Mood disorders polygenic scores influence clinical outcomes of major psychiatric disorders

Major Depression

The first application of PGS in treatment outcome was reported in 2013 (9). The study reported a meta-analysis of three genome‐wide pharmacogenetic studies—GENDEP (N = 672), MARS (N = 604), and STARD (N = 980)—in individuals with MDD treated with various antidepressants for up to 12 weeks, and derived PGS from the GENDEP and MARS cohorts which, when applied to the STARD sample, modestly predicted antidepressant response by explaining between 0.5% and 1.2% of the variance in percentage improvement and remission rates, the very small sample size of the original samples from which PGS have been calculated may explain the low predictive power in this early study. A subsequent study (10) examined two large cohorts, NEWMEDS (N = 1791) and again STARD (N = 1107), specifically testing whether BD PGS, derived from the largest Psychiatric Genomics Consortium (PGC) BD GWAS of that time (7481 cases and 9250 controls), could predict antidepressant response in MDD; however, the analyses revealed no significant association, with BD PGS explaining less than 0.01% of variance overall for selective serotonin reuptake inhibitors (p values ranging from 0.829 to 0.934) and only slightly higher yet nonsignificant variance (0.15%–0.34%, p ranging from 0.184 to 0.999) in the norepinephrine reuptake inhibitor subgroup. Also in this case, the GWAS sample where PGS have been calculated was relatively small and much smaller than the following ones that have been used later, which led to PGS with much less predictive power (11).

A few years later, García-González et al. (12) investigated MDD PGS in antidepressant treatment response across seven pharmacogenetic studies, with primary analyses in GENDEP (n = 736) and STARD (n = 1409) and validation in five additional independent samples totaling 3756 subjects, but found that MDD PGS did not significantly predicted symptom improvement or remission, as p values remained greater than 0.1 across nine significance thresholds, though rarer variants, p < 0.0001, showed a modest trend. Other than the relatively small original GWAS sample size (13), in this and many older studies PGS were calculated without more recently optimized Bayesian tools (14) and multiple thresholds were used, moreover the target samples heterogeneity may have influenced results. Indeed, MDD is recognized to be a more heterogeneous disorder when compared to other major psychoses, as evidenced by the lower genetic heritability, and this constitutes a challenge in biological studies. In fact, in a more homogeneous study some more significant results were reported. Ward et al. (15) analyzed 760 patients with MDD from three cohorts (GENDEP, AMPS‑1, and AMPS‑2) treated with escitalopram, nortriptyline, or citalopram/escitalopram over 8–12 weeks, computing MDD PGS and for neuroticism (NEU PGS) using GWAS data from the largest studies at the time (16, 17); in this study meta‐analyses revealed for the first time some nominal associations, such as MDD PGS at p < 5 × 10⁻⁵ showing a β of −0.019 (p = 0.009) for 4‑week response and NEU PGS at p < 0.1 showing a β of −0.017 (p = 0.03) for 8‑week response; however, the variance explained remained very low (≤1.2%), and the associations did not survive stringent correction for multiple testing.

The first community sample, including a larger and more powered sample, was studied in 2020, it examined antidepressant treatment resistance using prescription data from the Generation Scotland: Scottish Family Health Study (GS:SFHS) and the GENDEP cohort, with a meta‐analysis of 4213 individuals (358 cases and 3855 controls) and a separate GWAS on stages of resistance (n = 3452) (18). PGS using summary statistics for MDD and BD revealed again nominal associations between treatment resistance and MDD PGS at thresholds of <0.1, <0.5, and <1, whereas BD PGS showed no significant relationship. This study performed in a relatively large sample therefore confirmed previous MDD PGS results.

