Discovery of a potential novel pharmacogenomic biomarker on ANK3 gene for liafensine, a triple reuptake inhibitor for treatment-resistant depression Authors: Wen Luo, Gang Wang, John W. Whitaker, Zafrin Dhali, Hong Sun, Haiping Lu, Xiaojun Li, Tao Shi, Jinkun Zhang Published: 2026 ABSTRACT Liafensine is a triple reuptake inhibitor targeting transporters for serotonin, norepinephrine, and dopamine for treatment-resistant depression (TRD). It did not exhibit efficacy in non-biomarker-selected TRD patients in two Phase 2b studies. We utilized the blood samples from the patients enrolled in these two studies and extracted genomic DNA to conduct a genome‑wide association study aiming to find a biomarker which can predict liafensine response. A single single-nucleotide polymorphism (SNP), rs12217173, at ANK3 gene was identified as strongly associated with treatment response to liafensine (p = 6.61 × 10−8) in the discovery set (n = 186) and was further confirmed in the replication sample set (n = 47, p = 0.05, combined p = 1.27 × 10−8). In addition, this SNP was not associated with the efficacy of the duloxetine or escitalopram, suggesting it is a liafensine-specific biomarker. This finding was subsequently confirmed in a prospective clinical study. Thus, this study represents a novel approach to translating precision medicine into psychiatric diseases. Introduction Treatment-resistant depression (TRD) is a severe form of major depressive disorder (MDD), in which patients have had an inadequate response to previous antidepressant treatments. Up to 55% of people with MDD meet the criteria for TRD [1, 2]. It is a serious and life-threatening condition with high rates of individual and society-level morbidity, mortality, medical and economic costs. Although there are more than 20 drugs approved to treat MDD, the development of new drugs to treat TRD has been challenging. Only two drugs have been approved by the United States Food and Drug Administration: Symbyax® and Spravato®. Symbyax’s use is limited by adverse reactions such as metabolic dysfunction, weight gain, and extrapyramidal symptoms which are poorly tolerated by many patients. Spravato (esketamine) nasal spray has been the only new drug approved for TRD over the last 15 years. It demonstrates rapid onset of treatment effect, but is associated with significant adverse reactions including sedation, dissociation, respiratory depression, and abuse and misuse [3, 4]. The significant unmet needs in TRD patients highlight the importance of developing new classes of treatment that balance the risk-benefit profile to deliver an optimal approach to managing TRD effectively. The monoamine hypothesis of depression postulates primary dysfunction of serotonin (5-HT), norepinephrine (NE), and dopamine (DA) systems in the brain. This hypothetical framework has guided the development of antidepressants for many years. Selective serotonin reuptake inhibitors (SSRIs), and dual serotonin and norepinephrine reuptake inhibitors (SNRIs) are two commonly prescribed classes of antidepressants to treat MDD. Thus, a drug that can simultaneously modulate 5-HT, NE, and DA, a triple reuptake inhibitor (TRI), was postulated to be an attractive approach to treat TRD. Liafensine (formerly BMS-820836 and AMR-000013, also known as DB104) is a potent and selective TRI. It was evaluated as monotherapy in Phase 2b clinical studies for the treatment of TRD patients in two separate trials, CN162006 and CN162007 [5]. Both CN162006 and CN162007 have three phases: screening (Phase A), prospective treatment with duloxetine or escitalopram (Phase B), and randomized treatment (Phase C) (Fig. 1a). Liafensine was well tolerated and exhibited an excellent safety profile. However, it did not demonstrate efficacy compared to the control arms in the two Phase 2b TRD studies (Fig. 1b).Fig. 1CN162006 and CN162007 study design, results, and sample availability.a Both studies have three phases: In Phase A, Patients with a history of inadequate response to 1 to 3 adequate trials of antidepressant were enrolled; In Phase B, patients were prospectively treated with duloxetine (CN162006) or duloxetine or escitalopram (CN162007), and the inadequate responders were randomized; In Phase C, the randomized patients either continued with the drug being used in Phase B (as the control arm) or treated with liafensine. b Mean change from baseline for MADRS total score during Phase B and Phase C (separated by the vertical yellow line). c Number of subjects from BMS studies with samples available for biomarker study. BMS‑820836 = liafensine; SE = standard error. CN162006 and CN162007 study design, results, and sample availability. a Both studies have three phases: In Phase A, Patients with a history of inadequate response to 1 to 3 adequate trials of antidepressant were enrolled; In Phase B, patients were prospectively treated with duloxetine (CN162006) or duloxetine or escitalopram (CN162007), and the inadequate responders were randomized; In Phase C, the randomized patients either continued with the drug being used in Phase B (as the control arm) or treated with liafensine. b Mean change from baseline for MADRS total score during Phase B and Phase C (separated by the vertical yellow line). c Number of subjects from BMS studies with samples available for biomarker study. BMS‑820836 = liafensine; SE = standard error. In the current study, we conducted genome-wide association research with archived samples from the two failed Phase 2b studies to identify novel genetic biomarkers that correlated with liafensine efficacy. Materials and methods Study patients Both CN162006 (NCT01309945) and CN162007 (NCT01369095) were global, multicenter, randomized, double-blind, active controlled studies [5]. CN162006 was a flexible-dose study where patients with a history of inadequate response to 1–3 antidepressants and one prospective treatment (duloxetine) were randomized to be switched to flexibly dosed liafensine (0.5‑2 mg/day) or remain on duloxetine. CN162007 was a fixed-dose, dose-response study where patients with a history of inadequate response to 1–3 antidepressants and one prospective treatment (duloxetine/escitalopram) were randomized to be switched to a fixed dose of liafensine (0.25, 0.5, 1, or 2 mg/day) or remain on duloxetine/escitalopram (Fig. 1a). CN162006 was conducted in Canada, Finland, France, South Africa, Sweden, and the United States between April 2011 and January 2013; CN162007 was conducted in Argentina, Austria, Australia, Canada, France, India, Italy, South Africa, Spain, the United Kingdom, and the United States (including Puerto Rico) between July 2011 and May 2013. Both studies were conducted by BMS in accordance with the aims and principles of the International Council for Harmonisation (ICH), Good Clinical Practice (GCP), and the Declaration of Helsinki. All patients provided written informed consent before study participation. Both studies included male and female patients aged 18 to 65 years with a diagnosis of MDD who were currently experiencing a major depressive episode ( > 8 weeks but < 3 years in duration) as defined by the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, criteria (American Psychiatric Association, 2000). Patients entering the prospective phase of the study were required to have a Hamilton Rating Scale for Depression‑17 Item (HAMD-17) total score of 18 or higher at screening and baseline, as well as a history of inadequate response (<50% reduction in depressive severity) to treatment with 1 to 3 adequate trials (adequate dose and duration) of antidepressant medication (as assessed by the Antidepressant Treatment Response Questionnaire and clinician interview). In both studies, the primary efficacy endpoints were assessed from the end of the prospective phase to the end of the randomized phase. The primary endpoint was the change in MADRS total score from randomization to the end of treatment timepoint. Real world data from Beijing Anding Hospital Depression Cohort: Collecting between January 2017 and December 2022, all patients were diagnosed with major depressive disorder (MDD) using the Mini International Neuropsychiatric Interview (MINI) according to DSM-IV criteria. Psychiatric assessments, including the 17-item Hamilton Depression Rating Scale (HAMD-17), the Young Mania Rating Scale (YMRS), the 16-item Quick Inventory of Depressive Symptomatology-Self Report (QIDS-16), and the Patient Health Questionnaire-9 (PHQ-9), were conducted at baseline, week 8, and week 12. Detailed medication records were also collected during the study period. A total of 713 patients were included in the present study based on the following criteria: aged between 15 and 65 years, a baseline HAMD-17 score ≥14, no use of antidepressants prior to enrollment, initiation of antidepressant monotherapy after inclusion, and availability of outcome assessments at week 8. Patients were excluded if they had missing key clinical or medication data or if the efficacy of antidepressants treatment could not be determined, such as due to incomplete rating scale assessments or significant data loss. The study was approved by Ethic Committee with EC registration number of [2019]-006 by Beijing Municipal Health Commission Ethics Committee. ANK3 genotype is one of the many genetic analyses under this cohort study. Study design and oversight After the liafensine program at BMS was terminated due to lack of efficacy in the Phase 2b studies, we acquired the program, attempting to rescue the drug by finding a predictive biomarker. Of the 887 subjects who enrolled in CN162006 and received prospective treatment with duloxetine 60 mg/day in Phase B, 346 subjects with inadequate response were randomly assigned to double-blind treatment in Phase C with either liafensine (172 subjects, 1 subject was not treated) or continuing duloxetine (174 