Introduction

Cognitive decline in patients with mild cognitive impairment (MCI) and dementia is a complex trait resulting from the interaction between genetic and environmental factors. The E4 variant of apolipoprotein E (APOE4) is a well-known strong genetic risk factor for Alzheimer’s disease (AD) [1] and confers susceptibility to MCI [2]. In addition, genetic association studies performed using genome-wide association analysis and meta-analysis have shown that APOE ε4 (APOε4) confers susceptibility to vascular dementia and dementia with Lewy bodies [3–7]. APOE4 risk of cognitive decline may be modified by interaction with other factors; however, the role of risk modifiers or interacting factors is not well understood. In our previous longitudinal study, we found that high serum vitamin C (VC) levels during normal cognition reduce APOE4-associated risk of cognitive decline, especially in women [8].

VC is a well-known antioxidant in humans and other mammals [9]. VC is also required for diverse biological processes and is involved in several enzymatic reactions, including the synthesis of collagen and neurotransmitters as a cofactor [10]. Studies in animal models have revealed that VC is maintained at relatively high concentrations in the brain during systematic VC deficiency, suggesting the vital role of VC in the brain [11, 12]. Animal studies have showed that chronic VC deficiency causes cognitive dysfunction via mitochondrial dysfunction and increased oxidative stress [13, 14]. Acute administration of VC improved cognitive function in mouse models of AD and aging [15, 16].

As humans cannot biosynthesize VC, its intake through diet is essential for the maintenance of the normal VC levels. The ingested VC is transported from the digestive tract to the body. VC includes ascorbic acid (ASC; reduced form) and dehydroascorbic acid (DHA; oxidized form). ASC absorption occurs in the body via sodium-dependent VC transporter (SVCT), whereas that of DHA occurs via glucose transporters (GLUTs) [17]. A study using gene-knockout mice showed that SVCT1 (encoded by SLC23A1) plays an important role in intestinal absorption and renal reabsorption of ASC [18]. Additionally, the ASC level in neonatal mice with homozygous knockout of SVCT2 (encoded by SLC23A2) was undetectable or considerably reduced in various tissues including the brain, compared with that in heterozygous knockout or wild-type pups. Besides, compared with age-matched wild-type mice, newborn and adult mice with heterozygous knockout of SVCT2 have reduced ASC level in the brain and several other tissues [19, 20]. SVCTs and GLUTs transport VC to various organs. The transport of ASC into the brain is mediated by SVCT2 (encoded by SLC23A2) [21], whereas DHA uptake in the brain occurs via GLUT1 (encoded by SLC2A1) and GLUT3 (encoded by SLC2A3) [22, 23]. Depending on the ASC and DHA levels, the intracellular DHA-to-ASC recycling system is activated, and the amount of VC influx and efflux is regulated by SVCT2 and GLUT1/3 [17]. Thus, VC transporters are important in maintaining VC levels in the body, and in particular, SVCT2 and GLUT1/3 are thought to play important roles in the regulation of VC levels in the brain.

To investigate the relationship between cognitive decline and brain VC level in humans, we need to focus on the roles of brain-expressed VC transporters (SLC23A2, SLC2A1, and SLC2A3) in human cognitive function. We hypothesized that genetic variants of VC transporters (SLC23A2, SLC2A1, and SLC2A3) expressed in the brain could affect the risk of developing APOE4-associated cognitive decline. To test this hypothesis, we first selected functional variants of VC transporter genes from database searches, and then performed case–control association analysis to examine the effects of functional variants on the risk of developing APOE4-associated cognitive decline. In addition, we examined the relationship between functional variants of VC transporter genes and cognitive decline.