Treatment‐resistant depression (TRD) was then analyzed in the context of esketamine response by analyzing 527 European‐ancestry individuals from two phase III trials (SUSTAIN‑2 and TRANSFORM‑3, the latter restricted to those with age of onset <55 years), where the primary outcome was the percentage change in Montgomery–Åsberg Depression Rating Scale (MADRS) score at 4 weeks (19); the GWAS identified a significant Single Nucleotide Polymorphism (SNP) in IRAK3 (rs11465988, p = 3.57 × 10⁻⁸, β = −51.6, SE = 9.2) and a gene‐level association for NME7 (p = 1.73 × 10⁻⁶), and PGS based on depressive symptoms (20) were nominally associated with esketamine response (p = 0.001, β = −3.1, SE = 0.9), whereas BD PGS (21) was not significantly associated (p = 0.076 for MADRS change, p = 0.141 for remission), expanding the potential role for depressive symptom genetic loading in antidepressant outcome also to esketamine efficacy, but not for bipolar liability. This finding is of interest given that it was the first focusing on the narrow phenotype of TRD and it could reduce the heterogeneity of antidepressant response by focusing on esketamine treatment outcome (22–25).

Shortly after, the focus shifted to late‐life antidepressant response in 335 older adults (≥60 years) with MDD treated with venlafaxine XR over 12 weeks in the IRL‑GREY (26) clinical trial (27); although the GWAS did not reveal genome‐wide significant SNPs for remission or symptom improvement, and PGS constructed for depression and Alzheimer's disease were not significantly associated with treatment response, a PGS for cardioembolic stroke was significantly linked to nonremission [OR = 0.63, 95% confidence interval (CI) = 0.48–0.83, p = 0.001, permutation p = 0.006], suggesting that vascular factors, frequently associated with depression, might play a role in antidepressant resistance among older individuals, as suggested by recent genetic correlation studies between MDD and cardiometabolic factors (28). However a later reanalysis of the same sample investigated antidepressant response in late‑life depression in 342 adults aged ≥ 60 from the same IRL‑GREY study with the same treatment, and observed that while the BD PGS was nominally associated with better remission (OR = 0.75, 95% CI = 0.58–0.97, p = 0.031) and symptom improvement (β = 4.27, SE = 2.17, p = 0.049), the MDD was not associated with treatment outcomes, though with a trend in the same direction of the other papers here reviewed (p = 0.086) and intriguingly, the ADHD PGS was nominally associated with higher odds of remission (OR = 1.36, 95% CI = 1.07–1.73, p = 0.011), contrary to a previous finding (29). Though none of these associations survived Bonferroni correction, suggesting that in late‐life depression genetic predictors of treatment response may partially differ from those in younger populations (30), trends were mostly in line with other studies. The BD and MDD PGS results difference across the two studies on the same sample could be explained by the use in the second sample of PGS calculated excluding 23andme data, that may be less powerful in explaining phenotypic variance given the self-report bias (31).

In 2021, the European Group for the Study of Resistant Depression (GSRD) sample of 1148 patients with MDD was used to assess the relationship between PGS for BD, MDD, and neuroticism (NEU) with treatment nonresponse and treatment resistance (TRD defined as failure of two or more antidepressants) (32); we found that that MDD PGS was nominally associated with non‑response (p = 0.032) in the same direction of previous studies, while BD PGS and NEU PGS did not show significant effects. In a following meta‐analysis, we examined PGS for MDD, BD, and NEU in relation to antidepressant nonresponse and nonremission in a larger sample of 3637 and 3184 patients respectively from six European clinical samples, and confirmed that the MDD PGS was nominally associated with nonresponse (OR = 1.10, 95% CI = 1.02–1.19, p = 0.013, pseudo‑R² = 0.24%) and nonremission (OR = 1.14, 95% CI = 1.04–1.24, p = 0.004, pseudo‑R² = 0.57%) at specific p‑value thresholds, while BD PGS showed no significant association; however, though none of these associations survived correction for multiple testing, the observed trends of MDD PGS effects remained in a relatively large and heterogeneous target sample (33).

A smaller, but with a more complex design study, investigated predictors of clinical outcome in 174 patients with TRD admitted to a specialist inpatient unit where patients received a multimodal treatment regimen including pharmacotherapy, cognitive behavioral therapy (CBT), behavioral activation, and, if indicated, electroconvulsive therapy (34); while clinical predictors such as later age of onset, a higher number of previous depressive episodes, and lower treatment resistance (as measured by the Maudsley Staging Method) were significantly associated with a favorable response, PGS for MDD, BD, and schizophrenia (SCZ) did not predict treatment outcome, as none of the genetic variables reached significance in either univariate or multivariate analyses, suggesting that in the context of intensive inpatient treatment for TRD, clinical factors may overshadow the modest effects of genetic liability, the relatively small target sample may also have influenced results. Indeed, variance explained in the range of 1%–2% need larger samples to be detected.

