Genetic basis for plasma amino acid concentrations based on absolute quantification: a genome-wide association study in the Japanese population

To assess the use of plasma free amino acids (PFAAs) as biomarkers for metabolic disorders, it is essential to identify genetic factors that influence PFAA concentrations. PFAA concentrations were absolutely quantified by liquid chromatography–mass spectrometry using plasma samples from 1338 Japanese individuals, and genome-wide quantitative trait locus (QTL) analysis was performed for the concentrations of 21 PFAAs. We next conducted a conditional QTL analysis using the concentration of each PFAA adjusted by the other 20 PFAAs as covariates to elucidate genetic determinants that influence PFAA concentrations. We identified eight genes that showed a significant association with PFAA concentrations, of which two, SLC7A2 and PKD1L2, were identified. SLC7A2 was associated with the plasma levels of arginine and ornithine, and PKD1L2 with the level of glycine. The significant associations of these two genes were revealed in the conditional QTL analysis, but a significant association between serine and the CPS1 gene disappeared when glycine was used as a covariate. We demonstrated that conditional QTL analysis is useful for determining the metabolic pathways predominantly used for PFAA metabolism. Our findings will help elucidate the physiological roles of genetic components that control the metabolism of amino acids.

Recently, several genome-wide association studies (GWAS) also identified genetic variations associated with PFAAs in European populations [9-11, 13, 15]. However, the influence of heritability and whether these loci are shared among other human populations are still unknown. In addition, metabolite concentrations are influenced by other metabolites within the same metabolic pathway. Therefore, genome-wide quantitative trait locus (QTL) analyses conditioned on the other amino acids sharing the same pathway are necessary.
In this study, we sought to elucidate genetic determinants that influence PFAA concentrations. We conducted a QTL analysis of PFAAs measured by an absolute quantification method using plasma samples from 1338 Japanese individuals.

Subjects and ethics
Participants were recruited from the Nagahama Prospective Genome Cohort for Comprehensive Human Bioscience (the Nagahama Study). All of the subjects were approved by the Institutional Review Board and the ethics committees of each institute, to which donors gave written informed consent, in accordance with the national guidelines.

Absolute quantification of PFAA concentrations
The concentrations of 21 PFAAs from 2,084 individuals who participated in the Nagahama study in 2008 (n = 1124) and 2009 (n = 960) were quantified. Blood samples (5 ml) were collected from forearm veins after overnight fasting into tubes containing ethylenediaminetetraacetic acid (EDTA; Termo, Tokyo, Japan). The plasma was extracted by centrifugation at 2010×g at 4°C for 15 min and then stored at -80°C. After deproteinizing the thawed plasma samples using 80% acetonitrile, the samples were subjected to pre-column derivatization, then the absolute concentrations, the absolute concentrations of the PFAAs were measured by high performance liquid chromatography -electrospray ionization mass spectrometry (HPLC-ESI-MS). The methods were previously developed and verified by the authors [16][17][18][19].
Quantification was considered successful when the obtained value was within the determination range of the calibration curve.

PFAA concentrations for QTL analysis
For the QTL analysis, we prepared three adjusted PFAA concentrations from the measured absolute concentrations of PFAAs. The first adjusted concentration was adjusted for sex and age by linear regression after Box-Cox transformation. The second was adjusted by the other 20 PFAAs by multiple linear regression after the first adjustment. For this regression analysis, explanatory variables were selected by the step-wise function (stepwiseglm) in MATLAB, with P = 0.001 and P = 0.01 as inclusion and exclusion criteria, respectively. The third was adjusted by one of the other PFAAs by linear regression after the first adjustment.