Subjects and methods

Results

In the baseline survey, 923 individuals aged 65 years or older were included in the study. Among the 923 participants, by applying the previously mentioned diagnostic criteria, 730 were identified with normal cognitive function in the baseline study. In the follow-up survey, 400 participants were included, among which 252 were classified into the normal group and 141 in the cognitive decline group (MCI: 87 and dementia: 54), except for seven participants whose cognitive function could not be determined (Fig 1). The details of the characteristics of participants at baseline and follow-up surveys are summarized in Table 1. The participants in the cognitive decline group in follow-up survey were older than those in the baseline survey, and had a longer follow-up period, shorter education period, and lower MMSE score in both baseline and follow-up surveys, and a higher frequency of APOE4 phenotype than the normal cognition group in the follow-up survey. In female participants of the cognitive decline group, the blood VC level was significantly lower in the baseline survey than that in the follow-up survey. The multivariate logistic regression analysis adjusted for the effects of age, follow-up period, education period, MMSE points at baseline, frequencies of hypertension and hyperlipidemia, standardized VC levels, and sex showed that APOE4 is an independent risk factor for cognitive decline (MCI or dementia) (S1 Table: p = 0.027, OR = 1.91, and 95% CI, 1.10–3.33). In addition, the lost to follow-up participants (n = 330) were older, had lower MMSE scores at baseline, and had a shorter education period than the follow-up participants (n = 400) (S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Design of the case–controlled study.

Baseline survey was conducted between 2006 and 2008. The participants aged 65 years or older with normal cognition were selected. In the baseline survey, participants took cognitive battery, underwent blood sampling, and responded to a questionnaire, including medical history. DNA was extracted from the blood. Between 2014 and 2016, follow-up survey was conducted, and the cognitive status was evaluated again. Based on the cognitive status, the participants were classified into normal group (normal cognition) or cognitive decline group (MCI or dementia).

https://doi.org/10.1371/journal.pone.0259663.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the participants with normal cognition (n = 393) at baseline (2006–2008) who presented with normal cognition (n = 252) or MCI/dementia (n = 141) in the follow-up survey (2014–2016).

https://doi.org/10.1371/journal.pone.0259663.t001

In the case of SVs overlapped with SLC23A2, SLC2A1, and SLC2A3 in DGV, we found 1 “DGV Gold Standard” SV in SLC23A2, 0 SV in SLC2A13, and 7 SVs in SLC2A3, all of which did not meet the criterion of at least 15% frequency. In the search with the “Genome Variation” database at jMorp, we found no SNV and small indels altering the protein sequences of SLC23A2, SLC2A1, and SLC2A3, with MAF ≥ 15% in the Japanese population. Therefore, we focused on SNVs and small indels altering the transcript abundance of SLC23A2, SLC2A1, and SLC2A3 in the subsequent analyses.

Among 226 SNVs and small indels with MAF ≥ 15% in SLC23A2 from the “Genome Variation” database at jMorp, only one SNV (rs1279683) was classified as a functional variant with the “Regulome DB” score 1d, which had evidence of eQTLs and functional genomic regions such as transcriptional factor binding sites. Using the “GTEx” Portal database, rs1279683 was proven to be an eQTL of SLC23A2 in the spleen tissue, where the most significant association between the genotype of rs1279683 and the SLC23A2 transcription level was observed among 49 human tissues in the “GTEx” Portal database (Fig 2). We also found that the transcription level of SLC23A2 in several brain tissues was significantly different according to the genotype of rs1279683 (data not shown). In SLC2A1, we found two SNVs (rs710218 and rs841851) as functional variants with the “Regulome DB” score 1b among 86 SNVs and small indels with MAF ≥ 15% in the “Genome Variation” database at jMorp. The two SNVs were not in high linkage disequilibrium with each other (pairwise r² = 0.48) in the JPT database. The “GTEx” Portal database showed that rs710218 and rs841851 were eQTLs of SLC2A1 in the left ventricle tissue, with the lowest p value for the association between the genotype of each SNV and the SLC2A1 transcript level among the 49 human tissues (Fig 2). In the brain tissues, we confirmed that the transcription level of SLC2A1 in the frontal cortex and the nucleus accumbens tended to differ depending on the rs841851 and rs710218 genotypes, respectively, in the “GTEx” Portal database (data not shown). As for SLC2A3, we found six SNVs (rs4883461, rs933552, rs7975829, rs12313154, rs7309332, and rs11610602) as functional variants with the “Regulome DB” score 1f among 99 SNVs and small indels with MAF ≥ 15% in the “Genome Variation” database at jMorp. However, there was no evidence showing that the six SNVs were eQTLs of SLC2A3 in any tissues in the “GTEx” Portal database.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Genotype-specific differences in transcript levels of vitamin C transporter gene.

These data were cited from the GTEx Portal on 11/01/2020 (https://gtexportal.org/). In the “GTEx” Portal database, a significant correlation was observed between the genotype of each SNV and the transcript level of SLC23A2 or SLC2A1 in human tissues. The p value indicates a significant difference between gene expressions by genotype. The genotype-expression association data for tissues with the lowest p values are presented for the 49 human tissues in the “GTEx” Portal database. (A) The transcription level of SLC2A1 in the left ventricle by rs710218 genotype. (B) The transcription level of SLC2A1 in the left ventricle by rs841851 genotype. (C) The transcription level of SLC23A2 in the spleen by rs1279683 genotype.

https://doi.org/10.1371/journal.pone.0259663.g002

The database searches revealed rs1279683 (SLC23A2), rs841851 (SLC2A1), and rs710218 (SLC2A1) as functional variants of VC transporters in the brain. These three SNVs were genotyped in 388 participants, excluding five participants in the normal group who had poor DNA quality. The genotype frequencies of the three SNVs are listed in S3 Table. No deviation was found in the genotypic distributions of the three SNVs from the Hardy–Weinberg equilibrium. In the three SNVs, a minor allele of each variant was associated with the reduced expression of SLC23A2 or SLC2A1 (Fig 2). Using the dominant model, we defined homozygote carriers of the major allele as the high expression group, and minor allele carriers as the low expression group in the subsequent analyses.

To examine the impacts of three functional variants on the risk of developing APOE4-associated cognitive decline (MCI or dementia), we assessed the APOE4-associated risk by stratification based on the genotype of each of the three variants (S4 Table), although power calculations showed limited statistical power (ranging from 32% to 55%) in each stratum. In the stratified genetic association analysis by SNV genotype (Table 2), APOE4-associated risk of developing cognitive decline was significant in the low expression group of rs1279683 (AG+AA) (p = 0.035, OR = 2.02, 95% CI, 1.05–3.87), whereas APOE4-associated risk was not significant in the high expression group (GG) (p = 0.71, OR = 1.21, 95% CI, 0.44–3.37). In contrast, both high expression groups of SLC2A1 (rs710218; TT, rs841851; AA) showed a significant APOE4-associated risk of developing cognitive decline (rs710218: p = 0.037, OR = 2.35, 95% CI, 1.05–5.23, rs841851: p = 0.0012, OR = 3.2, 95% CI, 1.58–6.46), whereas none of the low expression groups (rs710218: TA+AA, rs841851: AG+GG) showed a significant APOE4-associated risk of developing cognitive decline (rs710218: p = 0.49, OR = 1.3 and 95% CI, 0.62–2.75, rs841851: p = 0.39, OR = 0.67, 95% CI, 0.27–1.67) (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Odds ratios of APOE4 for developing MCI and dementia, stratified by genotype of VC transporter genes.

https://doi.org/10.1371/journal.pone.0259663.t002

Table 3 shows the results of genetic association analysis of three functional variants of SLC2A1 and SLC23A2 with cognitive decline (MCI or dementia) in the dominant model and log-additive model. In the multivariate logistic regression analysis, none of the SNVs were significantly associated with cognitive decline in either models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results of association analysis of three functional variants of SLC2A1 and SLC23A2 with cognitive decline (MCI or dementia).

https://doi.org/10.1371/journal.pone.0259663.t003

Discussion

This case–control study involved the Japanese cohort population to determine the association between VC transporter genes (SLC23A2, SLC2A1, and SLC2A3) and APOE4-associated risk of developing cognitive decline (MCI or dementia). For this purpose, we found three functional SNVs associated with the changes in SLC2A1 and SLC23A2 expression by searching several publicly available databases, and analyzed the effects of these variants on APOE4-associated risk of developing cognitive decline. For the first time, we found subgroups in which APOE4 was not a significant risk factor for cognitive decline, stratified by the SNV genotype, although the statistical power was limited to detect the APOE4-associated risk in each stratified analysis. On the contrary, no association was found between functional SNVs and the risk of developing cognitive decline.