In a large population study, we analyzed TRD using primary care records from UK Biobank (n = 230,096, with MDD cases numbering 19,979 and TRD cases 2430) and the EXCEED cohort (n = 8926, with 1271 MDD cases and 159 TRD cases) (29), finding that while MDD PGS robustly predicted MDD diagnosis (p = 1.89 × 10⁻⁷¹ in UKB and p = 6.05 × 10⁻⁶ in EXCEED), PGS for BD were only weakly associated with MDD in UKB and not in EXCEED, and crucially, when comparing TRD with non‑TRD MDD, these PGS did not show significant differences after correction for multiple testing, although MDD PGS showed a nominal positive correlation (p = 0.028). This study underlines the importance of investigating large target samples that may overcome the outcome heterogeneity observed in MDD, though populations samples such as this one may add other stratification factors when compared to clinical samples.

Placebo response was examined for the first time, versus antidepressant treatment, in 1364 patients with MDD from seven randomized, double‑blind, placebo‑controlled vortioxetine trials, with an additional self‑reported validation sample from 23andMe (N = 642), constructing PGS for antidepressant response as well as for MDD and BD, NEU, subjective well‑being, and cognition (35); although no PGS reached significance after Bonferroni correction, analyses showed that higher MDD PGS was nominally linked to a better placebo response on measures of somatic anxiety (β = 0.54, p = 0.011), suggesting that genetic predisposition may influence not only drug response but also placebo effects, albeit with small effect sizes and limited clinical utility on their own. It remains to be investigated whether MDD PGS effects are limited to antidepressant treatment or have a disease course modulatory effect.

PGS for antidepressant response (PGS‑AR), computed from GWAS summary statistics (36), was then investigated with electroencephalogram (EEG) biomarker data in a sample of 1123 participants (including 1061 psychiatric patients and 62 healthy controls) to determine whether a specific EEG component (component 4) could predict treatment response in MDD (37); in men, PGS‑AR was significantly associated with EEG component 4 (β = 0.172, R² = 2.91%, p = 0.000567), and this EEG component significantly predicted symptom improvement in an independent iSPOT‑D sample (β = −0.153, R² = 2.3%, p = 0.019) as well as in a dataset of rTMS plus psychotherapy patients (β = −0.230, R² = 5.3%, p = 0.022), although these associations were not observed in women. This study investigated PGS-AR, which in theory may be more powerful than MDD PGS, but it should be considered that the origin samples to calculate PGS-AR are usually much smaller that MDD GWAS samples, therefore the power is much reduced. However, the value of this study is to be a proof‑of‑concept that combining genetic and neurophysiological markers may enhance the prediction of antidepressant outcomes.

PGS for antidepressant and lithium response were then investigated in a large sample of 4572 patients with MDD from three Swedish cohorts (PREFECT, iCBT, and STAGE) using three distinct definitions of TRD (38), and found that while the PGS-AR did not significantly differ between TRD and non‑TRD groups (e.g., broad definition: mean difference = −0.015, p = 0.631), the PGS for lithium response was significantly higher in TRD cases (e.g., broad definition: mean difference = 0.094, p = 0.003; logistic regression showed an OR of 1.12 per SD increase, 95% CI = 1.04–1.20, p = 0.003), with a dose–response effect evident in the top PGS quartile (P_trend < 0.005), suggesting that TRD may be characterized by a higher genetic predisposition to respond to lithium and emphasizing the pleiotropic effects of PGS in both MDD and BD and, from a clinical point, supporting the potential utility of a targeted lithium use in TRD.

We recently examined the interaction between PGS for mood disorders, including MDD and BD, and environmental factors in the UK Biobank (with sample sizes ranging from 33,000 to 380,000) (39), finding that while both PGS and environmental variables had additive effects on wellbeing, significant interactions emerged such that higher MDD PGS intensified the negative impact of recent stress on loneliness (β = 0.0156, SE = 0.0025, p = 3.20 × 10⁻¹⁰) and BD PGS interacted with stress to predict lower household income (p = 1.17 × 10⁻⁴), even though these interactions explained only an additional 0.01%–0.02% of variance, supporting the differential susceptibility hypothesis and suggesting that genetic liability for mood disorders can modulate the effects of environmental adversity, as it will be discussed in the conclusion section.

