The Oncogenic Role of γ-Aminobutyrate Aminotransferase in Human Tumor: A Pan-Cancer Analysis

Emerging evidence supports the correlation between γ-aminobutyrate aminotransferase (ABAT) and tumors, but few research groups used pan-cancer analysis to verify it previously. Therefore, this study used the Cancer Genome Atlas (TCGA) database and the Gene Expression Omnibus (GEO) to obtain information about the correlations between ABAT and tumor development, and to explore its potential effectiveness for genetic alterations in tumor prognosis. The reduced expression level of ABAT in a majority of tumors is signicantly associated with the poor prognosis. The genetic alteration of ABAT seems linked to the favorable prognosis of Uterine Corpus Endometrial Carcinoma (UCEC). Immune inltration analysis showed a signicantly positive correlation between ABAT and cancer-associated broblasts in the majority of tumors, but a highly negative correlation with Kidney renal clear cell carcinoma (KIRC), Kidney Renal Papillary cell carcinoma (KIRP), and Prostate adenocarcinoma (PRAD). Enrichment analysis showed that cell junction organization, amino acids metabolism, and neuronal system-involved behaviors might affect the pathogenesis or etiology of cancer. This study is the rst pancancer analysis that offers a detailed, comprehensive study of the process of the oncogenic roles of ABAT across different human tumors. This study is the rst pan-cancer analysis of ABAT. Novel effects of ABAT on tumor prognosis have been revealed. The relationship between the ABAT protein and gene has been displayed.

Tumors have complex regulation, and it's necessary to analyze the related gene of pan-cancer expression and determine the correlation between pre-and post-evaluation with the potential molecular mechanism [24]. The present study is the rst one using the Cancer Genome Atlas (TCGA) database and the Gene Expression Omnibus (GEO) project to conduct a pan-cancer investigation of ABAT [25][26][27]. The survival conditions, genetic expression, immune in ltration, genetic mutation, and associated cellular pathways will be used with different perspectives to identify the possible molecular mechanisms of ABAT in the clinical prognosis or pathogenesis of various human cancers.

Gene expression analysis
The authors used Tumor Immune Estimation Resource 2 nd edition (TIMER2.0) to clarify the expression of the ABAT gene in various cancers in the TCGA database [28]. Firstly, opening TIMER2.0's website http://timer.cistrome.org/ and choosing the module "Exploration" and clicking "Gene-DE" to input ABAT under "Gene Expression." After clicking "submit," the authors obtained the results of ABAT expression between various tumors or speci c tumor sub-types and the normal tissues of the TCGA project.
If some tumors couldn't present normal tissue to control, the differences in gene expression of ABAT between those tumors and normal tissues couldn't be obtained due to some de ciencies. We could combine the TCGA database and Genotype-Tissue Expression (GTEx) database and use Gene Expression Pro ling Interactive Analysis 2 nd Edition (GEPIA2.0) for data analysis to supplement those de ciencies [29]. Authors opened the website http://gepia2.cancer-pku.cn/#index and chose the module "Expression Analysis," and then clicked "Expression DIY" for selecting the box plot to express the difference between the corresponding normal tissues of the GTEx database and the tumors, such as ACC, BRCA, DLBC, LAML, LGG, OV, SARC, TGCT, THCA, and UCS. The present study set the |Log2FC| Cutoff at 1 and the p-value Cutoff at 0.01, chose "multiple datasets," then picked the preferred tumors and clicked "Match TCGA normal and GTEx data" to get them involved plot.
To determine the gene expression levels of ABAT, we used the UALCAN portal (a comprehensive, userfriendly, and interactive web resource for analyzing cancer OMICS data) to explore the ABAT protein expression level in the TCGA database [30]. The authors applied the website http://ualcan.path.uab.edu/index.html to achieve the UALCAN portal, and the module "CPTAC analysis (the Clinical Proteomic Tumor Analysis Consortium)" was chosen. Then, the authors inputted the gene name ABAT to determine the protein expression difference between normal tissues and primary tumors. Finally, we need to make sure the protein expression level of ABAT in different tumor stages. We entered the GEPIA2.0 website and chose the module "Stage Plot", and typed the gene name ABAT under "Gene", chose the cancer name, and got the violin plot of ABAT expression in all TCGA tumors at different pathological stages (stage I to IV). The box or violin plots selected the log2 [TPM (Transcripts per million)+1] to transform expression data.

