Comprehensive analysis of basement membrane-related genes showed that NELL2 is a new therapeutic target for colorectal cancer

Abstract

Background

The basement membrane (BM), an omnipresent extracellular matrix, plays a pivotal role as a physiological element in the process of tumor metastasis. However, given the heterogeneity of colorectal cancer (CRC), prognosis is challengingly predictive. Therefore, we aim to construct a prognostic model using BM-associated genes to assess patient prognosis and clinical drug treatment effects.

Method

The Non-negative Matrix Factorization (NMF) algorithm leverages the characteristics or categories of matrix rows and columns to achieve BMAG molecular classification and further develop a model for predicting patient prognosis. ssGSEA quantified the relatively abundance of 13 immune functionalities and 16 immune cell typologies. To predict the efficacy of immunotherapy, a comprehensive investigation was conducted on the correlations between riskScores and key factors such as TME, immune checkpoints, and MMR-related genes. The CCK8 method, plate cloning method and Cell Apoptosis Assessment were used to evaluate the ability of NELL2 to affect the proliferation.

Result

We developed a powerful riskScore to predict colorectal cancer prognosis and effectively differentiate the tumor microenvironment. In clinical practice, this riskScore can also be utilized to further assess patient prognosis, thereby facilitating personalized treatment strategies. In addition, downregulation of NELL2 expression inhibits CRC cell proliferation.

Conclusion

In summary, we constructed a novel riskScore using BMAG for predicting prognosis in patients with CRC and explored the efficacy of this riskScore in predicting patient response to clinical drug therapy. Most importantly, we have identified the oncogenic role of NELL2 in CRC. By inhibiting NELL2, we can further suppress the initiation and progression of CRC.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-01979-5.

Introduction

Colorectal cancer (CRC) stands as one of the most prevalent malignant tumors. In 2020 alone, it was estimated that there were approximately 1.9 million new cases worldwide, resulting in 900,000 deaths [1]. It is also one of the deadliest malignancies worldwide, characterized by high rates of morbidity and mortality. Furthermore, studies have found that colorectal cancer (CRC) exhibits significant heterogeneity, which leads to poor prognosis for CRC patients [2]. While various drug combination strategies have been developed for treating CRC, the outcomes for CRC patients have reached a bottleneck. Hence, it is imperative to establish a novel molecular classification to more accurately predict the prognosis of CRC.

Cancer cells encounter and must break through basement membranes (BM) several times during metastasis—during intrusion, introgression, and extravasation. Intrusion involves cancer cells breaking through surrounding tissues to invade adjacent normal tissues, whereas introgression signifies a deeper penetration and potential extensive tissue damage [3]. During EMT, cell–cell to cell and cell-extracellular matrix (ECM) exchanges are remodelled, resulting in epithelial cells detaching from each other and from the underlying basement membrane, and novel transcriptional programs are initiated to boost the metastatic fate [4]. ECM takes a central function in the tumor microenvironment [5], with the BM acting as a protective barrier against the long-range dissemination of cancer cells [3]. Consequently, aberrant modulation of the basement membrane can potentially contribute to cancer aggressiveness and metastasis [6]. Numerous researches have demonstrated BM-associated genes (BMAG) in the prognosis of various tumors, including renal cell cancer [7–9] and bladder cancer [10]. Furthermore, he distinct stromal environment and cellular mechanical properties of the basement membrane play a regulatory role in cell migration and contribute to the development of the invasive phenotype [11]. Because of the critical involvemen of BM in tumor metastasis, it represents a promising avenue for anticancer therapeutic strategies. Previous studies demonstrates that by utilizing distinct gene sets, including fibroblast and angiogenesis, constructing a risk index can aid in predicting the prognosis of CRC [12, 13]. Previous studies have shown that a riskScore based on BMAG can stably predict the prognosis of patients with hepatocellular carcinoma, lung adenocarcinoma, and gastric cancer [14–16]. However, there is currently relatively limited research on BMAG and CRC, which may be attributed to the high complexity and heterogeneity of colorectal cancer. Therefore, it is necessary to comprehensively analyze BMAG in CRC and find new prognostic markers to accurately predict the prognosis and treatment of CRC patients.

In our current study, we swept 222 BMAG and used Non-negative Matrix Factorization (NMF) techniques to structure the molecular typing of these genes to generate a riskScore. The derived riskScore can effectively prognosticate CRC prognosis, and we comprehensively assessed the association between the riskScore and the immune microenvironment. In addition, this study evaluated the efficacy of clinical drug therapy against this riskScore.

Finally, we determined that NELL2 plays a tumor-promoting role in CRC through in vitro studies, and determined that NELL2 is mainly distributed in the nucleus in HT29 and HCT116 cells.

Methods and materials

Data acquisition

We obtained RNA-seq and clinical data of CRC patients from the TCGA database. RNA-seq data and survival data of 556 CRC samples were obtained from GSE39582 for model validation. Samples with missing data and survival time were excluded during the survival analysis. Additionally, samples with a survival time of less than 30 days were also excluded. Additionally, we downloaded scRNA-seq data from four CRC patients from GSE161277.

Identification of CRC subtypes

NMF is a powerful machine learning algorithm that can be applied to cancer classification [17, 18] In this study, we utilized the "NMF" package to classify CRC based on BMAG gene expression. After analysis, it was determined that the optimal number of clusters was 2.

Development and validation of riskScore

Based on the NMF classification results, we first analyzed the differentially expressed genes (DEGs) among different subtypes. To identify key genes, we conducted LASSO-multivariable Cox analysis on the DEGs and successfully identified a set of 5 core genes. Utilizing the gene expression levels and their importance (Coef values), we constructed a riskScore. The GSE39582 cohort served as the validation set, which comprised 556 samples.

GO analysis and KEGG analysis, immune correlation analysis, drug susceptibility analysis

The methods involved in this part of the study have been described in detail in our previous research [19].

Clinical relevance of riskScore

To explore the independent protective/risk factors for CRC patients, we performed univariable Cox (uni-Cox) analysis and multivariable Cox (mti-Cox) analysis, considering riskScore and clinical parameters. Among these, clinical parameters include Age, Gender, stage, as well as T, N, and M classifications. Additionally, to facilitate clinical application, we constructed a column chart by combining riskScore and clinical parameters.

Cell culture and transfection, RNA extraction, and qRT-PCR and CCK8 assay

The specific steps of these experiments refer to our previous studies [19]. HCT116, HT29 and SW116 were obtained from the Chinese Academy of Sciences. The human CRC cell lines HCT116 and HT29 were cultured in McCoy’s 5A (Modified) Medium supplemented with 10% fetal bovine serum, while SW116 cells were cultured in Leibovitz's L-15 medium containing 10% fetal bovine serum. All cells were maintained in a humidified atmosphere at 37°C with 5% CO2. The siRNA sequences and primer information designed for this study can be found in Supplementary Table 1.

Colony formation assay

The HCT116 and HT29 cell lines were used for the Colony Formation Assay, where the control group received no treatment, while the experimental group had the target gene knocked down. Subsequently, 250 CRC cells were counted, cultured in 6-well plates, and the experiment was terminated on day 7. After fixation with paraformaldehyde, staining with crystal violet was performed. Estimation of cell colony number using Image J software.

Cell apoptosis assessment

The apoptosis ability of CRC cells was evaluated using Annexin V Cell Apoptosis Detection Kit (Thermo Fisher Scientific, USA). And flow cytometry analysis was performed using a FACS system to determine the cell apoptosis rate.

Immunofluorescence (IF)

Cells were seeded in small dishes. Firstly, paraformaldehyde was used for fixation. Then permeabilize with 0.5% Triton X-100 to disrupt the cell membrane. After blocking with sheep serum, the primary antibodies were incubated overnight at 4 °C with the cells. On the following day, the cells were incubated with fluorescent secondary antibodies. DAPI is utilized for fluorescent staining of cell nuclei. The DAPI working solution (1:100, Beyotime, China) and the primary antibodies employed were anti-NELL2 (1:100, Proteintech, China, cat no:11268–1-AP).

Statistical analysis

In this study, all bioinformatics analyses were conducted using the R language. For correlation analysis, Spearman's rank correlation method was utilized. The Wilcoxon test was employed to perform differential analysis on non-normally distributed data. p < 0.05 was considered to be statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001.

Result

Filtering for BM critical genes by NMF

Firstly, to facilitate readers' better understanding of the content of this paper, we have meticulously crafted a flowchart (Fig. 1). BMAG were extracted based on previous research, with a total of 222. We firstly conducted a NMF of 222 BMAG. On the basis of the expression data of the filtered basement membrane-related genes, molecular typing was preliminarily constructed by NMF, and the optimal number of subgroups was established to be 2 by the maximum covariance correlation factor descent method (Fig. 2A). We found that the overall survival (OS) period of C1 was obviously longer than that of C2, with a p-value of 0.024 (Fig. 2B). And the expression profiles of the differential expression genes among the per-patient phenotypes in Cohort 1 and Cohort 2 were visualized and analyzed by heatmap (Filter criteria: P < 0.05) (Fig. 2C). Differential analysis was then performed to determine genes that were markedly different between normal and cancer tissues (Fig. 2D). As shown in the Venn diagram, 52 key genes were obtained by identifying DEGs between different types, differentially expressed BMAG and the intersection with BMAGs of prognostic value (Fig. 2E).

Fig. 1.

A flow chart of the manuscript

Fig. 2.

Filtering for BM critical genes by NMF. A 2 was determined to be the best value for consensus clustering. B OS survival curves for the two clusters. C Expression profiles of BMAG among clusters. D Volcano maps for BMAG difference analysis. E Venn diagram indicating DEGs among subtypes, differentially expressed BMAG and shared genes of prognosis-related BMAGs

Development and validation of riskScore for CRC patients

To further investigate the interrelationships among the 52 key BMAG, we employed STRING to analyse their protein interaction network. This analysis effectively revealed the intricate interactions among the proteins (Fig. 3A). We further analyzed by Lasso (Fig. 3B, C) and mti-Cox regression, we identified “CTSD”, “NELL2”, “TGFBI”, “TIMP1” and “ADAM9” as key genes. Based on the mti-Cox analysis outcomes, a riskScore was formulated. Patients were categorized into low-risk cohorts (LRC) and high-risk cohorts (HRC) using the median value of the riskScore. Taking into account the correlation between the riskScore and prognosis, a validation of the riskScore was conducted utilizing two datasets. The TCGA-CRC cohort was allocated as the training cohort, whereas the GSE39582 cohort to serve as the validation cohort. It is noteworthy that in both datasets, patients classified as HRC had a significantly unfavorable prognosis (Fig. 3D, E).

Fig. 3.

Construction and verification of BMAG for CRC patient typing models. A PPI network demonstrating the interaction among proteins. B, C The Lasso algorithm was used to screen key genes. D, E Kaplan–Meier (KM) survival curves were drawn for HRC and LRC

Evaluation of the prognostic performance of the riskScore.

In both the training and validation cohort, the LRC exhibited a more favorable prognosis compared to the HRC. Further investigation into the survival of patients in both groups revealed a higher mortality rate in the HRC (Fig. 4A, B). Heatmap analysis demonstrated markedly higher expression levels of key genes in the HRC compared to the LRC (Fig. 4C, D). The ROC demonstrated that the prognostic prediction AUC reached 0.670 and 0.601 in the TCGA-CRC training cohort and the GSE39582 dataset validation cohort (Fig. 4E, F). These results indicate that our model serves as an excellent prognostic indicator for CRC.

Fig. 4.

RiskScore performance validation. A, B Distribution of patient riskScore and survival status in TCGA-CRC and GSE39582 cohort. C, D gene expression heatmap within the TCGA-CRC and GSE39582 cohort. E, F Time-dependent ROC curves for the riskScore to predict outcomes in the TCGA-CRC and GSE39582 cohort

Construction of nomogram

Uni-Cox analysis indicated that tumor TNM stage and riskScore were significant risk factors influencing CRC prognosis (Fig. 5A). Multi-Cox analysis showed that riskScore and T emerged as independent risk factors significantly influencing CRC prognosis (Fig. 5B). To construct a nomogram that is easy to use clinically, we included riskScore and several commonly used clinicopathological parameters. Among these, Age, Gender, stage, T, and N classifications are the clinical information to be studied (Fig. 5C). Furthermore, the calibration curves demonstrate the predictive performance of the nomogram for 1, 3, and 5 years. And ROC further showed the outstanding robustness and precision of the nomogram in assessing patient survival (Fig. 5D, E).

Fig. 5.

Construction of nomogram. A, B uni-Cox and mti-Cox regression analyses were performed to evaluate valuable independent prognostic indexes. C Nomograms were utilised to approximate the probability of 1-, 3-, and 5-year survival in patients with CRC. D, E Calibration and ROC curve demonstrated the nomogram's satisfactory stability, accuracies and predictive capabilities

Immunological analyses between LRC and HRC

We estimated the richness of immune cells and immune-related activities, investigating their correlation with the riskScores. The findings revealed that, on one hand, patients in the HRC exhibited higher levels of infiltration of various immune cells, including aDCs, B cells, and CD8 + T cells (Fig. 6A). On the other hand, compared to the LRC, the HRC demonstrated higher activity in multiple immune-related activities, including APC co-inhibition, checkpoints, and HLA (Fig. 6B). Additionally, we conducted an assessment of the ESTIMATEScore, StromalScore and ImmuneScore in the samples. It was observed that the ESTIMATEScore, StromalScore, and ImmuneScore scores of LRC were significantly lower compared to HRC (Fig. 6C). Simultaneously, we explored the correlation between riskScores and MMR-related genes (MLH1, MSH2, MSH6, PMS2 and EPCAM). The findings showed that riskScores were negatively correlated with all MMR-related genes (Fig. 6D). The analysis of immune checkpoints further identified the associations of the riskScores with immune checkpoint molecules (ICMs). Out of the 48 ICMs examined, the expression of 31 ICMs exhibited a significant correlation with the riskScore (Fig. 6E).

Fig. 6.

Immunological analyses between HRC and LRC. A, B The associations between riskScore and various immune cells and immune-related activities were evaluated. C The associations between riskScore and ESTIMATEScore, StromalScore, and ImmuneScore scores. D Mismatch repair (MMR) analysis was performed to show a significant relationship between riskScore and MMR -related genes. E The associations between riskScore and ICMs

GO and KEGG analyses associated with BMAG

We proceeded with a more detailed analysis of the genes that exhibited differential expression among the various risk groups. To understand the biological characteristics involved in the riskScore, we then performed GO and KEGG analysis based on these DEGs. The results of the GO analyses suggested that in terms of biological process (BP) analyses, we saw an enrichment of positive regulation of cell–cell adhesion (Fig. 7A). The molecular function (MF) analysis revealed enrichment in extracellular matrix structural constituent (Fig. 7B). The cellular component (CC) analysis indicated enrichment in MHC protein complex (Fig. 7C, D). Moreover, the findings from KEGG analysis reveal significant enrichment of DEGs in critical biological processes such as cell adhesion and cell cycle (Fig. 7E).

Fig. 7.

GO and KEGG analyses associated with riskScore. A–D GO analysis results revealing enriched functions in BP, CC and MF. E Exploration of latent pathways utilising the KEGG analysis

Drug sensitivity analysis

Subsequently, we conducted an investigation into the sensitivity of the LRC and HRC towards various chemotherapeutic agents. The lesser the IC50 value, the more sensitive to the drugs. Interestingly, we discovered that Galibenzquinazole, OF-1, SB505124, Ulisatinib, VX-11e, BI-2536, and Daporinad had higher IC50 in HRC (Fig. 8A–G). In the LRC, PLX-4720, PRT062607, PF-4708671, AZ960, JAK1_8709, Entospletinib, XAV939 8709, AZD1332, BMS-754807, and WZ4003 had higher IC50 values (Fig. 8H–Q).

Fig. 8.

Drug sensitivity analysis of the model in CRC patients. A–Q Differential analysis of sensitivity to 17 drugs between the HRC and LRC

ScRNA-seq data analysis

We applied downscaling and clustering to the preprocessed scRNA-seq data (GSE161277) by using tSNE, resulting in 15 clusters (Fig. 9A). Subsequent heatmap analysis revealed the most conspicuously expressed genes between these clusters (Fig. 9B). Subsequently, the SingleR package automatically anonymised the cell types and categorised these 15 clusters into 6 cell types (Fig. 9C). Briefly, the bubble plots clearly indicated that CTSD was highly expressed in epithelial cells, monocytes, NK cells and endothelial cells, TGF-β was strongly expressed in epithelial cells, monocytes, TIMP1 and ADAM9 were predominantly expressed in epithelial cells, monocytes, and endothelial cells, whereas NELL2 was lowly expressed in all cell types (Fig. 9D–I).

Fig. 9.

ScRNA-seq data analysis. A Cells were classified into 15 clusters with the usage of the t-SNE clustering algorithm. B Heatmap showing DEGs in the clusters which were identified. C The "SingleR" package was utilised to anoint different cell types. D–I Distribution and expression levels of key genes in 6 cell types

Exploration of the biological role of NELL2

We conducted further research on the role of model genes in CRC. After searching Pubmed, we found that there were no studies on NELL2 in CRC. Therefore, we focused our research on the biological role of NELL2 in CRC. Firstly, by assessing the expression of NELL2 in HCT116, HT29, and SW116 cells, we found that SW116 exhibited the lowest expression, while HCT116 and HT29 exhibited the highest expression levels (Fig. 10A). Therefore, we used siRNA performed NELL2 knockdown experiments in HCT116 and HT29 cells, and the efficiency was validated using qRT-PCR (Fig. 10B). CCK-8 and colony formation assays demonstrated a significant reduction in proliferative capacity of HCT116 and HT29 cells upon NELL2 knockdown (Fig. 10C–E). Apoptosis assays revealed a substantial decrease in apoptotic ability of HCT116 and HT29 cells after NELL2 knockdown (Fig. 10F). Furthermore, we also evaluated NELL2 distribution using immunofluorescence, and the results showed that NELL2 protein was distributed in the nucleus and cytoplasm of HT29 and HCT116, and was mainly localized in the nucleus (Fig. 10G).

Fig. 10.

Biological functions. A NELL2 mRNA expression levels in HT29, HCT116 and SW116. B Knockdown efficiency of siNELL2 in HT29 and HCT116. C–E CCK8 and colony formation assay to evaluate the effect of NELL2 on the proliferation of HT29 and SW116 cells. F Apoptosis experiment to evaluate the effect of NELL2 on the apoptotic ability of HT29 and SW116 cells. G subcellular localization of NELL2

Discussion

In the United States, CRC poses a significant health concern with an estimated annual incidence of around 130,000 new cases and approximately 50,000 deaths [1]. Finding or developing new therapeutic targets for CRC is critical to developing safe and effective prevention and treatment methods. The BM, being a crucial element of the ECM, represents a formidable obstacle that cancer cells must overcome in order to initiate metastasis formation [3, 11]. Considerable research has shown an association between major components of BM and CRC [20–22]. Despite significant progress, prognostic markers developed based on BMAG in CRC are still lacking. To bridge this gap, our research employed NMF to develop a robust prognostic riskScore utilizing BMAG as the basis.

The prognostic value of riskScores was comprehensively assessed. In the TCGA-CRC cohort and GSE39582 cohort, the prognosis of HRC was significantly worse than that of LRC. This emphasizes the high stability and accuracy of the riskScore in predicting the prognosis of CRC. Moreover, the predictive ability of the riskScore for CRC patient prognosis was validated in the GSE39582 cohort, further highlighting its consistent predictive efficacy in evaluating CRC patient prognosis. Mti-Cox analysis showed the riskScore was an independent risk factor affecting the prognosis of CRC patients.

Furthermore, we integrated commonly used clinicopathological parameters with riskScores to develop a user-friendly nomogram. This nomogram not only facilitates its clinical application but also enhances the accuracy of prognostic prediction for patients with colorectal cancer. In clinical practice, the prognosis of patients can be further predicted based on indicators such as age, gender, and staging, thereby enabling personalized treatment for different patients.

To gain more insight into the underlying biological characteristics within these subgroups, we conducted a comprehensive analysis of DEGs across various groups. Our comprehensive analysis, incorporating GO and KEGG, revealed that the disparities in biological processes among distinct riskScore subgroups primarily revolve around cell adhesion and immune-related mechanisms. These results indicate a close association between the riskScore constructed based on BMAG and metastasis as well as the immune response.

Furthermore, we explored differences in the tumor immune microenvironment (TIME) between subgroups with different riskScores. We first conducted a study utilizing the ssGSEA estimate the levels of 13 immune-related activities and 15 immune cell types, and explored the correlation between risk scores and 13 immune-related activities and 15 immune cell types.

Observations revealed a significant enrichment of 12 distinct immune cell types and 11 immune-related activities in the HRC in comparison to the LRC. Previous studies have shown that hot tumors represent higher immune cell infiltration, which means better immunotherapy effects [23]. This means that patients in the HRC are more likely to be diagnosed with hot tumors. Therefore, patients with a high HRC tend to have better prognosis.

Immunotherapy has become a revolutionary cancer treatment, but its efficacy varies depending on the type of solid tumour. Therefore, it is crucial to identify biomarkers that can accurately predict patient response to immunotherapy [24]. MMR status is additionally a commonly used predictor of response to immunotherapy [25]. We identified significant negative correlations between the riskScores and MMR-related gene, including MLH1, MSH2, MSH6, PMS2 and EPCAM. Notably, previous studies also showed that EpCAM was first described in 1979 as a humoral antigen expressed on colon cancer cells [26]. Additionally, we examined the association between the riskScore and ICMs. Our findings showed a strong correlation between the riskScore and 31 ICMs, 28 of which were positively correlated and 3 (CD44, TNFSF15 and HHLA2) were negatively correlated. This suggests that riskScores should be considered as immune checkpoints to guide immunotherapy when making clinical treatment decisions. Notably, the riskScore exhibited strong positive correlations with the well-known CD274 (PD-L1), PDCD1 (PD-1) and CTLA4 molecules. Collectively, these findings reveal that riskScores can serve as effective discriminators of TIME in CRC. Moreover, CRC patients classified in the HRC may exhibit greater suitability for immunotherapy. The riskScore exhibited strong positive associations with the well-known PD1 (PDCD1), PDL1 (CD274), and CTLA4 molecules. These results collectively imply that riskScores can effectively differentiate the Tumor Immune Microenvironment (TIME) in CRC. Furthermore, individuals with CRC identified in the high-risk category might display enhanced responsiveness to immunotherapy. We hypothesize that NELL2 may regulate the tumor microenvironment through immune checkpoints, thereby influencing the progression of CRC.

ScRNA-seq is a well-resolved genomics technology that can characterise cellular heterogeneity and developmental trajectories [27]. Previous studies have shown the presence of spatially organised multicellular immune centres in human CRC [28]. To explore the expression of key genes used for model construction in distinct cell types, we analyzed the scRNA data in the GSE161277 dataset.

We were able to observe that CTSD was strongly expressed in epithelial cells, monocytes, NK cells and endothelial cells, TGF-β was highly expressed in epithelial cells and monocytes, TIMP1 and ADAM9 were dominantly expressed in epithelial cells, monocytes and endothelial cells, whereas NELL2 was lowly expressed in all cell types.

In addition, we assessed the drug sensitivity of CRC patients in different riskScore subgroups towards 17 frequently prescribed therapeutic drugs. These findings offer valuable insights to inform clinical decision-making.

Most importantly, we conducted a more in-depth investigation into the role of NELL2 in CRC within our model. Our results revealed that knockdown of NELL2 in HCT116 and HT29 cells resulted in reduced proliferative capacity and increased apoptotic ability. Furthermore, in vivo experiments demonstrated that overexpression of NELL2 significantly enhanced tumor growth while diminishing apoptotic ability. Additionally, we conducted subcellular localization analysis of NELL2, which indicated its distribution throughout the cell but primarily concentrated within the cell nucleus. These results offer valuable insights into future the mechanistic study of NELL2.

To conclude, we have developed a stable and accurate riskScore based on BMAG that can stably for predicting the prognosis of CRC. Furthermore, the riskScore demonstrated effectiveness in predicting patient response to immunotherapy. More importantly, we identified the oncogenic role of NELL2 in CRC. The riskScore can accurately predict patients' response to immunotherapy. Importantly, we elucidated the oncogenic function of NELL2 in CRC.

Author contributions

CC and WW designed the study. WW and ZD analyzed the data. WW, ZD, and CG drafted the manuscript. CG, XZ, and YZ collected the literature. CC revised the manuscript. All authors reviewed the manuscript.

Funding

This research was supported by the Top Talent Support Program for Young and Middle-Aged People of Wuxi Health Committee (HB2023116).

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.