Exploring depression treatment response by using polygenic risk scoring across diverse populations

Sandra Lapinska; Aditya Pimplaskar; Zhuozheng Shi; Yi Ding; Clara Frydman-Gani; Kangcheng Hou; Vidhya Venkateswaran; Kristin Boulier; Loes M Olde Loohuis; Bogdan Pasaniuc

doi:10.1016/j.ajhg.2025.06.003

. 2025 Jun 27;112(8):1877–1891. doi: 10.1016/j.ajhg.2025.06.003

Exploring depression treatment response by using polygenic risk scoring across diverse populations

Sandra Lapinska ^1,^∗, Aditya Pimplaskar ², Zhuozheng Shi ¹, Yi Ding ³, Clara Frydman-Gani ⁴, Kangcheng Hou ⁵, Vidhya Venkateswaran ^4,⁶, Kristin Boulier ^2,⁷, Loes M Olde Loohuis ^8,^9,¹⁰, Bogdan Pasaniuc ^4,^9,^10,^11,^12,^13,^∗∗

PMCID: PMC12362525 PMID: 40580956

Summary

Treatment-resistant depression (TRD), usually defined as limited or no response to at least two antidepressants, occurs in approximately one-third of individuals diagnosed with major depressive disorder (MDD). Studies of individuals of European ancestry highlight a genetic overlap between TRD and MDD. We analyzed two large and diverse biobanks, the UCLA ATLAS Community Health Study (ATLAS) and the All of Us Research Program (AoU), to test for associations between a polygenic score for major depression (MDD-PGS) and TRD. Compared to treatment responders, TRD individuals have higher MDD-PGS across all ancestries. MDD-PGS was significantly associated with response to selective serotonin reuptake inhibitors in individuals of European and Hispanic/Latin American genetic ancestries in both biobanks. In AoU, a decreased MDD-PGS was observed in response to tricyclics or serotonin modulators in individuals of European American ancestry and in response to serotonin and norepinephrine reuptake inhibitors in individuals of African American ancestry. ATLAS found that MDD-PGS showed lower odds of responding to atypical agents than did TRD in MDD-affected individuals belonging to the Hispanic/Latin American group, MDD-PGS was associated with atypical agents. Overall, by leveraging larger sample sizes from two diverse biobanks, we provide new insights into antidepressant response and treatment specificity for MDD in individuals of diverse genetic ancestries.

Keywords: major depressive disorder, genetics, antidepressant response, treatment-related phenotypes, electronic health records, polygenic risk scores, biobanks, diverse ancestry

Ancestrally diverse biobanks where medication, disease, and genetics are linked provide an opportunity to understand treatment response to major depressive disorder (MDD). We developed polygenic risk scores for MDD to test the hypothesis of elevated genetic risk for MDD in various response classes of MDD-affected individuals stratified across genetic ancestries.

Introduction

Major depressive disorder (MDD) is a debilitating psychiatric disorder affecting more than 185 million people globally,¹ and antidepressants are a first-line treatment.² One-third of individuals with MDD achieve complete remission after first antidepressant treatment, whereas another third fail to respond to multiple different antidepressant treatments and are often labeled as having treatment-resistant depression (TRD).³^,⁴ To improve treatment response, TRD individuals are treated with alternative strategies, such as antidepressant, antipsychotic, or lithium augmentation; psychotherapy; electroconvulsive therapy; or transcranial magnetic simulation.⁵ Twin studies have shown MDD to be heritable (h² = 30%–40%); genome-wide association studies (GWASs) estimate SNP heritability ( $h_{SNP}^{2}$ ) on the liability scale of 0.057–0.167.⁶^,⁷^,⁸ Considering the large heritability of MDD, it has been hypothesized that individual drug-specific, antidepressant-class-specific response or resistance to antidepressants might be partially influenced by genetic factors⁹; indeed, 13.2% (SE = 0.06) of variance in symptom remission can be explained by common genetic variation.⁴^,¹⁰ The largest GWAS of clinical studies to date estimated SNP heritability on the liability scale of 0.396 (SE = 0.153) for remission, further demonstrating that antidepressant response is influenced by common genetic variation.¹¹ TRD also exhibited a significant genetic basis: $h_{SNP}^{2}$ was 0.08 (SE = 0.04) when TRD-affected individuals were compared to individuals without TRD and 0.17–0.19 (SE = 0.05) for TRD-affected individuals compared to healthy controls.⁹^,¹⁰

A major limitation of current studies on response and resistance to antidepressants for MDD is that they are primarily performed in individuals of European ancestries.¹²^,¹³^,¹⁴^,¹⁵ Genetic ancestry can impact MDD and its treatment. For example, STAR^∗D studies¹^,¹⁶^,¹⁷^,¹⁸ show poorer antidepressant treatment response for African Americans than for individuals who self-reported as “white.” However, after correction for baseline differences and sociodemographic and severity metrics, these differences tend to become attenuated.¹^,¹⁶^,¹⁷^,¹⁸ Furthermore, a STAR^∗D study on citalopram response for outpatients with European, African, Asian, and Native American genetic ancestries suggested a significant effect of genetic ancestry on antidepressant treatment response.¹ In addition, a multi-ancestry GWAS of MDD from 21 cohorts on individuals of European, African, East Asian, and South Asian ancestry and on a group with self-reported Hispanic or Latin American ancestry highlighted the importance of ancestral diversity in genetic studies for transferability of findings in light of the limited degree of transferability of GWAS loci across ancestry groups.¹⁹

Inadequate sample sizes and lack of diversity among participants in current studies cause decreased generalizability of findings to a population-wide level and can lead to incorrect interpretations. A systematic review discussed five studies¹⁰^,¹³^,¹⁴^,²⁰^,²¹ that were conducted on individuals of European ancestry and that aimed to assess the association between polygenic risk scores (PGSs) for MDD and antidepressant treatment response.¹² Because each study had inadequate power, no associations survived correction for multiple testing, but each observed that a higher polygenic loading for MDD was associated with poorer treatment response or resistance. In addition, small sample sizes in genetic studies with strict clinically assessed inclusion criteria result in limited power to detect significant signals.²² For instance, Pain et al. performed a GWAS on clinically assessed antidepressant response in 5,843 individuals treated for MDD and found no significant association between response to antidepressants and a genetic liability to MDD; however, it did demonstrate that antidepressant response has common genetic variation: PGS for remission predicted antidepressant response.¹¹ Utilizing prescription data from electronic health records (EHRs) linked to genetic data provides an opportunity to increase the power needed to characterize the relationship between antidepressant response and MDD.

We analyzed data from two large and diverse biobanks, the UCLA Community Health Initiative (ATLAS)²³^,²⁴^,²⁵^,²⁶^,²⁷ and the All of Us Research Program (AoU),²⁸^,²⁹^,³⁰ to investigate the genetic overlap between MDD risk and medication response. Utilizing ancestrally diverse biobanks where medication, disease, and genetics are linked within the same medical system provides an opportunity for understanding treatment response to MDD across diverse genetically inferred ancestries. We derived a TRD phenotype using medication records across different time points in addition to treatment responder and treatment non-responder (TND) phenotypes for various antidepressant medication classes (see methods) and employed a recently developed polygenic risk score for MDD (MDD-PGS)⁸ to test the hypothesis of elevated genetic risk for MDD in various classes of MDD-affected individuals stratified across genetic ancestries. MDD-PGS had significant predictive performance for MDD in European and non-European ancestry groups except for South Asian ancestry individuals in ATLAS. We found elevated MDD-PGS in individuals of European American ancestry (both TRD and TND) as compared to responders, consistent with a hypothesis of increased genetic risk for MDD in TRD and TND cases. MDD-PGS in individuals of European American ancestry has decreased odds of responding to selective serotonin reuptake inhibitors (SSRIs) when compared to TRD (odds ratio [OR] 0.85 [0.81, 0.89], p = 8.37e−11) and SSRI non-responders in addition to tricyclics/tetracyclics (as compared to both tricyclic non-responders and TRD) and to serotonin modulator medication classes when compared to TRD. In individuals belonging to the Hispanic/Latin American group, we found that MDD-PGS is associated with response to SSRIs (as compared to both SSRI non-responders and TRD) and atypical agents (as compared to TRD). Finally, we observed that individuals of African American ancestry have lower odds of responding to serotonin norepinephrine reuptake inhibitor (SNRI) treatment when compared to treatment-resistant individuals. Overall, we show that PGSs for MDD may be a useful biomarker for identifying drivers of response to antidepressants in MDD-affected individuals.

Methods

Datasets

The UCLA ATLAS Biobank is a community health initiative in and around the greater Los Angeles area embedded within the UCLA Health medical system.²³^,²⁴^,²⁵^,²⁶^,²⁷ It includes ancestrally diverse genomic data for 54,208 individuals, 14,381 of which had a phecode for major depressive disorder and were included in this study. These genotyped participants are linked to de-identified EHRs via the UCLA Data Discovery Repository (DDR), which includes basic information, diagnosis codes, laboratory tests, medications, prescriptions, and procedures.

The AoU initiative houses one of the largest, most diverse, and broadly accessible datasets aimed to recruit participants in under-represented demographic categories in biomedical research. To date, the research hub contains genomic data for 245,394 participants, of which 42,874 had a phecode for MDD. The AoU data repository includes >412,000 EHR data as well as physical measurements, digital health technology data, and health questionnaires containing information related to the lifestyle, socioeconomic factors, environment, and biological characteristics.²⁸ Controlled tier data were aggregated to include participant demographics, medication, survey information on health insurance, and genomic data from the allele count/allele frequency (ACAF) threshold callset for this study.

Both cohorts include participants with full information for the defined outcomes and covariates stratified by European American; Hispanic or Latin American; East Asian American,;and African American ancestry groups for sufficient sample sizes.

The individual-level genotype and phenotype data of AoU is available at https://www.researchallofus.org. Due to privacy concerns, de-identified individual-level data for UCLA ATLAS is only available to UCLA researchers and can be accessed through the Discovery Data Repository Dashboard (https://it.uclahealth.org/about/ohia/ohia-products/discovery-data-repository-dashboard-0).

Quality control and imputation

Detailed information on genotype, imputation, and quality control procedures can be found in previous publications.²³^,²⁴^,²⁵^,²⁶^,²⁷ Blood samples were collected from the UCLA ATLAS Community Health Initiative and genotyped using a custom genotyping array. Array-level genotype quality control inclusion criteria included removing unmapped SNPs, strand-ambiguous SNPs, missingness per SNP >5%, filtering SNPs that do not pass the Hardy-Weinberg equilibrium (HWE) p-value cutoff of >0.001, and restricting to unrelated individuals. After completing quality control on the array-level genotype, imputation was completed against the TOPMed Freeze5 imputation panel on the Michigan Imputation Server.³¹^,³² The final set of variants was generated after filtering by R² > 0.90 and minor allele frequency (MAF) > 1%. All quality control steps were conducted using PLINK 1.9.³³

For the AoU cohort, we utilize the ACAF callset of short-read whole-genome sequence (srWGS) genotypes, which retains variants that have a population-specific allele frequency >1% or a population-specific allele count >100 in at least one ancestry subpopulation. More information on AoU’ quality control procedures and genomic data organization can be found in prior publications and on their website.³⁰

Population stratification

In UCLA ATLAS, genetically inferred ancestry membership was estimated by computing the first ten principal components for the study population using FlashPCA 2.0 software.³⁴ We then used the K-nearest neighbors algorithm on the principal components to estimate genetically inferred ancestry (GIA) membership for each individual using the continental ancestry populations from the 1000 Genomes project.³⁵ This clustered the study population into European American (EA), Hispanic or Latin American (HL), East Asian American (EAA), South Asian American (SAA), or African American (AA) ancestry groups.