SNP genotyping and quality control (QC) process
A total of 1594 samples were genotyped using three commercially available Illumina genotyping arrays (Illumina, Inc., San Diego, CA): Human610-Quad BeadChip (610 K), HumanOmni-2.5-Quad BeadChip (2.5M-4), and HumanOmni-2.5-8 BeadChip (2.5M-8). The 1,124 subjects recruited in 2008 were genotyped using 610 K (n = 1,113) or both 610 K and 2.5M-4 (n = 11). The 470 subjects recruited in 2009 were genotyped using 610 K (n = 101), 2.5M-4 (n = 293), 2.5M-8 (n = 62), or both 610 K and 2.5M-4 (n = 14). In total, 2,638,338 SNPs were genotyped in the arrays. As shown in Fig. 1, through a sample QC Fig. 1 Flow diagram of the QC processes and QTL analyses using the PFAA concentrations of Japanese subjects from the Nagahama Study process, 256 samples were excluded from the analysis: 3 genetic outliers identified by principal component analysis (PCA), 138 relatives, and 115 samples with a call rate of SNPs < 0.95. Through a marker QC process, 1,200,574 SNPs were excluded: 138,990 SNPs with a call rate < 0.99, 1,187 SNPs with deviation from Hardy-Weinberg equilibrium (P < 1.0 × 10 -6 ), and 1,060,397 SNPs with a variant allele frequency < 1%. After the QC processes, 1,437,764 SNPs in 1338 samples remained for the GWA studies. Of these, 266,274 SNPs shared among all of the arrays were defined as intersectional SNPs. All of the QC procedures were processed using PLINK ver. 1.07 [20]. Both genotype and PFAA concentrations data of Nagahama study is deposited on the Japanese Genotype-phenotype Archive affiliated to the DDBJ (DNA Data Bank of Japan), via National Bioscience DataBase (NBDC), Japan. The data is accessible on hum0012 at https://ddbj.nig.ac.jp/jga/viewer/permit/dataset/ JGAD00000000012.

Imputation
The 1,437,764 SNPs from 1,338 samples used for the GWA studies were imputed using MACH ver. 1.0 [21]. An imputation panel was generated using the genotyping data of 665 Nagahama Study samples that were not used for the present GWA analyses and contained the results of 1,560,699 SNPs with Illumina HumanCoreExome Bead-Chip (Exome), HumanOmni2.5 S BeadChip (2.5 S), 2.5M-4, and 2.5M-8 arrays. Of these 665 samples, 478 were genotyped using all of the arrays, and 187 were genotyped using the Exome, 2.5 S, and 2.5M-4. Imputed SNPs with a variant allele frequency > 1% or an r 2 < 0.5 were excluded from the subsequent association analysis. Finally, 1,288,202 SNPs from the 1338 samples were fixed with 1,021,918 additional SNPs.

QTL analysis
For the three PFAA concentrations described above, QTL analysis was conducted with a an additive model implemented in PLINK [2.0]. The genome-wide significance threshold after Bonferroni correction was P < 3.88 × 10 -8 .

In silico analysis of genetic variants
The exome sequencing data of 300 Japanese individuals from the Human Genetic Variation Database (HGVD) were used to identify candidates for genetic variants with a functional impact on PFAA concentrations [22,23]. Pairwise linkage disequilibrium (LD) coefficients (r 2 ) were calculated using PLINK [20]. The impacts of the nonsynonymous variants were predicted using the Ensembl Variant Effect Predictor [24], which is based on the SIFT [25] and PolyPhen [26] algorithms. Expression QTL (eQTL) analysis data (release version 8.0) were also downloaded from the HGVD [22].

PFAA profiling
We measured the PFAA concentrations by an absolute quantification method using HPLC-ESI-MS [16][17][18][19]. The concentrations of 21 PFAAs were quantified successively in all of the samples (N = 2,094). The PFAA concentrations in the 1,338 samples used for GWAS are summarized with biochemical parameters in Table 1. The means, standard deviations, and ranges of the absolute concentrations were comparable to those obtained in an independent study in a Japanese population, except for arginine, glutamate, and ornithine [27]. The averaged levels of glutamate and ornithine higher, and that of arginine was lower, than in the previous study.

