Fatty acid-binding protein 4 as a biomarker for colon adenocarcinoma risk and prognosis: Challenges and future directions

Abstract

In this letter, we have commented on the study by Zhang et al, which utilized bioinformatics and immunohistochemistry to assess the value of fatty acid-binding protein 4 (FABP4) as a biomarker for colon adenocarcinoma (COAD). Their findings improve our understanding of FABP4 in cancer cell adhesion and immune cell infiltration. However, differential expression analysis was insufficient to demonstrate a direct association between FABP4 expression and the occurrence and progression of COAD. Using Mendelian randomization for causal inferences can provide a solid biological foundation for model construction. Furthermore, integrating machine and deep learning approaches may yield more robust and precise prognostic outcomes than using a single Cox regression model. In addition, integrating genome-wide association study data to identify additional pathogenic genes involved in the regulation of fatty acid metabolism may facilitate the development of a multi-target strategy. This approach could potentially mitigate the compensatory effects associated with targeting FABP4 alone, and enhance therapeutic efficacy. Enhancing experimental validation would further improve the reliability of the results. With the continuous advancement of machine learning, multi-omics technologies, and experimental techniques, future studies may systematically integrate diverse sequencing datasets to offer novel insights into the early diagnosis, individualized treatment, and prognostic evaluation of COAD.

Key Words: Fatty acid-binding protein 4; Colon adenocarcinoma; Mendelian randomization; Machine learning; Deep learning; Multiple targets

Core Tip: Fatty acid-binding protein 4 (FABP4) is a promising biomarker and potential therapeutic target for colon adenocarcinoma diagnosis and prognosis. Our letter calls for a stronger biological rationale for FABP4’s role and advocates using Mendelian randomization to confirm its causal links. Incorporating machine learning and deep learning can yield more precise prognostic models. We also propose exploring multi-target strategies related to FABP4 and encourage future studies to combine experimental validation to reduce bioinformatics false positives. This allows us to elucidate underlying molecular biological mechanisms, offering fresh insights for personalized colon adenocarcinoma treatment.

TO THE EDITOR

Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second leading cause of cancer-related deaths worldwide[1]. The current staging system relies predominantly on pathological assessment while overlooking molecular characteristics, which may result in over- or under-treatment[2]. Colorectal adenocarcinoma (COAD) is the most common subtype of CRC[3]. Identifying novel biomarkers to guide the diagnosis and treatment of COAD could enhance personalized clinical outcomes. Fatty acid-binding protein 4 (FABP4), primarily expressed in adipocytes and macrophages[4], not only solubilizes various long-chain fatty acids and facilitates their intracellular transport, but also plays a critical role in regulating metabolism and inflammation[5]. Previous studies have suggested that FABP4 promotes tumor growth by increasing the fatty acid supply and modulating both inflammatory responses and insulin resistance. In this context, we reviewed the study by Zhang et al[6], which integrated bulk RNA-seq data from The Cancer Genome Atlas and Gene Expression Omnibus datasets and employed differential analysis, weighted gene co-expression network analysis (WGCNA), enrichment analysis, immune infiltration analysis, Cox regression, and immunohistochemistry (IHC) to establish FABP4 as a biomarker for COAD screening, auxiliary diagnosis, and prognosis. However, the study did not provide a biological rationale for using FABP4 to predict COAD risk. The accuracy of the prognostic model requires improvement, and specific strategies for targeting FABP4 are lacking, with insufficient experimental validation. Considering these issues, we propose several enhancements.

MENDELIAN RANDOMIZATION CLARIFIES THE BIOLOGICAL BASIS FOR RISK PREDICTION

Zhang et al[6] constructed a receiver operating characteristic curve and concluded that FABP4 expression level has a strong predictive value for COAD risk. However, the FABP4-related biological mechanisms identified in their study, such as promotion of cell adhesion and modulation of the tumor microenvironment, primarily pertain to COAD development. The lack of exploration of the mechanisms by which FABP4 induces COAD calls into question the validity of using its expression level as a risk predictor. Fatty acid metabolism is closely associated with precancerous lesions as it affects inflammatory responses, endoplasmic reticulum stress, and hormone regulation[7]. Leukotriene receptor antagonists have been shown to inhibit the formation and proliferation of abnormal crypt foci in the colonic epithelium, suggesting their potential in preventing COAD[8]. As an important transporter of arachidonic acid, a precursor of leukotrienes, FABP4 is possibly involved in the development of precancerous lesions and may be important for early prevention strategies for COAD. Incorporating mendelian randomization (MR) analysis to examine the causal link between FABP4 and COAD could provide robust biological support for the predictive value of FABP4.