A secondary analysis of the Early Medication Change (EMC) trial was recently performed involving 481 patients with MDD (compared with 3215 controls from the Heinz Nixdorf Recall study) undergoing an 8‑week treatment algorithm starting with escitalopram and switching to venlafaxine or lithium (40), and although the MDD PGS was significantly associated with disorder status (Nagelkerke's R² = 2.48%, p < 1 × 10⁻¹²), it did not predict treatment outcomes such as early improvement, response, or remission (with Nagelkerke's R² values ranging from 0.007% to 0.256%); however, the relatively small sample size of the target sample may be considered as a limitation also of this study.

Following with the large series of very recent studies, Monistrol‑Mula et al. (41) explored the impact of polygenic liability to various mental disorders on COVID‑19 outcomes in 4405 individuals with a history of depression from the Australian Genetics of Depression Study (AGDS), and found that the MDD PGS was significantly associated with higher COVID‑19 burnout (β = 0.36, SE = 0.12, p = 0.003, adjusted R² = 0.089), with individuals in the top 10% having 4.17‑fold higher odds of burnout (95% CI = 1.47–11.86), while the BD PGS showed a trend toward a protective effect against burnout that did not survive multiple testing, suggesting that genetic liability for depression may predispose individuals to greater psychological distress during the pandemic through its influence on anxiety, which fully mediated the observed association.

Pregnancy and postpartum are relevant periods for depression (42) and they were the focus of another study that investigated whether PGS for MDD and BD predicted antidepressant treatment trajectories in a Danish cohort of 2316 women with mood disorders (43), but found no significant associations between these PGS and treatment trajectories (categorized as continuers, early discontinuers, late discontinuers, or interrupters), with clinical factors such as higher prepregnancy antidepressant dose, longer treatment duration, and prescription of multiple antidepressant classes being the primary predictors of continued antidepressant use, thus suggesting that, at least in the perinatal period, treatment trajectories are largely driven by clinical severity and possibly by other environmental factors rather than genetic liability.

Positive results have also been reported in another recent study aimed at validating an antidepressant response algorithm across multiple electronic health record (EHR) systems from Vanderbilt University Medical Center, the All of Us Research Program and the Mass General Brigham Healthcare System (44). It demonstrated that higher polygenic risk scores for MDD (OR = 1.07, p = 2.84 × 10⁻⁸), and BD (OR = 1.04, p = 1.99 × 10⁻³) were significantly associated with poorer antidepressant response, with an estimated heritability of antidepressant response of 3.84% (SE = 0.007) and significant genetic correlations with these psychiatric traits (rg = 0.23 for MDD, rg = 0.15 for BD), thereby supporting previous findings on MDD PGS though the overall predictive power of PGS remains modest.

Genetic analyses are mainly performed in Caucasian populations, but information on Asians is also needed, given the known differences in the genetic background (45, 46). Shao et al. (47) examined the association between MDD PGS and early antidepressant efficacy in 912 Han Chinese patients with nonpsychotic MDD (aged 18–65, with baseline HAM‑D17 ≥ 18 and medication‑free for at least 2 weeks), and reported that a higher MDD PGS was significantly associated with a lower percentage reduction in HAM‑D17 scores after 2 weeks (p = 0.009; Spearman r = −0.075, p = 0.024; in multivariate regression, β = −4.086, p = 0.039, adjusted R² = 0.086), no significant interaction with negative life events was observed, suggesting that the direction of the effect may be the same when compared to Caucasians and that the effect is quite high, this is quite encouraging for generalizability of the MDD PGS effect.