QTL analysis of PFAAs (GWAS-1)
A flow diagram of the QC processes and QTL analyses is shown in Fig. 1. The first QTL analysis was conducted for each PFAA concentration, which was adjusted for sex and age after Box-Cox transformation, with 266,274 intersectional SNPs of 1338 samples (GWAS-1). Twenty-eight SNPs in four loci were significantly associated with glycine, serine, glutamine, and phenylalanine ( Fig. 2a and Table 2). The strongest association was observed in the CPS1 locus on chromosome 2 for the glycine concentration (rs12613336, P = 2.07 × 10 -70 ). CPS1 encodes mitochondrial carbamoyl-phosphate synthase 1 (CPS-I), a key enzyme in the urea cycle, which generates carbamoylphosphate from H 2 O, CO 2 , and ammonia. Two chromosomal loci showing significant associations with the serine concentration were identified: the PSPH locus on chromosome 7 (rs13244654, P = 1.80 ×10 -21 ) and the CPS1 locus, which was also associated with the glycine concentration (rs12613336, P = 4.77 × 10 -12 ). PSPH encodes phosphoserine phosphatase (PSPH), which catalyzes the hydrolysis of 3-phosphoserine to generate serine. In addition, the GLS2 locus and the PAH locus on chromosome 12 were associated with the concentration of glutamine (rs7302925, P = 9.73 × 10 -11 ), and phenylalanine (rs17450273, P = 6.60 × 10 -10 ), respectively. GLS2 encodes glutaminase, which catalyzes the hydrolysis of glutamine to glutamate and ammonia, and PAH encodes phenylalanine hydroxylase.
Next, we performed imputation using the genotyping results of 665 samples that were unrelated to those used for the present GWA study. The additional imputed genotypes of 1,021,928 SNPs were used for the QTL analysis. We identified two additional chromosomal loci in which multiple SNP markers showed a significant association with PFAA concentrations (Fig. 2b and Table 2): the ASPG (putative asparaginase) locus on chromosome 14 for asparagine (rs1744297, P = 1.30 × 10 -51 ) and the PRODH (proline dehydrogenase) locus on chromosome 22 for proline (rs2238732, P = 5.96 × 10 -40 ). To confirm these associations, these two SNPs were genotyped for the same DNA samples using the Taqman assay (Table 3). We obtained P = 2.36 × 10 -48 for rs1744297 and P = 5.78 × 10 -36 for rs2238732, and the concordance rates between the imputed and directly genotyped SNPs were 98.6% (1300/ 1319) and 97.6% (1289/1321) for rs1744297 and rs2238732, respectively.
The imputation analysis identified six additional chromosomal loci with potential associations. However, only one SNP marker showed a significant association for each locus, so no further analysis was performed. The 151 SNPs showing significant associations either by genotyping or by imputation are listed in Table S1.

QTL analysis of PFAAs conditioned on the other amino acids (GWAS-2)
We next conducted QTL analysis using the concentration of each PFAA adjusted using the other 20 PFAAs as covariates (GWAS-2 in Fig. 1). The optimal regression models for each PFAA were constructed using a step-wise variable selection method (Table S2). Two additional association loci were identified by the conditional QTL analysis ( Fig. 2c and Table 2). The strongest association was observed at a non-synonymous variant, rs56335308, in the SLC7A2 (a solute carrier family 7, cationic amino acid transporter) gene on chromosome 8 for arginine (P = 2.64 × 10 -16 ) and ornithine (P = 4.70 × 10 -14 ). The other was the PKD1L2 locus on chromosome 16 for glycine (rs8059153, P = 1.46 × 10 -8 ). This gene encodes a member of the polycystin protein family. In addition, an association was found between rs2238732 in the PRODH locus and the alanine concentration (P = 7.10 × 10 -11 ). On the other hand, the significant association between serine and the CPS1 locus disappeared (P = 0.63). The other six associations maintained significant levels after the conditional analysis. The 209 SNPs identified as significant in GWAS-2 are listed in Table S3. QTL analysis of PFAAs adjusted conditioned on one of the other amino acids (GWAS-3) We also conducted QTL analyses of 21 PFAAs conditioned on one of the six PFAAs (asparagine, glutamine, glycine, phenylalanine, proline, and serine) identified as significant in the above studies (GWAS-3 in Fig. 1). The significant association between serine and the CPS1 locus disappeared when glycine was used for the adjustment (P = 0.002) (Fig. 3a). In contrast, significant associations between five PFAAs (arginine, asparagine, glutamine, ornithine, and threonine) and the CPS1 locus were still apparent when glycine was used as a covariate (Fig. 3a). When asparagine was used as a covariate, rs1744297 in the ASPG locus showed additional associations with ten PFAAs: alanine, glutamine, histidine, lysine, methionine, ornithine, serine, threonine, tryptophan, and tyrosine (Fig. 3b). Rs2238732 in the PRODH locus also showed an association with seven PFAAs (alanine, isoleucine, leucine, ornithine, threonine, tyrosine, and valine) (Fig. 3c) using proline as a covariate. Similarly, significant associations were observed in PSPH for threonine using serine (P = 3.82 × 10 -8 ) and in PAH for methionine using phenylalanine (P = 1.91 × 10 -9 ) as a covariate. All of the statistics in GWAS-3 are listed in Table S4.

