AMPK variant, a candidate of novel predictor for chemotherapy in metastatic colorectal cancer: A meta-analysis using TRIBE, MAVERICC and FIRE3

Abstract

AMP-activated protein kinase (AMPK) is a key sensor of energy homeostasis and regulates cell metabolism, proliferation and chemotherapy/radiotherapy sensitivities. This study aimed to explore the relationship between the AMPK pathway-related single nucleotide polymorphisms (SNPs) and clinical outcomes in patients with metastatic colorectal cancer (mCRC). We analyzed a total of 884 patients with mCRC enrolled in three randomized clinical trials (TRIBE, MAVERICC and FIRE-3: where patients were treated with FOLFIRI, mFOLFOX6 or FOLFOXIRI combined with bevacizumab or cetuximab as the first-line chemotherapy). The association between AMPK pathway-related SNPs and clinical outcomes was analyzed across the six treatment cohorts, using a meta-analysis approach. Our meta-analysis showed that AMPK pathway had significant associations with progression-free survival (PFS; p < 0.001) and overall survival (OS; p < 0.001), but not with tumor response (TR; p = 0.220): PRKAA1 rs13361707 was significantly associated with favorable PFS (log HR = −0.219, SE = 0.073, p = 0.003), as well as PRKAA1 rs10074991 (log HR = −0.215, SE = 0.073, p = 0.003), and there were suggestive associations of PRKAG1 rs1138908 with unfavorable OS (log HR = 0.170, SE = 0.083, p = 0.041), and of UBE2O rs3803739 with unfavorable PFS (log HR = 0.137, SE = 0.068, p = 0.042) and OS (log HR = 0.210, SE = 0.077, p = 0.006), although these results were not significant after false discovery rate adjustment. AMPK pathway-related SNPs may be predictors for chemotherapy in mCRC. Upon validation, our findings would provide novel insight for selecting treatment strategies.

Introduction

AMP-activated protein kinase (AMPK) is a key sensor of cellular energy homeostasis. In response to stress, such as low glucose and hypoxia, which increase intracellular AMP/ATP ratio, AMPK is phosphorylated by the upstream kinase LKB1 and is activated.¹ AMPK mainly induces a metabolic switch from anabolism to catabolism to regulate energy homeostasis: AMPK regulates cell metabolism to inhibit Acetyl-CoA carboxylase (which induces the biosynthesis of fatty acids), HMG-CoA-Reductase (which induces the biosynthesis of cholesterol),² and TORC2 (which induces glyconeogenesis).^3,4 Furthermore, AMPK suppresses cell proliferation through mTOR pathway by phosphorylating upstream tumor-suppressor protein TSC1 and TSC2,⁵ and cell cycle through p53–p21/p27 pathway.^3,6 (Fig. 1) In cancer, phosphorylated AMPK inhibits tumor growth,⁷ and the phosphorylated AMPK expression status in cancer tissue is associated with superior prognosis.^8,9 AMPK is activated in response to chemotherapy and radiotherapy: AMPK facilitates a DNA damage response-induced G2-M checkpoint, and mediates the antiproliferative effects of common cytotoxic cancer therapy.¹⁰ In addition, loss of LKB1/AMPK activation rapidly acquired resistance to antiangiogenic therapy.¹¹ Therefore, AMPK might be a promising target in cancer treatment.

Figure 1.

Schematic representation of the AMPK-related pathways. AMPK is regulated by activators, such as LKB1 and environmental factors, and suppressors, such as UBE2O and ALDOA. AMPK mainly increases chemo/radio-sensitivity, suppresses cell proliferation, and changes the cell metabolism from anabolism to catabolism, via various signaling pathways. The molecules enclosed in boxes are analyzed in our study. Abbreviations: ω3PUFAs, omega-3 polyunsaturated fatty acids.

