Polymorphic Genetic Markers of the GABA Catabolism Pathway in Alzheimer’s Disease

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline, and represents the most common cause of dementia in the elderly [1]. AD etiology involves polygenic and multifactorial causes, along with environmental risk factors. The main pathological hallmarks of AD are extracellular senile plaques containing amyloid-β (Aβ) peptides, deriving from the aberrant cleavage of the amyloid-β protein precursor (AβPP), surrounded by dystrophic neurites and intracellular neurofibrillary tangles that consist of abnormally hyperphosphorylated tau protein [2]. The presence of the misfolded Aβ peptides causes a cascade of partially still unknown events triggering reactive gliosis, a significant neuropathological feature in the brain of AD patients [3].

Several hypotheses have been put forward to explain the basis of disease onset and progression. A complex array of events and independent impaired pathways contribute to disease etiology. A number of studies have shown that, besides Aβ oligomers formation and precipitation, increased oxidative stress, mitochondrial dysfunction, altered energy metabolism, abnormal insulin signaling, and dysregulation in metal homeostasis seem to play a key role in the disease [4–8].

Early-onset familial AD represents a small proportion (<1%) of all AD cases [9]; the patients typically develop the disease before the age of 65. In the majority of these cases, the accumulation and aggregation of Aβ oligomers is due to fully penetrant variants in either APP or Presenilins (PSEN1 and PSEN2) genes, the latter encoding for essential components of the γ-secretase complexes responsible for the cleavage of AβPP and the release of Aβ.

For the more prevalent late-onset form of AD (LOAD), multiple genetic and environmental factors are involved in the pathogenesis. As reported for other multifactorial diseases, genome-wide association analyses have allowed the identification of several potential susceptibility loci for both early- and late-onset AD but, overall, a large fraction of the genetic variance beyond the Apolipoprotein E (APOE) risk still remains to be explained [10].

To date, the APOE gene represents the most robust genetic risk factor. In fact, the frequency of the ɛ4 allele is dramatically increased to around 40% in patients with AD across different human populations [11]. Several studies demonstrate that Aβ deposition in the senile plaques is more abundant in carriers of APOE ɛ4 with respect to the non-carriers. Gene association studies have confirmed that the ɛ4 allele is the strongest genetic risk factor not only for LOAD, but also for early-onset AD, mild cognitive impairment, and other central nervous system (CNS) diseases related to cognitive decline [11]. The suggestion that APOE ɛ4 impacts AD pathology mainly through impairment of astrocyte- and microglia-mediated Aβ clearance has been recently qualified as one of the few instances in which a mechanistic cause-effect relationship has been worked out to explain the observational association [12].

Onto this strong background, the compilation of a list of genetic modifiers is an open research field. The GABAergic system is affected in several neurological and psychiatric disorders but its role in AD is still largely understudied. Recent work reported a diffused dysregulation of the inhibitory GABA signaling system at the molecular level in AD brain [13]. Some regions of the cortex and hippocampus show alterations of the subunit composition of GABA A receptors [14, 15]. In addition, the hippocampus shows changes in GABA transporter expression. Some studies underline a decrease in GABA levels in the parietal and frontal region of the brain of AD patients [16]. Changes in the levels of some enzymes involved in GABA metabolism, namely Glutamic Acid Decarboxylase (GAD65, GAD67) and GABA transaminase (GABA-T), were found in postmortem AD brains and cerebellum [17, 18]. A possible impairment of GABA metabolism in AD patients is also suggested by the significant increase of 2,4-dihydroxybutyrate (2,4-DHBA) [19] and 3,4-dihydroxybutyrate (3,4-DHBA) [20] in the serum, both compounds deriving from the oxidation of the GABA catabolite γ-hydroxybutyrate (GHB).

In summary, genes and the corresponding enzymes of the GABA catabolism are good candidates as possible modifiers of AD risk. GABA degradation in mitochondria of neural cells is mediated by the sequential action of GABA-T (encoded by ABAT), which converts GABA in succinic semialdehyde (SSA), and by succinic semialdehyde dehydrogenase (SSADH, encoded by ALDH5A1), which further oxidizes SSA to succinate (Fig. 1 in [21]). These reactions take part in the GABA shunt, an alternative energy production pathway activated in condition of hypoxia and cellular stress, when the function of the Krebs (Tricarboxylic Acid, TCA) cycle is compromised. Indeed, increased oxidative stress and reduction of the function of the TCA cycle have been observed in AD patients [22]. A minor fraction of SSA undergoes a reductive cytosolic pathway and is converted into GHB by SSA reductase (SSAR), an enzyme belonging to the family of aldo keto reductases, of which the AKR7A2 member (coding for SSAR) is mainly expressed in brain [23]. When SSADH activity fails, as in the case of SSADH deficiency (OMIM #271980), the cytosolic pathway prevails, resulting in GHB accumulation [21, 24].