subjects) (Fig. 1a). Of the 979 subjects who enrolled in CN162007 and received prospective treatment with duloxetine/escitalopram (Phase B), 502 subjects with inadequate response in Phase B were randomly assigned to double-blind treatment (Phase C) as follows: 51 subjects to liafensine 0.25 mg/day, 51 subjects to liafensine 0.5 mg/day, 102 subjects to liafensine 1 mg/day, 100 subjects to liafensine 2 mg/day, and 198 subjects to continuation of the Phase B antidepressant (duloxetine 60 mg/day or escitalopram 20 mg/day) (Fig. 1a). From Phase C of CN162006 and CN162007, DNA or whole blood samples were available for 415 patients, out of whom 84 from CN162006 and 149 from CN162007 (totaling 233) were treated with liafensine. These 233 liafensine-treated patients were then randomly divided (4:1 ratio) into two independent sets: 186 for discovery and 47 for replication, stratified by race, treatment dose, study (CN162006 and CN162007) to preserve the balance (Fig. 1c). The discovery and replication set also shared similar distribution for MADRS total score change from baseline. A genome-wide association study was performed on the 186-patient discovery set, and the results were tested with the 47-patient replication set. All patients provided written informed consent for using the samples for pharmacogenomic study. Sample processing and genotyping Genomic DNA was extracted from 300 µL of whole blood samples using QIAamp DNA blood Mini Kit. The genomic DNA samples obtained from the 186 patients in the discovery set were genotyped by Illumina using their Infinium® Omni5Exome-4 v1.3 BeadChip array, which contains about 4.5 million SNPs across the human genome. Genotype-based sex predictions were compared with the reported sex to confirm sample fidelity and identity. SNPs from sex chromosomes, with missing call rates greater than 2%, or with minor allele frequency below 5% were removed, resulting in 1,789,293 SNPs for the subsequent genome-wide association analysis. The genotyping results of the selected top SNPs from the genome-wide association analysis of the 186-patient discovery set were confirmed by real-time polymerase chain reaction (RT-PCR) using Life Technologies TaqMan SNP genotyping assay kits and reagents. The TaqMan assay was also used to genotype the remaining 47-patient replication set, the 182 patients from the control arms in Phase C, and all other patients who participated in Phase B if their clinical samples were available. Statistical analysis The Montgomery-Åsberg Depression Rating Scale (MADRS) is a clinical assessment tool designed to measure the severity of depressive episodes in patients with mood disorders. The 10 items in the MADRS are each scored between 0 and 6. Zero represents normal or baseline, and 6 represents continuous or severe presence of symptoms. Thus, the 186‑patient discovery set was dichotomized into 92 responders and 94 non-responders using the overall median MADRS total score change, approximately 4.75-point MADRS total score reduction from the baseline, as the cutoff. This binary labelling was used as the phenotype variable for the genome-wide association analysis. The genome-wide scan results were analyzed using PLINK v1.90b4.5 for quality control (QC) and genotype-phenotype association analysis [6]. The significance of association between each SNP passed QC and the phenotype was assessed using the standard case-control model. The clinical data analysis on the selected SNPs was based on a multivariable mixed model for repeated measures with an unstructured covariance matrix, including fixed factors of baseline MADRS score, visit, treatment, interaction of treatment and visit, age, sex, race, region, and study by SAS software (SAS Institute). Results After performing genome-wide association analysis using the clinical sample from prior BMS studies, the strongest association for liafensine treatment effect is the SNP rs12217173 with a p-value of 6.61 × 10−8 (Fig. 2). This p-value almost reaches the genome‑wide significance cut-off of 5 × 10−8. We still considered this top SNP as significant given the small sample size of 186, which is far less than most genome‑wide association studies. This SNP is located on chromosome 10 in an intron of the ANK3 (Ankyrin 3) gene. Ankyrins are a family of proteins that are believed to link the integral membrane proteins to the underlying spectrin-actin cytoskeleton. The ANK3 gene is mainly expressed in the central and peripheral nervous system [7], and is associated with several neuropsychiatric disorders [8]. Thus, we named this biomarker as ANK3 biomarker.Fig. 2GWAS results from the discovery set.a Manhattan plot of genome-wide association study result for liafensine response in the discovery set. The figure illustrates the association between individual SNPs and a response to liafensine response in 186 TRD patients. The negative log of the P value for the association is plotted against the chromosomal