A larger longitudinal population study utilized data from the iPSYCH2015 sample in Denmark to investigate polygenic liabilities in early‑onset MDD (diagnosed between ages 10 and 25, N = 10,577) and identified four treatment trajectories over 7 years using latent class growth analysis, brief contact, prolonged initial contact, later re‑entry, and persistent contact, and found that the MDD PGS was nominally associated with the later re‑entry trajectory (OR = 1.09, 95% CI = 1.02–1.17, p = 0.01) and significantly associated with continued antidepressant treatment in primary care (OR = 1.11, 95% CI = 1.05–1.17, p = 0.0003), while BD PGS did not show significant associations, suggesting that the MDD PGS effect may be detected also in early‑onset depression and during longitudinal clinical course. This study also supports the usefulness of large population sample, that, despite the limitations inherent to registries, may well contribute to the definition of the genetic modulating effects (48).

Bipolar Disorder

MDD PGS were also studied on lithium response in patients with BD using data from the Consortium on Lithium Genetics (ConLi + Gen), comprising 2586 patients (with analyses stratified by ethnicity: multiethnic, European, and Asian subsamples) (49); employing weighted PGSs constructed from a large PGC GWAS (135,458 MDD cases and 344,901 controls), the authors found that higher MDD PGS were significantly associated with poorer lithium response, with continuous outcomes showing significant R² values of approximately 0.8% in the multiethnic sample (and similar findings in the European subsample). Stratified analyses showed that patients in the lowest quartile of MDD PGS had significantly better outcomes (e.g., OR = 1.54 in the multiethnic sample and OR = 1.75 in Europeans). This study suggested for the first time that patients with BD with lower polygenic liability for depression may represent a distinct lithium‐responsive biotype; sensitivity analyses using unrelated trait PGS (bone mineral density) confirmed the specificity of the effect.

Lithium response was again the focus in patients with BD in the same largest sample collected so far described before (N = 2283 from ConLi + Gen) and the study examined MDD PGS associated with treatment outcomes as measured by the Alda scale (50); the study found higher MDD PGS (OR = 1.61, p = 0.04) significantly associated with poorer lithium response, and that a meta‑analytic approach combining SCZ and MDD PGS into a MET2‑PGS improved prediction (OR = 2.54, p = 0.002, with Nagelkerke's R² of 0.91%), whereas BD PGS alone did not significantly predict response; functional pathway analyses of MET2‑PGS implicated histone modification and glucose metabolism pathways, suggesting epigenetic and metabolic mechanisms may underlie lithium efficacy. This result is in line with the previously reported detrimental effect of SCZ PGS (51) and suggests a synergic effect of both SCZ and MDD PGS on BD maintenance outcome.

The same ConLi + Gen sample was again analyzed with a machine‑learning approach in a subsample of 1034 patients with BD to predict lithium response using both clinical predictors and PGS for SCZ and MDD (52); the study demonstrated that while PGS alone explained only modest variance (1.2% in linear models and 2.0% in nonlinear models), combining PGS with clinical variables improved prediction to 4.7% (and up to 13.7% in PGS‑stratified models), with patients in the lowest quartile of MDD PGS being 67.7% more likely to respond to lithium than those in the highest quartile (OR = 1.68, 95% CI = 1.14–2.47, p = 0.009). This study clearly underlines the benefit of a combined model with the perspective of a potential clinical utility of integrating clinical and genetic risk information for personalized treatment stratification.

Lithium pharmacogenetics in BD was also investigated by developing a lithium response polygenic score (Li + PGS) in the mentioned large ConLi + Gen cohort (N = 2367) and replicating the findings in PsyCourse (N = 89) and BipoLife (N = 102) cohorts (53); lithium response, measured via the ALDA scale (both continuously and categorically with a cutoff of 7), was significantly predicted by Li + PGS (categorical outcome: p = 9.8 × 10⁻¹², R² = 1.9%; continuous outcome: p = 6.4 × 10⁻⁹, R² = 2.6%), with patients in the highest Li + PGS decile having 3.47‑fold higher odds of a favorable response (95% CI = 2.22–5.47), and gene‑based pathway analysis implicated cholinergic and glutamatergic systems, thereby reinforcing the notion that lithium responders may have a distinct genetic profile in addition to the previously reported lower polygenic liability for depression.

