Prognostic Value and Possible Mechanism of m6A Methyltransferase METTL3 in Prostate Cancer

Background: The N6-methyladenosine (m6A) methyltransferase METTL3 has been reported to be closely related to prostate cancer (PCa). Thus, we aimed to explore its prognostic value and possible mechanism in PCa. Methods: The METTL3 gene and protein expression status in between the PCa and normal tissues as well as the correlation between METTL3 and Gleason score (GS) were evaluated using The Cancer Genome Atlas Project (TCGA) dataset and the tissue microarray (TMA) dataset. The prognostic value of METTL3 was evaluated using 490 PCa patients from TCGA cohort followed by the verication using 515 PCa patients from the dataset integrated by TCGA cohort and International Cancer Genome Consortium (ICGC) cohort. The machine learning methods were used for the possible mechanism of METTL3 which was closely related to the prognosis of PCa. A nomogram was constructed to provide a quantitative approach to predict the prognosis of PCa. Results: The gene and protein expression levels of METTL3 in the PCa tissues were signicantly higher than those in the normal tissues, and the gene and protein expression levels of METTL3 in the high (GS>7) risk PCa tissues were signicantly higher than those in the low-moderate risk PCa (GS ≤ 7) tissues. The PCa patients with high expression level of METTL3 had higher risk for the progression-free survival (PFS) events and poorer short-term and long-term PFS than those with low expression level of METTL3. Through using machine learning, a METTL3 related risk model (M-RM) consisted of seven m6A methyltransferase METTL3 target genes was obtained. The high M-RM score was revealed to be signicantly related to the poor PFS of PCa as well as the high activity of the KEGG pathways closely related to the process of cell cycle. Conclusion: METTL3 has the potential to be the prognostic predictor of PCa. Its coded protein may affect the prognosis of PCa mainly through regulating the expression of their target genes closely related to the process of cell cycle. These outcomes will be benet to improve the prognosis of patients, as well as reduce the mortality of PCa.


Introduction
Prostate cancer (PCa) is the most common malignancy for male, especially in old age, as its incidence is second to lung cancer among men in western countries [1]. In our country, PCa has become the 6th incidence and 7th death of male malignant tumors. With the development of medicine, some curable therapeutic methods such as radical prostatectomy (RP) are proposed and widely used in clinical practice. However, a high recurrence rate still exists [2]. Salvage treatment at the early stage of recurrence is bene cial to prolong survival and even can cure tumor [3]. Therefore, recurrence prediction is of great signi cance to reduce mortality and improve prognosis of the patients with PCa.
Serum prostate-speci c antigen (PSA), Gleason score (GS) and pathological TNM (pTNM) staging are the common methods used to evaluate the recurrence and prognosis of PCa patients. However, there are still some shortcomings. The rising serum PSA level after curable treatment is unreliable to predict the prognosis of PCa patients, because it does not mean that the patients with rising serum PSA levels are all at a high risk of death from PCa [4]; both GS and pTNM staging are limited by subjective assessment, distant micro-metastasis, and the variations among patients with the same tumor stage or GS [4]. These above have provided the motivation and goal for further exploring a more credible prognostic predictor for PCa patients.
Recently, METTL3 (methyltransferase like 3), a critical N6-methyladenosine (m6A) methyltransferase, has gained increasing attentions. It can maintain the homeostasis of m6A methylation by methylating its target mRNAs, thus participating in diverse pathological processes [5][6][7]. It is reported that the upregulated METTL3 was closely related to the progression and poor prognosis of many kinds of cancers, such as gastric cancer [8], bladder cancer [9], oral squamous cell carcinoma [10], lung adenocarcinoma [11], ect. In case of PCa, Ma et al. and Li et al. proposed that the up-regulation of METTL3 was signi cantly related to the poor over survival (OS) of PCa by using The Cancer Genome Atlas Project (TCGA) cohort [12][13]. However, due to the few OS events in TCGA PRAD cohort (only ten OS events out ve hundred cases), OS is not recommended for PCa survival studies by using TCGA cohort [14]. In addition, how METTL3 affects the prognosis of PCa remains largely uncertain.
Therefore, in this study, we used progression-free survival (PFS) event, the recommended clinical outcome endpoint of PCa survival studies by using TCGA database [14], as a substitute for OS to evaluate the prognostic value of METTL3 for the PCa patients, which is veri ed subsequently by using the dataset integrated by TCGA cohort and International Cancer Genome Consortium (ICGC) cohort. Then, a series of machine learning methods such as gene co-expression analysis, single-sample gene set enrichment analysis (ssGSEA), etc. were performed to uncover the possible mechanism of METTL3 closely related to the prognosis of PCa. Finally, the nomogram of METTL3 combined with other clinical risk factors was constructed, which provided a quantitative approach to predict the prognosis of PCa. The detail strategy is showed in Figure 1.

