Genome-wide association study and polygenic risk score analysis of esketamine treatment response

To elucidate the genetic underpinnings of the antidepressant efficacy of S-ketamine (esketamine) nasal spray in major depressive disorder (MDD), we performed a genome-wide association study (GWAS) in cohorts of European ancestry (n = 527). This analysis was followed by a polygenic risk score approach to test for associations between genetic loading for psychiatric conditions, symptom profiles and esketamine efficacy. We identified a genome-wide significant locus in IRAK3 (p = 3.57 × 10–8, rs11465988, β = − 51.6, SE = 9.2) and a genome-wide significant gene-level association in NME7 (p = 1.73 × 10–6) for esketamine efficacy (i.e. percentage change in symptom severity score compared to baseline). Additionally, the strongest association with esketamine efficacy identified in the polygenic score analysis was from the genetic loading for depressive symptoms (p = 0.001, standardized coefficient β = − 3.1, SE = 0.9), which did not reach study-wide significance. Pathways relevant to neuronal and synaptic function, immune signaling, and glucocorticoid receptor/stress response showed enrichment among the suggestive GWAS signals.

Esketamine nasal spray has been shown to have rapidly-acting antidepressant effects in patients with treatment resistant depression (TRD) and in patients with major depressive disorder (MDD) at imminent risk for suicide [1][2][3][4][5][6][7][8] . Predictors for conventional oral antidepressant treatment outcome including sociodemographic, symptom profiles, genetics, and clinical comorbidities were systematically reviewed by Perlman et al. 9 In a small clinical study assessing the antidepressant efficacy of ketamine, a racemate consisting of two enantiomers, R-and S-ketamine, it was recently reported that body mass index (BMI) was associated with the remission rate, with greater BMI being associated with greater remission rate 10 . BMI and clinical comorbidities are influenced by both genetic and environmental factors. Genetic loading of such traits provides an objective way of measuring the relationship between these and other predictors with antidepressant treatment response.
In studies assessing individual genetic factors, the brain-derived neurotrophic factor (BDNF) Val66Met allele was reported to impair basal and ketamine-stimulated synaptogenesis in prefrontal cortex in vitro 11 , and a significant genetic association between Val66Met and ketamine treatment outcome at 4 h post treatment was reported in a candidate gene study of small sample size 12 . A more recent study further suggested that the BNDF Val66Met polymorphism may influence the improvement in suicide ideation following ketamine infusion in a sample of depressed participants from Taiwan 13 . In general, however, genetic associations with MDD disease susceptibility outcome reported in relatively small candidate gene studies have proven difficult to replicate in studies of larger samples 14 . Therefore, studies of genetic effects influencing antidepressant treatment outcome may particularly benefit from the use of genome-wide association analysis (GWAS) approaches in clinical trials of larger patient samples. Here, we assessed the genetic contributions to esketamine treatment response from patients with TRD who participated in two Phase III trials testing the efficacy and safety of esketamine, using both a genome-wide association analysis and a polygenic risk score (PRS) approach.

Results
Esketamine treatment response outcome was assessed at the 4 week study endpoint using one continuous variable ( www.nature.com/scientificreports/ defined by achieving a final MADRS score of < 12). The demographic and clinical characteristics of study participants are summarized in Table 1 and Supplemental Table 1. Participants of the randomized TRANSFORM-3 study were recruited from an elderly population and had a lower remission rate than participants of the openlabelled SUSTAIN-2 study. Gender and concomitant medication proportions were comparable between remitters and non-remitters. After controlling for study, the baseline demographic characteristics (age and baseline BMI) were comparable between remitters and non-remitters. As expected from the clinical literature, remitters had lower baseline depression symptom severity score than non-remitters. The genome-wide association analysis revealed one genome-wide significant association between an exonic synonymous variant (rs11465988, p = 3.57 × 10 -8 ) in the interleukin 1 receptor associated kinase 3 (IRAK3) gene and the percent change in MADRS score ( Table 2, Fig. 1 for Manhattan plot, Fig. 2A for regional plot and Supplemental Fig. 1A for QQ plot, Genomic Control lambda (λ) = 0.986). SNPs (e.g. rs115989442, rs144324167, rs79138866, rs116371327, rs150373274, and rs144520864) in linkage disequilibrium (r 2 = 0.64) with rs11465988 are part of the regions engaging in intra-chromosomal loop (Fig. 2B, Supplemental Table 2) and could potentially be regulatory elements. rs144520864 is in fact located in a region with an annotated enhancer. An additional regional plot using rs17767394 as index SNP is also shown as Supplemental Fig. 2.
Gene-level association analysis revealed one significant gene NME/NM23 family member 7 (NME7, p = 1.73 × 10 -6 , Supplemental Fig. 4A) for the percentage change in MADRS score. In the percent change in MADRS score GWAS, a pathway enrichment analysis revealed suggestive enrichments of genes involved in the negative regulation of glucocorticoid metabolic process (nominal p = 3.53 × 10 -5 ) and neuronal action potential  Table 2. Variants with association p-value less than 1 × 10 -6 in GWAS. Note that beta coefficient is reported for percentage change of MADRS score from baseline and OR is reported for responder status. www.nature.com/scientificreports/  Table 4). We did not identify an association between the change in MADRS score and the BDNF Val66Met polymorphism in the current study (p > 0.05). After applying corrections for multiple testing, none of the associations between esketamine's antidepressant efficacy and the PRS genetic loading for psychiatric conditions or symptom profiles was significant at the studywide level (which required p < 0.0004 for significance). In Table 3 and Supplemental Fig. 5 we list suggestive associations observed in these analyses, however, along with their nominal p-values. Thus we observed suggestive (i.e., p-values listed are nominal) negative correlations between the depressive symptom PRS 15 (p = 0.001, Table 3 and Supplemental Fig. 5) and the esketamine treatment response outcome as measured by percentage change from baseline in the MADRS score at the end of four week treatment period (Table 3). In addition, the depressive symptom PRS (p = 0.004) displayed suggestive positive correlations with esketamine responder status. Lastly, depressive symptoms PRS (p = 0.002) and insomnia 16 PRS (p = 0.003) exhibited suggestive positive correlations with esketamine remission status.