AoU has made publicly available the genetically predicted ancestry for all samples with srWGS data in the v.7 data release in the AoU Control Tier.²⁹^,³⁰ In the auxiliary files, AoU provides categorical genetic ancestry that corresponds to ancestry definitions used in gnomAD, the Human Genome Diversity Project, and 1000 Genomes. In addition, this ancestry prediction table located in the auxiliary path contained the principal components of the projection for each sample. For replication of ATLAS findings, we selected individuals in EA, HL, AA, EAA, and SAA ancestry categories.

Polygenic risk scores

We compared the performance of four different PGSs. The first is generated from a large genome-wide meta-analysis of Psychiatric Genomics Consortium (PGC), UK Biobank (UKB), FinnGen, 23andMe, Million Veteran Program (MVP), and iPSYCH containing 371,184 affected individuals and 978,703 controls, which we will refer to as Als et al.⁸ The second is generated from a genome-wide meta-analysis of PGC MDD phase 2, UKB, and 23andMe containing 246,363 affected individuals and 561,190 controls, which we will refer to as Howard et al.⁷ The genome-wide studies from Als et al. and Howard et al. are all of European ancestry. The third is generated from a multi-ancestry GWAS of major depression with 88,316 affected individuals and 902,757 controls.¹⁹ This included individuals of European, African, East Asian, and South Asian ancestry and Hispanic or Latin American participants. The fourth PGS was obtained from a publicly available polygenic score that uses PRSMix, a framework that leverages the PRS corpus of a target trait to improve prediction accuracy, for the trait “major depressive disorder” from the PGS catalog (PGS004759).³⁶^,³⁷ In a secondary analysis, we generated a PGS from a GWAS of remission from 3,299 individuals classified as non-remitting and 1,852 individuals classified as remitting, which we will refer to as Pain et al.¹¹ We evaluate its performance in predicting treatment response in AoU. For the GWAS by Howard et al., Als et al., Pain et al., and Meng et al.,¹⁹ quality control inclusion criteria included SNPs with MAF > 0.01, imputation information score (INFO) > 0.8, strand-ambiguous SNPs, and duplicated SNPs. Each GWAS was overlapped with a 3 cM banded linkage disequilibrium reference panel for 1,444,196 HapMap3+ variants based on European individuals from the UK Biobank downloaded from LDpred2.³⁸

We calculated PGSs for MDD with SBayesR, a method that extends individual-level data Bayesian multiple regression models (BayesR) to adjust summary statistics from GWAS for linkage disequilibrium.³⁹ The scores for both the Howard et al. and Als et al. PGSs were originally implemented using ATLAS genotype data for fewer than 40,000 participants referred to as ATLAS 40K. To utilize the newest version of the ATLAS genotype data, we computed the PGS from each ATLAS 60K and AoU participant by multiplying the individual risk allele dosages by their corresponding weights that were provided by SBayesR on the ATLAS 40K data. For PGS004759, we multiplied the individual risk allele dosages by their corresponding weights that are provided by the PGS catalog using pgsc_calc.⁴⁰ Each PGS was mean-centered and standardized by the standard deviation within each GIA group to generate a PGS Z score.

We then tested the predictive performance of each PGS using a GIA-stratified logistic regression model adjusted for the first five principal components of the genotype matrix, age, age², insurance class, and sex at birth. The model was stratified by EA, HL, AA, EAA, and SAA GIA groups. Comparing the performance of each PGS in predicting the phecode for MDD, the Als et al. PGS was selected for two reasons: (1) the PGS was developed on the largest genome-wide meta-analysis of depression to date; and (2) the PGS had better prediction accuracy than MDD PGSs from the PGS catalog and Howard et al. GWAS.

For all subsequent analyses, we stratified by EA, HL, AA, and EAA GIA groups due to the limited sample size in the SAA GIA group for our treatment phenotypes.

Prescription data curation

Prescription data were available through the UCLA DDR and the drug-exposure table in AoU. To define all prescription-related phenotypes, we curated a list of antidepressant and antipsychotic drugs prescribed for MDD based on treatment plan recommendations from UpToDate, a system that houses evidence-based clinical resources, and from previous work on TRD.⁵^,⁹ More specifically, in “Major depressive disorder in adults: initial treatment with antidepressants” under the classes of antidepressants tab, we define our antidepressant class treatment definition using the medication listed in Tables S6 and S7. More information on these antidepressants can be found in “Major depressive disorder in adults: approach to initial management.” We pulled the second-generation antidepressants and older, first-generation antidepressants listed in these tables for everyone identified as having MDD with phecode 296.22. In addition to antidepressants, we pulled second-generation antipsychotics for each MDD-affected individual according to the guidelines outlined in “Unipolar depression in adults: treatment with second-generation antipsychotics” and “Unipolar depression in adults: choosing treatment for resistant depression.” The list of medication used to define TRD, TND, and treatment responders are in Tables S3–S5. As an individual’s treatment history, only the drugs (or medication) that are considered second-generation antidepressants, first-generation antidepressants, and second-generation antipsychotics were considered.

As suggested by UpToDate, the antidepressants belonged to one of the following antidepressant class categories: SSRIs, SNRIs, serotonin modulators (serotonin), tricyclics/tetracyclics (tricyclic), atypical agents (atypical), and monoamine oxidase inhibitors (MAOIs). Antipsychotics only included second-generation antipsychotics. For this analysis, we did not exclude medication based on dosage amounts. We also used the prescription’s generic name to pull the prescription history from UCLA DDR and the drug-exposure table in AoU for each individual classified as having major depressive disorder by phecode 296.22 and ordered them in chronological order by prescription date for each person. Prescription history will be used to define treatment outcomes for MDD-affected individuals. Following curation of MDD medication history, we totaled the number of drugs in each person’s prescription history to visualize the differences in antidepressant class distribution between TRD, TND, and treatment responders.

Phenotype definition

In both cohorts, International Classification of Diseases 9 (ICD9) and ICD10 billing codes are grouped into clinically meaningful phenotypes called phecodes. To define phecodes, we used mappings derived from version 1.2 in the PheWAS catalog.⁴¹ Cases are defined as individuals containing an ICD code tagged by the respective phecode and controls as individuals who do not have an ICD code. Individuals diagnosed with MDD were identified as case-defined individuals for phecode 296.22. As a secondary analysis, we obtained individuals whose treatment response and pathogenic mechanisms are not influenced by common MDD comorbidities by excluding those with bipolar disorder (phecode 296.1), schizophrenia (phecode 295.1), alcohol-related disorder (phecode 317), or substance-use disorder (phecode 316), which can be found in Table S11. In a subsequent analysis, we implemented an eMERGE definition for MDD to subset to individuals who had more than one instance of an MDD diagnosis. The criteria to be considered an MDD-affected individual consisted of (1) have any instance of any diagnostic code for depression, (2) fail to qualify by the 2/30/180 rule for diagnosis of depression with diagnosis, and (3) qualify by the 2/30/180 rule for diagnosis of MDD. The 2/30/180 rule states that an individual had to have two diagnosis codes, and each code had to be at least 30 but no more than 180 days apart (see Tables S17–S20 and Figure S13).

To minimize instances of pseudo-TRD (a phenomenon whereby individuals are classified as TRD due to factors other than genuine inadequate response to treatment, such as inadequate dosages, treatment length, diagnoses, or individual adherence⁴²^,⁴³), we adapted several criteria from Fabbri et al.¹⁰ to filter out certain medication switches attributable to poor compliance, insufficient treatment duration, adverse side effects, and inadequate diagnoses.

(1)
Minimal treatment duration: in our definition of inadequate response to medications that included antidepressants and antipsychotics, we considered only treatments administered for a minimum of 6 consecutive weeks to rule out treatments that were not prescribed for a period sufficient to properly take effect or were suspended due to intolerance to side effects.¹⁰^,²¹^,⁴⁴
(2)
Maximal period between different treatments: when defining valid switches between medications, we allowed a maximum interval of 14 weeks between two consecutive antidepressant and or antipsychotic treatments. This threshold was designated to filter out instances of medication switches that were less likely to represent genuine inadequate responses to a medication and more likely stemmed from poor compliance or the discontinuation of treatment for other reasons such as unknown treatments due to potential inpatient (hospitalization periods).
(3)
General individual adherence: to minimize cases of pseudo-TRD due to poor adherence, all individuals were assigned a general adherence score, as in Fabbri et al.¹⁰ To calculate this score, an adequate prescription was defined as any prescription received within 14 weeks of the previous prescription. Unlike the maximal period threshold imposed between different treatments, this definition applied to any two consecutive prescriptions, regardless of whether they pertained to the same medication or not. For each person, the adherence score was calculated as the ratio between the number of adequate prescriptions and the number of total prescriptions (minus the first one, which did not have a preceding prescription), resulting in a score ranging from 0 to 1, which was utilized as an individual-level measure for general adherence and treatment continuity. To filter out individuals likely characterized by overall poor compliance, we filtered out individuals with a general adherence score lower than 0.75.
(4)
Treatment history length: lastly, individuals were required to have a treatment history spanning at least 6 months to be categorized as suffering from TRD. This criterion was imposed to minimize the risk of considering medication switches under an incorrect diagnosis, accounting for possible diagnostic instability during the early stages of severe mental illness, and since certain antidepressants and antipsychotics may be reserved for use after initial or secondary treatments fail.⁴⁵^,⁴⁶

In both cohorts, individuals considered TRD were defined as having two or more medication switches while satisfying the above-mentioned criteria.⁴⁷

TNDs were defined as having one or more valid medication switches while satisfying all above-mentioned criteria except for general adherence. The general adherence score was excluded from TND criteria because it helped to filter out TND individuals that were falsely being considered as TRD. Once individuals were defined as having TND, they were further broken down into antidepressant class non-responder groups. Each medication in an individual’s treatment history fell into a specific antidepressant class category: SSRI, SNRI, serotonin, tricyclic, and atypical. A person was considered a non-responder to a specific antidepressant class when the class of the first prescription was different than the class of the last prescription in their treatment history. This helped us ensure that no one fell into more than one antidepressant class non-responder group.

Treatment responders were defined as individuals who undergo antidepressant monotherapy, meaning that the individual took the same antidepressant class throughout their treatment history. Prior to assessing antidepressant monotherapy, we removed individuals whose treatment history spanned less than 6 months and who had any instance of second-generation antipsychotic from consideration. Since antipsychotics are typically used in combination with antidepressants to treat psychotic depression or TRD, individuals who have a history of antipsychotic use were not considered. Subsequently, we filtered out medications administered for less than 6 consecutive weeks from each person’s treatment history for the same reasons detailed in the TRD definition.⁴⁸ Since treatment responders have only one antidepressant class in their treatment history, it was further broken down into antidepressant class responder groups based on that antidepressant class.

