Genetic variations in genes of metabolic enzymes predict postoperational prognosis of patients with colorectal cancer

The distributions of 697 patients’ demographic and clinicopathologic characteristics by death or recurrence were summarized in Additional file 1: Table S1. During the median follow-up time of 35.7 months (range, 3.1-83.4 months), there were 205 patients (29.4 %) died and 240 patients (34.4 %) developed recurrence among CRC patients. Significant higher death and recurrence risks were observed in patients at late stage (stages III + IV) with HRs of 3.82(95 % CI 2.75-5.33, P < 0.001) and 3.13 (95 % CI 2.32-4.23, P < 0.001), respectively, when comparing to those at early stage (stages I + II). Expectably, patients with adjuvant chemotherapy exhibited significant protective effect on OS and RFS with HRs of 0.36 (95 % CI 0.25-0.52, P < 0.001) and 0.45 (95 % CI 0.32-0.63, P < 0.001), respectively, when comparing to those without adjuvant chemotherapy. There was no significantly different death or recurrence risk observed in patients with regard to gender, age, hospital site, tumor position, or tumor differentiation. Very similar demographic and clinicopathologic characteristics were observed in validation cohort of 256 CRC patients (Additional file 2: Table S2).

After excluding 2 SNPs with unqualified call rate (84.9 % for rs9708193 and 66.6 % for rs4932158), we eventually analyzed 16 SNPs in 7 genes encoding the TCA cycle core metabolic enzymes. We assessed the association of each individual SNP with CRC death or recurrence risk by multivariate Cox regression with adjustment for gender, age, hospital site, tumor position, clinical stage, tumor differentiation and adjuvant chemotherapy. As shown in Table 1, there were 2 SNP in SDHC gene and 2 SNPs in SDHD gene exhibited significant association with increased CRC death risk under additive and recessive models, respectively. And the most significant result was found in the association of rs4131826 of SDHC gene with OS (HR 0.61, 95 % CI 0.47-0.79, P < 0.001) under additive model. In addition, 2 SNPs in SDHD and FH genes exhibited borderline significant association with CRC death risk under additive model. In the RFS analysis, we found that 1 SNP (rs4131826) in SDHC gene, 3 SNPs (rs10789859, rs544184 and rs7121782) in SDHD gene, and 1 SNP (rs12071124) in FH gene had significant associations with CRC recurrence risk (Table 1). We calculated the false-positive report probability (FPRP) to validate all statistically significant findings. As shown in Additional file 3: Table S3, rs4131826 in SDHC gene and 3 SNPs in SDHD gene were considered to be noteworthy. Further analysis showed that 3 SNPs (rs10789859, rs544184 and rs7121782) in SDHD gene were located within the same haplotype block with strong linkage disequilibrium (LD = 0.947, 0.896 and 0.923 respectively). Therefore, we only select one (SNP rs544184) for further validation. As shown in Table 2 with a total of 4 validated SNPs, similar significant results were confirmed for rs4131826 in SDHC gene, rs544184 in SDHD gene and rs12071124 in FH gene, but not for rs12064957 in SDHC gene. We have further examined the expression of SDHC and SDHD in tissue samples from CRC patients with different genotypes of SNPs rs4131826 and rs544184 using immunohistochemical staining (IHC). As showed in Fig. 1, patients carrying VV or WV genotype of rs4131826 exhibited the significantly higher expression of SDHC when compared with those carrying WW genotype, and patients carrying WW or WV genotypes of rs544184 exhibited significantly higher expression of SDHD when compared with those carrying VV genotype.

To assess the cumulative effect of combined SNPs on CRC clinical outcomes, we grouped the patients by the number of unfavorable genotypes of SNPs with significant or borderline significant association under dominant model, and evaluated their associations with CRC death or recurrence risk by multivariate Cox proportional hazard model. As shown in Table 3, significant increased death risk were observed in patient groups with 2, 3, and 4 unfavorable genotypes, with HRs of 1.93 (95 % CI 1.18-3.18), 2.14 (95 % CI 1.28-3.58), and 3.17 (95 % CI 1.79-5.61), respectively, when comparing to group with no more than 1 unfavorable genotype. The cumulative death risk of CRC patients conferred by combined unfavorable genotypes exhibited a significant dose-dependent manner with P for trend < 0.001. Similarly, the significant increased recurrence risks were observed in patient groups with 2, 3, and 4 unfavorable genotypes, when comparing to those with zero or 1 unfavorable genotype. And also the increased combined recurrence risk exhibited a dose-dependent manner with P for trend = 0.001. As shown in Additional file 4: Table S4, very similar cumulative effect of unfavorable genotypes was also observed in validation cohort.

To evaluate whether the number of unfavorable genotypes could modify the protective effects of postsurgical adjuvant chemotherapy on CRC patients’ clinical outcomes, Kaplan-Meier curves analysis and log-rank test were performed to assess the difference of death and recurrence risks between patient groups with lower and higher number of unfavorable genotype. As shown in Fig. 2, more prominent protective effects conferred by chemotherapy on CRC patients were observed in patient group with higher number of unfavorable genotypes in both overall survival and recurrence-free survival analyses with HRs of 0.28 (95 % CI 0.15-0.52, P = 6.02 x 10^-5) and 0.35 (95 % CI 0.19-0.62, P = 3.76 x 10^-4), respectively (Fig. 2b and d). While less prominent protective effects were observed in patient group with lower number unfavorable genotype in both the overall survival analysis (Fig. 2a) and recurrence-free survival analysis (Fig. 2c).

