colonic and ileal), with the ileal cluster showing significantly upregulated metabolic pathways, including lipid metabolism (29). Studies have shown that endothelial cells (ECs) can metabolize fatty acids to acetyl-CoA to maintain the tricarboxylic acid (TCA) cycle and promote deoxynucleotide triphosphate (dNTP) synthesis, together with non-nucleotide substrates, to facilitate EC proliferation (30). This suggests that the Crohn’s disease cluster with active lipid metabolism may exhibit a higher degree of angiogenesis.

Further investigation revealed that the C2 cluster with active angiogenesis showed a higher rate of non-response to infliximab. This suggests that angiogenic responses may play a role in the mechanisms of infliximab non-response. A study by Alicia Algaba et al. found that elevated baseline levels of vascular endothelial growth factor (VEGF) are associated with poor response to anti-TNF treatment and VEGF can serve as a useful biomarker for anti-TNF treatment response (AUC=0.8) (12). This also suggests a potential connection between angiogenesis and anti-TNF treatment. In our exploration of the expression of angiogenesis-related genes in Crohn’s disease patients, we found upregulation of OSM, CXCL5, CEACAM3, ISG20, and DUOXA1 genes in Crohn’s disease patients. Among them, two genes, OSM and CXCL5, serve as good predictive markers for infliximab non-response (OSM, AUC=0.829; CXCL5, AUC=0.818). The finding was also validated in an external dataset, further supporting the relationship between angiogenesis and anti-TNF treatment response.

According to studies, Oncostatin M (OSM) belongs to the interleukin-6 (IL-6) family and has the same receptor subunit, gp130, as IL-6 and leukemia inhibitory factor (LIF) (31, 32). OSM has been found to induce angiogenesis both in vivo and in vitro, and its effect on angiogenesis is stronger than that of IL-6 and LIF. OSM can stimulate vascular smooth muscle cells to produce VEGF through various pathways such as PI3K, p38 MAPK, Erk1/2, STAT1, and STAT3, whereas other gp130 ligands have not been reported to regulate VEGF mRNA. However, it is important to note that promoting VEGF secretion is not the sole mechanism by which OSM promotes angiogenesis, as VEGF-neutralizing antibodies only inhibit 21% of OSM angiogenesis effects (33, 34).

The expression of the OSM gene is closely related to IBD. Nathaniel R. West et al. discovered that OSM is overexpressed in the tissues of active IBD patients and is strongly associated with the severity of tissue pathology (35), which is consistent with our findings. Furthermore, some studies also found a close correlation between baseline OSM expression and poor response to infliximab (36–38). Compared to responders to anti-TNF therapy, OSM is one of the top 20 genes most strongly expressed in non-responders to anti-TNF therapy. In addition, neutralizing OSM was found to significantly alleviate colitis in mice. Therefore, OSM has the potential to serve as both a biomarker and therapeutic target for IBD (35), which further bolsters our own research findings.

Similar to OSM, CXC motif chemokine ligand 5 (CXCL5) also plays a role in promoting angiogenesis (39). CXCL5, also known as epithelial neutrophil-activating peptide 78 (ENA-78), is a member of the ELR (+) CXC chemokine family and can mediate angiogenesis in the absence of inflammation. CXCR2 is the receptor for CXCL5 and mediates the angiogenic effects of ELR (+) CXC chemokines (40). Research has demonstrated that CXCL5 is involved in the angiogenic effects of multiple diseases. It orchestrates the modulation of FOXD1 expression through the activation of the AKT/NF-κB pathway, thereby fostering angiogenesis in colorectal cancer (41). It is also a crucial angiogenic factor in idiopathic pulmonary fibrosis (42). In gastric cancer, it can significantly impact the vascular distribution of tumors (43). In renal cell carcinoma, androgen receptor mediated CXCL5 signaling activates the AKT/NF-κB pathway to facilitate angiogenesis (43, 44). Blocking CXCL5 can effectively reduce tumors blood supply (45).

K Z’Graggen et al. found that CXCL5/ENA-78 is preferentially expressed in intestinal epithelial cells in IBD. Although CXCL5/ENA-78 protein is also present in some normal controls, CXCL5/ENA-78 mRNA is completely negative (46). In addition, the study also pointed out that CXCL5 may be involved in the early stages of IBD and may contribute to the continuous recruitment and activation of neutrophils (46). John H Kwon’s experiments showed that pretreatment with antisense oligonucleotides against lipopolysaccharide-induced CXC chemokine (LIX) reduced DSS-induced colitis in mice. LIX was identified as a murine homolog of CXCL5/ENA-78. This suggests that CXCL5 may be a potential therapeutic target for IBD (47). In our study, CXCL5 can also serve as a good predictive marker for infliximab non-response. This indirectly suggests a potential connection between OSM and CXCL5. Indeed, previous research on pneumonia has shown that OSM plays a significant role in inducing CXCL5 and can directly initiate the mRNA synthesis of CXCL5 (48). However, whether there is a connection between the two in CD patients still requires further experimental validation. In our study, the predictive model constructed by combining CXCL5 and OSM (AUC=0.876) outperforms the predictive capabilities of CXCL5 and OSM individually. Additionally, the predictive performance of this model is also superior to serum VEGF (AUC=0.8).