The aim of the present study is to gain further insight on the possible involvement of the alteration in GABA catabolism in AD. By using small-sized cohorts, we searched for possible associations between variants in genes encoding for the relevant enzymes and the late-onset form of the disease in Italian patients and controls. To this aim, we considered variants in the three genes ABAT, ALDH5A1, and AKR7A2 which satisfy two conditions: the first is that there is evidence for an effect on the gene product (e.g., SSADH activity), either because they are coding missense variants or significant markers of the gene expression (eQTL). The second is that they are polymorphic enough to be informative for the sample sizes here considered (see Materials and Methods).

Furthermore, we introduced a multivariate method to take into account the individual APOE genotype, thus controlling for the effect of the discrepant allele distributions in cases versus controls. This guided us in identifying specific genotypic interactions that represented a subtle but significant peculiarity of the AD cohort. We complemented this finding with a genotype-phenotype analysis, taking into account the age at onset (AAO) and the Mini-Mental State Evaluation (MMSE) score.

The main demographic and clinical characteristics of the individuals included in this study are reported in Table 1. The AD cohort was enriched in females and the control cohort was equalized for this parameter (p = 0.25). Age was significantly (p = 2E–16) older in AD patients, whereas education was significantly shorter (p = 1.8E–12).

Allele and genotype frequencies at all genetic systems were in HWE in both the AD and control cohorts (Table 2). The two SNPs in APOE were combined into a three-allele system (corresponding to ɛ2, ɛ3, and ɛ4). The four SNPs in ALDH5A1 were combined, and the haplotype frequencies obtained after phasing. The presence of a strong linkage disequilibrium in this genomic region resulted in six haplotypes out of the 16 possible (Table 2). The HWE was verified in both AD and controls.

As to the comparison between the AD and control groups, strongly divergent APOE allele distributions were obtained (p = 2.2E–16). In particular, the ɛ4 allele was over-represented in AD versus controls (0.312 versus 0.089), and both the ɛ3 and ɛ2 alleles were under-represented (Table 2, Supplementary Table 1). None of the SNPs and/or haplotypes in the GABA catabolism genes (ALDH5A1, AKR7A2, ABAT) individually displayed significant frequency differences between the two groups. This basic level of analysis ruled out simple association relationships between variation at these sites and AD.

In order to mine the data and obtain a synthetic view of the possible differential occurrence of multilocus genotypes in our cohorts, we used sparse Principal Component analysis using 13 variables, each describing the presence/absence of a specific allele or haplotype. Sparse PC1 (Supplementary Table 2) was poorly influenced by APOE alleles and mostly influenced by ALDH5A1 haplotypes 4 and 5, with negative and positive loadings, respectively. The rs1731017A allele (ABAT) contributed a moderately negative value. Sparse PC2 captured the negatively correlated occurrence of ALDH5A1 haplotype 1 versus 4 and 5 and was positively contributed by APOE ɛ4. The rs1731017A and G alleles were the main determinants of sparse PC3, with opposite contributions, as expected. Finally, the overwhelming determinant of sparse PC4 was APOE ɛ4. All together, these PCs accounted for 67% of variance.

When examining the sparse PC scores obtained by AD and control subjects, we observed significantly different distributions for both sPC2 and sPC4. Significantly more AD patients had negative scores on sPC4 (p = 2.2E–16), in line with the higher occurrence of the APOE ɛ4 allele in this group. Significantly more AD patients had positive scores on sPC2 (p = 0.002). The placement of AD and controls in the space of these two axes is depicted in Fig. 1. Sparse PC1 and PC3 did not discriminate AD versus controls (p = 0.270 and 0.139, respectively).

The above results prompted us to search in deeper detail for possible associations between alleles contributing to these PCs, which implicitly included the APOE genotype. The corresponding genotyping results are cross-tabulated in Supplementary Table 3. In particular, we focused on sPC2, which seemingly conveyed information on alternative effects of alleles at GABA catabolism genes. Two genotypic interactions emerged. The first is between ABAT and APOE: rs1731017 A/A homozygotes were under-represented in the pooled APOE ɛ4 carriers (p = 0.02), as compared to the pooled alternative genotypes rs1731017 A/G and G/G. rs1731017 A/A homozygotes were not found among APOE ɛ4/ɛ4 AD patients (Supplementary Table 3, columns Q-S), while the same ABAT genotype accounted for about 1/5 among AD patients with other APOE genotypes (p = 0.04).