In silico analysis for the functional interpretation of the association between identified SNPs and PFAA concentrations
In the above analyses, three of the eight loci that showed significant associations, namely, rs56335308 in GLS2, rs2657879 in PRODH, and rs450046 in SLC7A2, were nonsynonymous variations with potential functional effects on PFAA concentrations (Table S1). We tried to identify nonsynonymous SNPs that were in strong LD with the SNPs showing significant associations within the eight loci in the above analysis, by conducting an in silico analysis using pair-wise LD information in the HGVD exome sequencing data set [23]. Among the 223 SNPs with LD Fig. 2 Regional association plots of the six loci significantly associated with PFAA concentrations (n = 1338). a Association was significantly identified from genotyped data. b Association was significantly identified after imputation. Chromosomal positions and P values for genotyped SNPs (red) and imputed SNPs (blue) are shown. c Chromosomal positions and P values for genotyped SNPs (red) and imputed SNPs (blue) of the conditional (upper) and unconditional (lower) analyses are shown. Dotted lines indicate the genome-wide significance threshold after Bonferroni correction. Brightness of the red color in the linkage disequilibrium (LD) blocks corresponds to the strength of LD Table 2 Genetic variants associated with PFAAs in Japanese imputation was performed using the genotyping results of 665 samples that were unrelated to those used for the present study coefficients (r 2 ) greater than 0.8 with SNPs significantly associated with PFAAs, four non-synonymous variants were identified: rs1047891 in CPS1, rs8012505 in ASPG, rs8054182 in PKD1L2, and rs5747933 in PRODH (Table S5). The predicted functional impacts of these variants were not deleterious to the gene products except for ASPG. There were no non-synonymous variants for the PSPH or PAH gene.
We next investigated whether these SNPs within the eight loci identified in the current study or those that were in strong LD with them influenced gene expression using the HGVD expression QTL (eQTL) information [22]. We found that rs4948073, which was located approximately 240-kb downstream of the PSPH gene, showed the strongest association with the expression level of the PSPH gene (P = 3.03 × 10 -46 ). There were no significant eQTLs for the other six loci.