We hypothesized that genetic variants that alter the function of VC transporters in the brain also affect the VC level in the brain and APOE4-associated risk of developing cognitive decline. We applied strict criteria to identify gene variants that could affect gene functions and expression in the available databases. Similar to our study, methods for finding variants associated with the regulation of target gene expression have been reported using various databases [35]. We identified three functional SNVs in SLC23A2 and SLC2A1, with abundant evidence of eQTLs for the target genes. However, we could not find any functional SNVs of SLC2A3 using the present strict criteria. In addition, we found no large-scale SVs, SNVs, and small indels altering protein sequences in the target genes using 15% frequency threshold. The results of this database search do not imply that variants that alter the function and expression of SLC2A3 in humans are absent. Variant filtering based on population frequency was important for our cohort size, and future studies should examine low-frequency functional variants of the brain-expressed VC transporters in larger populations with well-powered replication strategies.

The effects of APOE4 on cognitive decline (MCI or dementia) were clearly different among subgroups stratified by genotypes rs1279683, rs710218, and rs841851. The results showed that the functional variants of SLC2A1 and SLC23A2 may affect or modify the risk of developing APOE4-associated cognitive decline. As for SLC23A2, in the stratified analysis by genotype, APOE4 had a significant risk of cognitive decline in the low expression group of rs1279683, but not in the high expression group. A previous study has shown that in transgenic mice expressing additional copies of SVCT2 (encoded by SLC23A2), the expression of SVCT2 mRNA and the VC levels in organs including the brain were increased accordingly, up to two-fold in the brain depending on the mRNA expression [36]. In addition, heterozygous knockout mice of SVCT2 (encoded by SLC23A2) showed VC deficiency in the brain [37], suggesting that the VC level in the brain changes according to SLC23A2 expression. Cognitive function was reduced in SVCT2 heterozygous knockout mice, and further reduced in APP_SWE/PSEN1_ΔE9 mice (animal AD model) crossed with SVCT2 heterozygous knockout mice [37]. These findings suggest that the VC level in the brain and cognitive function can be dependent on SLC23A2 expression, and cognitive function can be affected by the brain VC level, especially in animal models with a genetic risk of AD. Whether the rs1279683 genotype in SLC23A2 changes the VC level in the human brain is unclear, and differences in brain SLC23A2 expression might be the reason for susceptibility to cognitive decline in this study.

Contrary to SLC23A2, in the stratified analysis by genotype in SLC2A1, APOE4 was shown to have a significant risk of cognitive decline in the high expression groups of SLC2A1 (rs710218 and rs841851), but not in the low expression groups. SLC2A1 (also called GLUT1) is expressed in the blood–brain barrier (BBB) and transports DHA bidirectionally [38]; however, whether the DHA level in the brain increases or decreases depending on the amount of this transporter, is unclear. The effects of DHA on the brain tissue have not been fully elucidated. In addition, as SLC2A1 is a non-specific DHA transporter and transports not only DHA but also sugars [23], the present results might not be because of the DHA transport function of SLC2A1. To the best of our knowledge, no study has shown that SLC2A1 knockout in animal models alters the in vivo concentration and transport efficiency of DHA or ASC. Therefore, the effect of high expression of SLC2A1 on the amount of ASC or DHA in the brain tissues remains unknown. As the first step in interpreting these results, careful investigation of the relationship among SLC2A1, DHA, and cognitive function using genetically modified animal models is necessary.