location across the genome. The pink horizontal line indicates the genome-wide significance level of 5 × 10−8. b Linkage Disequilibrium structure around rs12217173 using CEU data from 1000 genome phase 3. GWAS results from the discovery set. a Manhattan plot of genome-wide association study result for liafensine response in the discovery set. The figure illustrates the association between individual SNPs and a response to liafensine response in 186 TRD patients. The negative log of the P value for the association is plotted against the chromosomal location across the genome. The pink horizontal line indicates the genome-wide significance level of 5 × 10−8. b Linkage Disequilibrium structure around rs12217173 using CEU data from 1000 genome phase 3. The efficacy of liafensine treatment in each of the three genotypes of rs12217173 (designated AA, AG, and GG) within the discovery were compared to each other and the control group (Fig. 3a). Liafensine‑treated patients with the GG genotype at ANK3 biomarker exhibited significant improvement in MADRS total score change from baseline compared to those with the AA or AG genotype (5.8‑point or 4.6‑point improvement in MADRS total score changes at Week 6, respectively) and compared to patients receiving duloxetine/escitalopram regardless of their genotype (4.2‑point improvement at Week 6). Thus, the GG genotype is defined as ANK3-positive, and the AA and AG genotypes are defined as ANK3- negative.Fig. 3Comparisons of efficacy between ANK3-positive and negative patients.a Mean MADRS total score change from baseline in patients with different ANK3 biomarker genotypes (AA/AG/GG) treated with liafensine vs duloxetine (Dul) or escitalopram (Escit) control in the discovery set. N in the box indicates the number of patients. b Similar analysis was conducted in the replication set. SE = standard error. c Treatment effects in ANK3-positive and ANK3-negative patients combing discovery set and replication set. Mean MADRS total score change from baseline for liafensine vs control (duloxetine [Dul] or escitalopram [Escit]) based on pooled data from CN162006 (flexible dose of 0.5‑2 mg averaging 1.5 mg) and CN162007 (fixed doses of 1 mg and 2 mg). SE = standard error. The least-square (LS) means and SE computed from MMRM (Mixed Model Repeated Measures) that included the following fixed factors with an unstructured covariance matrix: baseline MADRS score, visit, treatment, interaction of treatment and visit, age, sex, race, region, and study. d Comparisons of response rate between liafensine and control in ANK3-negative vs ANK3-positive patients. The error bars represent 95% confidence intervals. Comparisons of efficacy between ANK3-positive and negative patients. a Mean MADRS total score change from baseline in patients with different ANK3 biomarker genotypes (AA/AG/GG) treated with liafensine vs duloxetine (Dul) or escitalopram (Escit) control in the discovery set. N in the box indicates the number of patients. b Similar analysis was conducted in the replication set. SE = standard error. c Treatment effects in ANK3-positive and ANK3-negative patients combing discovery set and replication set. Mean MADRS total score change from baseline for liafensine vs control (duloxetine [Dul] or escitalopram [Escit]) based on pooled data from CN162006 (flexible dose of 0.5‑2 mg averaging 1.5 mg) and CN162007 (fixed doses of 1 mg and 2 mg). SE = standard error. The least-square (LS) means and SE computed from MMRM (Mixed Model Repeated Measures) that included the following fixed factors with an unstructured covariance matrix: baseline MADRS score, visit, treatment, interaction of treatment and visit, age, sex, race, region, and study. d Comparisons of response rate between liafensine and control in ANK3-negative vs ANK3-positive patients. The error bars represent 95% confidence intervals. The findings from the ANK3 biomarker discovery set were confirmed using the DNA samples from the replication set as shown in Fig. 3b. Liafensine‑treated patients with the GG genotype at ANK3 biomarker in the replication set exhibited a larger improvement in MADRS total score change from baseline compared to those with the AA or AG genotype (14.5‑point or 10.3‑point improvement at Week 6, respectively) and compared to patients from the control arm (10.5‑point improvement at Week 6). When the discovery data set and replication data set were combined, SNP rs12217173 showed even stronger association for liafensine treatment effect with p value 1.27 × 10−8. In the CN162006 study, patients received flexibly dosed liafensine 0.5–2 mg at an average dose of 1.5 mg; all 29 ANK3-positive patients from CN162006 were included in the analysis. There were 4 fixed doses of liafensine (0.25, 0.5, 1, and 2 mg) in the CN162007 study and there was no significant treatment effect in the non-biomarker-selected population in any dose group [5]. However, among ANK-positive patients, MADRS total