A novel machine learning model based on ubiquitin-related gene pairs and clinical features to predict prognosis and treatment effect in colon adenocarcinoma

Colon cancer is the most prevalent malignancy affecting the gastrointestinal tract, with an estimated 101,420 newly diagnosed cases and 51,020 deaths in the United States in 2019 [1]. Colon adenocarcinoma (COAD) accounts for over 80% of all colon malignancies, with sarcomas and squamous cell carcinomas responsible for the remaining 20% [1, 2]. COAD is typically treated with surgery, chemotherapy, radiation therapy, immunotherapy, and other treatments. Cass et al. found that following complete primary resection, 37% of patients suffered a local recurrence and distant metastases, with the most prevalent contributor to mortality within 5 years being local recurrence in the absence of clinical indication of distant metastases [3]. Chemotherapy can be administered as adjuvant therapy after surgery or as neoadjuvant therapy before surgery in advanced COAD patients to help decrease the tumor. Despite this, 40–50% of advanced COAD patients die as a result of disease recurrence or metastasis [4]. As a result, a distinct COAD prediction signature is critical, with the potential to identify new therapeutic targets and prognosis markers.

Tumor cells can effectively change their microenvironment by producing a variety of chemokines, cytokines, and other substances. Immunotherapy is commonly utilized for the treatment of malignancies in humans. Over the last several decades, many inhibitory receptors have been shown to perform an integral function in dampening anti-tumor immune responses. Included among these are programmed death-ligand 1 (PD-L1), cytotoxic T lymphocyte-associated antigen-4 (CTLA4), and programmed cell death protein-1 (PD-1). COAD-related early treatment and first-line therapy include immune checkpoint blockade (ICB) to serve a greater number of patients. Other immune-associated indicators have also been discovered, opening the way for more efficacious immunotherapy and demonstrating immunotherapy's promise as a COAD treatment regime.

Ubiquitin and ubiquitin-like (UB/UBL) conjugations are essential post-translation modifications that are required for almost all biological activities and pathways, particularly protein breakdown and turnover, DNA damage repair, and cell cycle, as well as intercellular signal transmission. Ubiquitin is a protein that is evolutionary conserved and is known to modify proteins post-translationally either for degradation or non-degradative signaling. In addition, it is covalently linked to lysine residues sequentially by 3 enzymes, namely ubiquitin-activating enzymes (E1s), ubiquitin-conjugating enzymes (E2s), and ubiquitin-protein ligases (E3s). UB's C-terminus is initially triggered by an E1 activating enzyme before being transported to the catalytic domain of an E2 conjugating enzyme. Furthermore, an E3 ubiquitin ligase connects the target protein and the E2-ubiquitin intermediate to act as a catalyst for the creation of an isopeptide bond between the UB C-terminal glycine and substrate lysine [7, 8]. Deubiquitinases (DUBs), which are Ub-specific proteases, act as catalysts for the elimination of UB from substrates since it is a reversible post-translational modification. There are roughly 100 distinct DUBs in humans, which may be roughly divided into seven structurally diverse superfamilies. DUBs regulate key cellular functions by cleaving UB bound to substrates or inside UB chains, acting as either switch to eliminate UB signals or rheostats to fine-tune the amount and kind of ubiquitylation [9, 10]. Furthermore, proteins with ubiquitin-like domains (ULDs) and ubiquitin-binding domains (UBDs) perform an integral function in ubiquitination regulation [9, 11]. Numerous human illnesses, including cancer and neurodegenerative disorder, have been linked to protein ubiquitination dysfunction, according to research [12]. However, no research has looked into the link between ubiquitin-related genes (URGs) and COAD patients' prognosis.

In this research, we analyzed COAD patients' gene expression patterns and clinicopathological data to develop a unique 6-ubiquitin-related gene pairs (URGPs) profile for predicting prognoses and immune responses. Our URGPs signature will offer an insightful comprehension of the tumor immune milieu and the treatment efficacy in COAD.