Survival prognosis analysis
To identify the Overall Survival (OS) and Disease-Free Survival (RFS) of ABAT, GEPIA2.0 was used to get the signi cance map data of ABAT among all TCGA tumors. We entered the GEPIA2.0 website and chose the module of "Survival Analysis". Firstly, we clicked "Survival map" and input gene name ABAT under "Gene or Transcript", then chose the "Overall Survival" or "Disease-Free Survival" and set the Cutoff-high and cutoff-low values at 50% to divide ABAT into low-and high-expression cohorts. After that, the authors obtained the integrated survival map of ABAT by adding all tumors' names. Secondly, we used the module of "Survival Analysis" to get the ABAT survival plot in different tumors. Finally, we input gene name ABAT, chose cancer according to the survival map, and then applied the log-rank test to get the hypothesis testing and the survival plots.

Genetic alteration analysis
The cBioPortal for Cancer Genomics website (http://cbioportal.org) was chosen to study the genetic alteration characteristics of ABAT [31]. We entered the cBioPortal homepage and chose the "quick selected" module, then the "TCGA Pan-Cancer Atlas Studies" module, and clicked "Query By Gene" for moving to the next step. We input the gene name of ABAT and clicked "Submit Query" to get the result of ABAT gene mutations in various tumors in the "Cancer Type Summery" module. The outcomes of the alteration frequency, mutation type, and CNA (Copy number alteration) among all TCGA tumors were observed in the "Cancer Types Summary" module. Next, we chose the "Mutations" module to get the mutated site information of ABAT. We continued to click "View 3D Structure" to get the protein structure or the 3D (three-dimensional) structure of ABAT. We also used the cBioPortal portal to get the results of survival analysis of ABAT genetic alteration. On the homepage of cBioPortal, we selected "Uterus" in the module "Query" and input the cancer name UCEC, and then chose the "TCGA, PanCancer Atlas" in the "Uterus" to query. Moving to the next page, "Mutations" and "Putative copy-number alterations from GISTIC" were selected, and the gene name ABAT was input for further inquiry. Finally, we clicked "survival" in the "Comparison/Survival" module to obtain the survival analysis graph.
Immune in ltration analysis TIMER2.0 (the Tumor-in ltrating immune cell analysis database) was chosen to explore the information of ABAT and tumor-associated broblast in ltration. We opened the homepage of TIMER2.0 ( http://timer.cistrome.org) and input the gene name of ABAT in the "Gene Expression" in the module "Immune". Then we selected the "Cancer associated broblast" in the "Immune In ltrates" module to get the correlation results. The XCELL, MCPCOUNTER, TIDE, and EPIC algorithms were applied for the immune in ltration estimations. P-values and partial correlation (cor) values were obtained through the purity-adjusted Spearman's rank correlation test. At last, the data were visualized as a hot plot and scatter plot.