score changes from baseline were 0.9, −1.0, −2.8, and −2.5 for the 0.25, 0.5, 1, and 2 mg groups, respectively. There appeared to be a dose response from 0.25 mg up to 1 mg, and the treatment effect appeared to reach a plateau from 1 mg to 2 mg Dose response curve (Suppl Fig. 1). In addition, the primary objective of CN162007 was to compare the combined 1 mg and 2 mg doses vs control based on the levels of SERT and DAT occupancy established by positron emission tomography. Thus, only data from the 18 (out of 25) ANK3-positive patients treated with the 1 mg and 2 mg doses were included in the analysis. The 29 ANK3‑positive patients from CN162006 and 18 ANK3‑positive patients from CN162007 made up the 47 total liafensine treated ANK3-positive patients shown in Fig. 3c. To match the analysis in the ANK3-positive patients, only data from the 74 (out of 124) ANK3-negative patients on the 1 mg and 2 mg arms in CN162007 were included in the analysis. The 55 ANK3-negative patients from CN162006 and 74 ANK3‑negative patients from CN162007 made up the 129 total liafensine treated ANK3‑negative patients shown in Fig. 3c. ANK3-positive patients, which account for about 20% of the total patients, exhibited a 4.7‑point greater reduction in MADRS total score change at Week 6 from baseline compared to control (p = 0.025). A similar comparison was performed in ANK3‑negative patients, which represents about 80% of the total patients, and no meaningful difference in treatment effect was observed between liafensine and control (p = 0.736). This analysis revealed why the BMS studies failed to show meaningful treatment effect in the non-biomarker-selected patient population as the treatment effect of the small subset of responders (ANK3-positive patients) was diluted by the larger subset of ANK3-negative patients. Each item of the MADRS total score was also analyzed and the results show the different impact of liafensine on these 10 items (Suppl Table 1). MADRS-6, which focuses on the core symptoms of depression (as assessed by the MADRS items: apparent sadness, reported sadness, inner tension, lassitude, inability to feel, and pessimistic thoughts) [9], is significantly different(p = 0.015) in ANK3-positive patients between liafensine treatment group vs. the control group. None of the 10 items nor MADRS-6 reaches statistical significance level in ANK3-negative patients between liafensine treatment and the control. The adverse events in ANK3-positive are numerically higher than ANK3-negative patients in some categories (Suppl. Table 2), but it is difficult to draw conclusion as the adverse events are so low and some could be the results of symptoms of depression, such as headache [10]. The response rate is defined as the percentage of patients who have more than 50% MADRS total score reduction from the baseline and is another commonly used measure of drug efficacy in depression studies. Based on the data from CN162006 and CN162007 study, the liafensine arm’s response rate was 59.1% vs 32% from the control arm (p = 0.031) in ANK3-positive patients, while the difference is only 1.7% in ANK3-negative patients (21.8% vs 23.5%, p = 0.749) (Fig. 3d). Patients whose MARDS total score at week 6 were reduced to below10 points are considered as remission. The remission rate from liafensine arm (43.2%) is more than double that of control arm (20.0%) in ANK3-positive patients (p = 0.052), while the remission rate is identical between liafensine arm and control arm in ANK-negative patients (p = 1) (Fig. 3d). To examine whether ANK3 biomarker can also predict the efficacy of other antidepressants, DNA samples from patients treated with duloxetine or escitalopram were also genotyped for their ANK3 biomarker status. In Phase B of both CN162006 and CN162007 studies, about half of the patients were responders from duloxetine or escitalopram treatment. Figure 4 shows that patients carrying GG genotype (ANK3‑positive) did not exhibit a better response compared to patients carrying AA or AG genotypes (ANK3‑negative) when treated with duloxetine (CN162006) or duloxetine/escitalopram (CN162007). The samples from BMS studies are mostly from Caucasian patients or patients with African ancestry. In addition, only two antidepressants, duloxetine and escitalopram, other than liafensine, were evaluated in BMS studies. To further explore the correlation of ANK3 biomarkers with the efficacy of other antidepressants, also in a different ethnic group, we examined the real-world data from 713 patients being treated at Beijing Anding hospital. Patients were categorized into two groups based on the reduction rate of HAMD-17 scores, which was used more often in China instead of MADRS, from baseline to week 8 after antidepressant monotherapy treatment. Both MADRS and HAMD are well accepted gold standard as primary endpoints for depression studies, but it may limit the comparability across the two studies. While comparing