Recent clinical trials have shown that the combinations of fluorouracil (5-FU), oxaliplatin and irinotecan (such as FOLFIRI, mFOLFOX6 and FOLFOXIRI) combined with molecular targeted agent (bevacizumab or EGFR inhibitor) are effective first-line chemotherapies for mCRC, based on patient characteristics and the molecular profiling of the tumor. Recently, immunotherapy has shown significant activity in mCRC patients with microsatellite instability-high (MSI-H) status.^12,13 However, there is clear need for effective predictors for chemotherapy to select the best optimal treatment strategy. Although anti-EGFR antibody would be better for left-sided, wild-type RAS and BRAF, and HER2 negative tumors,¹⁴ no effective predictor for conventional chemotherapy or bevacizumab has been identified. AMPK status may be an important pathway to predict the effects of chemotherapy and/or molecular targeted therapy, and the variants that influence the molecular function may be promising.

We thus investigated relationships between the AMPK pathway-related SNPs and clinical outcomes in patients with mCRC treated with chemotherapy in three randomized clinical trials (TRIBE, MAVERICC and FIRE-3), using a meta-analysis method. As a result of our data including the efficacy of representative chemotherapy (FOLFIRI, mFOLFOX6 or FOLFOXIRI plus bevacizumab or cetuximab) for mCRC, we could suggest that the AMPK pathway-related SNPs might be biomarkers for selecting treatment strategy for mCRC.

Materials and Methods

Baseline patients and study design

The study subjects were 884 patients with mCRC treated with chemotherapy. The patients underwent FOLFIRI, mFOLFOX6 or FOLFOXIRI combined with bevacizumab or cetuximab as the first-line chemotherapy in three prospective phases II or III clinical trials: TRIBE (NCT00719797),¹⁵ MAVERICC (NCT01374425)¹⁶ and FIRE-3 (NCT00433927).¹⁷ In TRIBE, patients underwent either FOLFIRI plus bevacizumab or FOLFOXIRI plus bevacizumab; in MAVERICC, either FOLFIRI plus bevacizumab or mFOLFOX6 plus bevacizumab; in FIRE3, either FOLFIRI plus bevacizumab or FOLFIRI plus cetuximab (detailed in Supporting Information Methods). Use of the clinical data and clinical samples for molecular analysis were approved by the institutional review boards of each participating institute. This study was conducted in accordance with the Declaration of Helsinki, Good Clinical Practice and REMARK guidelines.¹⁸ Patients without sufficient peripheral whole blood samples for analysis were excluded (the inclusion/exclusion criteria is shown in Supporting Information Fig. S1A).

Selection of single nucleotide polymorphisms and genotyping

Nineteen polymorphisms underlying AMPK pathway-related molecules (PRKAA1: rs461404, rs13361707, rs10074991; PRKAA2: rs10789038; PRKAB1: rs4213; PRKAB2: rs3766522; PRKAG1: rs1138908; PRKAG2: rs13224758, rs12703159; PRKAG3: rs6436094, rs692243; TSC1: rs13295634, rs1073123; TSC2: rs3087631; RHEB: rs6972955; MTOR: rs2295080; UBE2O: rs3803739; and ALDOA: rs9783783, rs2071390) were selected using the following predefined criteria: (i) biological significance according to published literature review; (ii) a cut-off of minor allele frequency is at least 10% in Caucasians (in the Ensemble Genome Browser: https://www.ensembl.org/index.html); and (iii) tag SNPs chosen by the HapMap genotype data with r² threshold = 0.8: https://snpinfo.niehs.nih.gov/snpinfo/snptag.html (Supporting Information Table S1). Genomic DNA was extracted from peripheral whole blood taken before initiation of chemotherapy, using the QIAmp Kit (Qiagen, Valencia, CA) according to the manufacturer’s protocol (www.qiagen.com). The OncoArray including 530K single nucleotide polymorphisms (SNPs) was used for genotyping in this study (Illumina, San Diego, CA).¹⁹ The SNPs of LKB1 were not included in our OncoArray data and excluded from this analysis.