For each treatment-related phenotype, we ensured that all individuals had an instance of phecode 296.22.

Statistical analysis

All analysis used Python 3 or R 4.2.1 for both cohorts.

A logistic regression model stratified by GIA group was used to estimate the associations between the different treatment-related phenotype comparisons and the standardized MDD-PGS adjusting for the first five principal components of the genotype matrix, age, age², insurance class, and sex at birth. This was performed on 12-phenotypes comparison: (1) TRD vs. responder, (2) TND vs. responder, (3) SSRI responder vs. SSRI non-responder, (4) SNRI responder vs. SNRI non-responder, (5) serotonin responder vs. serotonin non-responder, (6) tricyclic responder vs. tricyclic non-responder, (7) atypical responder vs. atypical non-responder, and (8–12) models 3–7 were repeated, but the different antidepressant class non-responders were replaced with TRD. We performed 48 comparisons in total.

Insurance class is included to account for bias related to participation and access to healthcare within the de-identified EHRs.⁴⁹ In ATLAS, the insurance class variables consist of the type of insurance used by the individual, which includes federal, private, state, or none. In AoU, response to the Basics survey question “Are you covered by health insurance or some other kind of health care plan?” id used to define the insurance class, which included Private, Public, Self-pay, Yes—has insurance, None, and Unknown. The Public class includes “Medicare” and “Medicaid.” The Private class includes “Employer or Union,” “Veteran Affairs (VA),” “Other health plan,” “Tricare or other military health care,” and “Indian Health Service.” The Self-pay class includes “I don’t have health insurance, self-pay.” The Yes—has insurance class includes all individuals who have selected more than insurance class or have answered yes to being covered by insurance with no additional information on the type.

ORs were calculated within each GIA group along with p values. We applied the Bonferroni correction for multiple comparisons when analyzing the predictive performance of the standardized MDD-PGS, considering as significant a p value of <9.43e−4 (0.05/53). We derived this number from the five models that were run for each GIA group (EA, HL, AA, EAA, and SAA) when comparing MDD vs. controls, both TRD and TND to responders stratified by four GIA groups (EA, HL, AA, and EAA), and the models comparing SSRI, SNRI, atypical agents, tricyclics/tetracyclics, and serotonin modulator responders with its corresponding antidepressant class non-response as well as with TRD, stratified by the four GIA groups. In addition, we performed two types of secondary analysis presented in Tables S12–S14. First, we corrected for the number of encounters in each significant association to ensure that these associations are due to treatment outcomes rather than severity of disease. Second, we re-ran the 48 models described above using the comorbidity-excluded MDD-affected individuals.

The procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national), and proper informed consent was obtained.

Results

Phenotyping MDD treatment response

Phenotypes were defined from EHR data using mappings derived from version 1.2 in the PheWAS catalog (methods). We selected all case-defined individuals for phecode 296.22 to represent individuals diagnosed with MDD. We used EHR-derived antidepressant prescription histories as a proxy to stratify MDD cases into TRD, treatment non-responder (TND), treatment responder, and antidepressant class responder or non-response (Figure 1). MDD-affected individuals had to meet specific selection criteria such as the amount of antidepressant classes in their prescription history, the duration of each prescription, and the number of switches between different antidepressant drugs (see methods; Figures 1 and 2). Of the 54,208 (ATLAS) and 245,394 (AoU) participants with genetic data, there were 14,381 participants with the MDD phecode (296.22) in ATLAS and 42,872 in AoU (Table S23), making the prevalence of MDD 26.5% and 17.5%, respectively. Of these MDD-diagnosed individuals, 68.1% (ATLAS) and 61.6% (AoU) were assigned to EA, 14.6% (16.5%) to HL, 4.85% (19.9%) to AA, 5.12% (0.99%) to EAA, and 1.03% (0.66%) to SAA GIA groups, respectively (Table S23 and methods).

Definition of MDD, TRD, treatment response, and treatment non-response in ATLAS and AoU

Selection of individuals diagnosed with major depressive disorder in the UCLA ATLAS and All of Us biobanks. We utilize All of Us to validate our findings in ATLAS. For every individual with MDD, we pulled their antidepressant prescription history and defined them as either an individual with treatment-resistant depression (TRD), a treatment non-responder (TND), or a treatment responder. TRD, TND, and treatment responder groupings are further stratified by genetically inferred ancestry (GIA) group and antidepressant class.

Definition of treatment-related phenotypes

Prescription history was pulled for every ATLAS and AoU individual diagnosed with MDD. A prescription with a crossout (X) was observed but was excluded due to insufficient treatment lengths. For instance, individual 4 had four different antidepressant drugs, each from a different medication class, throughout this individual’s medication history, which by TRD definition (two or more antidepressant switches) would classify this individual as being treatment resistant. However, because the drug within the atypical (green) and serotonin modulator (red) classes had a treatment gap of more than 14 weeks from the next prescription, the drug was excluded from consideration. In addition, our definition of a treatment responder requires an individual to have each prescribed drug be from the same medication class. This is not the case for individual 2. However, because the drug within the MAOIs class has a prescription length of less than 6 weeks, the MAOI drug was not considered. MDD, major depressive disorder; SSRI, selective serotonin reuptake inhibitors; SNRI, serotonin-norepinephrine reuptake inhibitors; atypical, atypical agents; tricyclic, tricyclic and tetracyclic antidepressants; MAOIs, monoamine oxidase inhibitors; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7; TRD, treatment-resistant depression.

As previously reported by the NIMH, we observed an increased percentage of females (ATLAS 64.2%, AoU 67.4%) in individuals diagnosed with MDD than overall (ATLAS 55%, AoU 59.3%) as well as for all treatment-related phenotypes. For instance, TRD, defined as two or more medication switches, is composed of 69.3% and 65.6% females in ATLAS and AoU, respectively. A study estimating the prevalence of TRD using two claim databases in the United States found that females had a higher risk of TRD than males, validating our higher percentage of females in TRD.⁵⁰ Non-responders, defined as one or more medication switches, is composed of 71.5% and 70.1% females, and responders, the same antidepressant class throughout entire prescription history, have 70.3% and 69.9% females in ATLAS and AoU, respectively. 6.6% and 13.5% were classified as belonging to the EHR-derived definition of TRD, while 31.0% and 24.5% were classified as TND to one or more antidepressants in ATLAS and AoU, respectively.

MDD cases had varying encounter frequencies depending on their treatment classification. For both biobanks, MDD-affected individuals grouped into the EHR-derived proxy of TRD (median [Q1, Q3]: ATLAS 31 [10, 75], AoU 18 [6, 55]) and TND (median [Q1, Q3]: ATLAS 16 [5, 41], AoU 9 [3, 24]) had more visits than EHR-derived proxy of treatment responders (median [Q1, Q3]: ATLAS 8 [3, 24], AoU 5 [2, 12]). HL individuals had higher median and upper quartiles when compared to the other GIA groups. Median number of visits for TRD groups of HL GIA ancestry was 46 [Q1 = 11, Q3 = 99] in ATLAS and 30 [Q1 = 9, Q3 = 109.25] in AoU. In the EA GIA group, the median was 29 [Q1 = 11, Q3 = 73] in ATLAS and 16 [Q1 = 5, Q3 = 46] in AoU for TRD individuals (Figure 3 and Table S1). The most prescribed antidepressant class was SSRIs in both biobanks across each treatment outcome, with the total prescribed for all treatment phenotypes being 46,215 in ATLAS and 106,507 in AoU. Atypical agents have a higher prescription rate for treatment responders in ATLAS (N = 9,012) than in AoU (N = 6,369) (Figure 3 and Tables S3–S5).

Distribution of the number of diagnostic codes for MDD and prescriptions stratified by treatment outcome and genetically inferred ancestry

Top panels illustrate the number of antidepressant and antipsychotic prescriptions in each treatment group separated by antidepressant class and antipsychotic. Antipsychotics are not included in the definition of a treatment responder and are not reported. The frequency of MAOIs was lower than 50 prescriptions in each treatment outcome in ATLAS and lower than 330 in AoU. See Tables S3–S5 for further details of the total prescriptions as well as reporting for the drug frequency in each antidepressant class. Bottom panels illustrate the number of visits with a diagnostic code for MDD in individuals classified as treatment resistant, treatment non-responder, and treatment responder. Each treatment group was stratified by GIA. Diagnostic codes used for MDD can be found in Table S15. Lines within boxes indicate medians, boxes indicate interquartile range (IQR), whiskers denote 1.5× IQR of the upper and lower quartiles, and circles denote outliers. Values were converted to the log scale. See Table S1 for further details. SSRI, selective serotonin reuptake inhibitors; SNRI, serotonin-norepinephrine reuptake inhibitors; atypical, atypical agents; tricyclics and tetracyclics, tricyclic and tetracyclic antidepressants; MAOIs, monoamine oxidase inhibitors; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7; GIA, genetically inferred ancestry; MDD, major depressive disorder; EA, European American ancestry; AA, African American ancestry; EAA, East Asian American ancestry; HL, Hispanic and Latin American group.

Association of MDD-PGS with MDD across genetically inferred ancestry groups

First, we compared the accuracy of four different PGSs for MDD within each cohort to determine the PGS that best predicts MDD across ancestries: from a large meta-analysis on >1.3 million individuals (Als et al.-PGS)⁸; a meta-analysis of 800,000 individuals (Howard et al.-PGS)⁷; a multi-ancestry GWAS of MDD of 88,316 individuals (Meng et al.-PGS)¹⁹; and a PRSmix that combined all PGSs for MDD from PGS catalog (PGS004759).³⁶^,³⁷ Als et al.-PGS explained a greater proportion of the phenotypic variance explained on the liability scale in each GIA group compared to the other PGSs (Figure 4 and Table S7). Given the better performance of Als et al.-PGS, we focus all subsequent analyses on this score, where it will be referred to as MDD-PGS. In ATLAS, MDD-PGS significantly predicts MDD for individuals of EA (OR 1.28, confidence interval [CI] [1.25, 1.31]), HL (OR 1.18, CI [1.12, 1.24]), and EAA (OR 1.34, CI [1.23, 1.45]) ancestry groups. The MDD-PGS is nominally associated with the AA ancestry group (OR 1.17, CI [1.07, 1.28]). Within the AoU biobank, the MDD-PGS significantly associated with MDD for individuals of EA GIA (OR 1.31, CI [1.30, 1.33]), HL GIA (OR 1.23, CI [1.20, 1.27]), and AA GIA (OR 1.11, CI [1.08, 1.14]) (Figure 4 and Table S7). For the remainder of the analysis, we exclude individuals in SAA ancestry groups due to the small sample size across all treatment phenotypes (Table 1).