Previous studies have reported various mechanisms underlying the effects of APOE4 on AD and other dementing disorders; APOE4 affects (1) Aβ aggregation and clearance, (2) tau phosphorylation and tangle formation, (3) lipid/cholesterol transport and clearance, (4) glucose and mitochondrial metabolism, (5) inflammation, (6) vascular function including BBB, (7) insulin and VEGF signaling, and (8) synaptic and neuronal function [39]. VC has been reported to affect cognitive function in animal models and humans. In a mouse model of AD and VC deficiency in the brain (transgenic APP/PSEN1 mice with heterozygous SVCT2 knockout), VC deficiency exacerbated cognitive performance and Aβ deposition [37]. In addition, higher VC could prevent Aβ deposition, oligomerization, BBB disruption, and mitochondrial oxidant stress in an AD mouse model [13, 14, 40, 41]. These data suggest that VC protects normal cognitive function, particularly in APOE ɛ4-positive individuals with Aβ deposition and oligomerization, BBB disruption, and mitochondrial oxidant stress. In humans, several prospective cohort studies have shown a significant association between higher blood VC level or antioxidant supplements, including VC, and less cognitive decline over time [42–49]. Another study reported that VC supplements in combination with NSAIDs reduced the cognitive decline in APOE4 carriers [49]. However, no randomized controlled studies have revealed the beneficial effects of VC supplementation on cognitive function [50, 51].

This study has some limitations. First, the number of participants was small as we set a high MAF (>15%) for the variants analyzed, which prevented us from testing the interaction effects in the multivariate analysis. Second, the cohort was single without replication. The third limitation is a low follow-up rate. Compared with the follow-up cases, lost cases were older, had lower MMSE scores, and had a shorter education period. The high nonparticipation and low follow-up rate suggest the potential for non-responder bias. Fourth, we did not identify the cause of dementia and MCI such as AD, dementia with Lewy bodies, and vascular dementia. Fifth, we did not include questionnaires on dietary nutrient intake, including VC. Sixth, we could not analyze data stratified by sex because of the small number of participants in this study. Additional large-scale studies in well-powered replication cohorts involving neuroimaging, neuropathology, and questions about dietary nutrients are required to understand the underlying mechanisms of the effects of functional SNVs of VC transporter genes on APOE4-associated risk of developing AD. In addition, as factors controlling the VC level in the body are thought to be complex, future analyses should take into account various factors including other genetic factors such as SLC23A1, which has been reported to be associated with blood VC levels in humans. These further studies will provide a deeper understanding of the relationship between VC and cognitive function in humans.

In conclusion, APOE4 significantly affects cognitive decline (MCI or dementia) in the Japanese population, and the effects are significantly different between subgroups stratified by genotype for each of the three SNVs (rs1279683, rs710218, and rs841851), which were eQTLs for the brain-expressed VC transporters (SLC23A2 or SLC2A1). Our results imply that the functional SNVs of VC transporters can affect APOE4-associated risk of developing cognitive decline via altered VC levels in the brain.

Supporting information

Acknowledgments

We wish to thank all the residents of Nakajima for their participation in this study. We thank Dr. Hiroaki Yoshikawa, Dr. Minako Hashii, Dr. Akiyoshi Morinaga, Dr. Chiho Ishida, Dr. Kazuya Takahashi, Dr. Akio Akagi, Dr. Taro Ozaki, Dr. Ai Shimizu, Dr. Miwako Asakawa, Dr. Hiroki Yamaguchi, Dr. Yoshihisa Ikeda, Dr. Shutaro Shibata, and Dr. Hidehiro Yokoji for their valuable help throughout this study. We thank the staff of the Department of Neurology of Kanazawa University Hospital, Ioh Hospital National Hospital Organization, Noto General Hospital, and Keiju Medical Center for their clinical and technical support. We also thank Ms. Yoko Iwauchi, a laboratory assistant in the Department of Bioinformatics and Genomics, for supporting the genotyping of the SNVs.