ABAT-related gene enrichment analysis
First of all, we screened 50 experimentally and veri ed ABAT-binding proteins through the STRING portal [a Protein Interaction (PPI Network) database] [32]. We opened the STRING website https://string-db.org, chose the module "Search", and then input the protein name ABAT and selected "homo sapiens". When the rst PPI network was achieved, we clicked "Settings" and set the following main parameters, such as active interaction sources ("Experiments"), the minimum required interaction score ["low con dence (0.150)"], max number of interactors to show ["1st shell (no more than 50 interactors)"], network display options "disable structure previews inside network bubbles". The authors obtained 50 proteins of the second PPI network after updating the settings. At last, we input the second PPI network into the Cytoscape 3.8.2 (an open-source software platform for visualizing complex networks and integrating these with any type of attribute data) with decoration to get the nal PPI network plot.
Secondly, the GEPIA2.0 was used to select the top 100 genes involved in ABAT for correlation analysis.
Authors opened the GEPIA2.0 website and selected "Similar Genes Detection" in the module of "Expression Analysis" and then input gene name ABAT and set 100 at the "Top # Similar Genes". After that, we added all the tumors and normal tissues to get a "List". The pair Pearson correlation analysis was performed for ABAT and the top 100 genes by applying the log2 TPM for the dot plot and indicating the P-value and the correlation coe cient (R). We selected the part of the top 100 genes to get scatter diagrams. We clicked the "Correlation Analysis" module in the "Expression Analysis" and input the gene names, and then added all tumors to get the "Plot".
Next, the hot plot of the correlation between ABAT and the top 100 genes was obtained by the TIMER2.0 portal. We opened the homepage of TIMER2.0 and selected the "Gene_Corr" in the "Exploration" module. The "Gene_Corr" module supplied the hot plot data of the selected genes and contained the partial correlation (cor) and P-value in the purity-adjusted Spearman's rank correlation test. Then, inputting "ABAT" and part names of the top 100 genes was conducted to get the correlation heat map.
To further screen genes, we explored the intersection of the top 100 genes and 50 ABAT-binding proteins by using Venn diagrams drawn on the Bioinformatics & Evolutionary Genomics website [33]. We opened the Venn website http://bioinformatics.psb.ugent.be/webtools/Venn/, uploaded two gene lists, and clicked "Submit" to get the intersection genes Venn diagram and information. Then the Metascape website https://metascape.org/gp/index.html#/main/step1 was applied for pathway analysis [34]. We combined the top 100 genes and 50 genes interacting with proteins, uploaded them to Metascape's homepage, and chose "H. sapiens" in both the "Input as species" and "Analysis as species" module, and then clicked "Express Analysis" to get the analysis reports. The clustering tree showed the same result: the more prominent and darker nodes indicate more genes and a more signi cant p-value.

Gene expression analysis data
To clarify the effects of ABAT in various human cancers, TIMER2.0 was applied to explore the expression levels of ABAT in multiple cancers in the TCGA database, as shown in Figure 1(a). The difference of ABAT expression between normal tissues and some cancers was shown with p< 0.001, including BRCA, COAD, KICH, KIRC, KIRP, LIHC, LUSC, THCA, and UCEC; but HNSC, PCPG, and STAD showed p<0.05.
To supplement the missing normal tissue control for some tumors, authors selected normal tissues corresponding to those tumors from the GETx database for comparison. Then they obtained the results in Figure 1(b), which shows the expression of ABAT of BRCA, LAML, TGCT have signi cant differences with normal tissues (p<0.05). However, we didn't nd signi cant differences in ACC, DLBC, LGG, OV, SARC THYM, and UCS.
CPTAC, which was used to express the protein level of ABAT, integrated genomic and proteomic data to identify and describe the proteins of tumors and normal tissues and explored the candidate proteins as tumor biomarkers. Figure 1(c) indicated that Clear cell renal cell carcinoma, LUAD, Breast cancer showed signi cant expression differences of protein level of ABAT with normal tissues, respectively (p<0.05 Figure 1(c)). However, there was no signi cant difference by comparing Ovarian cancer, Colon cancer, Uterine corpus endometrial carcinoma, and Pediatric Brain Cancer with normal tissues (p>0.05 Figure   1(c)). Moreover, the relationship between OS and ABAT expression levels in LIHC could be reserved for about 80 months (p=0.0021). As showed in Figure 2(b), the increased ABAT expression in CESC (p=0.39), DLBC (p=0.7), THCA (p=0.54), UCS (p=0.48) and UVM (p=0.22) seem to associate with increased OS rate, but not signi cantly. The DFS analysis of ABAT was shown in Figure 2