the data within the real-word cohort at Anding hospital, the ANK3 genotype ratios between the responders and non-responders are similar (Table 1). Therefore, ANK3 biomarker does not appear to be a common pharmacogenomic biomarker for antidepressants in general (e.g., escitalopram as a SSRI or duloxetine as a SNRI), but rather a specific pharmacogenomic biomarker for predicting the treatment effect of liafensine.Fig. 4Comparisons of efficacy among ANK3 biomarker genotypes in patients treated with duloxetine or escitalopram.Mean MADRS total score change from baseline in patients with different ANK3 biomarker genotypes (AA/AG/GG) in Phase B of BMS studies. a Patients treated with duloxetine in study CN162006. b Patients treated with Dul/escitalopram in study CN162007. SE standard error.Table 1Real World Data Evaluating Outcome of Antidepressants Among Three ANK3 Genotypes.a. Major Demographic Characteristics and ANK3 Genotype Analysis Results based Respond RateVariableResponse Group1 (n = 434)Non-response Group2 (n = 279)Sample size (n)434279Mean age (years)28.95 ± 8.1627.50 ± 7.40Mean age at onset (years)24.67 ± 7.8723.80 ± 7.38Sex distributionMale: 144 (33.2%)Female: 290 (66.8%)Male: 73 (26.2%)Female: 206 (73.8%)ANK3 genotypeAA: 100 (23.0%)AG: 217 (50.0%)GG: 117 (27.0%)AA: 65 (23.3%)AG: 132 (47.3%)GG: 82 (29.4%)b. Major Demographic Characteristics and ANK3 Genotype Analysis Results based on Remission RateVariableRemission Group (n = 339)Non-remission Group (n = 374)Sample size (n)339374Mean age (years)28.81 ± 8.1827.99 ± 7.62Mean age at onset (years)24.64 ± 7.7924.04 ± 7.60Sex distributionMale: 114 (33.6%)Female: 225 (66.4%)Male: 103 (27.5%)Female: 271 (72.5%)ANK3 genotypeAA: 77 (22.7%)AG: 167 (49.3%)GG: 95 (28.0%)AA: 88 (23.5%)AG: 182 (48.7%)GG: 104 (27.8%)1, The response group: patients achieving a ≥ 50% reduction in HAMD-17 total score at week 8 compared to baselin.2, The non-response group: patients with a reduction of <50% in HAMD-17 total score at week 8 compared to baseline.1, The remission group: patients achieving HAMD-17 total score <= 7 at week 8.2, The non-remission group: patients with HAMD-17 total score > 7 at week 8. Comparisons of efficacy among ANK3 biomarker genotypes in patients treated with duloxetine or escitalopram. Mean MADRS total score change from baseline in patients with different ANK3 biomarker genotypes (AA/AG/GG) in Phase B of BMS studies. a Patients treated with duloxetine in study CN162006. b Patients treated with Dul/escitalopram in study CN162007. SE standard error. Table 1Real World Data Evaluating Outcome of Antidepressants Among Three ANK3 Genotypes.a. Major Demographic Characteristics and ANK3 Genotype Analysis Results based Respond RateVariableResponse Group1 (n = 434)Non-response Group2 (n = 279)Sample size (n)434279Mean age (years)28.95 ± 8.1627.50 ± 7.40Mean age at onset (years)24.67 ± 7.8723.80 ± 7.38Sex distributionMale: 144 (33.2%)Female: 290 (66.8%)Male: 73 (26.2%)Female: 206 (73.8%)ANK3 genotypeAA: 100 (23.0%)AG: 217 (50.0%)GG: 117 (27.0%)AA: 65 (23.3%)AG: 132 (47.3%)GG: 82 (29.4%)b. Major Demographic Characteristics and ANK3 Genotype Analysis Results based on Remission RateVariableRemission Group (n = 339)Non-remission Group (n = 374)Sample size (n)339374Mean age (years)28.81 ± 8.1827.99 ± 7.62Mean age at onset (years)24.64 ± 7.7924.04 ± 7.60Sex distributionMale: 114 (33.6%)Female: 225 (66.4%)Male: 103 (27.5%)Female: 271 (72.5%)ANK3 genotypeAA: 77 (22.7%)AG: 167 (49.3%)GG: 95 (28.0%)AA: 88 (23.5%)AG: 182 (48.7%)GG: 104 (27.8%)1, The response group: patients achieving a ≥ 50% reduction in HAMD-17 total score at week 8 compared to baselin.2, The non-response group: patients with a reduction of <50% in HAMD-17 total score at week 8 compared to baseline.1, The remission group: patients achieving HAMD-17 total score <= 7 at week 8.2, The non-remission group: patients with HAMD-17 total score > 7 at week 8. Real World Data Evaluating Outcome of Antidepressants Among Three ANK3 Genotypes. Male: 144 (33.2%) Female: 290 (66.8%) Male: 73 (26.2%) Female: 206 (73.8%) AA: 100 (23.0%) AG: 217 (50.0%) GG: 117 (27.0%) AA: 65 (23.3%) AG: 132 (47.3%) GG: 82 (29.4%) Male: 114 (33.6%) Female: 225 (66.4%) Male: 103 (27.5%) Female: 271 (72.5%) AA: 77 (22.7%) AG: 167 (49.3%) GG: 95 (28.0%) AA: 88 (23.5%) AG: 182 (48.7%) GG: 104 (27.8%) 1, The response group: patients achieving a ≥ 50% reduction in HAMD-17 total score at week 8 compared to baselin. 2, The non-response group: patients with a reduction of <50% in HAMD-17 total score at week 8 compared to baseline. 1, The remission group: patients achieving HAMD-17 total score <= 7 at week 8. 