Materials And Methods
Gene expression dataset of the prostate tissues  Analysis of the mechanism of METTL3 closely related to the prognosis of PCa By using TCGA cohort and RNA modi cation data, three main steps were included in this part. Firstly, the m6A methyltransferase METTL3 target genes which were signi cantly related to the prognosis of PCa patients were selected. These genes should meet the following criteria: (1) the target genes should be signi cantly co-expressed with METTL3 in PCa (the absolute value of the Pearson's correlation coe cient should be more than 0.3, and P value should be less than 0.05.); (2) the target genes should have the m6A methyltransferase METTL3 protein binding site; (3) the target genes should be signi cantly related to the prognosis of PCa (P value should be less than 0.05.). Secondly, in order to eliminating the in uence of confounding factors, the multi-variable PFS analysis via Least absolute shrinkage and selector operation (LASSO) Cox regression was performed to establish METTL3-related risk model (M-RM) by using "glmnet" R package [16], which was followed by the veri cation by using 515 PCa patients from the cohort integrated by TCGA cohort and ICGC cohort. Thirdly, ssGSEA was performed to evaluate the activity of the KEGG pathways by calculating their estimate score (ES). Pearson's correlation test was performed to identify the KEGG pathways that were closely related to the M-RM (the absolute value of the correlation coe cient should be more than 0.3, and P value should be less than 0.05.).

Nomogram construction and evaluation
A nomogram that integrated the METTL3 and other clinical risk factors was constructed with TCGA cohort using the 'rms' R package [17]. Calibration plot was drawn to describe the degree to which the predicted PFS were consistent with the observed PFS. A concordance index (C-index) was calculated to determine the discrimination of the nomogram via a bootstrap method with 1000 re-samples.

Statistical analysis
Statistical analyses were performed with R software (version 4.1.2). Pearson's correlation test was performed to determine the correlation between two variables. The t-test was used for the comparison between two groups of variables, and the variance test was used for the comparison among three groups of variables. The "survival" R package [18] was used for Kaplan-Meier curve and uni-variate Cox regression analysis. The "maxstat" R package [19] was used to determine the optimal cutoff value of METTL3 for Kaplan-Meier curve. Log Rank (Mantel-Cox) test was used to evaluate long-term PFS, and Breslow (Generalized Wilcoxon) test was used to evaluate short-term PFS. The "pROC" R package [20] was used to draw the time-depend receiver operating characteristic (ROC) curve to evaluate the diagnostic performance for the PFS events of PCa. P value < 0.05 was regarded as statistical signi cance.

Results
Up-regulated METTL3 was signi cantly related to PCa as well as the high risk PCa A total of 496 PCa tissues and 51 normal prostate tissues from TCGA dataset were used to analyze the relationship between METTL3 and PCa. The result showed that the expression level of METTL3 in the PCa tissues was signi cantly higher than that in the normal prostate tissues (Figure 2A). According to GS, the 496 PCa tissues were divided into three groups including the low-risk (GS<7) (45) group, the moderate-risk (GS=7) (246) group and the high-risk (GS>7) (205) group. We compared the expression levels of METTL3 in these three groups and found that there was a signi cant difference among them (P<0.001). As can been seen in Figure 2B, the expression level of METTL3 in the high-risk group was signi cantly higher than that in the low-risk group (P<0.01) as well as the moderate-risk group (P<0.001). Furthermore, through observing the expression level of METTL3 protein by using the PCa TMA dataset, a similar result was got. It was not only that the expression level of METTL3 in the PCa tissues was visibly higher than that in the normal tissues and the AD tissues, but also that the expression level of METTL3 in the high risk PCa tissues (GS>7) was obviously higher than that in the low-moderate risk PCa tissues (GS≤7) ( Figure 2C). Taken together, these results suggest that the up-regulated METTL3 is showed close relationship to PCa as well as the high risk PCa.  Figure 2D). 515 PCa patients from the cohort integrated by TCGA cohort and ICGC cohort were used for veri cation. Consistently, the similar results were got ( Figure 2E). Above all, we suggest that METTL3 have the potential to be the prognostic predictor for PCa.

Mechanism of METTL3 that was closely related to the prognosis of PCa
In order to uncover the mechanism of METTL3 involved in the prognosis of PCa, rstly, twenty-ve m6A methyltransferase METTL3 target genes were identi ed (S- Table 1 Figure 3C). According to the optimal cutoff value of the M-RM score, the PCa patients were divided into the high M-RM score group and the low M-RM score group. As shown in Figure 3D, the gene expression levels of variables in the M-RM and the number of the PFS events in the same period of time in the high M-RM score group were both obviously higher than those in the low M-RM score group. In addition, Kaplan-Meier curve showed that the patients with high RM score had poorer both short-term and longterm PFS than those with low PFS score (Log Rank (Mantel-Cox): X 2 =46.131, P<0.0001; Breslow (Generalized Wilcoxon): X 2 =40.082, P<0.0001; Figure 3E). This result was con rmed by using the cohort integrated by TCGA cohort and ICGC cohort subsequently (Log Rank (Mantel-Cox): X 2 =22.123, P<0.0001; Breslow (Generalized Wilcoxon): X 2 =18.510, P<0.0001; Figure 3F). Thus, we believed that METTL3 might affect the prognosis of PCa mainly through regulating the expression of these seven m6A methyltransferase METTL3 target genes in the M-RM. Finally, in order to further understand how the M-RM affected the prognosis of PCa, ssGSEA was performed. And based on it, we found that the M-RM score was signi cantly positively correlated to the activity of the cell cycle related KEGG pathways, such as cell cycle, DNA replication, base excision repair, etc. (Figure 4). Taken together, we suggest that METTL3 may affect the prognosis of PCa mainly through regulating the expression levels of their target genes which were closely related to the process of cell cycle.  Figure 5A). The nomogram based on the RM was drawn to provide the quantitative approach to predict the probability of 1-year, 3year and 5-year PFS for the PCa patients ( Figure 5B). The C-index of the RM was 0.729 (95% CI: 0.651-0.808; P<0.0001). And the calibration curve showed that the RM-based nomogram-predicted 3-year and 5year PFS of the PCa patients was highly t with the observed ones ( Figure 5C).

