A Novel BCAA-catabolism Related Gene Signature for Overall Survival Prediction of Pancreatic Carcinoma

Branched-chain amino acid (BCAA) metabolism plays an important role in the pancreatic carcinogenesis, but its mechanism remains unclear. Hence, this study was performed to investigate comprehensively the value of genes related to BCAA catabolism in pancreatic cancer. The online GEO, TCGA, and ICGC datasets were searched for bioinformatic analysis. Univariate Cox and Lasso regression were applied to construct a predictive model. Human cancer cell lines and tissue microarray (TMA) were applied for validation.


Introduction
According to the 2021 Cancer Statistics, pancreatic cancer remains as one of the top ten most common cancers and the fourth leading cause of cancer death worldwide.(1) Two major obstacles for the improvement of prognosis are the diagnosis of early-stage resectable cancer and response rate for therapy. Resistance against rst-line chemotherapy treatments always induces the progression and recurrence of tumors. Hence, identifying patients with poor prognosis is critical for the management of pancreatic cancer.
Cancer harbors a reprogrammed metabolic pro le to meet the energetic needs of the aberrant and neoplastic regeneration. (2) The pancreas is the foremost exocrine and endocrine organ, and its metabolic function is speculated to be critical in carcinogenesis. Hence, the metabolism of branched-chain amino acid (BCAA) has been paid great attention. Numerous studies have demonstrated that BCAA metabolism is in uenced by cancer, to a great extent. (3) Two recent prospective cohort studies have indicated that the serum BCAA level in the peripheral blood is a high-risk factor for the development of pancreatic cancer.(4-6) However, the exact mechanism of BCAA in carcinogenesis is unclear. In the human body, 40 BCAA-catabolism enzymes (BCE) have been identi ed, some of which, such as the branched-chain aminotransferase (BCAT), have emerged as useful markers for tumor prognosis. The genomic landscapes of pancreatic cancer have been well-described and studied. (10,11) The representative sequencing databases are Gene Expression Omnibus database (GEO), The Cancer Genome Atlas (TCGA), and International Cancer Genome Consortium (ICGC). To elucidate the mechanism of the BCAA metabolic enzymes, the current study applied three online databases to investigative role of BCE genes and construct model for prognosis prediction. Further, the bioinformatic ndings were validated in human cancer cell lines and tissue microarray (TMA) of human pancreatic cancer. Immune in ltration analysis was also performed to check the relationship between BCE genes and macrophage.

Data sources and processing
To identi ed the prognosis related genes, genomic data of RNA sequence and micro-array for pancreatic cancer were queried from TCGA cohort, ICGC cohort, and GEO cohort on Sep 1, 2020. PKM value of TCGA cohort, normalized counts value of ICGC cohort, and RAM value of GEO cohort were extracted and applied for further analysis. Survival data was available for TCGA cohort, ICGC cohort, and one GEO cohort (GSE57495). To investigate the expression level of BCE genes in pancreatic cancer, three GEO databases (GSE15471, GSE16515, and GSE28735) which included both tumor and normal control samples. Totally, 48 BCE genes were retrieved from literature and KEGG database (Supplemental Table 1).(10) This is a secondary analysis based on the open online database, and the informed consent was waived.

Survival Signature development and validation
In the study, we set the TCGA cohort as train cohort, while ICGC and GSE57495 as test cohorts. All the analyses were performed in Rstudio software with implemented packages. All BCE genes were treated as continuous variable in signature development. Firstly, All BCE genes were veri ed in three GEO datasets by t.test between tumor and normal control samples. Secondly, BCE genes in TCGA were screened by univariate Cox analysis of overall survival (OS) with survival package. Finally, LASSO-penalized regression analysis for model construction with candidate genes with glmnet/Survival package. After prognostic genes with altered expression in tumor were identi ed, the risk-score was calculated based on the formula: . In the equation, n represented selected gene number, exp(G i ) represented the expression of gene i, while β i represented the coefficient for gene i. Based on the risk-score mean value, TCGA and ICGC patients were both strati ed as high-risk and low-risk groups. For survival analysis, we applied Kaplan-Meier method to calculate the OS plot in different risk groups with survminer package. The log-rank test was performed to check the statistical signi cance. Forest plots were as draw to demonstrate the Hazard ratio (HR) of selected prediction genes.

Protein-protein interaction (PPI) analysis
The PPI analysis of the prognostic genes was performed using the STRING database (http://stringdb.org), which provides critical assessment and integration of protein interactions.

Immune in ltration analysis
The immune in ltration score of immune cells between two groups were calculated with single-sample gene set enrichment analysis (ssGSEA) with gsva package.