2, The non-remission group: patients with HAMD-17 total score > 7 at week 8. Discussion Precision medicine using biomarkers to tailor drug treatment to those patients who can benefit the most is revolutionizing modern drug development. Using archived clinical samples from prior studies had proved to be a fruitful way to discover novel biomarkers in oncology, such as EGR mutations for Iressa and K-Ras mutations for Erbitux [11–13]. Thus, this study aims to find a potential biomarker which could predict a subset of responders from archived clinical samples for an abandoned psychiatric drug. TRIs, including liafensine, were expected to provide new hope for the treatment of TRD, as they can simultaneously target the transporters for all three key monoamines which are believed to be critical in depression [14]. However, none of the TRIs have been approved thus far to treat depression, including liafensine. Utilizing the previous BMS Phase 2b clinical data and bio-samples, we identified a novel pharmacogenomic biomarker (ANK3 biomarker) that was highly correlated to liafensine treatment effects. Our discovery set p-value of 6.61 × 10−8 did not reach the genome‑wide significance cut-off of 5 × 10−8, which could be due to the small sample size of 186. Our study purposely set aside a cohort of patients for independent validation to assess the robustness of the associations identified in the discovery stage, thereby mitigating the risk of false positives. Notably, when we combined the discovery and validation cohorts, the association between SNP rs12217173 and DB104 response reached a p-value of 1.27×10⁻⁸, surpassing the conventional genome-wide significance threshold. Despite the fact that the patients from our discovery and validation sets consist of majority white patients (~79.8%), with approximately 16.7% of African ancestry and smaller proportions from other ethnic groups, we found ethnicity is not statistically associated with DB104 response when it was incorporated as a covariate. There is much lower frequency of ANK3-positive among Black vs White patients (~3.5% vs ~18.1%, according to 1000 Genomes Project) [15]. However, due to the small sample size of black patients in our study, it is challenging to assess the association between rs12217173 and DB104 response within this subgroup. The ANK3 biomarker is a single SNP (rs12217173) located in the ANK3 gene. ANK3 expression is mostly restricted to the nervous system and plays an important role in neuronal development, cell motility, proliferation, and signaling. ANK3 gene encoding ankyrin G, which links spectrin-actin cytoskeleton to membrane proteins including ion channels and transporters [16–19]. Since liafensine targets transporters of 5-HT, NE, and DA, there is a potential mechanistic link for the associated treatment response between liafensine and ANK3 gene. Variants in ANK3 have been linked to a broad range of neurological diseases, including depression, autism spectrum disorder, schizophrenia, and bipolar disorder [8, 20–23]. Additionally, in a genetic mouse model with conditional disruption of forebrain ANK3 pyramidal neurons, it was found that repeated social defeat stress resulted in repeated episodes of dramatic behavioral changes ranging from hyperactivity to “depression-like” behavior [24]. The ANK3 biomarker is located in an intron of the ANK3 gene and it is not surprising as 90% of genetic variants associated with complex traits lie in non-coding regions of the genome [25]. There is increasing evidence that non‑coding DNA sequences, e.g., introns, can play a major role in splicing, gene expression, mRNA transport, etc. [26–29]. It is intriguing that eQTL analysis using QTLbase shows rs12217173 is highly correlated with ANK3 expression in brain and brain-prefrontal cortex (Suppl Fig. 2a) [30]. The methylation QTL analysis from QTLbase suggests that the G allele of rs12217173 is the effective allele for ANK3 expression brain-prefrontal cortex (Suppl Fig. 2b) [30]. In addition, rs12217173 is found to be located within an enhancer which regulates ANK3 gene expression using Open Targets Platform (Suppl Fig. 2c) [31]. Thus, rs12217173 might be involved in the regulation of ANK3 gene expression in certain regions of the brain Even though the strong correlation rs12217173 with antidepressant activity of liafensine (TRI), this SNP shows little correlation with efficacy of duloxetine (SNRI) or escitalopram (SSRI) (Fig. 4). It suggests that rs12217173 may not be a pharmacogenomic biomarker for SNRI or SSRI. The real-world data from 713 patients at Beijing Anding hospital covers much broader classes of antidepressants being used except TRIs (Table 1), and there is little correlation between rs12217173 genotype with the efficacy of these antidepressants. Therefore, ANK3 biomarker appears to be a specific pharmacogenomic biomarker for liafensine, a TRI, which may hint that ANK3 biomarker may be related to the activity of dopamine transporters (DAT). Thus, it is possible that patients carrying ANK3-positive biomarker may have increased expression of ANK3 gene in certain regions of the brain, which in turn can anchor more DAT to the membrane of presynaptic neurons modulate and lead to more efficient reuptake of dopamine