Genetic alteration analysis data
Previous researches have reported the single-nucleotide polymorphisms (SNPs) of ABAT were associated with some diseases, i.e., affective disorder [35]. Here, the cBioportal was selected to process the tumor data of TCGA for exploring ABAT genetic mutation levels in various cancers. As Figure 3(a) shows, the top 1 alteration of frequency of ABAT was BLCA (>5%) with "Ampli cation" as the primary type. The main component of SKCM and UCEC is "Mutation" at about 4% alteration frequency. Interestingly, UCS, ASCC, DLBC, and ESCA have all gene "Ampli cation". The types of "Mutation" and "Ampli cation" were the majority part of ABAT genetic alteration.
The sites, types, and case number of ABAT genetic alterations are shown in Figure 3(b), which offers 106 genetic alteration data, including 88 "Missense", 8 "Truncating", 5 "Splice", and 5" SV/Fusion". The alteration of site R436*/Q has been found in 2 UCEC cases, 1 SKCM case, and 1 HNSC case, which the missense mutation may cause. The missense mutation of ABAT was the primary type of genetic alteration. Figure 3 The EPIC, MCPCOUNTER, XCELL, and TIDE algorithms were used through TIMER2.0 to clarify the correlation between ABAT and tumor-in ltrating immune cells in various cancers from TCGA. Figure 4(a) shows a signi cant positive correlation between ABAT and cancer-related broblasts of BRCA, CESC, HNSC, HNSC-HPV-, LUSC, SKCM, SKCM-Metastasis, and TGCT. The negative correlation between ABAT and cancer-associated broblasts of ESCA, KIRC, KIRP, PCPG, and PRAD can also be observed in Figure  4(a). The cancer-associated broblasts (CAFs) of KIRC, KIRP, and PRAD are negatively associated with ABAT, and these tumors show a low OS and DFS when ABAT expression decreases. The CAFs of CESC, THCA, UVM and a few tumors positively correlate with ABAT expression. The survival-related analyses show that high expression of ABAT in these tumors seems linked to a not signi cant favorable prognosis in these tumors in this study.
One of the algorithms was used to get the scatter plot of the relationship between cancer-related broblasts and ABAT in individual tumors. For instance, the ABAT expression level in BRCA is positively associated with the in ltration level of the cancer-related broblasts (Figure 4(b) Rho=0.237, p=4.09e-14) based on the EPIC algorithm.

Enrichment analysis of SND1-related partners
The different pathway enrichment analysis was conducted to identify targeted ABAT combining proteins and their corresponding expression-related genes for exploring the molecular mechanism of ABAT during tumor development. We selected the 50 ABAT-binding proteins with experimental evidence from STRING and used Cytoscape software for decoration to get Figure 5 (a). The tool GEPIA2 was used to select the top 100 genes most close to ABAT from the TCGA database and draw scatter plots. Figure 5(b) indicated some genes that have a positive association with the ABAT expression level, such as ASTN1 (Astrotactin 1) (R=0.81), APC2 (Adenomatous polyposis coli 2) (R=0.82), ATCAY (caytaxin) (R=0.77), etc. And then, we used the gene and TIMER2.0 to draw the heatmap of the correlation between ABAT and those genes in cancers ( Figure 5(b)). As Figure 5(c) shows, most speci c cancers positively associate ABAT and the above genes. The intersection analysis of the 50 ABAT-binding proteins and the top 100 genes showed a joint member, ALDH5A1, in Figure 5(d).
The above two groups have combined in Metascape for exploring the results of Gene Ontology annotation. Figure 5(e, f, g) indicated the cell junction organization might produce the essential bene ts of ABAT during the tumor pathogenesis. The majority of genes may also be associated with cell behaviors, such as the biosynthesis and metabolism of amino acids, carbohydrate metabolic process, regulation of trans-synaptic signaling, etc.