The second interaction was between some ALDH5A1 genotypes and APOE. ALDH5A1 genotypes characterized by homozygosity at rs4646828C and heterozygosity or homozygosity for rs2760118T (4/5, 4/6, 5/5, and 5/6 at haplotype level, Supplementary Table 3, columns AB-AE) were enriched among AD patients carriers of APOE ɛ4 (p = 0.003). Notably, both these interactions were peculiar to the AD cohort, and were not replicated in the control group.

In order to understand the origin of these under- and over-representation of two-locus genotypes among AD patients, we considered the MMSE score and AAO as indicators of the disease severity. For each of the genotype combinations reported in Supplementary Table 3 we calculated the means and s.d. of both indicators (Supplementary Table 4).

For MMSE, APOE ɛ4/ɛ4 patients displayed the lowest values, in line with the cognitive decline often reported for this genotype [11]. However, within-genotype variation was large, and between-genotype heterogeneity was not significant (Kruskal-Wallis p = 0.1219). When the ɛ4/ɛ4 group was partitioned according to the rs1731017 (ABAT) genotype, we observed a sharp decline in the order G/G>G/A (shaded blue in Supplementary Table 4), with borderline significance (p = 0.07517). No measurements are available for the A/A genotype, as this was not observed, but it is possible that homozygosity might be associated with a further reduction in MMSE. As to the ALDH5A1 overrepresented genotypes among APOE ɛ4 carriers (see above), they did not display any relevant alteration of MMSE scores as compared to other APOE or ALDH5A1 genotypes (p = n.s. in both cases).

Overall, AAO in APOE ɛ4/ɛ4 patients was the earliest, with a significant heterogeneity among APOE genotypes (Kruskal-Wallis p = 0.0045). In this case, we observed that the rs1731017 A/A genotype (shaded pink in Supplementary Table 4) was associated with earlier AAO than both A/G and G/G, though not significantly. Also for AAO, the ALDH5A1 overrepresented genotypes among APOE ɛ4 carriers (shaded purple in Supplementary Table 4) did not display any relevant alteration as compared to both other APOE or ALDH5A1 genotypes (p = n.s. in both cases).

Finally, we wanted to assay the power of sPC scores in predicting the AD status in our dataset. We calculated the ROC curves based on sPC4 and sPC2. The curve based on PC4, mostly contributed by APOE ɛ4 (Supplementary Figure 1), displayed a marked predictive power (Area under the curve = 0.687, 95% CI: 0.6444–0.7310). The curve based on sPC2, strongly contributed by ALDH5A1 haplotypes (Supplementary Figure 2), showed a minute but significant departure from 0.5 (Area under the curve: 0.5791, 95% CI: 0.5335–0.6237). The curves based on sPC1 and sPC3 showed a null predictive power.

In this study, we evaluated the possible association between AD and six SNPs occurring in the three genes responsible for GABA degradation: ABAT (coding for GABA-T), ALDH5A1 (coding for SSADH), and AKR7A2 (coding for a brain specific SSAR). The SNPs were selected for their purported effects on gene expression and functionality of gene products. In particular, for the ALDH5A1 we considered three SNPs for which the alternative allele is associated with a reduced SSADH activity with respect to the wild type (wt), when expressed in vitro. These SNPs were combined with a fourth, non-coding 5’ SNP which showed up as an eQTL in multiple tissues [34].

The allele frequencies obtained for all SNPs were in agreement with those reported (Table 1) for the closest proxy Italian population, i.e., the Tuscans [35]. No significant difference was observed between AD patients and controls in allele and genotype frequencies for all the above SNPs.

We also genotyped both AD patients and controls for the APOE polymorphism, confirming also in this Italian AD cohort the genetic risk associated with the presence of the ɛ4 allele. ApoE has a significant role in the regulation of the metabolism of low-density lipoproteins (LDL), cholesterol and triglycerides. In the CNS, it is highly expressed in astrocytes where it contributes to the transport of cholesterol to neurons via ApoE specific receptors [36]. Two variable coding positions in APOE give rise to three polymorphic pseudo-haplotypes, often referred to as alleles ɛ2, ɛ3, and ɛ4 (Table 2). The corresponding amino acid substitutions affect the structure of the isoforms, which differ for their capability of binding lipids [11]. Recently it has been proposed that blood-brain barrier breakdown contributes to cognitive decline in APOE ɛ4 carriers independent of AD pathology [37], with Aβ protein accumulation and aggregation as subsequent events. Moreover, APOE ɛ4, because of the substitution of two fundamental redox cysteines, shows lower antioxidant capacity and may contribute to increased oxidative stress, as observed in AD brain [38]. Population studies have shown a very strong APOE ɛ4-AD association in Caucasian patients (OR = 12.5 for the ɛ4/ɛ4 genotype) and indicate that APOE ɛ4 increases the risk of development of AD and of earlier age of onset, in an allele dose-dependent manner [11].