Furthermore, research has shown that CEACAM3 is significantly upregulated in pediatric Crohn’s disease patients’ intestinal biopsies. The study also found that various IBD triggers can regulate CEACAM3. This suggests that trigger-induced CEACAM3 may play a role in the pathogenesis of IBD (49). Although there is a lack of research on DUOXA1 in CD, literature search revealed that one of the splice variants of DUOXA1 can be functionally associated with DUOX2 (50), and in active ulcerative colitis, DUOX2 is significantly upregulated (51). Since Crohn’s disease and ulcerative colitis both belong to inflammatory bowel diseases, DUOXA1 has potential research value. ISG20 has been reported to be a major effector of innate immune responses against various pathogens, including viruses, bacteria, and parasites, and ISG20 is also involved in angiogenesis in liver cancer (52, 53). However, whether ISG20 is involved in the pathogenesis of Crohn’s disease requires further research.

The analysis of immune infiltration reveals that the C2 cluster has a higher abundance of M1 macrophages and plasma cells. According to the study by Renaud Gaujoux (54), the abundance of inflammatory macrophages and plasma cells is significantly higher in non-responders to anti-TNFα treatment compared to responders, and this difference persists after treatment. Previous studies have indicated that during intestinal inflammation, an increase in the abundance of inflammatory macrophages can lead to elevated secretion of TNFα at the site of inflammation (55), and elevated levels of TNFα can support the survival of plasma cells (56). Therefore, if both cell subtypes increase simultaneously, it may indicate that pathological tissue has a higher concentration of TNFα. Hence, the elevated levels of TNFα could be one of the reasons for the lower response rate in the C2 cluster.

In terms of clinical applications, a nomogram is a practical and intuitive tool widely employed in predictive models in the field of medicine (57). When prediction is required, one simply needs to locate the value of the corresponding gene on the nomogram, then use connecting lines to find the corresponding scale above, obtaining the individual gene’s points. By summing the individual gene points, the total points are obtained. Locating the corresponding scale for the total points provides the probability of the outcome occurrence. Our study includes two nomograms. For diagnosis, measuring the expression levels of the OSM, CXCL5, CEACAM3, ISG20, and DUOXA1 genes and using the first nomogram allow us to determine the probability of disease occurrence. This contributes to early disease diagnosis and provides opportunities for early intervention, thereby improving treatment outcomes. Meanwhile, the expression levels of OSM and CXCL5 can be further used in the second nomogram to predict the probability of non-response to infliximab. Based on the predictive results, doctors can adjust the treatment plan, such as adjusting dosage, incorporating other therapeutic medications, or changing biologics (58). This aids in avoiding ineffective treatment plans for patients, enhancing the individualized level of treatment, and reducing unnecessary drug exposure and potential side effects. Therefore, our research results are beneficial for both disease diagnosis and decision-making regarding subsequent treatment strategies. Furthermore, predictive models help increase patients’ understanding of their own diseases and treatment plans. Patients can actively participate in medical decision-making and gain a more comprehensive understanding of the risks and benefits of treatment plans.

In summary, our comprehensive analysis of angiogenesis-related genes in CD patients identified two distinct clusters. Further analysis revealed hub genes involved in the pathogenesis of CD, with OSM and CXCL5 being associated with non-response to infliximab. Simultaneously, we constructed disease diagnostic and infliximab non-response predictive models, validating their good performance. This may contribute to the diagnosis and personalized treatment of CD patients.

Although we have examined our research results using multiple datasets and achieved relatively satisfactory validation performance, our study still possesses certain limitations. Firstly, the sample size for investigating the response to infliximab treatment is limited, and the clinical information associated with the samples is inadequate, potentially impacting the precision of our study. Secondly, our study only included microarray data, which may impact the generalizability of the model. Thirdly, our research predominantly relies on database samples and bioinformatics analysis, necessitating further prospective experiments to enhance the robustness of our research findings. Lastly, we observed a correlation between angiogenesis-related genes and the non-response to infliximab in CD patients. However, whether there is a causal relationship between angiogenesis response and the therapeutic effectiveness of infliximab still requires further experimental confirmation.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material .

CZ proposed the research topic and designed the study protocol. XC were responsible for the paper writing. YK collected and analyzed the data. XX contributed to the data visualization. CW and LJ reviewed and revised the manuscript. All authors contributed to the article and approved the submitted version.