Statistical analysis

The primary purpose of our study was to evaluate the associations of selected SNPs in AMPK pathway-related molecules with tumor response (TR), progression-free survival (PFS) and overall survival (OS). Patients were defined as responders when achieving complete (CR) or partial response (PR) and nonresponders when stable (SD) or progressive disease (PD) occurred as defined by Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 criteria. TR was calculated from the number of patients who achieved CR and PR to treatment. PFS was defined as time from randomization until disease progression, death or until last follow-up in patients who were alive and remained free of disease progression. OS was defined as time from randomization until death. Patients still alive were censored at the last date of follow-up. Allelic distribution of genetic variants was tested for deviation from Hardy–Weinberg equilibrium (HWE) using the Chi-squared test and all genetic variants examined were within the HWE in each cohort (p > 0.05). The comparison of baseline patient characteristics between the three cohorts was analyzed using Chi-square test. Multivariable logistic regression model was performed to test the associations between the SNPs and TR, and multivariable Cox proportional hazards regression model and Wald tests were performed to test the associations between the SNPs and survival outcomes (detailed in Supporting Information Methods).

Each SNP was evaluated across all treatment arms using a meta-analysis approach implemented in the METASOFT software. SNPs were coded using an additive genetic model for the number of variant alleles, that is, the common homozygote is represented by 0, the heterozygote by 1 and the variant homozygote by 2. The prognostic value of each SNP across all arms was quantified using the inverse-variance-weighted effect size.²⁰ We also analyzed heterogeneity of effects across all arms using Cochran’s Q statistic. We conducted pathway analysis to identify effects across multiple SNPs and treatment arms using a statistically powerful approach called Pegasus.²¹ The Pegasus analysis requires a p value for each SNP and linkage disequilibrium (LD) estimates; the p values corresponded to meta-analysis common effects, and LD was estimated through the SNAP search (http://archive.broadinstitute.org/mpg/snap/ldsearchpw.php) using 1000 Genomes data (Supporting Information Table S2). p values were adjusted for multiple testing using the BH false discovery rate (FDR) approach. All analyses were performed by SAS 9.4 (SAS Institute, Cary, NC) and R v.3.4.0 (R Development Core Team). All tests were two-sided at a significant level of 0.05. Genotype-Tissue Expression (GTEx; https://www.gtexportal.org/home/) Portal demonstrated the associations between selected SNPs and the gene expression status in colon tissue using GTEx Analysis Release V7 (dbGaP Accession phs000424.v7.p2).²²

Results

Patient characteristics

Supporting Information Table S3 shows baseline patient characteristics in six treatment arms. Briefly, TRIBE patients have significantly better performance status and more RAS mutant tumors; MAVERICC patients have significantly more right-sided tumors, less number of metastases and less status with primary tumor resected; and FIRE3 has significantly older patients. In MAVERICC cohort, there are no data about liver-limited disease, RAS status and BRAF status. The median follow-up and survival time in each cohort were summarized in Supporting Information Table S4.

Associations between selected SNPs and clinical outcomes in each treatment arm

Figure 2 and Supporting Information Table S5 summarize associations between the selected AMPK pathway-related SNPs and clinical outcomes (TR, PFS and OS) in each treatment arm.

Figure 2.

Associations between selected SNPs and clinical outcomes in six treatment arms. p values are plotted for associations of each selected SNP with clinical outcomes, where analyses were performed separately for each treatment arm. TR was analyzed using multivariate logistic regression model, and PFS and OS were analyzed using multivariate Cox proportional regression model. Abbreviations: BEV, bevacizumab; CET, cetuximab; OS, overall survival; PFS, progression-free survival; SNP, single nucleotide polymorphism; TR, tumor response.

In the TRIBE, FOLFIRI plus bevacizumab arm, there were significant associations for PRKAA1 rs461404, PRKAA1 rs13361707 and PRKAA1 rs10074991 with PFS (p < 0.001, p = 0.003 and p = 0.003) and in the FOLFOXIRI plus bevacizumab arm, there were significant associations for PRKAB2 rs3766522 with PFS (p = 0.025) and OS (p = 0.047); PRKAG2 rs13224758 with PFS (p = 0.039); PRKAG3 rs6436094 with OS (p = 0.047); PRKAG3 rs692243 with OS (p = 0.032); UBE2O rs3803739 with OS (p = 0.022); and ALDOA rs2071390 with OS (p = 0.021).