Discussion
In this study, we conducted a QTL analysis of the absolute concentrations of PFAAs quantified by LC-MS technology. The concentration of a PFAA can be influenced by other PFAAs within the same metabolic pathway. Therefore, it is also important to perform conditional QTL analysis considering the amino acids' metabolic pathways when identifying the genetic determinants of PFAA concentrations. Notably, here we identified two additional genetic loci associated with the concentrations of arginine, ornithine, and glycine (Table S2). One of the identified genes, SLC7A2 was associated with arginine and ornithine. This protein is known to transport plasma arginine into cells for protein synthesis and to convert arginine into ornithine or nitric oxide [28]. SLC7A family members, such as SLC7A5, SLC7A6, and SLC7A9, are associated with plasma tryptophan, lysine, and arginine, respectively [9,13,29,30]. Thus, it is likely that genetic variations of SLC7A2 would affect the plasma concentration of arginine and ornithine. The other identified gene, PKD1L2, which showed an association with glycine concentration, encodes polycystic kidney disease protein 1-like 2. Previous studies suggested that genetic variation of the PKD1L2 gene may be associated with highdensity lipoprotein cholesterol [31,32]. Rs8054182, which has strong LD with rs8059153 (r 2 = 0.989), introduces an amino acid change from methionine to isoleucine at position  The six PFAAs (red) were associated with genotypes in the genes (yellow). THF tetrahydrofuran, 5-CH 3 -THF 5-methyltetrahydrofolate, 5,10-CH 2 -THF 5,10-methylenetetrahydrofuran, NH 4 + ammonium ion, GCS glycine cleavage system, SDH serine dehydratase, SHMT serine hydroxymethyltransferase 1630, which is in the conserved ion channel pore region [33]. It is therefore possible to speculate that PKD1L2 acts as part of a glycine transporter (Fig. 4).
Significant associations of CPS1 with the plasma levels of arginine, asparagine, glutamine, ornithine, and threonine were observed only after being conditioned on glycine (Fig. 3a). Asparagine and glutamine syntheses have CPS-I in their upstream pathway (Fig. 4). Threonine is also involved in the ammonia-generating reaction mediated by L-serine dehydratase/L-threonine deaminase [34]. Both arginine and ornithine are involved in the urea cycle, which is the downstream pathway of CPS-I (Fig. 4). These mechanisms suggest that the plasma concentrations of these five PFAAs are influenced by the enzymatic activity of CPS-I. Similarly, associations exist for ASPG for ten PFAAs were obtained only after being conditioned on asparagine (Fig. 3b). Rs8012505, a non-synonymous SNP that has strong LD with rs1744297, is located at the provisional cytoplasmic asparaginase I (ansA) domain and changes serine to arginine at position 344. We speculate that ASPG can use these ten PFAAs as substrates for deamination. Similarly, significant associations of PRODH for seven PFAAs were obtained only after being conditioned on proline, suggesting that PRODH can use them as substrates (Fig. 3c). In some situations, use of heritable covariates might introduce unintended bias into estimate [35]. Direct enzymatic verification whether ASPG and PRODH can catalyze other amino acids than asparagine and proline, respectively, will be desirable to confirm our speculations.
We also demonstrated that the conditional QTL analysis is useful for determining the metabolic pathway predominantly used for PFAA metabolism. For example, the concentration of plasma glycine is correlated with that of serine (r = 0.54), and the significant association between serine and CPS1 in the unconditional QTL analysis disappeared when conditioned on glycine (Fig. 3a). In contrast, the significant association between glycine and CPS1 was not affected by the analysis conditioned on serine (Fig. 3a). CPS-I is considered an entrance to the urea cycle, which detoxifies the ammonia that is produced by amino acid degradation. Two separate pathways that generate ammonia are likely to be involved in this process. The first is the conversion of glycine to ammonia catalyzed by the glycine cleavage system (GCS) with tetrahydrofolate production [36]. The second is the conversion of serine to ammonia and pyruvate, which is catalyzed by serine dehydratase [37]. In addition, glycine and serine are reversibly converted to each other via serine hydroxymethyltransferase [38]. The study of hyperglycinemia, an inborn deficiency of GCS, revealed that GCS plays a critical role in both glycine and serine catabolism in the liver [36]. The results of the conditional QTL analysis in the present study were consistent with previous clinical observations.
It is still unknown whether common variants that influence the concentrations of plasma amino acids are associated with risks of lifestyle-related metabolic diseases. For example, although the plasma glycine concentration is associated with an increased risk of diabetes [7,8,39], no genetic variants that are significantly associated with a risk of diabetes have been identified within the CPS1 locus [12]. Further longitudinal studies with increased sample sizes are needed to assess whether the PFAA concentrations can be used as intermediate biomarkers for metabolic disease risk under a variety of genetic backgrounds.
Acknowledgements This research was partially supported by the Center of Innovation Program from MEXT and JST. We thank the Non-profit Organization "Zero-ji Club for Health Promotion," the staff and associates of the Nagahama City Office, Kohoku Medical Association, Nagahama City Hospital, Nagahama Red Cross Hospital, Nagahama Kohoku City Hospital, and the participants of the Nagahama Study, for data collection.

Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of interest. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons. org/licenses/by/4.0/.