The outcomes of MR analyses are often considered comparable to those of randomized trials because they employ instrumental variables to infer causal relationships between risk factors and outcomes[9]. Originally developed to investigate the connection between modifiable exposure or biomarkers and diseases, MR is now widely applied in this field[10]. Zhou et al[11] used multivariate MR to identify body fat percentage and omega-3, omega-6, and omega-3 to omega-6 ratios as potential mediators for both chronic obstructive pulmonary disease and CRC, suggesting a causal link between fatty acid metabolism and CRC. Moreover, by mapping single-nucleotide polymorphism to genes and performing a cross-trait meta-analysis, their study found that GNAS, FAM163B, RHPN2, and STARD3 might contribute to CRC pathogenesis by influencing fatty acid and lipid metabolism. Recent MR analyses focusing on cis-eQTLs have indicated a positive association between FABP4 and CRC risk in women, although a causal relationship between FABP4 and CRC has not been established[12]. These findings indicate that the association between FABP4 and CRC is more complex than initially anticipated. Larger-scale genome-wide association study (GWAS) data integration, pQTL MR analysis, and combination with other fatty acid-regulated proteins may be feasible approaches for the causal inference of FABP4 and CRC pathogenesis in the future.

MACHINE LEARNING ENHANCES THE ACCURACY OF PROGNOSTIC MODELS

Zhang et al[6] computed risk scores based on FABP4 and its 15 co-expressed genes, and then integrated them with clinicopathological factors, such as age and tumor stage, to construct a Cox regression model for predicting the survival of patients with COAD. However, the model validation set exhibited a suboptimal C-index, indicating that its predictive accuracy requires improvement. Variable selection in Cox regression relies on the subjective judgment of the researcher, which may lead to the omission of important prognostic factors and the inclusion of irrelevant variables. In contrast, machine and deep learning methods can automatically select appropriate variables through feature selection and effectively handle nonlinear features in high-dimensional data, thereby enhancing model performance.

A previous study screened genes associated with ferroptosis and fatty acid metabolism, compared 117 machine learning algorithms, and ultimately selected a combination of CoxBoost and StepCox to construct a prognostic model for CRC. Notably, among the 15 genes used in this model was FABP1—a family member with functions similar to FABP4[13].

A retrospective article published in 2025 on the application of machine learning in medicine found that most omics studies have employed traditional machine learning and linear regression models, with deep learning primarily applied to image analysis[14]. However, recent studies have increasingly integrated Cox regression into deep learning frameworks for survival analysis[15]. For instance, Cox-nnet employs deep features extracted from hidden layers as input for the Cox regression model, whereas AECOX uses an autoencoder to compress gene expression data into a low-dimensional embedding vector that serves as the model input, thereby achieving superior predictive accuracy compared with the traditional Cox proportional hazards model[16,17]. Moreover, deep learning facilitates integrative analysis of diverse types of sequencing data. For example, a network topology-based deep learning framework called NETTAG can integrate GWAS and multi-omics data to identify risk genes associated with Alzheimer’s disease[18]. However, the application of deep learning to sequencing data for constructing prognostic models for COAD remains limited.

Future research should focus on integrating various types of sequencing data and further optimizing deep learning models to extract FABP4-related fatty acid metabolism regulatory genes with prognostic value more precisely, thereby constructing more accurate and robust prognostic models for COAD and facilitating their widespread clinical application.

FINDING RELATED GENES TO DEVELOP MULTI-TARGET STRATEGIES

Zhang et al[6] reported that FABP4 may be a potential therapeutic target for COAD but did not provide a specific strategy for targeting it. Existing FABP4 inhibitors, such as BMS309403 and carbazole-based compounds, are yet to enter extensive clinical trials[19]. Despite significant advances in high-throughput screening methods based on genomics and proteomics as well as rational drug design, the number of successful single-target drugs has not increased substantially[20]. Cancer is not caused by a single molecular factor; rather, it involves multiple intervention points, each affecting a part of its etiology[21]. Multi-target strategies can be implemented by combining several approaches or using polypharmacological agents, a strategy that has shown superior efficacy compared with single-target drugs in the targeted treatment of lung and breast cancers[22,23]. In recent years, an increasing number of multi-target compound design methods, such as POLYGON and FSCA, have emerged[21,24]. However, the optimal method for screening for receptor ensembles of multitarget drugs to maximize their efficacy remains unclear. Zhang et al[6] used WGCNA to identify FABP4 co-expressed genes associated with COAD prognosis, which may aid in the development of multitarget therapies for this disease. Nevertheless, the parameter selection process in the WGCNA network construction can introduce subjective bias, and because WGCNA relies solely on statistical theory, it faces challenges in biological interpretation. Integrating comprehensive human gene databases, such as Genecards, to identify genes involved in regulating fatty acid metabolism, or employing summary-data-based MR to evaluate, on a large scale, the causal relationships between genes and COAD may help in screening for targets with clearer mechanistic underpinnings. In addition, the roles of other family members in COAD should not be overlooked. Given that the three-dimensional structures of different FABP family members are highly similar[19], targeted drug development may encounter challenges such as cross-reactivity, which can reduce efficacy and increase side effects. This structural similarity may facilitate the development of polypharmacological drugs, thereby opening new avenues for COAD treatment.