In the MAVERICC, FOLFIRI plus bevacizumab arm, there were significant associations for PRKAB1 rs4213 with TR (p = 0.008) and OS (p = 0.041); PRKAG2 rs13224758 with TR (p = 0.026) and PFS (p = 0.033); and PRKAG2 rs12703159 with PFS (p = 0.033) and in the mFOLFOX6 plus bevacizumab arm, there were significant associations for PRKAA2 rs10789038 with PFS (p = 0.031) and OS (p = 0.023); PRKAG2 rs13224758 with TR (p = 0.019); TSC1 rs13295634 with TR (p = 0.026); UBE2O rs3803739 with OS (p = 0.019); and ALDOA rs9783783 with TR (p = 0.003).

In the FIRE3, FOLFIRI plus bevacizumab arm, there were significant associations for PRKAB2 rs3766522 with OS (p = 0.030); PRKAG3 rs6436094 with TR (p = 0.018); TSC1 rs13295634 with PFS (p = 0.039); MTOR rs2295080 with TR (p = 0.014); ALDOA rs9783783 with TR (p = 0.034); and ALDOA rs2071390 with PFS (p < 0.001) and in the FOLFIRI plus cetuximab arm, there were significant associations for PRKAA1rs461404 with OS (p = 0.027); PRKAA1 rs13361707 and PRKAA1 rs10074991 with PFS (p = 0.006 and p = 0.006); and PRKAG1rs1138908 with OS (p = 0.026).

Notably, although many significant differences were observed, they were dependent on the six treatment arms and clinical outcomes (e.g., the relationship between PRKAA1 rs13361707 and PFS was depending on the treatment arms: Supporting Information Fig. S1B). We thus proposed a meta-analysis approach to determine the overall effect of selected SNPs on mCRC patients.

Associations between selected SNPs and clinical outcomes in a meta-analysis

Pathway analysis across multiple SNPs using the Pegasus approach showed that the AMPK pathway was significantly associated with both PFS (p < 0.001) and OS (p < 0.001), but not with TR (p = 0.220). As shown in Table 1, PRKAA1 rs13361707 and PRKAA1 rs10074991 (which were in perfect LD; Supporting Information Table S2) significantly associated with PFS, using the fixed effects model based on inverse-variance-weighted effect size (p = 0.003) and FDR correction (adjusted p = 0.032). There was no evidence of heterogeneity of effects for these SNPs for PFS across the six treatment arms according to Cochran’s Q statistic (p = 0.099 and p = 0.091). Overall, PRKAA1 rs13361707 was significantly associated with favorable PFS (log hazard ratio [HR] = −0.219, standard error [SE] = 0.073; p = 0.003), as well as PRKAA1 rs10074991 (log HR = −0.215, SE = 0.073; p = 0.003; Fig. 3). There were suggestive associations of PRKAG1 rs1138908 with unfavorable OS (log HR = 0.170, SE = 0.083; p = 0.041) and of UBE2O rs3803739 with unfavorable PFS (log HR = 0.137, SE = 0.068; p = 0.042) and OS (log HR = 0.210, SE = 0.077; p = 0.006), although these results were not significant after FDR adjustment. In addition, there was a significant difference in the effects of ALDOA rs9783783 for TR across the six treatment arms after Cochran’s Q statistic (p = 0.002, FDR correction [adjusted p = 0.032]; Table 1, Supporting Information Fig. S2).

Table 1.