Associations between MDD-PGS and MDD across GIA groups within the ATLAS and AoU biobanks

Top panels show the prediction results for MDD using different PGSs. PGSs were calculated using SBayesR³⁹ from GWASs from Howard et al.,⁷ Als et al.,⁸ and Meng et al.¹⁹ We compare the performance of PGSs derived from different GWA studies with a PGS from the PGS catalog. We select PGS004759, which uses PRSmix to aggregate all PGSs for MDD from the catalog into a single score. We multiplied the individual risk allele dosages by their corresponding weights that are provided by the PGS catalog via pgsc_calc.⁴⁰ All genome-wide studies for MDD were of European ancestry, except that Meng et al. was a multi-ancestry GWAS of European, African, East Asian, and South Asian ancestry as well as Hispanic and Latin American participants. Each bar reflects the proportion of variation explained by the PGS on the scale of liability, under the assumption of a population lifetime risk of 20%. See Table S7 for further details on the performance of the different MDD-PGS. The bottom panel shows the association of MDD-PGS, constructed from the Als et al. GWAS summary statistics, with MDD in UCLA ATLAS biobank and All of Us. The odds ratios (ORs) on x axis and 95% confidence interval for PGS standardized within each GIA group. The y axis represents the genetically inferred ancestry groupings. The dashed line indicates OR = 1. In both cohorts, we see significant associations after Bonferroni correction (p = 9.61 × 10⁻⁴) represented by asterisks (^∗∗∗∗). Stars below ^∗∗∗∗ means nominal significance, and ns means not significant. ∗∗∗ indicates p < 0.001, and ∗∗ indicates p < 0.01. GIA, genetically inferred ancestry; MDD, major depressive disorder; EA, European American ancestry; AA, African American ancestry; EAA, East Asian American ancestry; HL, Hispanic/Latin American group; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7.

Table 1.

Sample size of treatment outcomes for individuals with major depressive disorder

	TRD		TND		Treatment responder
	ATLAS	AoU	ATLAS	AoU	ATLAS	AoU
n	944	5,790	4,463	10,490	2,344	5,959
Female, n (%)	654 (69.3%)	3,800 (65.6%)	3,193 (71.5%)	7,356 (70.1%)	1,648 (70.3%)	4,165 (69.9%)
Age, median [Q1, Q3]	57.0 [42.5, 67.1]	60.0 [48.0, 68.0]	55.7 [42.3, 65.9]	60.0 [48.0, 70.0]	55.0 [40.5, 65.2]	62.0 [46.0, 71.0]
GIA: European American (EA)	669	3,975	3,439	6,668	1,765	4,120
GIA: Hispanic or Latin American (HL)	150	592	628	1,413	341	750
GIA: African American (AA)	73	1,154	189	2,265	94	985
GIA: East Asian American (EAA)	44	44	168	72	129	51
GIA: South Asian American (SAS)	8	25	44	72	15	53

Open in a new tab

The sample size and demographics for each treatment-related definition are described. For additional information on the sample-size breakdown for MDD cases, controls, and total population, see Table S23.

TRD, treatment-resistant depression; TND, treatment non-responder; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7; GIA, genetically inferred ancestry.

Elevated MDD-PGS is associated with decreased treatment response for MDD

The prevalence of TRD was 6.6% and 13.5% for all MDD-diagnosed individuals in ATLAS and AoU, respectively, which is in line with previous studies⁴^,¹⁰^,²¹ confirming the validity of the TRD phenotype defined using criteria outlined in the Fabbri et al. study.³ We next focused on TRD, TND, and treatment response to find that MDD-PGS associates with TRD (OR 1.20 [1.14, 1.26], p = 1.46e−14) and TND (OR 1.11 [1.06, 1.15], p = 5.82e−7) demonstrating a significantly higher MDD-PGS than responders after Bonferroni correction. The same direction of effect was observed in ATLAS for both TRD (OR 1.16 [1.05, 1.27], p = 2.24e−3) and TND (OR 1.10 [1.04, 1.17], p = 1.33e−3), but these associations did not remain after multiple test correction. These results show that individuals with elevated genetic risk for MDD have higher odds of poorer response to antidepressant treatment (Figure 5). We observed no association between MDD-PGS and response in other GIA groupings, likely due to limited sample size (Table S9). In addition, when stratifying MDD-PGS by sex across GIA groups in both biobanks, the PGS distribution did not differ between females and males for TRD (see Figure S2).

Higher genetic predisposition to MDD increases resistance to treatments in individuals within EA ancestry in both cohorts

(A) Comparison of polygenic risk scores between MDD-affected individuals classified as treatment resistant, treatment non-responders, responders, MDD individuals, and population-based controls in both cohorts. Error bars represent the distribution of polygenic risk scores for each treatment-related phenotype. p values on the bars are derived from logistic regression model comparing (1) TRD vs. treatment responder and (2) TND and treatment responder in UCLA ATLAS and All of Us.

(B) From these two logistic regression models, ORs and 95% confidence intervals for the standardized MDD-PGS were plotted.

TRD, treatment-resistant depression; TND, treatment non-responder; MDD, major depressive disorder; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7.

As a secondary analysis, we repeated these analyses using Pain et al.-PGS¹¹ for remission to further investigate the genetics of antidepressant response. Pain et al.-PGS associates with treatment response compared to TND at nominal significance in AoU (Table S24).

Elevated MDD-PGS is associated with reduced response to SSRIs, SNRIs, atypical agents, tricyclics, and serotonin modulators

Next, we stratified the TND and treatment-response phenotypes into antidepressant class response or non-response to evaluate associations between class response with MDD-PGS across ancestry groups. In both cohorts, MAOIs were the least prescribed antidepressant class, with less than 34 prescriptions in ATLAS and less than 314 in AoU across each treatment outcome (Figure 3 and Tables S3–S5). Due to the low sample size for MAOIs, it was excluded from the antidepressant class analysis. SSRI responders had significantly lower MDD-PGS than TRD (OR 0.85 [0.81, 0.89], p = 8.37e−11) for EA ancestry groups in AoU. The same direction of effect was observed in ATLAS when compared to TRD (OR 0.81 [0.73, 0.90], p = 6.02e−5). With nominal significance, SSRI responders had lower MDD-PGS when compared to SSRI non-responders (OR 0.88 [0.79, 0.96], p = 6.96e−3) in ATLAS. We found similar but non-significant direction of effect for SSRI non-responders in AoU. Additionally, in AoU, MDD-PGS was associated with lowers odds of tricyclics response when compared to TRD (OR 0.72 [0.61, 0.85], p = 6.68e−5) and tricyclic non-responders (OR 0.74 [0.62, 0.90], p = 1.77e−3). Similar results are observed for serotonin modulators (OR 0.83 [0.71, 0.99], p = 3.54e−2) when compared to TRD. Although direction of effect was consistent in ATLAS the associations were not significant, likely due to small sample sizes (Figure 6A; Tables S9 and S10).Within the HL GIA in ATLAS, an MDD-PGS showed lower odds of responding to the following antidepressant classes: SSRIs (OR 0.76 [0.59, 0.97], p = 2.94e−2), atypical agents (OR 0.58 [0.38, 0.86], p = 7.75e−3), and serotonin modulators (OR 0.53 [0.28, 0.99], p = 4.98e−2) when compared to TRD with nominal significance. The same direction of effect was observed in AoU, albeit not nominally significant for all except SSRI responders when compared to SSRI non-responders (OR 0.76 [0.59, 0.97], p = 2.94e−2) (Figure 6B). Repeating the analysis in AA ancestry groups, we observed that SNRI responders had lower MDD-PGS than TRD (OR 0.78 [0.60, 0.94], p = 1.37e−2). The same direction of effect was observed in ATLAS, but there were no significant associations, likely due to small sample sizes (Figure 6C; Tables S9 and S10).

Higher genetic liability for MDD impacts antidepressant class-specific response in individuals of EA, HL, and AA ancestry

(A) Comparison of polygenic risk scores between MDD-affected individuals with EA ancestry classified as belonging to SSRI, tricyclic, or serotonin-modulator-antidepressant-response or non-response groups. Model comparisons were made between antidepressant class responder and non-responder groups as well as antidepressant class responder and treatment-resistant depression. Tricyclic responder/non-responder refers to MDD-affected individuals who were prescribed a tricyclic and/or tetracyclic antidepressant.

(B) Comparison of polygenic risk scores between MDD-affected individuals with HL ancestry classified as SSRI or atypical responder/non-responder in both cohorts.

(C) Comparison of polygenic risk scores between MDD-affected individuals with AA ancestry classified as SNRI responder/non-responder in both cohorts.

(D) Odds ratio for the comparisons that had significant associations and were presented in (A)–(C).

Error bars in (A)-(C) represent the distribution of polygenic risk scores for each treatment-related phenotype. p values on the bars are derived from logistic-regression models comparing antidepressant class responder and non-responder groups as well as antidepressant class responders and individuals with treatment-resistant depression.

SSRI, selective serotonin reuptake inhibitors; SNRI, serotonin-norepinephrine reuptake inhibitors; Atypical, atypical agents; Tricyclics and Tetracyclics, tricyclic and tetracyclic antidepressants; MAOIs, monoamine oxidase inhibitors; ATLAS, UCLA ATLAS Community Health Study 60K individuals; AoU, All of Us Research Program v.7; GIA, genetically inferred ancestry; MDD, major depressive disorder; EA, European American ancestry; AA, African American ancestry; HL, Hispanic and Latin American group.

We tested the robustness of results to MDD definition after removing individuals with psychiatric comorbidities. To obtain affected individuals whose treatment response and pathogenic mechanisms are not influenced by common MDD comorbidities, we excluded those with bipolar disorder, schizophrenia, alcohol-related disorder, or substance-use disorder. Significant associations remain for all above except SSRI responders vs. SSRI non-responders in EA and HL ancestry groups as well as SSRI responders and serotonin modulator responders vs. TRD in the HL ancestry group (Tables S11–S13). Additionally, for each significant association, we corrected for the number of encounters to ensure that these associations are due to treatment outcomes rather than severity of disease. Most associations remained significant except for those observed, when compared to TRD, to have poorer response to either SSRI or serotonin modulators in the HL ancestry group in ATLAS as well as serotonin modulators in the EA ancestry group in AoU (Table S14).

As a sensitivity analysis, we repeated these analyses using a more stringent definition of MDD in AoU. We implemented the eMERGE definition for MDD that required at least two diagnosis codes for MDD in AoU (see methods). MDD cases decreased by 53.9% (Table S17). MDD-PGS predicts the eMERGE-defined MDD for all ancestry individuals with similar direction of effect as the phecode-defined MDD (Table S18 and Figure S13). In addition, significant associations remain for all above except for SNRI responders vs. TRD in the AA ancestry group, but the direction of effect remains the same (Table S19). Additionally, we repeated these analyses with Pain et al.-PGS and observed no significant associations after multiple test correction (Tables S25 and S26).