Discussion
ABAT is a well-established crucial participant in the GABAergic system, neurological diseases, and mental disorders [40]. It is an experimentally primary target in drug design for Refractory epilepsy and Osteoarthritis (OA) and the potential targets with speci city in many diseases, including cancer [19,41]. Li and co-authors reported that the negatively prognosis of ESCA was associated with the high expression level of ABAT [42]. We assumed that the expression level of ABAT may be the independent factor for prognosis in tumors. The decline of ABAT expression levels may lead to a bad ending in most tumors. Consistent with our results, the low ABAT expression is highly related to poor prognosis in estrogenreceptor-positive (ER+) and estrogen-receptor-negative (ER-) breast cancer [43][44][45][46]. The prognosis status of MESGBM, HCC, ccRCC, and lung cancer was similar to breast cancer [47][48][49][50]. A previous study showed that the essential role of ABAT in the mitochondrial nucleoside salvage pathway could maintain the function of mitochondrial that may lead to the favorable prognosis of the patient with tumors [7,21].
Chen showed that the ABAT might negative Ca 2+ -NFAT1 axis and prevent the aggressive behavior of BLBC [10]. The high expression of ABAT would increase GABA signaling to modulate the cell cycle and promote synergistic interactions with glutamate signaling, and may lead to a survival advantage [51]. The evidence above shows that ABAT may provide potential prognostic indicators and therapeutic targets for BLBC and other tumors.
Nemen et al. reported the opposite conclusion that a high ABAT level would accelerate breast cancer metastases to the brain [52]. Moreover, evidence shows that the elevated ABAT expression in MB is vital for cancer cells' leptomeningeal dissemination and may affect the survival rate of cancer cells in cerebrospinal uid [19]. Elevated ABAT expression may induce transcriptional and chromatin changes in metastatic tumor cells, leading to an upregulation of mitochondrial oxidative phosphorylation (OXPHOS)related genes in metastatic tumor cells and promoting survival of cancer cells in the cerebrospinal uid [53][54][55][56][57]. The altered GABA function and metabolic behaviors of tumor cells in proliferation and migration may be the reason for these two contradictory phenomena [58,59]. But the related evidence is limited and more research is needed to explore the molecular mechanism.
The present study shows that UCEC has about 4% ABAT genetic alteration and has a favorable prognosis. Moreover, this study shows that the reduced expression level of ABAT in UCEC was not associated with the prognosis. However, we couldn't nd evidence about the association between ABAT genetic alteration and other tumors' prognoses. Zheng et al. reported that the genetic variants in the 3'-UTR region of ABAT might potentially regulate ABAT expression and lead to disease risk [14]. However, a previous study reported that the low ABAT expression levels resulted from epigenetic silencing rather than a mutation at the gene locus [19]. DNA methyltransferase enzyme, Dnmt3b, binds to the ABAT promoter to increase Methylation upstream of the transcriptional start site lead to a negative ABAT expression [57]. Similarly, a few research reported that ABAT methylation might be an indicator of poor prognosis of hematological malignancies and related to the pathogenesis of MDS and AML[60-62]. ABAT methylation plays a signi cant role in the DNA methylation pattern of Autism Spectrum Disorder (ASD) cortical neurons, either[63]. Li reported that the demethylation treatments could induce re-expression of ABAT and improve tumor patients' prognosis [64]. Authors speculated that ABAT methylation might be an essential factor in the pathogenesis of some tumors and in uence their prognosis rather than genetic alteration. But less worthy evidence could be found. The treatment of ABAT demethylation in different cancers may provide new opportunities for cancer therapy.
This study also provided evidence of the correlation between ABAT expression and cancer immunity. This study shows that ABAT expression in some tumors would increase immune in ltration and lead to adverse outcomes. Oppositely, Sato et al. reported that immune in ltration resulted in a better prognosis [65]. Bartoschec et al. reported that CAFs have different subpopulations, and each of them is an independent prognostic factor in patients with UCEC[66]. CAF is a major stromal cell type in the cancer microenvironment, can promote some tumors' progression, such as CRC, HCC, thyroid cancer, and pancreatic tumor [67][68][69][70]. Authors hypothesized that ABAT expression could in uence tumors' prognoses by affecting speci c CAFs' subpopulation in ltration. Through exploring the correlation between ABAT and CAFs subpopulation may help to identify new targets of tumor treatment. Regrettably, the related research has not to be found in the publication. More exploration is needed.
ALDH5A1 is close to ABAT, and it has a similar function to ABAT in the development of neurological diseases and mental disorders [12,40,71]. The genetic alteration of ALDH5A1 and ABAT causes inherited disorders of gamma-aminobutyric acid (GABA) metabolism and may develop Alzheimer's Disease and paranoid schizophrenia [40,71,72]. Moreover, previous studies show that the decreased expression of ALDH5A1 in patients with ovarian cancer or papillary thyroid carcinoma may lead to a poor prognosis [73,74]. The interaction of ABAT and ALDH5A1 in tumor treatments may have a synergistic effect. Li et al. had the similar conclusion that the interaction effect may work in the clinical e cacy of valproic acid (VPA) [12]. But authors cannot nd a similar study focusing on tumor therapy. This may be a novel area waiting for exploration.
Through various enrichment analyses, authors discovered that cell junction structure, amino acids metabolism, and neuronal system-related behaviors might affect cancer pathogenesis or etiology. Garcia concluded that defects of cell junction structure result in a wide range of tissue abnormalities that are common in genetic abnormalities and cancers [75]. A few pieces of research support that the migration and metastasis of tumor cells depend on the cadherin-mediated cell-cell junctions [76][77][78]. Han et al. reported that ABAT overexpression could upregulate E-cadherin in liver cancer cells [79]. The authors assumed that ABAT might induce the cell junction structure stabilization and inhibit tumor cells migration by increasing cadherin. However, mentioned in previous studies, the energy metabolism of tumor cells may be associated with ABAT, and the cancer cell's migration and invasiveness behaviors may be induced by high ABAT expression levels [10,80]. The difference of primary molecular mechanisms (such as changes in energy metabolism and induces the transcription of speci c proteins) in different tumor migrations may be the reason for these contradictory conclusions. The targeted regulation between ABAT and cadherin may be a new therapy to inhibit tumor cell migration.