Associations between selected SNPs and clinical outcomes, using meta-analysis

	TR			PFS			OS
SNPs	p value for FE1	Q Statistics	p value for Q2	p value for FE1	Q Statistics	p value for Q2	p value for FE	Q Statistics	p value for Q2
PRKAA1 rs461404	0.652	7.537	0.184	0.066	10.002	0.075	0.068	6.108	0.296
PRKAA1 rs13361704	0.937	6.094	0.297	0.003 (0.032)3	9.275	0.099	0.806	4.053	0.542
PRKAA1 rs100749914	0.930	6.132	0.294	0.003 (0.032)3	9.492	0.091	0.756	4.145	0.529
PRKAA2 rs10789038	0.168	7.333	0.197	0.372	5.876	0.318	0.343	5.817	0.324
PRKAB1 rs4213	0.691	10.866	0.054	0.352	1.653	0.895	0.286	9.592	0.088
PRKAB2 rs3766522	0.796	2.470	0.781	0.500	7.676	0.175	0.340	11.686	0.039
PRKAG1 rs11389084	0.404	6.978	0.222	0.073	1.815	0.874	0.041	4.735	0.449
PRKAG2 rs132247585	0.989	12.799	0.025	0.975	11.720	0.039	0.716	6.384	0.271
PRKAG2 rs12703159	0.401	8.147	0.148	0.283	8.893	0.113	0.701	4.764	0.445
PRKAG3 rs6436094	0.714	6.074	0.299	0.836	1.641	0.896	0.319	6.767	0.239
PRKAG3 rs692243	0.618	4.305	0.506	0.209	5.346	0.375	0.305	5.764	0.330
TSC1 rs132956345	0.466	14.467	0.013	0.569	9.600	0.087	0.515	4.102	0.535
TSC1 rs1073123	0.632	3.683	0.596	0.446	4.719	0.451	0.356	1.715	0.887
TSC2 rs3087631	0.464	4.644	0.461	0.896	2.828	0.727	0.760	6.963	0.223
RHEB rs6972955	0.570	3.607	0.607	0.757	7.839	0.165	0.670	3.348	0.647
MTOR rs22950805	0.954	11.768	0.038	0.875	5.314	0.379	0.534	5.630	0.344
UBE2O rs38037394	0.952	4.706	0.453	0.042	5.130	0.400	0.006	5.771	0.329
ALDOA rs97837835	0.919	19.283	0.002 (0.032)3	0.262	2.775	0.735	0.190	1.643	0.896
ALDOA rs20713905	0.336	2.822	0.727	0.244	13.268	0.021	0.639	9.894	0.078

¹FE indicates fixed effects model based on inverse-variance-weighted effect size.

²Significant p values for Q indicate the heterogeneity of the effect of the SNPs among the six arms.

³Significant FDR adjusted p values were shown in parentheses.

⁴SNPs which had significant relationships with clinical outcomes.

⁵SNPs which had heterogeneity according to the clinical outcomes among the six treatment arms.

Bold means that the values have significant differences.

Abbreviations: FDR, false discovery rate; OS, overall survival; PFS, progression-free survival; SNP, single nucleotide polymorphism; TR, tumor response.

Figure 3.

Forest plots of meta-analysis for PRKAA1 rs13361707. Log OR or log HR are shown with SE. The summary row shows the inverse-variance-weighted effect size with SE, combining the six estimates for the individual arms into a single summary measure. A positive log OR or HR implies a negative influence on TR, PFS and OS, respectively. Abbreviations: BEV, bevacizumab; CET, cetuximab; HR, hazard ratio; OR, odds ratio; OS, overall survival; PFS, progression-free survival; SE, standard error; TR, tumor response.

Associations between selected SNPs and their corresponding gene expression status from GTEx analysis

Finally, the relationship between selected SNPs and their corresponding gene expression status in colon tissue was examined using data downloaded from GTEx. PRKAA1 rs13361707 and PRKAA1 rs10074991 were associated with higher expression status of PRKAA1 in transverse colon (p < 0.001 and p < 0.001), and PRKAG1 rs1138908 was correlated with lower expression status of PRKAG1 in sigmoid colon (p = 0.032; Supporting Information Fig. S3).

Discussion

This is the first study we know of that evaluate the associations between the AMPK pathway-related SNPs and clinical outcomes in patients with mCRC treated with chemotherapy in three randomized clinical trials (TRIBE, MAVERICC and FIRE-3), using a meta-analysis approach. Notably, we found that the SNP pair PRKAA1 rs13361707 and PRKAA1 rs10074991 (which were in perfect LD) was a significant predictor for the clinical benefit of chemotherapy as measured by PFS in mCRC. Our results might suggest novel treatment strategies for chemotherapy in mCRC.