MORE COMPREHENSIVE EXPERIMENTS TO VERIFY THE RESULTS AND DEEPEN THE RESEARCH

Experimental validation is crucial for enhancing the accuracy and reliability of bioinformatics analysis and helps reveal biological mechanisms that are difficult to elucidate solely through bioinformatics analysis. Currently, bioinformatics research on FABP4 in colon adenocarcinoma has employed this gene to construct COAD prognostic models based on immune and mRNA stemness index, as well as models for predicting COAD recurrence[25-27]. Other studies have found that, in COAD, FABP4 is associated with immune regulation, mA6-related ferroptosis, angiogenesis, and neutrophil extracellular traps[28-30]. However, these bioinformatics investigations generally lack sufficient experimental validation.

Although Zhang et al[6] used IHC to verify the differential expression of FABP4 between colon adenocarcinoma and normal tissues, they did not validate the accuracy of RNA-seq data from public databases, nor did they further investigate the mechanism by which FABP4 participates in cell adhesion, as revealed by enrichment analysis. In functional genomics studies involving large-scale differential gene analysis, the limited throughput of bulk RNA-seq hampers its sensitivity and specificity. Issues such as restricted detection and quantification of transcript isoforms, as well as biases in sample preparation[31] may lead to false positives in subsequent bioinformatics analyses. Experimental methods such as quantitative polymerase chain reaction (qPCR) can be employed to confirm the transcription levels of FABP4 and its co-expressed genes. Moreover, traditional IHC suffers from interobserver variability and limited labeling capability; however, when combined with digital image analysis, it allows for a more precise assessment of target proteins[32]. Emerging techniques, such as multiplex IHC and immunofluorescence, which enable the simultaneous detection of multiple markers in a single tissue sample while providing comprehensive data on cell composition and spatial arrangement[33], further promote in-depth research into FABP4-related biological mechanisms in colon adenocarcinoma.

Yu et al[34] employed various experimental methods, including flow cytometry, electron microscopy, qPCR, and western blotting, to assess cellular fatty acid uptake, mitochondrial mass, and fatty acid oxidation rates. Their results demonstrated that FABP4 maintains mitochondrial function and ROS production by promoting fatty acid uptake and oxidation, thereby supporting the progression of triple-negative breast cancer (TNBC). They also identified CPT1b as a key downstream effector of this process, offering critical insights into the molecular mechanisms by which FABP4 regulates TNBC progression[34]. Additionally, the NR1H3-SREBP1-FABP4 regulatory axis promotes tumor cell proliferation in obesity-associated breast cancer[35]. Furthermore, FABP4 induces epithelial-mesenchymal transition in pancreatic cancer cells through regulation of the NLRP3-IL-1β axis, which in turn enhances the migration, invasion, and metastasis of pancreatic cancer[36].

Although notable progress has been made in understanding the role of FABP4 in other cancers, strong experimental evidence supporting its biological mechanism in promoting the initiation and progression of COAD is lacking. Future bioinformatics research on FABP4 in COAD should incorporate more comprehensive experimental approaches to validate these findings and further investigate the underlying mechanisms. This will help elucidate the specific pathways through which FABP4 regulates biological processes, such as cell adhesion and immune modulation, thereby providing a stronger and more robust theoretical foundation for developing FABP4-targeted therapeutic strategies.

CONCLUSION

Zhang et al[6] demonstrated the potential of FABP4 as a biomarker for the diagnosis and prognosis of COAD. We commend the authors for their meticulous research and encourage future studies to place greater emphasis on the biological interpretation and mechanistic exploration of the role of FABP4 in COAD. By leveraging multi-omics approaches and machine learning techniques, future research is expected to achieve breakthroughs in clarifying the causal relationships between FABP4 and COAD, optimizing prognostic models, and developing multitarget treatment strategies, which will provide a more robust theoretical foundation and practical guidance for the early diagnosis, precision treatment, and prognosis evaluation of COAD.

ACKNOWLEDGEMENTS

We thank the reviewers for their comments that helped to improve the manuscript.