from the synapse (lower dopamine level is believed to be one of the key factors of depression). The increased level of DAT on neurons presents more targets for liafensine, a potent DAT inhibitor, which can effectively block the reuptake of dopamine by DAT and results in more dopamine remaining in synapse to alleviate depressive syndrome. The correlation of liafensine activity with ANK3 biomarker is mostly mediated via DAT because ANK3 biomarker status is not correlated with efficacy of duloxetine (SNRI) or escitalopram (SSRI). Despite this possible hypothesis, the precise mechanism of ANK3 biomarker affecting ANK3 gene function and subsequent modification of liafensine efficacy remains to be elucidated in future research. A number of studies have been conducted to discover predictive biomarkers for psychiatric drugs, but most of them have not been validated in a subsequent prospective study [32]. Due to the potential bias and high false positive rate of genome scale retrospective analysis, it is vital to confirm the potential biomarkers in a new clinical trial and eventually develop it as a companion diagnostic test to select drug responders, i.e., true precision medicine. Certainly, one of the limitations of the current study is that this is also a retrospective finding. In addition, the sample size is relatively small compared to most genome‑wide studies due to the limited number of samples available from the prior clinical trials. While this retrospective study is mostly translational research leading to the discovery of an intriguing novel for liafensine. In order further validate ANK3 as a pharmacogenomic biomarker, we have subsequently conducted a prospective, randomized, double‑blind global Phase 2b study to evaluate liafensine’s efficacy and safety in ANK3 ‑positive TRD patients [10]. The prospective study enrolled 189 ANK3 -positive TRD patients. The topline results from this prospective clinical trial confirmed the key findings of the current study: MADRS total score change from baseline at Week 6 in ANK3 -positive TRD patients treated with liafensine has shown a 4.4-point improvement vs the placebo, which is statistically and clinically significant, and close to the 4.7 found in this retrospective study. All of the secondary clinical depression measure outcomes also demonstrated significant improvements in ANK3 -positive patients. In addition, consistent with what we discovered in this study, the 1 mg and 2 mg liafensine arms exhibited similar efficacy, with improvement in MADRS total score changes from baseline of 4.4 and 4.5 respectively. When treatment groups are compared to each other, a 2-point difference in MADRS between groups has been found to be clinically meaningful [9, 33]. Having 4-point or higher MADRS total score improvement in TRD represents significant and meaningful clinical improvement. Among three phase 3 short term efficacy studies of esketamine in its NDA package, the MADRS improvement of esketamine arm ranges from 3.2 to 4-point MADRS improvement vs. control [34]. Liafensine as an oral, non-psychedelic drug demonstrates 4-point or higher MADRS improvement and with a great safety profile is extremely compelling to psychiatrists and patients. Excessive DAT inhibitor, such as cocaine, may lead to abuse potential. Liafensine’s DAT occupancy is only between 20–30% at therapeutical doses and lack of DA releasing activity at this range should mitigate against abuse liability effects. Liafensine was evaluated in a cocaine discrimination assay in rats, and it shows that liafensine partially substituted for the cocaine stimulus in the rat discrimination model (unpublished data). In addition, the onset of this partial effect was slower than the onset of the subjective effects of cocaine in this model. As most drugs of abuse have a rapid onset of central nervous system (CNS) effects, the data obtained with liafensine suggest that this compound will have a lower abuse potential than a potent psychomotor stimulant. A phase 1 study, CN162005, assessing the abuse potential of liafensine in healthy recreational polydrug using subjects did not demonstrate increased potential for abuse (unpublished data). This work shows that a single SNP can be effectively used as a pharmacogenomic biomarker for a psychiatric drug. ANK3 biomarker may eventually become one of the first true companion diagnostics for a psychiatric drug and bring long-awaited precision medicine to the treatment of depression. Compared to the oncology biomarkers which often need to be assayed using tumor biopsies, germline SNP can be conveniently genotyped using blood, buccal swab, or even saliva. Future implementation of biomarkers like ANK3 will have potential to transform the clinical practice of psychiatric disease management which traditionally relies on personal experience instead of patients’ genetic makeup. Supplementary information Supplementary Materials Supplementary Materials