AMPK, a heterotrimer consisting of a catalytic α-, a regulatory β-, and a AMP/ADP/ATP binding γ- subunit (α1 or α2, β1 or β2 and γ1, γ2 or γ3: encoded by separate genes PRKAA1–2, PRKAB1–2 and PRKAG1–3) is a serine/threonine protein kinase that is highly conserved across most eukaryotic organisms.²³ Accumulating evidences suggest that AMPK acts as a tumor suppressor. Under conditions of metabolic stress, the tumor suppressor LKB1 phosphorylates AMPKα1/2 Thr-172 kinase, which enhances AMPK activity,^24,25 and the activated AMPK inhibits tumor growth via mTOR signaling and cell cycle checkpoints.^3,5,6 All of those signaling are essential to maintain genomic stability and defend against carcinogenesis.²⁶ In consensus with that, AMPK acts as a sensor of genomic stress and a participant of the DNA damage response (DDR) pathway. Following DNA damage, ATM serine/threonine kinase (ATM), a key DNA damage sensor, phosphorylates AMPK in cancer cells, independent of LKB1, which mediates cell cycle arrest via cell cycle regulators including p53, p27 and p21.^3,6 Inhibition of ATM and AMPK in cancer cells attenuate the ionizing radiation-induced G2/M cell cycle arrest.²⁷ Loss of LKB1 (with reduced AMPK activation) lacks the benefit of bevacizumab combined with chemotherapy in lung cancer,²⁸ and low phosphorylated AMPK level was associated with worse survival in mCRC treated with FOLFIRI plus bevacizumab.²⁹ These findings highlight the importance of the status of AMPK pathway-related molecules for predicting the efficacy of conventional chemotherapy in combination with bevacizumab.

To the best of our knowledge, we are the first to report that PRKAA1 rs13361707 and rs10074991 may predict chemotherapy benefit in mCRC patients. PRKAA1 codes a catalytic α1-subunit of the AMPK and has a significant role for cellular homeostasis. In embryonic fibroblasts, lack of AMPK α-subunit expression induced lack of a G2-M checkpoint response and radio-resistance.³⁰ Furthermore, increased AMPK α-subunit expression suppressed cancer cell proliferation and showed increased survival in mice model treated with chemotherapy,³¹ suggesting that AMPK α-subunit plays in stabilizing the basal activity of DDR and survival signaling pathways. rs13361707 is located in the first intron of PRKAA1 at the 5p13.1 site, and it has been reported that this polymorphism is associated with the risk of gastric cancer.³² While AMPK activates the related-pathway by being phosphorylated, GTEx analysis showed that carrying the allele genotypes led to higher gene expression in colon tissue. Considering that the SNPs are predictors for favorable PFS in our analysis, these SNPs might upregulate AMPK activation. In addition, rs10074991 may act as tag SNPs, influencing functional effects through related polymorphisms at other nearby loci. Although the influence PRKAA1 SNPs on phenotypic change is still unknown, it is considered that loss of function conveys resistance to conventional chemotherapy in mCRC. Additional experimental studies are necessary for validation of our results.

The clinical effects of other PRKAG1 and UBE2O polymorphisms on benefit of chemotherapy also remain unknown. PRKAG1 codes a γ1-subunit of the AMPK, and binding of AMP to the γ subunit allosterically makes the complex more attractive for phosphorylation on Thr172 in α subunit.²³ UBE2O is upregulated in human cancers and induces tumorigenesis targeting AMPKα2 for ubiquitination and degradation.³³ PRKAG1 rs1138908 is in the 5′-UTR region of the gene, and there have been no reports suggesting associations between the variant and cancer status. In general, SNPs in 5′-UTR may alter the start/stop codon of an upstream open reading frame, thereby affecting the translation efficiency. In addition, PRKAG1 rs1138908 may act as tag SNP, influencing functional effects through related polymorphisms at other loci of PRKAG1. UBE2O rs3803739 is a common missense SNP: G3684A as a base pair change and Gly1207Ser as an amino acid change, and had no report about the variant and cancer status. GTEx analysis showed that carrying the PRKAG1 rs1138908 allele led to lower gene expression level in colon tissue (consistent with our results: rs1138908 was a predictor for worse OS), and UBE2O rs3803739 was not significantly associated with gene expression. Furthermore, although we could not detect a clinical benefit, ALDOA rs9783783 did exhibit evidence of heterogeneity of effects for the clinical outcomes across the six treatment arms. As for PRKAA1 rs13361707, these results also need confirmation from further experimental studies.

