CELF2 is a candidate prognostic and immunotherapy biomarker in triple‐negative breast cancer and lung squamous cell carcinoma: A pan‐cancer analysis

Abstract CUGBP Elav‐like family member 2(CELF2) plays crucial roles in the development and activation of T cell. However, the impacts of CELF2 on tumour‐infiltrating immune cells (TIICs) and clinical outcomes of tumours remain unclear. In this study, we found that elevated CELF2 expression was markedly correlated with prolonged survival in multiple tumours, particularly in breast and lung cancers. Notably, CELF2 only impacted the prognosis of triple‐negative breast cancer (TNBC) with lymph node metastasis. Further investigation showed CELF2 expression was positively correlated with the infiltration abundance of dendritic cells (DCs), CD8+ T cells and neutrophils in breast invasive carcinoma (BRCA) and DCs in lung squamous cell carcinoma (LUSC). CELF2 also had strong correlations with markers of diverse TIICs such as T cells, tumour‐associated macrophages and DCs in BRCA and LUSC. Importantly, CELF2 was significantly associated with plenty of immune checkpoint molecules (ICMs) and outperformed five prevalent biomarkers including PD‐1, PD‐L1, CTLA‐4, CD8 and tumour mutation burden in predicting immunotherapeutic responses. Immunohistochemistry also revealed lower protein levels of CELF2 in TNBC and LUSC compared to normal tissues, and patients with high expression showed significantly prolonged prognosis. In conclusion, we demonstrated that increased CELF2 expression was closely related to better prognosis and superior TIIC infiltration and ICM expression, particularly in BRCA and LUSC. CELF2 also performed well in evaluating the immunotherapeutic efficacy, suggesting CELF2 might be a promising biomarker.


| INTRODUC TI ON
With the change in disease spectrum, tumours have become a major threat to people's health in recent years, placing a heavy burden on global public health. The latest statistical report shows that in 2021, there will be more than 1,898,160 new cancer cases and 608,570 cancer deaths in the United States, which has become the leading cause of death in developed countries. Among them, the incidence of lung cancer ranks second and has becoming the number one killer in tumour-related disease for both men and women. In terms of women, breast cancer is the most common tumour type, with the second highest mortality rate for a long time, after lung cancer. 1 Although the mortality rate of breast and lung cancer has decreased with the improvement of tumour diagnosis and treatment, the clinical outcomes remain unsatisfactory. For patients with lung cancer, the 5-year survival rate is only 21%. Breast cancer exhibits a 5-year survival rate of nearly 90%, while the 5-year survival rate of distant metastatic breast cancer is only 28%. 1 Hence, a better understanding the specific mechanisms of breast and lung cancer oncogenesis and progression, as well as to find more accurate novel biomarkers that can be used for clinical and therapeutic management, is urgently needed.
Tumour microenvironment (TME) is the cellular environment for tumour cell growth, in which tumour-infiltrating immune cells (TIICs), an important component, play a dominant role. 2 For example, tumour-associated macrophages (TAMs) can exacerbate tumour progression by promoting tumour angiogenesis, metastasis and immune escape. 3 Dendritic cells (DCs) conduce to tumour metastasis by reducing CD8+ T-cell cytotoxicity and enhancing regulatory T (Treg) cell responses. 4,5 The past decade has witnessed encouraging advances in immunotherapy represented by immune checkpoint inhibitors (ICIs) has revolutionized the therapeutic paradigm of most tumours, especially non-small-cell lung cancer, triple-negative breast cancer (TNBC), advanced melanoma and bladder cancer. [6][7][8][9] While ICIs target the interactions between immune and tumour cells within the TME, certain alterations that occur in the TME can also affect the responsiveness to immunotherapy. 3 Recent studies have confirmed that key biological processes such as autophagy, hypoxia and ferroptosis, as well as some molecular alterations, can contribute to the immunotherapeutic efficacy and prognosis of cancer patients by affecting the distributions and interactions of distinct immune cell subsets in the TME. [10][11][12][13] To date, there are still few cancer patients who can benefit from immunotherapy, and thus, it is essential to explore additional therapeutic targets.
CELF (CUGBP Elav-like family) proteins are RNA binding protein of shuttle nucleoplasm characterized by three RNA recognition motifs. In humans, CELF contains six known isoforms, CELF1 to CELF6, which are further divided into two subgroups according to their amino acid sequencing similarity. One group consists of CELF3-6 and is largely restricted to neurons and a few other tissues. The other group includes CELF1 and CELF2, which are commonly expressed in most tissues, but are expressed differently during development and differentiation. 14 Previous studies have demonstrated that CELF2 could regulate multiple steps of RNA processing, such as pre-mRNA splicing, RNA editing, polyadenylation, mRNA stability and translation. 14,15 Over the past decade, substantial studies have confirmed that CELF2 played a tumour suppressor role in breast cancer, lung cancer, hepatocellular carcinoma, gastric cancer, ovarian cancer, glioma and acute myeloid leukaemia, suggesting that it can be used as a candidate biomarker to predict cancer prognosis. [16][17][18][19][20][21][22][23] Additionally, CELF2 expression was significantly elevated in developing thymocytes and activated T cells and promoted T-cell receptor expression and signalling through alternative splicing. [24][25][26] However, the comprehensive understanding of the impacts of CELF2 on the tumour immune microenvironment remains unknown.
Herein, we delineated the expression and prognostic landscape of CELF2 across human cancers. The relationships between CELF2 and TIICs as well as immune checkpoint molecules (ICMs) were further explored. In addition, we also recruited two immunotherapeutic cohorts and evaluated the power of CELF2 for predicting the responses to immunotherapy. Overall, our study provides a reference and direction for understanding the crucial role of CELF2 in the immune microenvironment of pan-cancer, as well as reveals the potential mechanism thereby CELF2 affects anti-tumour immunity and cancer immunotherapy.