Discussion

We present analyses suggesting that MDD-affected individuals with a higher genetic risk of MDD have poorer response to multiple classes of antidepressant treatment in individuals of EA, HL, and AA ancestries. EA ancestry individuals had significantly higher levels of MDD-PGS for TRD and TNDs than did treatment responders. For these individuals, a higher genetic liability for MDD decreased the response to drugs belonging to SSRI, tricyclic/tetracyclic, and serotonin modulator classes. In addition, an elevated MDD-PGS means individuals of HL ancestry have lower odds of responding to SSRI or atypical agents, and AA-ancestry individuals have lower odds of responding to SNRI treatment. Results for AA-ancestry individuals should be interpreted cautiously because the associations did not remain after repetition of the analyses with a stricter definition for MDD. Second, given an overlap in CIs for treatment-related phenotypes in AA- and HL-ancestry individuals, it is important to note that associations are not being driven by ancestry-specific loci but rather by an increase in power. Multiple studies have demonstrated that a higher polygenic loading for MDD is associated with worse treatment response or antidepressant treatment resistance in individuals of European ancestry; however, these associations did not survive multiple testing.¹²^,¹³^,²¹ Ward et al. found no significant association between MDD-PGS and response to SSRIs but reported a greater loading for MDD in individuals with less favorable response to SSRIs.¹⁴ Our analysis validates the direction of effect of treatment resistance while finding significant associations for poorer response to antidepressants across diverse populations, indicating the importance of conducting studies with larger sample sizes and diverse populations.

Few studies have had sufficient power to investigate the associations of antidepressant class response with MDD-PGS in individuals of European ancestries, and none demonstrate significant findings in diverse populations.⁹^,¹²^,¹³^,¹⁴^,²¹ By leveraging diverse biobanks, we observed associations with nominal significance between high MDD-PGS and antidepressant class response in individuals with HL and AA ancestry. For individuals of EA ancestry, we observed associations with nominal significance and ones that survived multiple-testing correction between MDD-PGS and TRD, TND, and antidepressant class response. The prevalence of MDD found in ATLAS and AoU was greater than the reported lifetime MDD prevalence of 10.8%.⁵¹ However, this lifetime prevalence was based on retrospective studies that, as a result of under-reporting and recall bias, most likely underestimated true prevalence. Prospective studies have suggested a lifetime prevalence of MDD of >30%,¹ whereas a study in EXCEED estimated it closer to 14.2%.³ Higher prevalence in these cohorts is most likely caused by enrollment of individuals with a higher burden of illness. It has been demonstrated that most diseases have a higher prevalence within AoU than in the UK Biobank.⁵² A higher burden of illness in AoU provides a possible reason for observing fewer TRD cases in ATLAS: 13.5% compared to 6.6%, respectively.

Even with current efforts to enroll a diverse group of individuals into ATLAS and AoU, the sample size remained relatively small for individuals who were diagnosed with MDD and were of AA, HL, and EAA ancestry; therefore, we had inadequate power to identify significant associations for TRD, responders, and TND. However, even with small sample sizes, we were able to observe significant, previously unreported associations in individuals of non-EA ancestry . In addition, the strict definition for treatment responder did not allow us to investigate the effects of the antidepressants not commonly prescribed at the first visit. For instance, SSRIs are commonly used as a first-line pharmacotherapy, but some individuals switch to a different antidepressant as a result of ineffective response, adverse effects, or drug-drug interactions. The treatment responder definition does not consider this case where individuals can be responders for antidepressants not prescribed at the first visit. Depression rating scales, such as the DSM-V, PHQ-9, or HAM-D, are a common method and useful tool for following an individual’s treatment trajectory to determine response or non-response to a prescribed antidepressant after the first prescription. Because of the lack of information on depression rating scales in ATLAS and AoU, we treated these individuals as being TND for their initial antidepressant. Considering that the prevalence for responders and TND was in the range of previous studies, we are confident that these cohorts are accurate representations of treatment response.

We conclude with several limitations of our work. First, we rely on EHRs to phenotype individuals for MDD and treatment, which carries bias in favor of healthcare access and can provide limitations in the interpretation of this work. We addressed this bias by incorporating insurance class into the regression models. In addition, because ICD codes are used to define phecodes, we may not be able to capture the full extent of an individual’s disease history, which suggests that the risk of having a phecode does not equate to the risk of having the disease.⁵³ Second, we were not able to account for psychotherapy or electroconvulsive therapy (ECT), which are typically prescribed in individuals with more severe forms of MDD. Typically, a person becomes eligible for ECT after TRD diagnosis; therefore, by the definition in this study, they would have been considered treatment resistant. However, inclusion of these therapies could have helped in the stratification of non-response and treatment-resistant phenotypes into augmentation categories to allow further investigation into whether adjunctive therapy indicates a higher genetic loading of MDD. Third, prescription history accounts for antipsychotics, but we did not include other psychotropic medications such as mood stabilizers, which could have provided more information for the antidepressant response definitions. Lastly, even though antidepressants might be prescribed for other conditions, such as anxiety, insomnia, or pain,⁵⁴ this study does not ensure that diagnostic codes for MDD and prescriptions were assigned on the same visit. However, two secondary analyses that (1) excluded participants with psychiatric comorbidities and (2) subset to individuals with at least two instances of MDD diagnostic codes found that most associations remain significant. In addition, an investigation into how including a 30-day prescription window from MDD diagnostic codes in our treatment responder definition impacts our findings found that associations for TRD vs. treatment responder and TND vs. treatment responders remained, thus providing reasonable confidence that using EHR-derived antidepressant prescription histories as a proxy for treatment-related phenotypes allows for the detection of meaningful associations.

Data and code availability

Code for this project including treatment-related phenotype and major depressive disorder curation is available at https://github.com/slapinska/TreatmentAssociation.

Acknowledgments

We gratefully acknowledge the Institute for Precision Health and participating individuals from the UCLA ATLAS Community Health Initiative. The UCLA ATLAS Community Health Initiative in collaboration with UCLA ATLAS Precision Health Biobank is a program of the Institute for Precision Health, which directs and supports the biobanking and genotyping of biospecimen samples from participating individuals from UCLA in collaboration with the David Geffen School of Medicine, UCLA Clinical and Translational Science Institute, and UCLA Health. The ATLAS Community Health Initiative is supported by UCLA Health, the David Geffen School of Medicine and a grant from the UCLA Clinical and Translational Science Institute (UL1TR001881). The All of Us Research Program is supported by the National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549, 1 OT2 OD026554, 1 OT2 OD026557, 1 OT2 OD026556, 1 OT2 OD026550, 1 OT2 OD 026552, 1 OT2 OD026553, 1 OT2 OD026548, 1 OT2 OD026551, and 1 OT2 OD026555; IAA#: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205 and 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277, 3 OT2 OD025315, 1 OT2 OD025337, and 1 OT2 OD025276. In addition, the All of Us Research Program would not be possible without the partnership of its participants. The procedures followed were in accordance with the ethical standards of the responsible committee, and proper informed consent was obtained.

Declaration of interests

The authors declare no competing interests.

Published: June 27, 2025

Footnotes

Supplemental information can be found online at https://doi.org/10.1016/j.ajhg.2025.06.003.

Contributor Information

Sandra Lapinska, Email: sandra.lapinska@pennmedicine.upenn.edu.

Bogdan Pasaniuc, Email: bogdan.pasaniuc@pennmedicine.upenn.edu.

Web resources

Depression phenotype, https://phekb.org/phenotype/1095

Infinium global screening array-24 kit, https://www.illumina.com/products/by-type/microarraykits/infinium-global-screening.html

International genome sample resource data, https://www.internationalgenome.org/data

Major depression: NIMH statistics, https://www.nimh.nih.gov/health/statistics/major-depression

Major depressive disorder in adults: approach to initial management, https://www.uptodate.com/contents/major-depressive-disorder-in-adults-approach-to-initial-management

Major depressive disorder in adults: initial treatment with antidepressants, https://www.uptodate.com/contents/major-depressive-disorder-in-adults-initial-treatment-with-antidepressants

Unipolar depression in adults: choosing treatment for resistant depression, https://www.uptodate.com/contents/unipolar-depression-in-adults-choosing-treatment-for-resistant-depression

Unipolar depression in adults: treatment with second-generation antipsychotics, https://www.uptodate.com/contents/unipolar-depression-in-adults-treatment-with-second-generation-antipsychotics#H2410409989