We further explored the higher order gene-gene interactions to reveal whether complex interactions among these significant SNPs could potentially predict the CRC outcomes by survival tree analysis. In the OS analysis, 4 SNPs, including SDHC: rs4131826, SDHC: rs12064957, SDHD: rs544184, and FH: rs12071124, exhibited gene-gene interactions, resulting in 5 terminal nodes with different OS times (Fig. 3a). SNP SDHC: rs4131826, which initially split the survival trees under dominant model, was the primary factor contributing to the survival difference. Patient groups with lower risk were composed of Node 1 and Node 2 with longest median survival time (MST) of 42.9 months, while patient groups with higher risk were composed of Node 3 with shorter MST of 35.6 months, and Node 4 or Node 5 with shortest MST of 31.5 months (Fig. 3b). In the RFS analysis, another set of 3 SNPs, including FH: rs12071124, SDHC: rs4131826 and SDHD: rs544184, exhibited gene-gene interactions, resulting in 4 terminal nodes with different RFS times (Fig. 3c). The initial split site on the recurrence-free survival tree was due to FH: rs12071124, suggesting that this SNP was the primary factor contributing to the recurrence risk difference in our cohort. Longest MST for recurrence was 41.0 months observed in Node1 CRC patients, and significant shorter recurrent MST were observed in Node 2, 3, and 4 patient groups with a P value of 0.001 (Fig. 3d).

In the present study, we evaluated the effects of SNPs in genes encoding TCA cycle key metabolic enzymes on the clinical outcomes of 697 CRC patients. We identified 4 SNPs to be significantly associated with CRC death risk and 5 SNPs to be significantly associated with CRC recurrence risk. Cumulative effect analysis showed that these SNPs exhibited significant dose-dependent effects on CRC clinical outcomes. Similar significant results were confirmed in a validation cohort. Further survival tree analysis revealed that SNP rs4131826 in SDHC gene and SNP rs12071124 in FH gene were the primary factors contributing to OS and RFS, respectively. As the best of our knowledge, our study for the first time comprehensively assessed the association of SNPs in TCA cycle metabolic key enzyme genes and CRC prognosis.

Accumulating evidences have suggested that genetic backgrounds may affect the risk and prognosis of CRC [18‐20]. Previous studies have reported that SNPs in genes of most metabolic enzymes, such as glutathione S-transferases, N-acetyltransferase, and cytochrome P450 superfamily of enzymes, can affect enzyme activities and be linked to CRC susceptibility [21‐23]. TCA cycle in mitochondria is the core metabolic pathway, which consists of three key metabolic enzymes complexities (SDH, FH and IDH). Recent metabonomic studies have reported that altered metabolite profiles of TCA cycle are observed in serum and urinary samples from CRC patients [24, 25]. Despite the extensive investigations of metabolic enzyme and TCA cycle metabolite markers in CRC, there is no study focusing on the association between SNPs in TCA cycle metabolic enzyme genes and CRC prognosis so far.

In our study, SNP rs4131826 in SDHC gene, SNP rs544184 in SDHD gene and SNP rs12071124 in FH gene were found to be significantly associated with the OS or RFS of CRC patients. IHC analysis further indicated the potential effect of SNPs rs4131826 and rs544184 on the expression of SDHC and SDHD in tissue samples, respectively. Our findings was supported by previous studies showing that the expression of SDHC and SDHD was reduced in tumor tissues and their roles as tumor suppressors have been identified [26‐28]. The mutations in genes encoding enzymes of SDH complex and FH in TCA cycle pathway have previously been linked with mitochondrial dysfunction and cancers [11]. Mutations in SDH complex have been initially observed in paraganglioma [29, 30], and subsequent studies have reported those mutations to be involved in T-cell leukemia and gastrointestinal stromal tumor [31, 32]. The potential mechanism of genetic alterations in SDH genes associated with cancers has been proposed that these mutations cause the reduction of SDH activity, increase the mitochondrial succinate levels and thus increase mitochondrial reactive oxygen species and oxidative stress [28]. Mutations in FH metabolic enzyme have been observed in multiple uterine leiomyomata and aggressive renal cell carcinoma [33, 34]. FH converts fumarate to malate in the TCA cycle, and FH mutations lead to the accumulation of oncometabolite furmarate [35]. Although the specific mechanisms by which SDH/FH is connected with malignancies are probably multifactorial, strong evidences have indicated that dysfunctional cell signaling stimulated by oncometabolites such as succinate and fumurate may be the primary mechanism [6, 8, 13, 36]. In our study, the significant SNPs identified in SDH and FH genes were all located in the predicted transcriptional factor binding sites (TFBS), suggesting that these SNPs may affect the expression of SDH and FH enzymes. Different enzyme levels may lead to different intracellular concentrations of oncogenic metabolites, resulting in different characteristics of cancer cells which may further affect the prognosis of patients. However further functional studies are needed to reveal the underlying mechanism of our observational results.

Additionally, further stratified analysis indicated that more prominent protective effects conferred by adjuvant chemotherapy after surgical resection on CRC patients were observed in subgroup with higher number of unfavorable genotypes (Fig. 2), in both the overall survival and recurrence-free survival analyses. These results suggest that the SNPs in TCA cycle metabolic enzymes may serve as potential surrogate biomarker for individualized CRC therapy. Although some histological factors such as tumor stage, differentiation and grade, have been identified to be associated with CRC prognosis, there is still a large lack of specific and sensitive biomarker for predicting CRC outcomes [37]. Our findings, together with other molecular biomarker such as KRAS, BRAF and p53 mutations, may be incorporated into the histological prognostic factors and improve the predictive value in outcomes and treatment response.

Conclusively, our study indicates that genetic variations in TCA cycle metabolic enzyme genes are significantly associated with OS and RFS of CRC patients. Further functional studies are needed to reveal the underlying mechanisms implied in our observational findings.