Discussion
METTL3 is a critical m6A methyltransferase, which can participate in diverse pathological processes through maintaining the homeostasis of m6A methylation by methylating its target mRNAs [5][6][7].
Recently, a few studies revealed that the up-regulation of METTL3 was signi cantly related to the poor OS of the PCa patients by using TCGA cohort [12][13]. However, there were only ten OS events out ve hundred cases in TCGA cohort, so that, according to the recommendations of TCGA-CDR, PFS was identi ed to be more suitable than OS for PCa survival studies [14]. Thus, in this study, by using TCGA database, PFS event was used as the substitute for OS to further evaluate the prognostic value of METTL3 for the PCa patients, and the results showed that the patients with high expression level of METTL3 had poorer both short-term and long-term PFS than those with low expression level of METTL3. Furthermore, the result of the PFS analysis by using 515 PCa patients from the cohort integrated by TCGA cohort and ICGC cohort con rmed it. Taken together, these results suggest that METTL3 had a good predictive ability for the prognosis of PCa, which is consistent with previous studies [12][13]. In addition, in order to provide a quantitative approach to predict the probability of 1-year, 3-year and 5-year PFS for the PCa patients, the nomogram based on the RM which integrated by the METTL3 and pT was constructed with TCGA cohort. The RM had a high ability to distinguish the occurrence of the PFS events, and the RM-based nomogram-predicted 3-year and 5-year PFS of PCa patients was also highly t with the observed ones, indicating the good predictive ability of it for the prognosis of PCa.
It has been reported that the role of METTL3 in the process of tumorigenesis and tumor development is tumor-speci c [22][23][24][25][26][27]. Some studies showed that METTL3 played an oncogenic role in myeloid leukaemia [22], breast cancer [23], bladder cancer [24], colorectal carcinoma [25], etc. Other studies indicated that METTL3 played a tumor suppressor in renal cell carcinoma [26] and glioblastoma [27]. In PCa, a recent study suggested that METTL3 was an oncogenic factor [28]. However, the mechanism of METTL3 that affected the prognosis of PCa were still unclear. Our results revealed that after eliminating the interference of confounding factors, m6A methyltransferase METTL3 target genes SNRNP70, CLDN15, NCAPH, DDX39A, RAD54L, EME2, and MC1R were identi ed to signi cantly affect the prognosis of PCa patients. So far there was no related studies to illuminate the mechanism of these seven genes in PCa. However, we noticed that except CLDN15, the rest m6A methyltransferase METTL3 target genes might be closely related to the process of cell cycle. SNRNP70 and DDX39A participated in transcription, NCAPH participated in DNA replication, RAD54L and EME2 participated in DNA repair, as well as MC1R promoted the process of DNA repair in skin [29]. Furthermore, by using ssGSEA and Pearson's correlation test, we noticed that the activity of some cell cycle related KEGG pathways such as cell cycle, DNA replication, and basic resection repair was signi cantly enhanced with the increasing of the M-RM scores calculated from the expression levels of these seven m6A methyltransferase METTL3 target genes. These results suggest that METTL3 might participate in PCa progression mainly through regulating the expression of their target genes which were closely related to cell cycle.
In conclusion, METTL3 had the potential to be developed as a PCa prognostic predictor. The process of cell cycle may be the possible mechanism of it in the prognosis of PCa. Our outcomes will be helpful to improve the prognosis of patients, as well as reduce the mortality of PCa.

Declarations
Ethics approval and consent to participate  Figure 1 Detail analysis strategy   Correlations between the activities of KEGG pathways and the M-RM score (Top) Heatmap of the top ten KEGG pathways that were signi cantly correlated to the M-RM score; (Bottom) Scatter diagrams of the top ten KEGG pathways that were signi cantly correlated to the M-RM score.

Figure 5
Correlations between the activities of KEGG pathways and the M-RM score (Top) Heatmap of the top ten KEGG pathways that were signi cantly correlated to the M-RM score; (Bottom) Scatter diagrams of the top ten KEGG pathways that were signi cantly correlated to the M-RM score.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.