Collectively, these studies (Table 1), spanning from 2013 to 2025 and incorporating diverse samples, from large-scale meta-analyses and clinical trials to population-based and EHR studies, strongly suggest that MDD PGS show modest but consistent associations with both disorder severity susceptibility and treatment outcomes (with higher MDD polygenic load generally predicting poorer antidepressant or lithium response). BD PGS have shown a less robust predictive value for antidepressant response in MDD, but they may play a role in influencing lithium response in BD when considered in conjunction with other PGS. In any case the overall variance explained by these genetic predictors remains low, underscoring the complexity of treatment response phenotypes and the need for integrating genetic information with clinical, environmental, and other biological markers to enhance predictive accuracy.

Outcome-Related Traits

A series of papers investigated clinical aspects that, though not directly measuring short-term treatment outcome, may inform about the possible influence of PGS. Kowalec et al. (54) analyzed 24,706 individuals with SCZ from the Swedish national registers, with a genomic subset of 4936 cases, to investigate clinical, demographic, and genetic factors associated with treatment‐resistant schizophrenia (TRS), defined both by clozapine prescription (N = 4813) and by clozapine prescription or antipsychotic polypharmacy for ≥ 90 days (N = 13,779); although they found that a higher SCZ family history burden [highest quartile vs. lowest quartile: OR = 1.31, 95% CI = (1.19–1.42), p = 4.8 × 10⁻⁸] and lower premorbid IQ in males (per 1 SD decrease: OR = 0.94, 95% CI = [0.90–0.98], p = 0.002) were robust predictors of TRS, none of the PGS, MDD or BD reached significance (p > 0.1 for both), though both PGS showed a nonsignificant trend in the direction of increasing TRS.

A smaller but well-designed study conducted an integrative genomic–epigenomic analysis in 44 patients with refractory psychosis treated with clozapine [31 with SCZ (70.45%), 9 with schizoaffective disorder (20.45%), and 4 with BD (9.09%)], computing PGS for BD and MDD, and found that BD PGS was significantly associated with clozapine metabolic ratio (pseudo-R² = 0.2080, p = 0.0008, adjusted p = 0.0189), while MDD PGS was only nominally associated with clozapine dose (pseudo-R² = 0.386, p = 0.0035, adjusted p = 0.0759), suggesting that bipolar genetic liability could influence clozapine metabolism, and, probably more interesting, that MDD genetic risk may lead treating clinicians to raise the dose possibly for an observed poor response (55).

A more recent study focused more directly on the interplay of MD and BD PGS with clinical and environmental factors (56). The study examined data from the AGDS (N = 14,146; 75% female, mean age = 44.0 years) to assess associations between PGSs for multiple mental disorders, MDD and BD, and exposure to 32 stressful life events (SLEs) categorized by childhood, past-year, lifetime, and cumulative events. Using logistic and linear regression models adjusted for age and sex with false discovery rate (FDR) correction, they found that higher MDD PGS was significantly associated with increased odds of exposure to all childhood SLEs (ORs = 1.07–1.12, p’s < 0.013, FDR-corrected), as well as with specific adverse events such as physical assault [OR = 1.06, 95% CI = (1.02–1.11), p = 0.006], unwanted or uncomfortable sexual experiences, sexual assault [OR = 1.10, 95% CI = (1.05–1.16), p < 0.001], severe human suffering [OR = 1.17, 95% CI = (1.05–1.30), p = 0.003], life-threatening illness or injury [OR = 1.09, 95% CI = (1.03–1.15), p = 0.003], and assault with a weapon [OR = 1.12, 95% CI = (1.04–1.21), p = 0.003]; additionally, higher MDD PGS was associated with increased cumulative SLEs (ORs = 1.05–1.24, FDR-corrected p’s < 0.05), whereas higher BD PGS was associated with lower odds of experiencing physical assault [OR = 0.95, 95% CI = (0.91–0.99), p = 0.014], major financial troubles [OR = 0.93, 95% CI = (0.88–0.98), p = 0.004], and living in unpleasant surroundings [OR = 0.92, 95% CI = (0.87–0.98), p = 0.008], as well as with fewer reported childhood SLEs [OR = 0.97, 95% CI = (0.95–0.99), p = 0.01]. Results from this complex study may suggest that, while genetic liability for depression may predispose individuals to greater exposure to stress, BD genetic risk appears inversely related to such exposure. The large sample and the evaluation of stressful life events are positive aspects of the study that add to the overall outcome domain and are in line with previous evidence therefore starting to identify specific modulating effects of MDD versus BD PGS.