Our cohort sizes were adequate for a multivariate analysis of all loci to orient us in searching for gene interactions. We observed that, among AD patients, the occurrence of APOE genotypes and specific genotypes at GABA degradation loci departed from independence. Using MMSE and AAO as indicators of disease severity we suggest that, among APOE ɛ4 carriers, some ALDH5A1 genotypes characterized by homozygosity at rs4646828C and heterozygosity or homozygosity for rs2760118T might have undergone over-sampling within our particular AD cohort. On the contrary, the APOE ɛ4/ɛ4-rs1731017 A/A genotype might be associated with under-sampling, possibly because of extreme severity.

We discuss here the rationale for considering the GABA degradation pathway in general, and the three genes here considered in particular, as good candidates as modifiers of AD, in light of our results. GABA metabolism stood out as the identifier of a network of 267 genes coherently expressed in the transcriptome of prefrontal cortex, and whose connectivity is severely disrupted in AD brain [10]. Both APOE and ABAT take part in this network, opening the possibility that ABAT-APOE interactions have peculiar effects in APOE ɛ4 carriers. The highly brain- and liver-expressed [34] gene ABAT encodes for the first enzyme in GABA degradation, i.e., GABA-T, which catalyzes the production of the highly reactive compound SSA. Association studies with epilepsy have been addressed at the identification of the possible role of SNPs in the variation of the pharmacodynamics and pharmacokinetics of antiepileptic drugs, such as vigabatrin and valproate (VPA) [39, 40]. Indeed, variants at rs1731017 have been shown to modulate the response to VPA among Chinese epilepsy patients, with a higher ratio of VPA resistance in the carriers of G/A and A/A genotypes (38.2 and 48.7%, respectively), with respect to G/G (20%) [39].

The second enzyme involved in GABA catabolism is SSADH, encoded by ALDH5A1, which is mainly expressed in the brain and liver. SSADH deficiency is characterized by the accumulation of neurotoxic metabolites, such as GABA and GHB, and a clinical picture of highly heterogeneous neurological symptoms [21, 42]. When overexpressed in vitro, the alternative alleles of the most common SNPs in ALDH5A1, c.106G>C (p.G36R), c.538C>T (p.H180Y) and c.545C>T (p.P182L), displayed lower levels of enzyme activity with respect to the wt allele (86.7%, 82.5%, and 46.7%, respectively) [21]. No association with subclinical phenotypes has been associated with the c.106G>C or c.545C>T variants. Conversely, the c.538C>T polymorphism has been suggested to be related to cognitive performance, as the homozygous T/T genotype was over-represented in a sample of Italian elderly with impaired cognitive function [43].

Strong linkage disequilibrium exists between the three SNPs, with the presence of only three haplotypic arrangements (GCC, GTC, and CTT). Interestingly, when the cDNA construct harboring both c.538T and c.545T was expressed, the enzyme activity dropped to 36% [33], probably due to an impairment of the oligomerization domain where the two corresponding amino acids reside [24], CTT was deemed as one of haplotypes responsible for the pathology in two SSADHD patients [33, 44]. It is likely that the presence of the c.106C variant could result in a further reduction in activity, because the corresponding amino acid resides in the mitochondrial leader peptide, possibly affecting its removal. To this haplotype definition, we added rs4646828. Allele T at this site was shown to increase the ALDH5A1 expression levels as compared to the C allele, with an allele effect of 0.25–0.30 in multiple CNS tissues (https://www.gtexportal.org/home/). Thus, in the ALDH5A1 genotypes here found to be over-represented in APOE ɛ4 AD carriers, multiple variants sum their effects towards a reduction of SSADH activity.

GABA is produced from glutamate by GAD. Further downstream, the key enzymes of GABA metabolism act sequentially in the GABA shunt: GABA-T produces SSA from GABA (subtracting alpha-ketoglutarate to the TCA cycle), which is in turn oxidized by SSADH to succinate, which returns to the TCA cycle. Thus, the SSADH-catalyzed reaction represents an anaplerotic reaction of the TCA cycle in neurons and astrocytes. The brain has a high requirement of energy in the form of ATP, produced via glycolysis, the TCA cycle and oxidative phosphorylation [45]. A bulk of evidence shows that glucose metabolism, TCA cycle and ATP production are impaired in the AD brain [46], with mitochondrial damage being an early sign of the disease [47]. Many mitochondrial enzymes, such as pyruvate dehydrogenase, alpha ketoglutarate dehydrogenase, citrate synthase and cytochrome c oxidase are decreased in the AD brain, with subsequent slowdown of TCA cycle efficiency, and reduction of energy production. In this context, reduced SSADH activity due to SNPs-encoded isoforms, could be an additional cause of energy impairment and mitochondrial distress in the AD brain. Indeed, GABA catabolism occurs in mitochondria in both neurons and astrocytes [48, 49] and is intrinsically related to intermediate metabolism and energy production. The GABA shunt increases the energy that brain obtains mainly from glycolysis, which is very small [50].

Furthermore, the SSADH-catalyzed reaction also produces NADH, which could further contribute to energy production [51]. Therefore, if SSADH activity is decreased, this could further affect the energy supply to the AD brain, being among the causes contributing to the cognitive decline.

An additional reason for energy depletion in AD brain may be hypoxia, caused by angiopathy consequent to deposition of Aβ aggregates on the walls of cerebral capillaries [52]. Concomitantly, the production of reactive oxygen species and subsequent oxidative stress increase [53], which is possibly also caused by inefficient glucose utilization [46]. Oxidative and nitrosative stress results from the imbalance between reactive oxygen and nitrogen species production and antioxidant defense, and is relevant to the onset and progression of AD [22]. Indeed, both glycolytic and mitochondrial enzymes are oxidatively damaged in the AD brain, showing carbonylated amino acids. Studies on SSADHD patients and the murine Aldh5a1–/– model highlighted that the absence of SSADH activity significantly contributes to oxidative stress and determines mitophagy and autophagy [54].

Membrane polyunsaturated fatty acids are peroxidized and degraded to toxic aldehydes [such as 4-hydroxynonenal (4-HNE)], able to covalently bind proteins, changing their structure and affecting their activity [22]. Indeed, 4-HNE is also a substrate alternative to SSA for SSADH [54–56]. It has been reported that 4-HNE adducts accumulation is a common feature of numerous neurodegenerative diseases, including AD [46, 57–59].

AKR7A2 encodes for the brain specific SSAR, that converts SSA into GHB, a compound with neurotransmitter and neuromodulator properties [60, 61]. Experimentally, AKR7A2 was found to belong to a cluster of genes whose expression changed coherently with APOE, when this latter was engineered to ɛ4 from an ɛ3 precursor in cultured stem cells and the derived brain-like types. This model was also found to parallel APOE ɛ4-dependent changes in human brain samples [12]. The SSAR-catalyzed reaction, leading to the formation of GHB, is relevant when SSADH activity decreases, yielding insufficient SSA oxidation to succinate. Therefore, the observed increase of GHB in AD [19], might be caused by decreased SSADH activity, while SSAR is still active.

GHB has neuropharmacological properties [62], and its administration improves neuropathological and cognitive symptoms in rat and mouse models of AD. The modification of the brain transcriptome observed in these models reveal the induction of several proteins exerting a beneficial role against proteinopathies [63–65]. These results suggest GHB as a neuroprotective treatment that may help to prevent or reduce, or delay the development of irreversible brain lesions, as observed in AD [66].

In conclusion, this work identifies the difference between the AD and control cohorts not in a shift of the crude allele frequencies at genes of the GABA catabolism pathway, but rather in gene interactions that are peculiar of the AD cohort. The emerging view is that of a multifactorial contribution to the disease, with a main risk factor, i.e., the APOE genotype, and additional contributions by common variants in the genes here addressed, among others. For example, the excess of AD carriers of APOE ɛ4 with decreased SSADH activity may represent an additional suggestion for increased oxidative damage in AD. Our analysis does not allow to distinguish between a direct causative role of the variants here typed versus genetic linkage disequilibrium with the true (yet unknown) causative variants. At any rate, we complement our statistical findings with biochemical arguments which link the properties of the examined gene products to alterations reported in the literature for AD patients. The above findings and their implications on prediction of disease progression urgently need a replication in larger, independent series, with higher statistical power.

This study was supported by the Italian Ministry of Health, Ricerca Corrente, to IRCCS Fatebenefratelli, and University Tor Vergata annual allowance for PhD students (GM).

Authors’ disclosures available online (https://www.j-alz.com/manuscript-disclosures/20-0429r1).

The supplementary material is available in the electronic version of this article: https://doi.org/10.3233/JAD-200429.