Real-Time PCR
Total RNA was isolated using TRIZOL Reagent (Invitrogen, Life Technologies) and was converted to cDNA using the PrimeScriptTM RT reagent Kit (Takara, Japan). Expression levels of mRNA were measured by real-time PCR (Applied Biosystems, 7500, USA) using SYBR Premix Ex TaqTM II (Takara, Japan). Total amount of mRNA was normalized to actin mRNA. The primer sequences were shown in Supplementary Table 2.
Immunohistochemical staining (IHC) with TMA A tissue microarray (Shanghai Outdo Biotech Co., Ltd. Shanghai, China) with 90 pancreatic carcinoma samples and paired adjacent samples were applied for IHC. Rabbit antihuman polyclonal ABAT antibody (Sigma-Aldrich, Cat# HPA041690, USA) and BCAT2 antibody (Sigma-Aldrich, Cat# HPA054091, USA) were used at a 1:500 dilution. Mouse antihuman CD68 antibody (Servicebio, Cat# GB14043, China) was used at a 1:200 dilution. Quantitative analysis of the staining was performed with histochemistry score (Hscore), which was determined based on the intensity of staining and the proportion of labeled tumor cells as previously described. (12) The Hscore was calculated based on the formula: . In the equation, i represented the graded of staining intensity, which included no staining (i=0), weak (i=1), moderate (i=2), and strong (i=3), while p i represented the percentage of labeled tumor cells with the corresponding stating intensity. To quantify the immune cell in ltration, positive cell in the tumoral and stromal compartments were enumerated separately and normalized per unit area as cells/mm 2 . (13) Statistics.
All analyses were set at two-sided p value <0.05 as the threshold for statistical significance. The data were expressed as mean ± standard deviation (SD).
Only SDSL showed no alteration in any of the three GEO cohorts and was excluded from further analysis. ABAT, ACAT1, and BCAT2 were signi cantly downregulated in pancreatic cancer in all three GEO cohorts.
BCAT1, DBT, and HMGCLL1 were downregulated in at least one cohort. PPI analysis with the six BCE genes indicated that BCAT2 was the hub gene (Fig. 1C).
Next, Lass-Cox analysis was applied to nd out independent predictors for prognosis. Five of the six BCE genes were identi ed and the risk-score was calculated as follows: BCAT1×0.18 -ABAT×0.06 -ACAT1×0.21 -BCAT2×0.40 + DBT×0.13. According to the risk-score, the high-risk and low-risk groups were strati ed in TCGA, ICGC, and GSE57495 cohorts. The Kaplan-Meier plot indicated that high-risk group in TCGA had signi cantly worse prognosis than low-risk group (p = 0.02), with median survival times of 1.62 and 1.90 years, respectively ( Fig. 2A). Moreover, the risk-score was validated using ICGC (high-risk vs. lowrisk: 1.25 years vs. 1.87 years; p = 0.01, Fig. 2B) and GSE57495 (high-risk vs. low-risk: 1.49 years vs. 2.63 years; p = 0.02, Fig. 2C). The association between risk groups and clinical parameters was shown in Table 2. The Kaplan-Meier plot of ve BCE genes with the mean expression as cut-off value were shown in Fig. 2D. With the above survival analysis, we might nd ABAT, ACAT1, and BCAT2 were favorable factors for a good prognosis in pancreatic cancer. The predictive value of the risk-score was further investigated by considering the clinical parameters. In the TCGA cohort, the age, gender, AJCC stage, histological grade, tumor location, and risk-score were included in the survival analysis. Multivariate Cox analysis indicated that risk-score was an independent predictor for prognosis (HR = 3.16, 95%CI = 1.63-6.12, p < 0.01; Fig. 3A). In ICGC, the age, gender, AJCC stage, and risk-score were included in the survival analysis. The results from the multivariate Cox analysis indicated that the risk-score was also an independent predictor for prognosis (HR = 3.16, 95%CI = 1.04-2.35, p = 0.03; Fig. 3B). The 5-gene risk-score showed robustness in the prognostic prediction of pancreatic cancer.
Expression of the ve BCE genes were veri ed in several human cancer and normal cell lines by RT-PCR. As shown Fig. 4A, mRNA expression of ACAT1 and DBT was undetermined, while ABAT and BCAT2 were selected for further experiments due to the stable expression in all cancer cells. Consistently with expression level in three GEO cohorts, Hscore of ABAT and BCAT2 signi cantly downregulated in carcinoma than adjacent tissues by quantitative analysis of IHC staining on TMA (Fig. 4B). Tumors with T3 stage also shown lower Hscore of ABAT and BCAT2 than T1-2 stage (Fig. 4C). The Hscore of ABAT was also linearly correlated with BCAT2 (R² = 0.875, p < 0.001, Fig. 4D). Further, cases were divided into high-expression and low-expression groups, based on mean level of Hscore. As shown in Kaplan-Meier plot, both ABAT and BCAT2 with high-expression shown worse survival than low-expression cases (Fig. 4E).
Given the reported immunomodulatory function of BCAA, the immune in ltration scores were compared between high-risk with low-risk groups. As shown in Fig. 5, dendritic cells, macrophages, neutrophils, and Th2 cells were enriched in high-risk group both for TCGA cohort (Fig. 5A) and ICGC cohort (Fig. 5B). Here, we focused the function of macrophages in BCAA metabolism. To distinguish the exact location, CD68 + total macrophage density was evaluated independently in TAM tumoral and stromal compartments. As shown in Fig. 5C, decreased positive cell density of stromal macrophage was observed in BCAT2 with low-expression than high-expression (p < 0.05), instead of tumoral macrophage. Besides, no signi cant association were detected for both tumoral and stromal macrophages with ABAT expression (Fig. 5D).