Discussion
In this investigation of genetic associations with the antidepressant outcome to esketamine treatment, two findings remained significant after applying corrections for multiple testing. From the GWAS, a genome-wide significant association was identified with the percent change in MADRS score in an exonic SNP in IRAK3. IRAK3 encodes a member of the interleukin-1 receptor-associated kinase protein family that is primarily expressed in monocytes and macrophages, where it functions as a negative regulator of Toll-like receptor signaling. In addition, the gene-level association analysis revealed one significant gene NME7 for the percentage change in MADRS score. NME7 is a γ-tubulin ring complex component that regulates the microtubule-nucleating activity of this complex 17 .
A suggestive signal observed in both the continuous endpoint GWAS and the responder status GWAS was an expression quantitative trait loci (eQTL) for GGH, an enzyme that regulates intracellular folate concentrations. Folate deficiency has been linked to oxidative stress 18 . Meta-analysis showed that individuals with depression had lower folate levels than those without depression 19 . Folic acid administration was also shown to ameliorate depression-like behavior in rats subjected to chronic unpredictable mild stress, a putative rodent depression model 20 .
The previously reported association between the Val66Met BDNF variant and antidepressant response to IV ketamine was not replicated in the current study. A significant methodological difference between studies, however, was that in the previous trial that reported this association the antidepressant response was assessed 4 h post ketamine infusion 12 , whereas in the current study, the antidepressant outcome was assessed after 4 weeks of repeated esketamine nasal spray administration. While the importance of this timing difference in detecting an association with the Val66Met BDNF variant remains unclear, it is noteworthy that hypotheses generated from relatively small candidate gene studies in MDD have typically proven difficult to replicate in larger samples 14 , emphasizing the importance of studying larger sample sizes to detect reliable genetic signals. Nevertheless, the findings from the current study also warrant replication in larger sample sizes given the relatively modest number of participants included. www.nature.com/scientificreports/ It has been shown previously that apoptotic biochemical cascades can exert local actions on the functions and structural dynamics of growth cones and synapses 21 . In this context, it is of interest that several apoptotic signaling pathways were identified as suggestive enriched gene sets. In preclinical models ketamine enhances structural plasticity in mouse mesencephalic and human iPSC-derived dopaminergic neurons via AMPARdriven BDNF and rapamycin kinase (mTOR) signaling 22 . In a rat traumatic brain injury (TBI) model, posttraumatic administration of a sub-anesthetic dose of ketamine exerts neuroprotection via attenuating inflammation and autophagy 23 . There have been conflicting reports as to whether ketamine induces apoptosis, which might reflect a dependence on dose and developmental period. It was reported that ketamine induced apoptosis in human uroepithelial SV-HUC-1 cells 24 and in the neonatal rat brain 25 , while a study in chronic unpredictable stress model of depression suggested an anti-apoptotic and antidepressant effects of ketamine 26 . In the exposure range encompassing concentrations at which esketamine nasal spray has been tested for antidepressant effects in humans, no evidence of neuronal toxicity was identified in experimental animals 27 . Notably, in a putative rodent depression model involving chronic mild stress that produces dendritic atrophy in the medial prefrontal cortex, a single ketamine administration restored synaptic density and function toward normative levels 28 . Such changes in synaptic plasticity are hypothesized to underlie the relatively long-lasting antidepressant effects of ketamine and esketamine following single or pulsed intermittent doses 29 , and the pathophysiology of MDD is associated with regional atrophy in the medial prefrontal cortex and other anatomically related structures 30 .
In clinical studies the antidepressant response to ketamine has been predicted by peripheral blood evidence of low-grade inflammation at baseline or by the enhancement of stimulus-evoked somatosensory cortical responses (a putative in vivo measure of long term potentiation effects mediated via changes in synaptic plasticity) at 4 h post-administration 31,32 . Pathways relevant to neuronal and synaptic function, immune signaling, and glucocorticoid receptor/stress response showed enrichment among the GWAS suggestive signals. These findings are consistent with the hypotheses that inflammation and synaptic plasticity play a role in differentiating esketamine responders from non-responders.
This study suggests that PRS for psychopathology/symptom profiles may influence the antidepressant treatment outcome for esketamine. Although the PRS for depressive symptoms 15 did not reach study-wide significance, the suggestive associations were consistent across multiple p-value thresholds (p T ) used to construct PRS and across three esketamine efficacy endpoints. In addition, the condition of "depression" 33 PRS (PGC2_ MDD + UKB in Table 3) constructed by the summary statistics from the GWAS meta-analysis by Howard et al.
(2019) (without the 23andMe cohort 34 ) showed suggestive associations that did not reach significance for the remission endpoint. The GWAS meta-analyses for depressive symptoms 15 and that for the condition, "depression" 33 differ in two respects. First, while both studies included PGC phase 1 samples 35 , the GERA 36 samples (7,231 cases and 49,316 controls), and the UK Biobank (UKB) samples, the "depression" GWAS meta-analysis 33 additionally included samples from other cohorts, e.g. iPSYCH, deCODE, GenScot, and the incremental core samples from PGC MDD Working Group phase 2 analysis 36 . Second, the phenotypic definition differed in that "depressive symptoms" in the UKB cohort from the Okbay et al. (2016) study employed a continuous phenotype measure by combining responses to two mental health questionnaire (MHQ) questions deployed to UKB participants, which asked about the frequency in the past two weeks with which the respondent experienced feelings of lack of enthusiasm/interest and depression/hopelessness, whereas the Howard et al. (2019) study used a "broad depression" phenotype 37 , e.g. self-reported past help-seeking for problems with "nerves, anxiety, tension or depression.
Finally, genetic loading of BMI was not associated with esketamine remission status, in contrast to the previous report of ketamine 10 . In a correlation analysis of the SUSTAIN-2 clinical data, baseline BMI was also not associated with remission status, either in the entire sample irrespective of race (p = 0.365, n = 667) or in the subsample with European ancestry (p = 0.742, Supplemental Table 1).