The death rate associated with COAD is among the highest of all malignancies affecting the digestive system. It is more prevalent in guys older than 40. Nonetheless, early detection of COAD is extremely challenging, and the majority of patients who were identified with COAD have progressed malignancy, which resulted in a poor prognosis. It is possible to remarkably improve the COAD patients' prognoses through early detection and treatment, which also reduces the patients' financial burden and enhances their quality of life. Due to the advancement of RNA sequencing technologies, several molecular markers have been described as predictors of prognosis and therapy success in colorectal cancer (CRC). Recent research has found that epigenetic alterations, such as gene malfunction and aberrant expression, are implicated in the onset and advancement of numerous human cancers. Some research has looked into whether core epigenetic modification-related genes could be employed as cancer biomarkers. UB is the founding member of the structurally conserved protein family that is responsible for the modulation of a wide variety of functions in eukaryotic cells, particularly, substrate activation or inactivation, protein activation, and protein–protein interactions [22]. Protein ubiquitination and deubiquitination play a critical role in protein stability, localization, and signal pathway regulation, and disruptions in protein homeostasis can lead to a range of illnesses, including neurological disorders, autoimmune disorders, and cancers. It has been discovered that aberrant E3s expression and DUBs might impact human cancers by altering the activities of tumor-associated proteins. RING-finger E3 ubiquitin ligase MDM2 is primarily responsible for the monoubiquitination of p53, which influences p53 activity by modulating p53 localization and transcription functions. Numerous E4 ubiquitin ligases (E4s), which are responsible for the extension of these monoubiquitin chains, have also been identified [23, 24]. Nonetheless, only a limited number of ubiquitin molecules have been investigated in-depth, with the majority of studies focusing on the function of specific genes. Using expression profile information, few research reports have comprehensively investigated the molecular features and prognostic potential of URGs. This is the first research that we know that utilizes URGPs to design a risk model for predicting COAD patients' prognoses.

In this work, several URGs were identified by conducting an analysis of the COAD dataset included within the TCGA database. Following that, the gene modules associated with the onset and progression of COAD were filtered by WGCNA in a systematic manner. As an unsupervised algorithm, WGCNA can build a relationship between gene expression and clinical traits [25]. Rather than focusing solely on differentially expressed genes, WGCNA identifies gene sets of interest and performs extensive association analysis with phenotypes, transforming the problem of multiple hypothesis testing corrections by transforming the correlation of thousands of genes with phenotypes into the association of several gene sets. Then an in-depth examination of the genes contained within the module was performed. Finding the optimum approach still poses challenges despite the fact that numerous research have employed various machine learning techniques to choose the appropriate variables. In this investigation, we employed the most traditional, widely applied technique [26, 27].The LASSO regression technique is a penalized form of regression that reduces the size of certain coefficients to produce a more accurate model by the construction of a penalty function. It is an estimator that is biased because it processed data that have complicated collinearity. However, it is frequently employed in high-dimensional regression and may help compensate for the deficiencies of univariate Cox regression analysis [28, 29]. Following additional processing using LASSO regression, 6 URGPs linked to the advancement of the tumor were ultimately found. These gene pairs were subsequently subjected to the multivariate Cox regression to establish a risk model for predicting the patients' prognoses. Additionally, it was discovered through ROC curve validation of the model's effectiveness that its prediction capacity of COAD patients' survival over 1, 3, and 5 years in the dataset was moderately accurate. In summary, both the univariate and the final multivariate Cox regression analyses illustrated that the risk model independently functioned as prognostic indicators. We observed that the established nomogram premised on this model performed well when we calibrated it. Numerous earlier research reports have constructed risk models that could accurately anticipate the COAD prognosis, and the majority of these models incorporate multiple functional gene sets. For instance, Chen et al. developed and validated a COAD predictive risk model using a total of 8 lncRNAs that are associated with endoplasmic reticulum (ER) stress [30]. Rong H and colleagues discovered a novel genetic signature that is associated with the invasion of COAD. This study created a risk model for forecasting COAD prognoses as well as a nomogram based on this model to collectively examine the prognosis of patients with TNM staging, providing insights and guidance for fundamental COAD research [31]. Many of these studies have been reported in other types of cancer, such as lung adenocarcinoma (LUAD) [32, 33].

In our study, 6 URGPs were incorporated in our signature, of which OTUB2|DTX1, PSMD7|TFG, ATG3|ATG16L1, RASD2|BRSK2 were determined to be risk protective markers. The remaining 2 pairs (MINDY1|CSTF1, RASD2|WDR76) were determined to be risk-related indicators. Previous research has established that several genes within each of these gene pairs perform an integral function in the advancement of COAD. One study suggests that deubiquitinase OTUB2 exacerbates colorectal cancer growth by increasing PKM2 activity and glycolysis [34]. Lower levels of DTX1 could promote breast cancer (BC) cell proliferation and migration and are associated with advanced BC [35]. While MINDY1, a member of the motif interacting with Ub-containing novel DUB family, has been identified as a potential estrogen receptor α (ERα) deubiquitylase in BC. High MINDY1 expression was linked to a poor prognosis of BC [36]. In LUAD, PSMD7 expression was linked to not only tumor laterality, but also lymph node invasion. In LUAD patients, an elevated level of PSMD7 was linked to the unfavorable OS and disease-free survival, and PSMD7 silencing considerably attenuated the proliferative ability of cells and triggered the G0/G1-phase cell cycle arrest, cell senescence, and apoptosis [37]. ATG3 and ATG16L1 were key players with important roles in different stages of autophagy. The researchers Huang et al. found that overexpression of ATG3, which was caused by downregulation of miR-435-5p, increases proliferative and invasive capacities in CRC via an autophagy-dependent process [38]. The results from Florin et al. suggested that ATG16L1 T300A polymorphism may be associated with gastric carcinogenesis [39]. RASD2 encodes a Ras-related GTP-binding protein and involves in the development and metastasis of Uveal melanoma [40]. In CRC, RAS stabilization is a critical event for hyperactivation of Wnt/-catenin signaling and cancer stem cell activation. WDR76 has been shown to destabilize RAS and serves as a tumor inhibitor in CRC by suppressing cancer stem cell activation [41]. BRSK2, which belongs to the serine/threonine-protein kinase of the AMPK family, was recognized to be a risk factor for pancreatic ductal adenocarcinoma (PDAC). BRSK2 was induced by nutritional deprivation in PDAC cells, which inhibited TORC1 activity through tuberous sclerosis complex 2 (TSC2) phosphorylation [41].