Supplemental information

Document S1. Figures S1–S13 and Tables S1–S26

mmc1.pdf^{(3.2MB, pdf)}

Document S2. Article plus supplemental information

mmc2.pdf^{(9.1MB, pdf)}

References

1.Murphy E., Hou L., Maher B.S., Woldehawariat G., Kassem L., Akula N., Laje G., McMahon F.J. Race, genetic ancestry and response to antidepressant treatment for major depression. Neuropsychopharmacology. 2013;38:2598–2606. doi: 10.1038/npp.2013.166. [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Nakano W., Mehta D., Ising M., Pfister H., Czamara D., Holsboer F., Lucae S., Erhardt-Lehmann A., Binder E.B. P.2.b.010 Genetic predictor of antidepressant response for major depressive disorder: a genome-wide association study and pathway analysis. Eur. Neuropsychopharmacol. 2013;23:S324–S325. doi: 10.1016/s0924-977x(13)70509-1. [DOI] [Google Scholar]
3.Fabbri C., Corponi F., Souery D., Kasper S., Montgomery S., Zohar J., Rujescu D., Mendlewicz J., Serretti A. The genetics of treatment-resistant depression: A critical review and future perspectives. Int. J. Neuropsychopharmacol. 2019;22:93–104. doi: 10.1093/ijnp/pyy024. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Rizvi S.J., Grima E., Tan M., Rotzinger S., Lin P., Mcintyre R.S., Kennedy S.H. Treatment-resistant depression in primary care across Canada. Can. J. Psychiatr. 2014;59:349–357. doi: 10.1177/070674371405900702. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Voineskos D., Daskalakis Z.J., Blumberger D.M. Management of treatment-resistant depression: Challenges and strategies. Neuropsychiatric Dis. Treat. 2020;16:221–234. doi: 10.2147/NDT.S198774. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Sullivan P.F., Neale M.C., Kendler K.S. Genetic epidemiology of major depression: review and meta-analysis. Am. J. Psychiatr. 2000;157:1552–1562. doi: 10.1176/appi.ajp.157.10.1552. [DOI] [PubMed] [Google Scholar]
7.Howard D.M., Adams M.J., Clarke T.-K., Hafferty J.D., Gibson J., Shirali M., Coleman J.R.I., Hagenaars S.P., Ward J., Wigmore E.M., et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nat. Neurosci. 2019;22:343–352. doi: 10.1038/s41593-018-0326-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Als T.D., Kurki M.I., Grove J., Voloudakis G., Therrien K., Tasanko E., Nielsen T.T., Naamanka J., Veerapen K., Levey D.F., et al. Depression pathophysiology, risk prediction of recurrence and comorbid psychiatric disorders using genome-wide analyses. Nat. Med. 2023;29:1832–1844. doi: 10.1038/s41591-023-02352-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Li Q.S., Tian C., Hinds D., 23andMe Research Team Genome-wide association studies of antidepressant class response and treatment-resistant depression. Transl. Psychiatry. 2020;10 doi: 10.1038/s41398-020-01035-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Fabbri C., Hagenaars S.P., John C., Williams A.T., Shrine N., Moles L., Hanscombe K.B., Serretti A., Shepherd D.J., Free R.C., et al. Genetic and clinical characteristics of treatment-resistant depression using primary care records in two UK cohorts. Mol. Psychiatr. 2021;26:3363–3373. doi: 10.1038/s41380-021-01062-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Pain O., Hodgson K., Trubetskoy V., Ripke S., Marshe V.S., Adams M.J., Byrne E.M., Campos A.I., Carrillo-Roa T., Cattaneo A., et al. Identifying the common genetic basis of antidepressant response. Biol. Psychiatry Glob. Open Sci. 2022;2:115–126. doi: 10.1016/j.bpsgos.2021.07.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Meerman J.J., Ter Hark S.E., Janzing J.G.E., Coenen M.J.H. The potential of polygenic risk scores to predict antidepressant treatment response in major depression: A systematic review. J. Affect. Disord. 2022;304:1–11. doi: 10.1016/j.jad.2022.02.015. [DOI] [PubMed] [Google Scholar]
13.Fanelli G., Benedetti F., Kasper S., Zohar J., Souery D., Montgomery S., Albani D., Forloni G., Ferentinos P., Rujescu D., et al. Higher polygenic risk scores for schizophrenia may be suggestive of treatment non-response in major depressive disorder. Prog. Neuropsychopharmacol. Biol. Psychiatry. 2021;108 doi: 10.1016/j.pnpbp.2020.110170. [DOI] [PubMed] [Google Scholar]
14.Ward J., Graham N., Strawbridge R.J., Ferguson A., Jenkins G., Chen W., Hodgson K., Frye M., Weinshilboum R., Uher R., et al. Polygenic risk scores for major depressive disorder and neuroticism as predictors of antidepressant response: Meta-analysis of three treatment cohorts. PLoS One. 2018;13 doi: 10.1371/journal.pone.0203896. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Xiong Y., Karlsson R., Song J., Kowalec K., Rück C., Sigström R., Jonsson L., Clements C.C., Andersson E., Boberg J., et al. Polygenic risk scores of lithium response and treatment resistance in major depressive disorder. Transl. Psychiatry. 2023;13 doi: 10.1038/s41398-023-02602-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Lesser I.M., Castro D.B., Gaynes B.N., Gonzalez J., Rush A.J., Alpert J.E., Trivedi M., Luther J.F., Wisniewski S.R. Ethnicity/race and outcome in the treatment of depression: results from STAR∗D. Med. Care. 2007;45:1043–1051. doi: 10.1097/MLR.0b013e3181271462. [DOI] [PubMed] [Google Scholar]
17.Lesser I.M., Myers H.F., Lin K.-M., Bingham Mira C., Joseph N.T., Olmos N.T., Schettino J., Poland R.E. Ethnic differences in antidepressant response: a prospective multi-site clinical trial. Depress. Anxiety. 2010;27:56–62. doi: 10.1002/da.20619. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Lesser I.M., Zisook S., Gaynes B.N., Wisniewski S.R., Luther J.F., Fava M., Khan A., McGrath P., Warden D., Rush A.J., Trivedi M. Effects of race and ethnicity on depression treatment outcomes: The CO-MED trial. Psychiatr. Serv. 2011;62:1167–1179. doi: 10.1176/ps.62.10.pss6210_1167. [DOI] [PubMed] [Google Scholar]
19.Meng X., Navoly G., Giannakopoulou O., Levey D.F., Koller D., Pathak G.A., Koen N., Lin K., Adams M.J., Rentería M.E., et al. Multi-ancestry genome-wide association study of major depression aids locus discovery, fine mapping, gene prioritization and causal inference. Nat. Genet. 2024;56:222–233. doi: 10.1038/s41588-023-01596-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.García-González J., Tansey K.E., Hauser J., Henigsberg N., Maier W., Mors O., Placentino A., Rietschel M., Souery D., Žagar T., et al. Pharmacogenetics of antidepressant response: A polygenic approach. Prog. Neuropsychopharmacol. Biol. Psychiatry. 2017;75:128–134. doi: 10.1016/j.pnpbp.2017.01.011. [DOI] [PubMed] [Google Scholar]
21.Wigmore E.M., Hafferty J.D., Hall L.S., Howard D.M., Clarke T.-K., Fabbri C., Lewis C.M., Uher R., Navrady L.B., Adams M.J., et al. Genome-wide association study of antidepressant treatment resistance in a population-based cohort using health service prescription data and meta-analysis with GENDEP. Pharmacogenomics J. 2020;20:329–341. doi: 10.1038/s41397-019-0067-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Lo C.W.H., Gillett A.C., Iveson M.H., Kamp M., Fabbri C., Wong W.L.E., Handley D., Pain O., Vassos E., Wray N.R., et al. Antidepressant switching as a proxy phenotype for drug non-response: investigating clinical, demographic and genetic characteristics. medRxiv. 2024 doi: 10.1101/2024.11.09.24316987. Preprint at. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Johnson R., Ding Y., Venkateswaran V., Bhattacharya A., Boulier K., Chiu A., Knyazev S., Schwarz T., Freund M., Zhan L., et al. Leveraging genomic diversity for discovery in an electronic health record linked biobank: the UCLA ATLAS Community Health Initiative. Genome Med. 2022;14 doi: 10.1186/s13073-022-01106-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Johnson R., Ding Y., Bhattacharya A., Knyazev S., Chiu A., Lajonchere C., Geschwind D.H., Pasaniuc B. The UCLA ATLAS Community Health Initiative: Promoting precision health research in a diverse biobank. Cell Genom. 2023;3 doi: 10.1016/j.xgen.2022.100243. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Chang T.S., Ding Y., Freund M.K., Johnson R., Schwarz T., Yabu J.M., Hazlett C., Chiang J.N., Wulf D.A., et al. UCLA Precision Health Data Discovery Repository Working Group Pre-existing conditions in Hispanics/Latinxs that are COVID-19 risk factors. iScience. 2021;24 doi: 10.1016/j.isci.2021.102188. [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Lajonchere C., Naeim A., Dry S., Wenger N., Elashoff D., Vangala S., Petruse A., Ariannejad M., Magyar C., Johansen L., et al. An integrated, scalable, electronic video consent process to power precision health research: Large, population-based, cohort implementation and scalability study (preprint) JMIR Preprints. 2021 doi: 10.2196/preprints.31121. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Naeim A., Dry S., Elashoff D., Xie Z., Petruse A., Magyar C., Johansen L., Werre G., Lajonchere C., Wenger N. Electronic video consent to power precision health research: A pilot cohort study. JMIR Form. Res. 2021;5 doi: 10.2196/29123. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.All of Us Research Program Investigators. Denny J.C., Rutter J.L., Goldstein D.B., Philippakis A., Smoller J.W., Jenkins G., Dishman E. The “all of us” research program. N. Engl. J. Med. 2019;381:668–676. doi: 10.1056/nejmsr1809937. [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Ramirez A.H., Gebo K.A., Harris P.A. Progress with the all of us research program: Opening access for researchers. JAMA. 2021;325:2441–2442. doi: 10.1001/jama.2021.7702. [DOI] [PubMed] [Google Scholar]
30.Bick A.G., Metcalf G.A., Mayo K.R., Lichtenstein L., Rura S., Carroll R.J., Musick A., Linder J.E., Jordan I.K., et al. All of Us Research Program Genomics Investigators Genomic data in the All of Us Research Program. Nature. 2024;627:340–346. doi: 10.1038/s41586-023-06957-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Das S., Forer L., Schönherr S., Sidore C., Locke A.E., Kwong A., Vrieze S.I., Chew E.Y., Levy S., McGue M., et al. Next-generation genotype imputation service and methods. Nat. Genet. 2016;48:1284–1287. doi: 10.1038/ng.3656. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Taliun D., Harris D.N., Kessler M.D., Carlson J., Szpiech Z.A., Torres R., Taliun S.A.G., Corvelo A., Gogarten S.M., Kang H.M., et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature. 2021;590:290–299. doi: 10.1038/s41586-021-03205-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Chang C.C., Chow C.C., Tellier L.C., Vattikuti S., Purcell S.M., Lee J.J. Second-generation PLINK: rising to the challenge of larger and richer datasets. GigaScience. 2015;4 doi: 10.1186/s13742-015-0047-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Abraham G., Qiu Y., Inouye M. FlashPCA2: principal component analysis of Biobank-scale genotype datasets. Bioinformatics. 2017;33:2776–2778. doi: 10.1093/bioinformatics/btx299. [DOI] [PubMed] [Google Scholar]
35.Auton A., Abecasis G.R., Brooks L.D., Korbel J.O., Kang H.M., Garrison E.P., Abecasis G.R., McCarthy S., McVean G.A., et al. 1000 Genomes Project Consortium A global reference for human genetic variation. Nature. 2015;526:68–74. doi: 10.1038/nature15393. [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Lambert S.A., Gil L., Jupp S., Ritchie S.C., Xu Y., Buniello A., McMahon A., Abraham G., Chapman M., Parkinson H., et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nat. Genet. 2021;53:420–425. doi: 10.1038/s41588-021-00783-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Truong B., Hull L.E., Ruan Y., Huang Q.Q., Hornsby W., Martin H., van Heel D.A., Wang Y., Martin A.R., Lee S.H., Natarajan P. Integrative polygenic risk score improves the prediction accuracy of complex traits and diseases. Cell Genom. 2024;4 doi: 10.1016/j.xgen.2024.100523. [DOI] [PMC free article] [PubMed] [Google Scholar]
38.Privé F., Albiñana C., Arbel J., Pasaniuc B., Vilhjálmsson B.J. Inferring disease architecture and predictive ability with LDpred2-auto. Am. J. Hum. Genet. 2023;110:2042–2055. doi: 10.1016/j.ajhg.2023.10.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Lloyd-Jones L.R., Zeng J., Sidorenko J., Yengo L., Moser G., Kemper K.E., Wang H., Zheng Z., Magi R., Esko T., et al. Improved polygenic prediction by Bayesian multiple regression on summary statistics. Nat. Commun. 2019;10 doi: 10.1038/s41467-019-12653-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Lambert S.A., Wingfield B., Gibson J.T., Gil L., Ramachandran S., Yvon F., Saverimuttu S., Tinsley E., Lewis E., Ritchie S.C., et al. The Polygenic Score Catalog: new functionality and tools to enable FAIR research. medRxiv. 2024 doi: 10.1101/2024.05.29.24307783. Preprint at. [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Denny J.C., Bastarache L., Ritchie M.D., Carroll R.J., Zink R., Mosley J.D., Field J.R., Pulley J.M., Ramirez A.H., Bowton E., et al. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nat. Biotechnol. 2013;31:1102–1110. doi: 10.1038/nbt.2749. [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Howes O.D., Thase M.E., Pillinger T. Treatment resistance in psychiatry: state of the art and new directions. Mol. Psychiatr. 2022;27:58–72. doi: 10.1038/s41380-021-01200-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Kornstein S.G., Schneider R.K. Clinical features of treatment-resistant depression. J. Clin. Psychiatry. 2001;62:18–25. https://pubmed.ncbi.nlm.nih.gov/11480880/ [PubMed] [Google Scholar]
44.Braund T.A., Tillman G., Palmer D.M., Gordon E., Rush A.J., Harris A.W.F. Antidepressant side effects and their impact on treatment outcome in people with major depressive disorder: an iSPOT-D report. Transl. Psychiatry. 2021;11 doi: 10.1038/s41398-021-01533-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Lage I., McCoy T.H., Jr., Perlis R.H., Doshi-Velez F. Efficiently identifying individuals at high risk for treatment resistance in major depressive disorder using electronic health records. J. Affect. Disord. 2022;306:254–259. doi: 10.1016/j.jad.2022.02.046. [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Baca-Garcia E., Perez-Rodriguez M.M., Basurte-Villamor I., Fernandez del Moral A.L., Jimenez-Arriero M.A., Gonzalez de Rivera J.L., Saiz-Ruiz J., Oquendo M.A. Diagnostic stability of psychiatric disorders in clinical practice. Br. J. Psychiatry. 2007;190:210–216. doi: 10.1192/bjp.bp.106.024026. [DOI] [PubMed] [Google Scholar]
47.Al-harbi Treatment-resistant depression: therapeutic trends, challenges, and future directions. Patient Prefer. Adherence. 2012;6:369–388. doi: 10.2147/ppa.s29716. [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Wang P., Si T. Use of antipsychotics in the treatment of depressive disorders. Shanghai Arch. Psychiatry. 2013;25:134–140. doi: 10.3969/j.issn.1002-0829.2013.03.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Institute of Medicine (US) Committee on Monitoring Access to Personal Health Care Services. Millman M. National Academies Press; 1993. A Model for Monitoring Access. [DOI] [PubMed] [Google Scholar]
50.Liu X., Mukai Y., Furtek C.I., Bortnichak E.A., Liaw K.L., Zhong W. Epidemiology of treatment-resistant depression in the United States. J. Clin. Psychiatry. 2021;83 doi: 10.4088/JCP.21m13964. [DOI] [PubMed] [Google Scholar]
51.Lim G.Y., Tam W.W., Lu Y., Ho C.S., Zhang M.W., Ho R.C. Prevalence of depression in the community from 30 countries between 1994 and 2014. Sci. Rep. 2018;8 doi: 10.1038/s41598-018-21243-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Zeng C., Schlueter D.J., Tran T.C., Babbar A., Cassini T., Bastarache L.A., Denny J.C. Comparison of phenomic profiles in the All of Us Research Program against the US general population and the UK Biobank. J. Am. Med. Inf. Assoc. 2024;31:846–854. doi: 10.1093/jamia/ocad260. [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Venkateswaran V., Boulier K., Ding Y., Johnson R., Bhattacharya A., Pasaniuc B. Polygenic scores for tobacco use provide insights into systemic health risks in a diverse EHR-linked biobank in Los Angeles. Transl. Psychiatry. 2024;14 doi: 10.1038/s41398-024-02743-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Wong J., Motulsky A., Eguale T., Buckeridge D.L., Abrahamowicz M., Tamblyn R. Treatment indications for antidepressants prescribed in primary care in Quebec, Canada, 2006-2015. JAMA. 2016;315:2230–2232. doi: 10.1001/jama.2016.3445. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Document S1. Figures S1–S13 and Tables S1–S26