A converging evidence comes from another study (57), that used data from two population-based cohorts, the Avon Longitudinal Study of Parents and Children (ALSPAC; N = 5521, mean age = 11.8 years, SD = 0.14, 50.3% female) and the Twins Early Development Study (TEDS; N = 4625, mean age = 11.27 years, SD = 0.69, 53.2% female), to compute MDD PGS among other traits and to examine their associations with psychopathology symptoms measured by the Short Mood and Feelings Questionnaire (SMFQ) and the Strength and Difficulties Questionnaire (SDQ); the study found that the depression PGS was significantly associated with the symptom “not enjoying anything” (r = 0.04) and with “being bullied” (r = 0.06) on the peer problems subscale, supporting the evidence that genetic risk for depression may be broadly influencing the complex interplay with environmental stressors and possibly reducing the heterogeneity of treatment outcome (58).

Another interesting possible effect, diagnostic transition, was recently investigated (59). The study included 10,565 individuals from a Danish registry with eating disorders (Anorexia nervosa [AN], n = 6901; Bulimia nervosa [BN], n = 1417; Eating Disorder Not Otherwise Specified [EDNOS], n = 2247) and calculated PGS for 422 traits including MDD and BD using LDpred2 and meta-PGS approaches; the study found that a higher PGS for MDD was significantly associated with a 15% greater hazard of transitioning from anorexia nervosa to either bulimia nervosa or EDNOS (HR = 1.15 per SD increase, p < 1.57 × 10⁻⁴), whereas the BD PGS was not significantly associated with diagnostic transitions. Though not directly investigating outcome, results from this study are interesting because report for the first time that that depression genetic liability may play a role also in the clinical evolution of eating disorders.

Nguyen et al. (60) then used Swedish and Danish national registries to study psychotic MDD, defined using ICD-10 subcodes F32.2/F32.3 and analyzing approximately 30,000 genotyped MDD cases from the UK Biobank and a Swedish clinical cohort. It reported that the heritability of psychotic MDD was estimated at 30.17% (95% CI = 23.53–36.80), with individuals with psychotic MDD having higher mean BD PGS [OR = 1.28, 95% CI = (1.20–1.36)], while the MDD PGS was associated with lower odds of psychotic MDD [OR = 0.93, 95% CI = (0.88–0.99)], this study is of particular interest given that it may suggest that, as expected, BD PGS may influence the risk of psychotic behavior, given its higher correlation with SCZ PGS (61, 62), while PGS MDD could define a less severe depressive subtype (31).

On the other hand, in patients with SCZ MDD PGS may have a similar effect. In fact, a study of first-episode psychosis cases examined 583 individuals from the EU-GEI study and derived transdiagnostic symptom dimensions via a bifactor model from measures such as the Positive and Negative Syndrome Scale (PANSS), finding that MDD PGS was significantly associated with lower positive [β = −0.48, 95% CI = (−0.765, −0.200), p = 0.002] and negative symptom scores [β = −0.48, 95% CI = (−0.754, −0.199), p = 0.002] but that it interacted with childhood trauma [as measured by the CTQ (Childhood Trauma Questionnaire)] to amplify positive symptoms [interaction β = 0.42, 95% CI = (0.155–0.682), p = 0.004], while BD PGS showed a trend toward association with lower positive symptoms [β = −0.49, 95% CI = (−0.875, −0.102), p = 0.021] and a significant interaction with childhood trauma on positive symptoms [β = 0.45, 95% CI = (0.106–0.798), p = 0.010], suggesting that genetic liability for mood disorders can influence SCZ symptomatology in a similar direction to the one observed in MDD and possibly also modulate the impact of adverse early-life experiences on psychosis symptomatology (63).