Accumulating evidence shows that environmental factors also regulate the function of AMPK pathway. AMPK can be activated by exercise,³⁴ low-calorie intake,³⁴ omega-3 polyunsaturated fatty acids (ω3PUFAs; through the activation of LKB1)³⁵ and adipocyte-derived antidiabetic hormones such as adiponectin^7,36 or leptin.³⁷ It has also been reported that the AMPK-activators may work well in combination with chemotherapy,³⁸ indicating that AMPK activity is associated with antitumor efficacy. Recently, biguanides, the first line pharmacotherapy for glucose control, appeared to activate AMPK⁷ and improve response to chemotherapy by removing cancer stem cell in a variety of cancers^39,40: Also, thiazolidinediones (TZDs), a class of insulin-sensitizing drugs, activate AMPK both directly and indirectly via the induction of adiponectin, leading to tumor suppression.^41,42 Due to these findings, according to ClinicalTrials.gov (https://clinicaltrials.gov/ct2/home), there are currently many trials of biguanides or TZDs for cancer mainly in combination with chemotherapy. As for predicting the effect of these drugs, AMPK pathway-related SNPs may be promising markers.

Genome-wide association studies (GWAS) have led to the identification of novel disease susceptibility genes in many complicated diseases in which genetic factors and environmental factors interact.⁴³ However, in more targeted analyses, due to the limitation of the number of patients and types of clinical samples available, even clinically relevant data can be judged not statistically significant. We have overcome these challenges by using a meta-analysis approach, identifying potential relationships between promising AMPK pathway-related SNPs and clinical outcomes in clinical trials. Although this candidate gene approach does not comprehensively search for SNPs useful for chemotherapy in mCRC, it could increase our statistical power to identify individual SNPs predictive of therapeutic effect across multiple treatment arms, here TRIBE, MAVERICC and FIRE-3.

Some limitations of this meta-analysis should be acknowledged. First, the number of cases remains small for a comprehensive analysis. Second, due to the sample loss, the direct association of our findings with clinical outcomes in each trial is unclear. Third, our GTEx analysis was performed on normal colon tissue. Fourth, although PRKAA1 rs13361707 (PRKAA1 rs10074991) had statistically significant interactions with patients who were more likely to benefit from first-line chemotherapy, there was no prognostic value for TR and OS. This means that treatments other than first-line chemotherapy must also be taken into account. More comprehensive analysis is necessary to translate our findings into clinical practice. Despite these limitations, by using the three prospective clinical trials for chemotherapy in mCRC in a meta-analysis approach, we were able to find a difference in the effect of chemotherapy in the patients with differing PRKAA1 rs13361707 (PRKAA1 rs10074991) alleles. In addition, we demonstrated the feasibility of the meta-analysis approach for discovery of associations between gene polymorphisms and clinical outcomes.

In conclusion, we have shown that AMPK pathway-related SNPs can be predictors for the effects of chemotherapy in mCRC, while the SNPs had some heterogeneity among six treatment cohorts. Additional studies are necessary to validate our observations and to elucidate the influence of the SNPs on phenotypic change in cancer. The possible correlation between AMPK pathway-related SNPs and the effects of chemotherapy in mCRC patients may have considerable implications for treatment strategies.

What’s new?

Using data from three randomized clinical trials (TRIBE, MAVERICC, and FIRE-3), this study investigates whether AMP-activated kinase (AMPK)-associated genomic markers are predictors for clinical outcome in patients with metastatic colorectal cancer. The authors find two single-nucleotide polymorphisms (PRKAA1 rs13361707 and PRKAA1 rs10074991) that predicted favorable outcome after chemotherapy while two others (PRKAG1 rs1138908 and UBE2O rs3803739) predicted unfavorable outcome although the latter result did not reach significance. Targeting AMPK in addition to chemotherapy could be beneficial for some patients with metastatic colorectal cancer.

Abbreviations:

AMPK

AMP-activated protein kinase

ATM

ATM serine/threonine kinase

DDR

DNA damage response

FDR

false discovery rate

GTEx

genotype-tissue expression

GWAS

genome-wide association study

HR

hazard ratio

HWE

Hardy–Weinberg equilibrium

LD

linkage disequilibrium

mCRC

metastatic colorectal cancer

MSI-H

microsatellite instability-high

OS

overall survival

PFS

progression-free survival

SNP

single nucleotide polymorphism

TR

tumor response

TZD

thiazolidinedione