mmc1.pdf^{(3.2MB, pdf)}

Document S2. Article plus supplemental information

mmc2.pdf^{(9.1MB, pdf)}

Data Availability Statement

Code for this project including treatment-related phenotype and major depressive disorder curation is available at https://github.com/slapinska/TreatmentAssociation.

[bib1] 1.Murphy E., Hou L., Maher B.S., Woldehawariat G., Kassem L., Akula N., Laje G., McMahon F.J. Race, genetic ancestry and response to antidepressant treatment for major depression. Neuropsychopharmacology. 2013;38:2598–2606. doi: 10.1038/npp.2013.166. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib2] 2.Nakano W., Mehta D., Ising M., Pfister H., Czamara D., Holsboer F., Lucae S., Erhardt-Lehmann A., Binder E.B. P.2.b.010 Genetic predictor of antidepressant response for major depressive disorder: a genome-wide association study and pathway analysis. Eur. Neuropsychopharmacol. 2013;23:S324–S325. doi: 10.1016/s0924-977x(13)70509-1. [DOI] [Google Scholar]

[bib3] 3.Fabbri C., Corponi F., Souery D., Kasper S., Montgomery S., Zohar J., Rujescu D., Mendlewicz J., Serretti A. The genetics of treatment-resistant depression: A critical review and future perspectives. Int. J. Neuropsychopharmacol. 2019;22:93–104. doi: 10.1093/ijnp/pyy024. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] 4.Rizvi S.J., Grima E., Tan M., Rotzinger S., Lin P., Mcintyre R.S., Kennedy S.H. Treatment-resistant depression in primary care across Canada. Can. J. Psychiatr. 2014;59:349–357. doi: 10.1177/070674371405900702. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] 5.Voineskos D., Daskalakis Z.J., Blumberger D.M. Management of treatment-resistant depression: Challenges and strategies. Neuropsychiatric Dis. Treat. 2020;16:221–234. doi: 10.2147/NDT.S198774. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] 6.Sullivan P.F., Neale M.C., Kendler K.S. Genetic epidemiology of major depression: review and meta-analysis. Am. J. Psychiatr. 2000;157:1552–1562. doi: 10.1176/appi.ajp.157.10.1552. [DOI] [PubMed] [Google Scholar]

[bib7] 7.Howard D.M., Adams M.J., Clarke T.-K., Hafferty J.D., Gibson J., Shirali M., Coleman J.R.I., Hagenaars S.P., Ward J., Wigmore E.M., et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nat. Neurosci. 2019;22:343–352. doi: 10.1038/s41593-018-0326-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] 8.Als T.D., Kurki M.I., Grove J., Voloudakis G., Therrien K., Tasanko E., Nielsen T.T., Naamanka J., Veerapen K., Levey D.F., et al. Depression pathophysiology, risk prediction of recurrence and comorbid psychiatric disorders using genome-wide analyses. Nat. Med. 2023;29:1832–1844. doi: 10.1038/s41591-023-02352-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib9] 9.Li Q.S., Tian C., Hinds D., 23andMe Research Team Genome-wide association studies of antidepressant class response and treatment-resistant depression. Transl. Psychiatry. 2020;10 doi: 10.1038/s41398-020-01035-6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib10] 10.Fabbri C., Hagenaars S.P., John C., Williams A.T., Shrine N., Moles L., Hanscombe K.B., Serretti A., Shepherd D.J., Free R.C., et al. Genetic and clinical characteristics of treatment-resistant depression using primary care records in two UK cohorts. Mol. Psychiatr. 2021;26:3363–3373. doi: 10.1038/s41380-021-01062-9. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] 11.Pain O., Hodgson K., Trubetskoy V., Ripke S., Marshe V.S., Adams M.J., Byrne E.M., Campos A.I., Carrillo-Roa T., Cattaneo A., et al. Identifying the common genetic basis of antidepressant response. Biol. Psychiatry Glob. Open Sci. 2022;2:115–126. doi: 10.1016/j.bpsgos.2021.07.008. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib12] 12.Meerman J.J., Ter Hark S.E., Janzing J.G.E., Coenen M.J.H. The potential of polygenic risk scores to predict antidepressant treatment response in major depression: A systematic review. J. Affect. Disord. 2022;304:1–11. doi: 10.1016/j.jad.2022.02.015. [DOI] [PubMed] [Google Scholar]

[bib13] 13.Fanelli G., Benedetti F., Kasper S., Zohar J., Souery D., Montgomery S., Albani D., Forloni G., Ferentinos P., Rujescu D., et al. Higher polygenic risk scores for schizophrenia may be suggestive of treatment non-response in major depressive disorder. Prog. Neuropsychopharmacol. Biol. Psychiatry. 2021;108 doi: 10.1016/j.pnpbp.2020.110170. [DOI] [PubMed] [Google Scholar]

[bib14] 14.Ward J., Graham N., Strawbridge R.J., Ferguson A., Jenkins G., Chen W., Hodgson K., Frye M., Weinshilboum R., Uher R., et al. Polygenic risk scores for major depressive disorder and neuroticism as predictors of antidepressant response: Meta-analysis of three treatment cohorts. PLoS One. 2018;13 doi: 10.1371/journal.pone.0203896. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] 15.Xiong Y., Karlsson R., Song J., Kowalec K., Rück C., Sigström R., Jonsson L., Clements C.C., Andersson E., Boberg J., et al. Polygenic risk scores of lithium response and treatment resistance in major depressive disorder. Transl. Psychiatry. 2023;13 doi: 10.1038/s41398-023-02602-3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib16] 16.Lesser I.M., Castro D.B., Gaynes B.N., Gonzalez J., Rush A.J., Alpert J.E., Trivedi M., Luther J.F., Wisniewski S.R. Ethnicity/race and outcome in the treatment of depression: results from STAR∗D. Med. Care. 2007;45:1043–1051. doi: 10.1097/MLR.0b013e3181271462. [DOI] [PubMed] [Google Scholar]

[bib17] 17.Lesser I.M., Myers H.F., Lin K.-M., Bingham Mira C., Joseph N.T., Olmos N.T., Schettino J., Poland R.E. Ethnic differences in antidepressant response: a prospective multi-site clinical trial. Depress. Anxiety. 2010;27:56–62. doi: 10.1002/da.20619. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib18] 18.Lesser I.M., Zisook S., Gaynes B.N., Wisniewski S.R., Luther J.F., Fava M., Khan A., McGrath P., Warden D., Rush A.J., Trivedi M. Effects of race and ethnicity on depression treatment outcomes: The CO-MED trial. Psychiatr. Serv. 2011;62:1167–1179. doi: 10.1176/ps.62.10.pss6210_1167. [DOI] [PubMed] [Google Scholar]

[bib19] 19.Meng X., Navoly G., Giannakopoulou O., Levey D.F., Koller D., Pathak G.A., Koen N., Lin K., Adams M.J., Rentería M.E., et al. Multi-ancestry genome-wide association study of major depression aids locus discovery, fine mapping, gene prioritization and causal inference. Nat. Genet. 2024;56:222–233. doi: 10.1038/s41588-023-01596-4. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib20] 20.García-González J., Tansey K.E., Hauser J., Henigsberg N., Maier W., Mors O., Placentino A., Rietschel M., Souery D., Žagar T., et al. Pharmacogenetics of antidepressant response: A polygenic approach. Prog. Neuropsychopharmacol. Biol. Psychiatry. 2017;75:128–134. doi: 10.1016/j.pnpbp.2017.01.011. [DOI] [PubMed] [Google Scholar]