This URGPs signature exhibited an excellent diagnostic capacity and may be utilized to distinguish COAD patients with an unfortunate prognosis, as determined by the survival and ROC curve analyses in the TCGA dataset. Additionally, the URGPs may predicted the OS of COAD patients in various clinical and pathological stratifications, and results showed that the signature was substantially linked to advanced clinical and pathologic stage. In the end, a nomogram was designed to fulfill the need for an easy-to-understand and practical scoring system, as well as to facilitate clinical decision-making. In addition, the gene functional enrichment analysis has shown that the URGPs are implicated in the onset and progression of COAD by engaging in a range of critical biological processes. Immunotherapy is a field that is now undergoing intensive research and innovation in COAD. Recent research has revealed that although TME performs an instrumental function in immunotherapy, the particular processes implicated are not yet completely understood [43]. Therefore, it is vital to conduct further studies on the involvement of the TME to increase the immunotherapeutic efficacy. COAD's TME consists of stromal cellular components, tumor cells, and immune cells. There is significant evidence that immune cells present in the TME impact carcinogenesis. The dysfunction of immune cells can have a range of outcomes, including those that are antitumorigenic or protumorigenic.

In this investigation, we adopted the CIBERSORT algorithms to ascertain the infiltration levels of immune cells correspondingly. In addition, we discovered that COAD samples having low-risk scores were linked to greater infiltration levels of follicular helper T cell, and CD4 memory activated T cell, as well as higher M1 macrophages. Immune checkpoint molecules, which function as inhibitory receptors, are detected on the immune cells' surface and are responsible for regulating the immune response. According to a growing body of research, the expression of immune checkpoint biological markers may serve as a positive predictive indicator for the effectiveness of immunotherapeutic interventions. As a result, we subsequently evaluated the relationship of the URGPs signature with the expression of immune checkpoint biological markers and discovered that COAD patients who were in the low-risk group exhibited elevated expression levels of immune checkpoint biological markers in contrast with those who were in the high-risk group. To summarize, "immune active" was the term used to describe the immunological landscape of low-risk tumors, which were characterized by a significant infiltration of immune cells. This was accomplished by cleaving UB that was attached to substrates or contained inside UB chains. In the regulation of key cellular functions, DUBs serve critical functions, functioning either as switches that eliminate UB signals or as rheostats to fine-tune the degree and kind of ubiquitylation that takes place. Furthermore, the immune milieu of high-risk cancers was described as "immune inactive" with limited infiltration levels. This suggests that the current URGPs signature may accurately predict the immunological aspects of COAD, and patients who are within the low-risk group have a greater likelihood of gaining benefit from anti-tumor immunotherapeutic intervention as opposed to those within the high-risk group.

According to our current knowledge, a predictive model that is premised on URGPs and the related nomogram in COAD are yet to be investigated. This model showed an excellent prediction accuracy, and it might help distinguish patients who have a high recurrence risk and choose the appropriate treatment. Nevertheless, there are a few drawbacks to consider. Firstly, this work is a retrospective examination of publicly available datasets, which makes it prone to bias. As a result, a large-sample prospective clinical investigation needs to be carried out to evaluate the robustness of the signature model. Secondly, further research needs to be done to better understand the specific molecular processes and biological activities of the URGs. Lastly, the methodology that is dependent on gene-level prognostic features to anticipate the cost of samples is expensive, and there is significant clinical promotion resistance.

Using bioinformatics approaches, we thoroughly evaluated the prognostic value of URGPs and developed a predictive model for COAD. The nomogram integrating clinicopathological data and the proposed URGPs signature can accurately predict COAD patients’ prognoses and aid doctors in the selection of individualized treatments. Additionally, it describes the connection between our signature and the sensitivity of immunological checkpoints as well as targeted medicines. These findings will aid in enhancing the predictive value of conventional clinical detection, which may be used to evaluate COAD clinical outcomes and contribute to precision medicine. Prospective research is required to confirm its validity.