Methods
PsychArray genotyping data were generated using blood DNA samples collected from the SUSTAIN-2 8 (NCT02497287, n = 598) and TRANSFORM-3 2 (NCT02422186, n = 95; only participants with age of onset less than 55 years were included) phase III pivotal clinical studies. All subjects genotyped were of European ancestry. The clinical studies were carried out in accordance with the ethical principles outlined in the Declaration of Helsinki, Good Clinical Practices guidelines, and applicable regulatory requirements. The study protocols were approved by the local, regional, or central Institutional Review Board (IRB)  The analysis was composed of TRD patients who received esketamine combined with a newly initiated oral antidepressant treatment (SSRI or SNRI) either in an open labelled (for SUSTAIN-2) or in a randomized (for TRANSFORM-3) fashion. A total of 527 samples were included in the final analysis. Treatment response endpoints were defined as follow: (1) a quantitative trait using percentage of change of MADRS score at the end of study compared to baseline; (2) response defined as ≥ 50% improvement from baseline in the MADRS Score; (3) remission defined as MADRS ≤ 12 at study endpoint. Additional details of these clinical studies are provided in the Supplemental Text or could also be found in https ://clini caltr ials.gov/.
Genotypes were imputed based on the 1000 Genome Project 38 Phase I reference panel. A SNP-wise genomewide association analysis was performed using PLINK 39,40 . In addition, a gene-wise genome-wide association followed by pathway enrichment analysis was performed using MAGMA 41 . In all analyses the models corrected for gender, study ID, baseline symptom severity, and 5 principal components representing the population substructure. Detailed methods are described in the Supplemental Text.
Polygenic risk scores (PRS) were constructed based on well-powered genome-wide association studies (GWAS) of 15 PRS phenotypes, of which six were constructed for psychiatric conditions (depression 33 , bipolar disorder 42 , schizophrenia 43 , autism 44 , ADHD 45 , anxiety 46 ) and seven psychiatric characteristics (history of suicide attempt 47 among depressive subjects or among schizophrenia, bipolar, and depressive subjects), depressive symptoms 15 , subjective well-being 15 , neuroticism 15 , insomnia 16 , education attainment 48 , and cognitive performance 48 ), and BMI 49 . To correct the resulting p-values for performing comparisons in multiple PRS phenotypes and at 8 p-value thresholds assessed (i.e., 5e−08, 0.001, 0.05, 0.1,…0.5, 1), the association p-value < 0.05/ (15 × 8) ~ 0.0004 (for 15 phenotypes and 8 P T bins) between PRS and any esketamine treatment response outcome was considered to be study-wide significant. To balance Type 2 error, nonsignificant associations that reached nominal p < 0.005 were considered "suggestive". The PRS analysis was performed using PRSice-2 50 . All p-values reported in this study were uncorrected p-values.