Conclusion
This study is the rst research in a pan-cancer analysis that systematically evaluated the potential role ofABAT in progression and prognosis in various cancers, providing the assumption of potential therapeutic targets and evidence of cancer immunity related to ABAT expression. Therefore, it is worthy of exploring more possibilities of ABAT further. and ovarian cancer based on the CPTAC data set (*P < 0.05, ***P < 0.001). (d) Based on TCGA data, the prime pathological stages (stages I to IV) was analyzed to identify ABAT gene expression levels for RCA, ESCA, KIRC, KIRP, LIHC, LUAD, and PAAD. The logarithmic scale was produced using log2 TPM + 1.

Figure 2
Page 18/18 The association between ABAT expression level and various tumors' prognosis. TCGA database was used to discover the relationships between the gene expression of ABAT and the prognosis of multiple cancers. The Gene Expression Pro ling Interactive Analysis 2nd Edition (GEPIA2) database was used to analyze different tumors in the TCGA project for (a), (b)overall survival (OS) in ABAT gene expression, and (c), (d) disease-free survival (DFS) analyses. OS is the time from the onset of a condition to the death from any cause. DFS is the time from beginning to the rst tumor recurrence/metastasis or death for any reason. Progression-free survival is the time from onset to the rst tumor progression or death. Blue and red squares show negative and positive associations of ABAT gene expression with the prognosis, respectively. Positive results for survival and Kaplan-Meier curves are shown.

Figure 3
The genetic alteration of ABAT in different tumors. TCGA was used to obtain the mutation effects of ABAT in different tumors via the cBioPortal. This gure shows (a) the alteration frequency in different mutation molds, (b) mutation sites, (C) the 3D structure of ABAT protein, and (D) the potential links between mutation condition and versions of UCEC and BRCA survival curves, as obtained by using the cBioPortal tool. Relationships between ABAT expression and tumor-associated broblast in ltration. The various algorithms were used to identify any links between ABAT expression and immune cells. The expression level of ABAT and the tumor-associated broblast in ltration status have been explored within all cancer types in the TCGA project ( Fig. 4(a, b)).

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