Discussion
In this study, BCAT1 and DBT were independent predictors for worse prognosis, while BCAT2, ACAT1, and ABAT were independent predictors for better prognosis. The BCAT enzymes, including BCAT1 and BCAT2, catalyzed the initial step of BCAA catabolism. Several studies have reported that BCAT1 promoted tumor proliferation and predicted poor survival in tumors, such as glioblastoma, gastric cancer, and non-small cell lung cancer. (14)(15)(16) Recent studies have shown that BCAT2-BCAA catabolism was critical in the development of pancreatic cancer, and the loss of BCAT2 expression signi cantly hampered the proliferation of this carcinoma. (17)(18)(19) A study on BCAA metabolism has indicated that pancreatic cancer showed decreased BCAA uptake, and BCAT1 and BCAT2 were not required for tumor formation.
(16) In the present results using human pancreatic cancer sample, BCAT1 was positively associated with poor survival, but BCAT2, as the hub gene of the predictive model, showed otherwise. The ACAT1 enzyme acted in the degeneration of ketone, which is the last step in the catalysis of BCAA into cholesterol. Li et al. found that the upregulation of ACAT1 was positively correlated with the survival from pancreatic cancer. (20) In clear-cell renal-cell carcinoma, ACAT1 favored the survival prognosis, which was not consistent with the current data. (21) Such discrepancy between the animal experimental results with clinical ndings indicated the BCAT1/2 and ACAT1 regulated the cancer in a tissue-speci c manner. This result needed further investigation. DBT was one of the three enzymatic components of the branchedchain alpha keto dehydrogenase complex (BCKDH), which was also the key enzyme in BCAA metabolism.
No study has directly investigated the function of DBT in tumors. However, the BCKDH-a was shown to sustain the growth of pancreatic cancer by regulating the lipid metabolism from BCAA. (17) Hence, the function of DBT should be further elucidated on the basis of the current nding.
Prospective studies have already validated the predictive value of serum BCAA for the risk of pancreatic carcinogenesis.(4-6) As described above, many enzymes for the metabolism of BCAA have been proved to be important in tumor progression and predictive for prognosis. This study was rstly focused on the immune in ltration of the BCE genes. The results showed that high-risk patient, who was determined by the 5-gene, harbored signi cantly enriched macrophage. The relation between BCAA metabolism with macrophage was poorly described. Theoretically, BCAA regulates the cell function not only by cellular signal transduction, similar to the mTOR pathway, but also triggers the metabolic reprogramming and cascades metabolite signal transduction. Ikeda et al. reported that BCAA sustained the balance of Treg cells by activating the mTOR pathway. (22) In glioblastoma, the phagocytic activity of macrophages was proved to be reduced by the tumor-excreted branched-chain ketoacids, the initial metabolites of BCAA catalyzed by BCAT1/2.(8) In addition, in macrophages, the metabolic pro le was controlled by BCAT1 and associated with in ammatory conditions.(23) No studies have reported BCAA metabolism regulated macrophage functions in pancreatic cancer. Here, the current ndings provided clues for the relation between BCAT2 and stromal macrophage in ltration. Whereas, further experiments to investigate the detailed mechanism were warranted.
In conclusion, a risk-score involving 5 BCE genes was proposed to predict the poor prognosis of pancreatic cancer in TCGA cohort, and the risk-score was validated in ICGC and GEO cohort. Based on the immune in ltration analysis, the underlying mechanism might be BCAT2 associated stromal macrophage in ltration.