| CELF2 expression across human cancers in Oncomine
Oncomine (https://www.oncom ine.org/resou rce/login.html) is a web-based data mining platform that assembles 86,733 samples and 715 gene expression data sets together. 27 The mRNA In conclusion, we demonstrated that increased CELF2 expression was closely related to better prognosis and superior TIIC infiltration and ICM expression, particularly in BRCA and LUSC. CELF2 also performed well in evaluating the immunotherapeutic efficacy, suggesting CELF2 might be a promising biomarker.

K E Y W O R D S
CELF2, immune infiltration, immunotherapy, lung squamous cell carcinoma, prognosis, triplenegative breast cancer expression levels of CELF2 in various cancer types were detected using Oncomine database with the following conditions: p-value: 0.001, fold change: 1.5, and gene rank: all.

| Survival analysis in PrognoScan and Kaplan-Meier plotter
The biological correlations between CELF2 expression and patient survival in pan-cancer were evaluated via PrognoScan (http://dna00. bio.kyute ch.ac.jp/Progn oScan/ index.html) and Kaplan-Meier plotter (http://kmplot.com/analy sis/) 28,29 . The PrognoScan database is publicly available to assess the prognostic value of genes by metaanalysing a large collection of published cancer microarray data. The Kaplan-Meier plotter, which includes data from the Gene Expression Omnibus (GEO), The Cancer Genome Atlas (TCGA) and European Genome-phenome Atlas (EGA), provides an easy way to explore the impact of 54,000 genes on survival in 21 human cancers, with a large cohort of breast (n = 7830), ovarian (n = 2190), lung (n = 3452) and gastric (n = 1440) cancers. We therefore evaluated the correlations between CELF2 expression and patient survival in breast, ovarian, lung and gastric cancers and further analysed the impacts of CELF2 expression on the outcomes of breast cancer patients with different clinicopathological characteristics. The hazard ratios (HR) with 95% confidence intervals (95% CI) and log-rank p-value (< 0.05 is considered to be significant) was also calculated.

| Correlations between CELF2 expression and immune cell infiltration in TIMER
TIMER (https://cistr ome.shiny apps.io/timer/) is a comprehensive resource that applies a deconvolution method to infer the abundance of TIICs from the TCGA database. 30 [31][32][33] In this part, we focused on analysing breast invasive carcinoma (BRCA) and lung squamous cell carcinoma (LUSC), with lung adenocarcinoma (LUAD) as a control. Finally, we further explored the correlations between CELF2 and ICMs in various cancer types using the Gene_Corr module of the TIMER2.0 website. 34 The ICMs were derived from previous studies. 35,36 Notably, the Spearman correlation coefficients presented in the heatmap were adjusted for tumour purity.

| Gene correlation analysis in GEPIA
The online database GEPIA (http://gepia.cance r-pku.cn/index.html) is an interactive analysis tool that contains RNA-seq data from 9736 tumour and 8587 normal samples from the TCGA and Genotype-Tissue Expression (GTEx) data set. 37 We used GEPIA to explore the relationships between CELF2 and TIIC-related markers in BRCA, LUSC and LUAD. The Spearman method was applied to determine the correlation coefficient.

| Evaluation of immunotherapeutic biomarkers
We finally recruited two immunotherapeutic cohorts: (1) a cohort of 38 metastatic melanoma patients treated with anti-PD-1 monoclonal antibody (GSE78220 cohort) 12,38 and (2) a cohort of 144 melanoma patients treated with anti-PD-1 monoclonal antibody (Van Allen cohort). 39 In addition, to explore the power of CELF2 as an immunotherapeutic biomarker, we evaluated the performance of CELF2 in predicting immunotherapy response in the two cohorts and further compared with five other well-studied biomarkers, including PD-1, PD-L1, CTLA-4, CD8 and tumour mutation burden (TMB). [40][41][42] We used the receiver operator characteristic (ROC) curves and the area under the ROC curve (AUC) to measure the predictive accuracy of different biomarkers for predicting the responses to immunotherapy.

| Statistical analysis
Differential expression of CELF2 in TIMER was explored using the Wilcoxon rank-sum test. The results produced by Oncomine were displayed with P-value, fold change and gene rank. Survival was as- performed by survival R package, and the optimal cut-off value was determined by survminer R package. The ROC curves were plotted by pROC R package. p < 0.05 was considered statistically significant.

| The mRNA expression levels of CELF2 in pancancer
We first analysed the expression levels of CELF2 mRNA in pan-cancer using Oncomine database. The results demonstrated that CELF2 was significantly elevated in colorectal, gastric, kidney, leukaemia, liver and melanoma cancers relative to their matched normal tissues. In contrast, we also found that CELF2 was lower in bladder, brain and central nervous system, breast, head and neck, lung, lymphoma, ovarian, prostate and sarcoma cancers compared with normal tissues ( Figure 1A).
Detailed expression results of CELF2 in specific tumours are shown in Supplementary Table 1.
To further assess CELF2 expression in different cancer types, we used the TIMER tool to analyse RNA-seq data from the TCGA database.
We found that CELF2 expression was significantly down-regulated in

| Potential prognostic value of CELF2 in pancancer
We next investigated the impact of CELF2 expression on the prognosis of different cancers using PrognoScan (Supplementary Table 2). The results revealed a significant correlation between CELF2 expression and the survival of patients with a variety of tumours, including blood, brain, breast, colorectal, eye, lung, ovarian, skin and soft tissue cancers. Representative survival curves for each tumour are shown in Figure Figure 2S). Taken together, these results in PrognoScan and Kaplan-Meier plotter simultaneously illustrated that CELF2 was related to its better survival in breast and lung cancers.

| Elevated CELF2 expression impacted the prognosis of triple-negative breast cancer patients with lymphatic metastasis
TNBC is a special type of breast cancer in which estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) are negative, which is characteristic by the lack of effective therapeutic targets, and a high degree of malignant, easy to metastasis and relapse. 43  found CELF2 was significantly down-regulated in breast cancer, and its expression was closely related to better prognosis.
To better understand the specific mechanism by which CELF2 affects prognosis in breast cancer, by integrating clinicopathological information from the Kaplan-Meier plotter database, we explored the prognostic value of CELF2 in stratified populations. We found that CELF2 expression exerted a positive effect on both OS and RFS in breast cancer and was significantly correlated with ER, PR, HER2 status, subtype, grade, lymph node status and TP53 status of patients (Table 1)

| CELF2 expression correlated with the immune infiltration in breast cancer and lung squamous cell carcinoma
Multiple studies have confirmed that CELF2 exerts tumour suppressive effects in most tumours, which is significantly increased in developing thymocytes and activated T cells, but its comprehensive understanding within the TME remains unknown. [16][17][18][19][20][21][22][23][24][25][26] Therefore, it is necessary to investigate the relationships between CELF2 expression and TIIC infiltration in the TME. In this study, we assessed  Figure 3C). However, in LUAD, although the correlations between CELF2 with TIICs were also statistically significant, their correlations were very weak ( Figure 3B). Of particular note where the correlation coefficients were greater than 0.6 between CELF2 with CD8+ T cells, DCs and neutrophils in BRCA, as well as DCs in LUSC, indicating CELF2 plays a crucial role in immune infiltration of these cells.

| Expression correlations between CELF2 and immune marker sets
In view of the relationships between CELF2 expression and multiple in BRCA and LUSC (p < 0.0001 and r > 0.6; Figure 4A-J). In contrast, the majority of correlations between CELF2 and individual immune cell markers in LUAD were not statistically significant, and the rare statistically significant coefficients were also less than 0.3, suggesting a relatively weak correlations (Table 2 and Figure 4K-O).
We further assessed the relationships between CELF2 and these markers in BRCA, LUSC and LUAD using GEPIA, revealing similar correlations between CELF2 and T cell (general), monocyte, TAM, M2 macrophage and DC markers to those TIMER (Table 3). Previous studies have demonstrated that the proportion of TAMs in the TME and their polarization status have important effects on cancer growth, invasion, metastasis and drug resistance. 44,45 Our findings suggested that infiltration. Notably, the marker levels of Th1 (TBX21 and STAT4) and Treg (FOXP3 and CCR8) cells, which are primarily immunosuppressive, were also significantly associated with CELF2 expression. It has been shown that DCs promote tumour metastasis by decreasing CD8+ Tcell cytotoxicity and enhancing Treg responses. 4,5 Recent studies have also showed that an autologous dendritic cell vaccine can kill breast cancer cells by polarizing the Th1 response, which raised new hopes for the treatment and prevention of breast cancer. 46 However, whether CELF2 mediates the progression and metastasis of BRCA and LUSC via affecting DC infiltration remains to be further investigated.

| Correlations between CELF2 expression and the responses to immunotherapy
The above showed that CELF2 expression was significantly associated with marker gene sets of T-cell exhaustion, such as PD-1, CTLA-4 and TIM-3, implying that CELF2 might play crucial roles in immune tolerance and immune evasion (  12,35,36 The results revealed that CELF2 expression was significantly positively correlated with the expression of these molecules in BRCA and LUSC ( Figure 5A). In addition, we also observed the significant correlations between CELF2 and ICM expression  Figure 5E). Overall, our study strongly confirmed that CELF2 could be used to evaluate prognosis and responses to immunotherapy in cancer patients and is superior to remaining prevalent biomarkers.

| Experimental verification of CELF2 expression and prognosis in TNBC and LUSC tissue microarrays
The above analysis suggested that CELF2 has significant prognostic significance in TNBC and LUSC, and is closely related to the TME and immunotherapy efficacy. We further used TNBC (n = 80) and LUSC (n = 90) tissue microarrays combined with immunohistochemistry for experimental validation at the protein level. The results showed that the protein expression levels of CELF2 in TNBC and LUSC were significantly reduced compared with the normal tissues ( Figure 6A-D). In the TNBC tissue microarray, 44 patients (55%) were classified as the high CELF2 expression group and 36 patients (45%) as the low CELF2 expression group according to the optimal cut-off point of the pathological score, and survival analysis revealed that TNBC patients with high CELF2 expression had longer OS (p = 0.004; Figure 6E). Similarly, the results of LUSC tissue microarray also indicated that the OS of LUSC patients with high CELF2 expression was significantly prolonged (p = 0.0018; Figure 6F). Collectively, these two independent cohorts highlight the significant prognostic significance of CELF2 in TNBC and LUSC, and the potential possibility of being a candidate biomarker.

| DISCUSS ION
In this study, we systematically summarized the expression levels and prognostic value of CELF2 in diverse cancer types using Oncomine and TIMER database, revealing prominent differences between tumours and adjacent normal tissues. Oncomine analysis showed elevated CELF2 expression in colorectal, gastric, kidney, leukaemia, liver and melanoma cancers compared with matched normal tissues, whereas CELF2 expression was down-regulated in bladder, brain and central nervous system, breast, head and neck, lung, lymphoma, ovarian, prostate and sarcoma cancers ( Figure 1A). Based on the RNA-seq data from the TCGA database, we found that CELF2  Table 2). This was consistent with our immunohistochemistry results ( Figure 6). Furthermore, high CELF2 expression was associated with significantly prolonged OS and RFS of breast cancer patients with ER, PR and HER2 negative, lymph node metastasis and higher grade. Notably, CELF2 expression was much higher in basal-like breast cancer (also known as TNBC) than in luminal and HER2 subtype ( Figure 1B). Briefly, these findings supported that CELF2 was a potential prognostic biomarker in breast and lung cancers, and might influence TNBC development and metastasis.
Another key finding of this study was that CELF2 expression cor-  Figure 3B). These results indicated that CELF2 expression and the levels of immune infiltration were closely correlated in BRCA and LUSC, but not in LUAD. In addition, we simultaneously observed significant correlation between CELF2 and certain immunological markers using TIMER and GEPIA database, hinting that CELF2 could regulate TIIC infiltration and interaction within the TME in BRCA and LUSC (Tables 2 and 3). For example, the markers of monocytes (CD86 and CSF1R) and M2 macrophages (CD163 and MS4A4A) showed strong correlations with CELF2, while TAM markers (CCL2, CD68 and IL10) showed moderate correlations (Tables 2 and   3). These results revealed a potential role of CELF2 in modulating TAM polarization. Remarkably, immune infiltration and immune marker set analysis consistently showed a strong correlation between CELF2 and DCs in BRCA and LUSC ( Figure 3A,C, Tables 2 and 3). Combined with the indispensable role of DCs in anti-tumour immunity and the promising future of DC vaccines in tumour treatment, we are confident that clarifying the mechanism by which CELF2 interacts with DCs in the TME may provide a new target for immunotherapy. [46][47][48] Moreover, we further found that CELF2 in BRCA and LUSC were moderately to strongly correlated with the main immunosuppressive Th1 and Treg cell marker sets (TBX21 and STAT4; FOXP3 and CCR8) as well as T-cell exhaustion markers (PD-1, CTLA-4 and TIM-3) within the TME (Tables 2 and 3). These results might indicate that CELF2 could regulate T cell-mediated immunity via Treg and Th1 cell in BRCA and LUSC. Given the lack of reliable diagnostic and prognostic biomarkers as well as therapeutic targets, TNBC and LUSC treatment remains challenging, whereas immunotherapy offers patients new hope. 7,8 A recent phase III clinical study on advanced TNBC showed that the levels of stromal tumour-infiltrating lymphocytes (sTILs) were correlated with PD-L1 status, and obvious improvements in the efficacy of immunotherapy were observed only in CD8+ and sTILs+ patients who were also PD-L1+. 40 Of particular interest was the significant positive correlations between CELF2 and ICMs in BRCA, LUSC as well as digestive and urinary tumours ( Figure 5A). Further exploration also demonstrated that CELF2 was more accurate than five prevalent indicators including PD-1, PD-L1, CTLA-4, CD8 and TMB in predicting the responses to immunotherapy, hinting that CELF2 was a promising biomarker for selecting immunotherapy-sensitive patients ( Figure 5D,E).
To the best of our knowledge, our study is the first to systematically address the expression and prognostic landscape of CELF2, which plays an indispensable in RNA processing, and to explore its potential relationship with immune infiltration in pan-cancer. Second, we comprehensively analysed a large amount of data from the multiple public databases and our two tissue microarrays and validated the results by integrating immune infiltration analysis and correlation analysis of immune marker sets to increase the reliability of our conclusions.
Most importantly, we confirmed that CELF2 could effectively predict the prognosis and responses to immunotherapy in TNBC and LUSC patients, and had significant clinical translational value for TNBC and LUSC that owing poor prognosis due to lack of effective biomarkers and targets. Nevertheless, this study also had several limitations. For example, although we found that CELF2 expression was associated with the abundance of TIIC infiltration in BRCA and LUSC patients, we could not conclude whether CELF2 directly affected patient survival through immune infiltration. To overcome these limitations, future detailed molecular and cellular mechanistic studies of CELF2 and prospective studies including CELF2 expression, immune cells infiltration, and efficacy of immunotherapy in tumour patients will help provide clear answers.
In summary, elevated CELF2 expression is correlated with better prognosis and higher TIIC infiltration in a variety of tumours. Especially, for BRCA and LUSC, CELF2 may contribute to TAM polarization, participate in the interaction between DCs and TME, and regulate immune tolerance and immune escape through Treg and Th1 cells. Furthermore, we demonstrated that CELF2 is strongly correlated with ICMs in various tumours, and significantly outperforms five prevalent biomarkers in predicting the responses to immunotherapy. Therefore, CELF2 may be a crucial regulator of tumour immune cell infiltration and serve as a prognostic and immunotherapeutic biomarker in TNBC and LUSC.