[bib21] 21.Wigmore E.M., Hafferty J.D., Hall L.S., Howard D.M., Clarke T.-K., Fabbri C., Lewis C.M., Uher R., Navrady L.B., Adams M.J., et al. Genome-wide association study of antidepressant treatment resistance in a population-based cohort using health service prescription data and meta-analysis with GENDEP. Pharmacogenomics J. 2020;20:329–341. doi: 10.1038/s41397-019-0067-3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] 22.Lo C.W.H., Gillett A.C., Iveson M.H., Kamp M., Fabbri C., Wong W.L.E., Handley D., Pain O., Vassos E., Wray N.R., et al. Antidepressant switching as a proxy phenotype for drug non-response: investigating clinical, demographic and genetic characteristics. medRxiv. 2024 doi: 10.1101/2024.11.09.24316987. Preprint at. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] 23.Johnson R., Ding Y., Venkateswaran V., Bhattacharya A., Boulier K., Chiu A., Knyazev S., Schwarz T., Freund M., Zhan L., et al. Leveraging genomic diversity for discovery in an electronic health record linked biobank: the UCLA ATLAS Community Health Initiative. Genome Med. 2022;14 doi: 10.1186/s13073-022-01106-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib24] 24.Johnson R., Ding Y., Bhattacharya A., Knyazev S., Chiu A., Lajonchere C., Geschwind D.H., Pasaniuc B. The UCLA ATLAS Community Health Initiative: Promoting precision health research in a diverse biobank. Cell Genom. 2023;3 doi: 10.1016/j.xgen.2022.100243. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib25] 25.Chang T.S., Ding Y., Freund M.K., Johnson R., Schwarz T., Yabu J.M., Hazlett C., Chiang J.N., Wulf D.A., et al. UCLA Precision Health Data Discovery Repository Working Group Pre-existing conditions in Hispanics/Latinxs that are COVID-19 risk factors. iScience. 2021;24 doi: 10.1016/j.isci.2021.102188. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib26] 26.Lajonchere C., Naeim A., Dry S., Wenger N., Elashoff D., Vangala S., Petruse A., Ariannejad M., Magyar C., Johansen L., et al. An integrated, scalable, electronic video consent process to power precision health research: Large, population-based, cohort implementation and scalability study (preprint) JMIR Preprints. 2021 doi: 10.2196/preprints.31121. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib27] 27.Naeim A., Dry S., Elashoff D., Xie Z., Petruse A., Magyar C., Johansen L., Werre G., Lajonchere C., Wenger N. Electronic video consent to power precision health research: A pilot cohort study. JMIR Form. Res. 2021;5 doi: 10.2196/29123. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib28] 28.All of Us Research Program Investigators. Denny J.C., Rutter J.L., Goldstein D.B., Philippakis A., Smoller J.W., Jenkins G., Dishman E. The “all of us” research program. N. Engl. J. Med. 2019;381:668–676. doi: 10.1056/nejmsr1809937. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib29] 29.Ramirez A.H., Gebo K.A., Harris P.A. Progress with the all of us research program: Opening access for researchers. JAMA. 2021;325:2441–2442. doi: 10.1001/jama.2021.7702. [DOI] [PubMed] [Google Scholar]

[bib30] 30.Bick A.G., Metcalf G.A., Mayo K.R., Lichtenstein L., Rura S., Carroll R.J., Musick A., Linder J.E., Jordan I.K., et al. All of Us Research Program Genomics Investigators Genomic data in the All of Us Research Program. Nature. 2024;627:340–346. doi: 10.1038/s41586-023-06957-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib31] 31.Das S., Forer L., Schönherr S., Sidore C., Locke A.E., Kwong A., Vrieze S.I., Chew E.Y., Levy S., McGue M., et al. Next-generation genotype imputation service and methods. Nat. Genet. 2016;48:1284–1287. doi: 10.1038/ng.3656. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib32] 32.Taliun D., Harris D.N., Kessler M.D., Carlson J., Szpiech Z.A., Torres R., Taliun S.A.G., Corvelo A., Gogarten S.M., Kang H.M., et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature. 2021;590:290–299. doi: 10.1038/s41586-021-03205-y. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib33] 33.Chang C.C., Chow C.C., Tellier L.C., Vattikuti S., Purcell S.M., Lee J.J. Second-generation PLINK: rising to the challenge of larger and richer datasets. GigaScience. 2015;4 doi: 10.1186/s13742-015-0047-8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] 34.Abraham G., Qiu Y., Inouye M. FlashPCA2: principal component analysis of Biobank-scale genotype datasets. Bioinformatics. 2017;33:2776–2778. doi: 10.1093/bioinformatics/btx299. [DOI] [PubMed] [Google Scholar]

[bib35] 35.Auton A., Abecasis G.R., Brooks L.D., Korbel J.O., Kang H.M., Garrison E.P., Abecasis G.R., McCarthy S., McVean G.A., et al. 1000 Genomes Project Consortium A global reference for human genetic variation. Nature. 2015;526:68–74. doi: 10.1038/nature15393. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib36] 36.Lambert S.A., Gil L., Jupp S., Ritchie S.C., Xu Y., Buniello A., McMahon A., Abraham G., Chapman M., Parkinson H., et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nat. Genet. 2021;53:420–425. doi: 10.1038/s41588-021-00783-5. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib37] 37.Truong B., Hull L.E., Ruan Y., Huang Q.Q., Hornsby W., Martin H., van Heel D.A., Wang Y., Martin A.R., Lee S.H., Natarajan P. Integrative polygenic risk score improves the prediction accuracy of complex traits and diseases. Cell Genom. 2024;4 doi: 10.1016/j.xgen.2024.100523. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib38] 38.Privé F., Albiñana C., Arbel J., Pasaniuc B., Vilhjálmsson B.J. Inferring disease architecture and predictive ability with LDpred2-auto. Am. J. Hum. Genet. 2023;110:2042–2055. doi: 10.1016/j.ajhg.2023.10.010. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib39] 39.Lloyd-Jones L.R., Zeng J., Sidorenko J., Yengo L., Moser G., Kemper K.E., Wang H., Zheng Z., Magi R., Esko T., et al. Improved polygenic prediction by Bayesian multiple regression on summary statistics. Nat. Commun. 2019;10 doi: 10.1038/s41467-019-12653-0. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib40] 40.Lambert S.A., Wingfield B., Gibson J.T., Gil L., Ramachandran S., Yvon F., Saverimuttu S., Tinsley E., Lewis E., Ritchie S.C., et al. The Polygenic Score Catalog: new functionality and tools to enable FAIR research. medRxiv. 2024 doi: 10.1101/2024.05.29.24307783. Preprint at. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib41] 41.Denny J.C., Bastarache L., Ritchie M.D., Carroll R.J., Zink R., Mosley J.D., Field J.R., Pulley J.M., Ramirez A.H., Bowton E., et al. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nat. Biotechnol. 2013;31:1102–1110. doi: 10.1038/nbt.2749. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib42] 42.Howes O.D., Thase M.E., Pillinger T. Treatment resistance in psychiatry: state of the art and new directions. Mol. Psychiatr. 2022;27:58–72. doi: 10.1038/s41380-021-01200-3. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib43] 43.Kornstein S.G., Schneider R.K. Clinical features of treatment-resistant depression. J. Clin. Psychiatry. 2001;62:18–25. https://pubmed.ncbi.nlm.nih.gov/11480880/ [PubMed] [Google Scholar]

[bib44] 44.Braund T.A., Tillman G., Palmer D.M., Gordon E., Rush A.J., Harris A.W.F. Antidepressant side effects and their impact on treatment outcome in people with major depressive disorder: an iSPOT-D report. Transl. Psychiatry. 2021;11 doi: 10.1038/s41398-021-01533-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib45] 45.Lage I., McCoy T.H., Jr., Perlis R.H., Doshi-Velez F. Efficiently identifying individuals at high risk for treatment resistance in major depressive disorder using electronic health records. J. Affect. Disord. 2022;306:254–259. doi: 10.1016/j.jad.2022.02.046. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib46] 46.Baca-Garcia E., Perez-Rodriguez M.M., Basurte-Villamor I., Fernandez del Moral A.L., Jimenez-Arriero M.A., Gonzalez de Rivera J.L., Saiz-Ruiz J., Oquendo M.A. Diagnostic stability of psychiatric disorders in clinical practice. Br. J. Psychiatry. 2007;190:210–216. doi: 10.1192/bjp.bp.106.024026. [DOI] [PubMed] [Google Scholar]

[bib47] 47.Al-harbi Treatment-resistant depression: therapeutic trends, challenges, and future directions. Patient Prefer. Adherence. 2012;6:369–388. doi: 10.2147/ppa.s29716. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib48] 48.Wang P., Si T. Use of antipsychotics in the treatment of depressive disorders. Shanghai Arch. Psychiatry. 2013;25:134–140. doi: 10.3969/j.issn.1002-0829.2013.03.002. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib49] 49.Institute of Medicine (US) Committee on Monitoring Access to Personal Health Care Services. Millman M. National Academies Press; 1993. A Model for Monitoring Access. [DOI] [PubMed] [Google Scholar]

[bib50] 50.Liu X., Mukai Y., Furtek C.I., Bortnichak E.A., Liaw K.L., Zhong W. Epidemiology of treatment-resistant depression in the United States. J. Clin. Psychiatry. 2021;83 doi: 10.4088/JCP.21m13964. [DOI] [PubMed] [Google Scholar]

[bib51] 51.Lim G.Y., Tam W.W., Lu Y., Ho C.S., Zhang M.W., Ho R.C. Prevalence of depression in the community from 30 countries between 1994 and 2014. Sci. Rep. 2018;8 doi: 10.1038/s41598-018-21243-x. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib52] 52.Zeng C., Schlueter D.J., Tran T.C., Babbar A., Cassini T., Bastarache L.A., Denny J.C. Comparison of phenomic profiles in the All of Us Research Program against the US general population and the UK Biobank. J. Am. Med. Inf. Assoc. 2024;31:846–854. doi: 10.1093/jamia/ocad260. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib53] 53.Venkateswaran V., Boulier K., Ding Y., Johnson R., Bhattacharya A., Pasaniuc B. Polygenic scores for tobacco use provide insights into systemic health risks in a diverse EHR-linked biobank in Los Angeles. Transl. Psychiatry. 2024;14 doi: 10.1038/s41398-024-02743-z. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib54] 54.Wong J., Motulsky A., Eguale T., Buckeridge D.L., Abrahamowicz M., Tamblyn R. Treatment indications for antidepressants prescribed in primary care in Quebec, Canada, 2006-2015. JAMA. 2016;315:2230–2232. doi: 10.1001/jama.2016.3445. [DOI] [PubMed] [Google Scholar]

PERMALINK

Exploring depression treatment response by using polygenic risk scoring across diverse populations

Sandra Lapinska

Aditya Pimplaskar

Zhuozheng Shi

Yi Ding

Clara Frydman-Gani

Kangcheng Hou

Vidhya Venkateswaran

Kristin Boulier

Loes M Olde Loohuis

Bogdan Pasaniuc

Summary

Introduction

Methods

Datasets

Quality control and imputation

Population stratification

Polygenic risk scores

Prescription data curation

Phenotype definition

Statistical analysis

Results

Phenotyping MDD treatment response

Figure 1.

Figure 2.

Figure 3.

Association of MDD-PGS with MDD across genetically inferred ancestry groups

Figure 4.

Table 1.

Elevated MDD-PGS is associated with decreased treatment response for MDD

Figure 5.

Elevated MDD-PGS is associated with reduced response to SSRIs, SNRIs, atypical agents, tricyclics, and serotonin modulators

Figure 6.

Discussion

Data and code availability

Acknowledgments

Declaration of interests

Footnotes

Contributor Information

Web resources

Supplemental information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases