Discussion
The present study investigated the genetic markers of antidepressant response/remission in two independent genome-wide datasets through SNP-, gene- and pathway-based analyses. The original sample included Korean MDD subjects and the STAR*D study served as independent replication on a different ethnicity. Despite the limited sample size of the Korean sample, we underline that relatively limited data exist on Asian populations [
8,
10] and only one previous study was performed on a Korean sample [
9]. The use of three levels of analysis, i.e. SNPs, genes and molecular pathways, was expected to reduce the risk of false positive findings despite our results should be considered as exploratory given the sample size issue. The number of SNPs that showed
p < 0.05 in both GWAS was approximately the same expected by chance, but SNP-level analysis was not intended to discover associated SNPs but to provide a relaxed criteria to select findings for subsequent levels of analysis. Gene- and pathway-based analyses were corrected for multiple-testing (Bonferroni correction and permutations), thus providing an acceptable support for the results of these analysis that can be hints for future studies.
SNP-level analysis did not retrieve any genome-wide significant result, but some potential interesting findings emerged when looking at the SNPs showing
p < 0.05 in both the samples. The top finding for remission was rs9315310 which nearest gene is
NBEA (neurobeachin).
NBEA codes a multidomain scaffolding protein primarily expressed in the brain where it is involved in trafficking of vesicles containing neurotransmitter receptors, specifically GABA and glutamate receptors [
35,
36]. No previous study suggested the involvement of this gene in antidepressant efficacy or MDD, but the glutamatergic and gabaergic systems are known to be involved in antidepressant mechanisms of action [
37,
38]. Another interesting finding for remission was rs4737771 that is located 100 Kbp far from the
CRH (corticotropin releasing hormone) gene. Indeed, the inhibition of
CRH gene expression by antidepressant drugs was demonstrated [
39] and polymorphisms within
CRHR1 and
CRHR2 (
CRH receptors 1 and 2) were associated with antidepressant response [
26]. The other SNPs with
p < 0.05 in both samples (for remission) are shown in Additional file
7: Table S5 and we underline the presence of the following genes: 1)
SLC6A3 (solute carrier family 6 member gene 3), that codes for the dopamine transporter and other polymorphisms within it were previously associated with antidepressant response by candidate gene studies [
40,
41]); 2)
CACNA1A (calcium channel, voltage-dependent, P/Q type, alpha 1A subunit) and
CACNB2 (calcium channel, voltage-dependent, beta 2 subunit), that were demonstrated as the top signals in a GWAS investigating risk loci with shared effects among major psychiatric disorders [
42]; and 3)
NRG3 (neuroregulin 3) gene that is a ligand of
ERBB4 which signaling is involved in neurotransmission, synaptic plasticity, and ketamine antidepressant action [
43]. Interestingly, a signal in
NRG1 was recently identified as promising marker of antidepressant response by a meta-analysis [
10].
The most interesting SNPs showing
p < 0.05 for response in both samples were in the
CTNNA3 (catenin (cadherin-associated protein), alpha 3) and
CACNA1A genes (Table
1).
The
CTNNA3 (catenin alpha-3) gene was associated with the risk of schizophrenia [
44] and intracellular signaling through beta-catenin translocation and AKT/PKB pathways have implicated in antidepressant-induced hippocampal cell proliferation [
45]. Most importantly, a suggestive signal in the
CTNNA3 gene was recently associated with antidepressant response in a GWAS including mainly patients of Asian ancestry [
10]. Among SNPs with
p < 0.05 for response (Additional file
9: Table S7), a SNP near to
SLC6A3 was again present (see results referred to the remission phenotype). Further, the intronic rs582854 polymorphism in the serotonin receptor 2A (
HTR2A) gene was only at 17 Kbp from a previous signal found in this gene [
46] and in linkage disequilibrium (
r
2 = 0.6 and D’ = 0.97) with rs6313 and rs6311 in the CEU population (1000 Genomes phase 3 data).
HTR2A is a replicated candidate gene for involvement in antidepressant response, rs6313-rs6311 were associated with this phenotype by candidate gene studies [
26] and rs6313 was confirmed by a recent meta-analysis [
47].
Compared to a previous GWAS of antidepressant response in a Japanese sample reporting the
CUX1 rs365836 and rs201522 as top findings [
8], we found that weak signals came from this gene (remission: rs10240601
p = 0.028; rs73412036
p = 0.039; rs12668172
p = 0.026; rs11971570
p = 0.025; response: rs2694159
p = 0.040). The
CUX1 rs365836 and rs201522 are ~ 99 far from rs2694159 and ~ 88 Kbp from rs11971570, despite no evidence of linkage disequilibrium among these SNPs was found in Asian populations according to 1000 Genomes data. No
CUX1 SNP with
p < 0.05 was found in the STAR*D.
Only one previous GWAS investigated antidepressant response in a Korean population [
9] and two SNPs in the
AUTS2 gene (rs7785360 and rs12698828) were genome-wide significant. In the present study several SNPs in the
AUTS2 gene showed p values < 0.05 in the Korean sample (remission: rs17141924
p = 0.01; rs4718974
p = 0.03; response: rs77393802
p = 0.03; rs10950209
p = 0.008; rs10234816
p = 0.04), and one SNP showed
p < 0.05 in both samples (rs7459368:
p = 0.0007 and
p = 0.03 in the STAR*D and the Korean sample, respectively), but these SNPs are about 700 Kbp from the previous findings and no evidence of linkage disequilibrium exists among them.
Gene-based analysis and pathway analysis provided some additional findings compared to SNP-level analysis.
AGBL1, CYB5A, MTRF1L, and
RGS22 emerged from SNP-based analysis and they were associated with remission also in gene-based analysis, as well as for
CTNNA3 and
HS6ST3 genes and response. A role of
CTNNA3 in antidepressant response is supported by literature as we discussed, while no previous study supporting the involvement of the other findings of gene-based analysis was published. Anyway, several of these genes are related to processes involved in cell survival, proliferation and migration, that are encompassed in the neural plasticity theory of depression [
48].
AGBL1 has a role in controlling the length of the polyglutamate side chains on tubulin and this process is critical for neuronal survival and the lack of such control results in neurodegeneration in mice [
38].
HS6ST3 generates structures required for interactions between heparan sulfate and a variety of proteins. These interactions are implicated in cell proliferation, differentiation, adhesion and migration [
49].
RGS22 (regulator of G-protein signaling 22) has been implicated in the processes of cell migration in cancer [
48].
CYB5A (cytochrome B5 type A (microsomal)) was related to autophagy induction, concomitant with reduced proliferation and migration/invasion in cancer cells [
50].
Despite their limited evidence due the small sample size, some genes were significant (
p < 0.0006) in the Korean sample but they were not replicated in the STAR*D. They included
NR3C2 (nuclear receptor subfamily 3 group C member 2),
SLC6A4 (solute carrier family 6 member 4),
HTR2A,
SLC25A4 (solute carrier family 25 member 4),
SLC6A3, and
CACNA1A for both response and remission.
NR3C2 codes for mineralocorticoid receptor 1 (MR1) and antidepressants were shown to modulate MR hormone-binding [
51] and expression [
52] in the context of the corticosteroid receptor hypothesis of depression.
SLC6A4 is a known candidate gene for involvement in antidepressant response [
26] even if the most studied polymorphism is an insertion/deletion thus it was not available in this sample. Quantitative proteomic analyses on mice hippocampal tissue implicated
SLC25A4 product in serotonergic antidepressant action [
53].
HTR2A, SLC6A3, and
CACNA1A are discussed elsewhere in this paragraph.
HTR2A (response,
p = 0.03),
SLC6A3 and
CACNA1A (remission,
p = 0.05 and
p = 0.03, respectively) gene-based analyses showed non-significant trends in the STAR*D.
The only pathway that survived multiple-test correction in both samples was GO:0022890 related to inorganic cation transmembrane transporter activity (Table
3). This pathway included several genes coding for calcium channels (
CACNA1A,
CACNA1C,
CACNB1, CACNB2). As we discussed above, polymorphisms in
CACNA1A and
CACNB2 were demonstrated to be trans-diagnostic markers of major psychiatric disorders [
42].
CACNB1 expression was modified in response to nortriptyline in hippocampal mice tissues [
53]. Preclinical studies supported a role of
CACNA1C in the pathogenesis of mood disorders [
54] and the gene was associated with MDD, bipolar disorder, schizophrenia and autism spectrum disorders [
55]. These genes code for subunits of the L-type voltage-gated calcium channel (LTCC) that is mainly involved in coupling of cell membrane depolarization to transient increase of the membrane permeability for calcium, leading to potential changes in intracellular signaling, gene transcription, and synaptic plasticity. These functions of LTCC involved brain regions that are pivotal in MDD pathogenesis such as the hippocampus and amygdale. LTCC antagonists were suggested as potential antidepressant molecules alone or in combination with SSRIs [
55] and imipramine was also demonstrated to modulate Ca(2+) intracellular rise in rat hippocampus [
56].
In the Korean sample only, the GO:0015844 pathway (monoamine transport) survived after multiple-testing correction for both response and remission (p = 7.00e-05 and p = 0.0006, respectively). This result is in line with the gene-based analysis in the Korean sample, but it was not replicated and thus poorly relevant given the small sample size.
Some limitations of the present study should be considered. First, the limited size of the Korean sample, thus the present findings should be interpreted cautiously. Quite large samples of white race were previously collected for GWAS on antidepressant efficacy, while less data are available on samples of Asian ancestry [
8‐
10], thus this study can contribute to expand our knowledge in the field and provide data for future meta-analysis (no meta-analysis including only subjects of Asian ancestry was performed yet). The risk of false positive findings was faced through the use of multi-level analysis (SNP-, gene, and pathway-based), permutation, and replication to corroborate findings. The imputation of data only at level of gene- and pathway analyses have limited the comparability of the two datasets at SNP-level, but this choice was based on resource/benefit ratio considerations. Indeed, it would have been more resource consuming and a Korean reference panel provided by the 1000 Genomes or HapMap projects is not available. The Korean population was demonstrated to have a distinctive genetic architecture [
57] and decreased imputation quality due to the lack of a Korean reference panel [
58] can be balanced in the context of multimarker tests (gene- and pathway-based tests) while it could affect the results at SNP level. The use of the STAR*D as replication sample can be seen as a further limitation because subjects were mainly of non-Hispanic white origin. Some polymorphisms were found overrepresented in certain ancestry groups [
59] but similar genes have been often implicated in the same phenotype across populations (e.g. rheumatoid arthritis [
60] and also antidepressant response [
26]). Given that the present study was not limited to the analysis of individual polymorphisms but included a gene-based analysis and pathway analysis, this approach is expected to increase the comparability between the two datasets. Anyway, we examined the MAFs of the SNPs reported in Table
1 across the main ethnic groups of the STAR*D (white non-Hispanic, white Hispanic and African-American). For 5/12 SNPs and 9/12 SNPs the MAF difference was less than 10 % and less than 15 %, respectively. Other stratification factors between the two samples should be considered, such as treatment and other clinical variables. Regarding treatment, it was not standardized, but the most part (76 %) of patients included in the Korean sample were treated with the SSRI paroxetine and all patients in STAR*D level 1 were treated with the SSRI citalopram. Inter-class pharmacogenetic differences between antidepressant drugs are possible [
61] while differences among drugs of the same class are less known. Anyway, non-standardized treatment provides the advantage of being more similar to real clinical settings. Treatment duration was also not standardized, but it was within a narrow range of 4–6 weeks. Finally, the Fisher’s exact test method that we applied to perform the GSEA was based on the distribution of SNPs with two a priori defined thresholds (0.05 and 0.01), thus we cannot exclude that the use of different p thresholds would have provided higher statistical power. On the other hand, this method avoided the restriction of the focus on the “top” findings of the pathway/gene and it provided an internal term of comparison in the real data (the random pathway) to test the null hypothesis.
Competing interests
AS is or has been consultant/speaker for: Abbott, Abbvie, Angelini, Astra Zeneca, Clinical Data, Boheringer, Bristol Myers Squibb, Eli Lilly, GlaxoSmithKline, Innovapharma, Italfarmaco, Janssen, Lundbeck, Naurex, Pfizer, Polipharma, Sanofi, Servier. PSM is or has been consultant/speaker for: Actavis, Forest, Merck, Glaxo Smith Kline, Lundbeck, Merck, Pamlabs, Pfizer, Sunovion, and Takeda; has received Research Support from Actavis, and has stock ownership for Global Medical Education. AAP is a Consultant/Advisory Board for Cubist Pharma; BDSI, Titan Pharma, Kaleo Pharma, is on the Speaker’s Bureau and received honoraria from Otsuka, Alkermes, Sunovion and BDSI; has received Grant Support from National Institutes of Health (NIDA, NIAAA), SAMHSA, AstraZeneca, Bristol-Myers Squibb, Cephalon, Daiichi Sankyo, Envivo Pharma, Forest, J & J, Jazz Pharmaceuticals, Lundbeck, Merck, Organon, Pfizer, Sunovion, Shire and Titan; is a major shareholder in Generys Biopharmaceuticals, and is not employed by or received other material support from pharmaceutical companies. CUP has been consultant/speaker for: Pfizer Korea, Lundbeck Korea, Sandoz Korea, OIAA, Otsuka Korea, Otsuka Japan, Daiichisankyo, GSK Korea; has received research grants from OIAA, Otsuka Korea, Eisai Korea, Korean Research Foundation, and Korean Ministry of Health and Welfare. CH has been consultant/speaker for: Pfizer Korea, Lundbeck Korea, Sandoz Korea, OIAA, Otsuka Korea, Otsuka Japan, Otsuka Turkey, Dongwha pharmaceuticals, GSK Korea; has received research grants from OIAA, Otsuka Korea, Eisai Korea, Korean Research Foundation, and Korean Ministry of Health and Welfare. Remaining authors do not have competing interests to report.
Authors’ contributions
EC and CF performed the statistical analysis and wrote the first draft of the paper. CUP is the principal investigator of this study, designed the study, and prepared the funding proposal. CH, SJL, AAP, PS M and AS contributed to the design of the study, gave intellectual endeavor, supervised the analyses and participated in revision of the paper. All authors read and approved the final manuscript.