A converging evidence comes from a population study. In the Norwegian MoBa cohort, Bakken et al. (64) assessed 54,839 children at ages 1.5, 3, 5, and 8 years using latent growth models and latent profile analysis to characterize trajectories of emotional and behavioral difficulties, finding that the PGS for depression was significantly associated with higher baseline emotional difficulties [β = 0.029, 95% CI = (0.018–0.041), p < 0.001] and with a steeper increase in behavioral difficulties [β = 0.041, 95% CI = (0.024–0.058), p < 0.001], whereas the BD PGS was not significantly associated with overall trajectories but was specifically associated with a latent profile characterized by severe behavioral dysregulation [OR = 1.52, 95% CI = (1.21–1.90), p = 0.001]. This large and well powered study adds to the potentially broad effect of MDD PGS that may apply also to subjects not affected by MDD or other major psychoses.

The complex interplay with genetic liability and trauma was very recently investigated in a large study including 96,002 individuals from hospital-linked biobanks at VUMC and MGB to investigate the interaction between sexual trauma and PGSs (65). The results suggest that in individuals without sexual trauma, BD PGS was significantly associated with BD diagnosis [OR = 1.36, 95% CI = (1.31–1.42), p < 0.002] and MDD PGS with MDD diagnosis [OR = 1.20, 95% CI = (1.17–1.22), p < 0.002], while in those with documented sexual trauma, the association for BD PGS was attenuated [OR = 1.11, 95% CI = (0.99–1.23), p = 0.072] yet the MDD PGS association remained robust [OR = 1.21, 95% CI = (1.08–1.37), p < 0.002], suggesting that severe trauma may diminish the predictive power of bipolar genetic liability but not the one of depressive genetic liability, which may have a synergic contribution with trauma.

In Taiwan, Wu et al. (66) used data from 106,806 participants to examine associations between BD PGS with educational attainment and cognitive aging (assessed via the mini–mental state examination (MMSE) in 27,005 individuals aged ≥ 60 with longitudinal data from 6194 participants over a mean follow-up of 3.9 years), and found that BD PGS was significantly associated with higher educational attainment (OR = 1.021 per SD increase, p = 0.001) and that its concordant variants explained 0.48% of variance (vs. 0.39% overall), and in terms of cognitive aging, BD PGS was associated with better cognitive performance (β = 0.054, p = 0.020). Indeed, this study supports the complex effect of BD PGS, an effect that is not unequivocally detrimental, possibly depending on other clinical and environmental factors.

Similarly, Jiang et al. (67) investigated cardiovascular disease risk in 345,169 European-ancestry individuals from the UK Biobank and found that each 1-SD increase in MDD PGS was significantly associated with increased risk of atrial fibrillation [HR = 1.04, 95% CI = (1.02–1.06), p = 1.5 × 10⁻⁴], coronary artery disease [HR = 1.07, 95% CI = (1.04–1.11), p = 2.6 × 10⁻⁶], and heart failure [HR = 1.09, 95% CI = (1.06–1.13), p = 9.7 × 10⁻¹⁰] in females, whereas BD PGS showed no significant associations. This finding supports the previously discussed correlation between MDD and cardiovascular diseases, that is an area of relevant clinical and research interest.

Another piece of evidence comes from the study by Scott et al. (68) which examined 1473 individuals aged 15–25 from the Brisbane Longitudinal Twin Study and, through principal component analysis of four PGSs (for MDD, BD, SCZ, and NEU), derived a BD-SCZ dimension (explaining 35.7% of variance) and an MDD-NEU dimension (34.2% variance), finding that the BD-SCZ dimension was significantly higher in individuals meeting Composite International Diagnostic Interview (CIDI) criteria for a full-threshold mood or psychotic disorder (p = 0.005) and was significantly associated with help-seeking behavior (p = 0.02), while the MDD-NEU dimension was only associated with help-seeking (p = 0.003). One interesting aspect of this study is the further evidence of a similarity between BD and SCZ liability as well as between MDD and NEU liability while they are quite independent from each other.

An onset focused study compared 207 older adults with BD from the PsyCourse Study, distinguishing 144 early-onset BD cases (onset <50 years) from 63 late-onset cases (onset ≥ 50 years) (69), and found that BD PGS was significantly higher in early-onset BD (p = 0.005), explaining a quite relevant 6.0% of the variance (Nagelkerke's pseudo-R² = 6.0%), whereas MDD PGS (p = 0.66) were not associated with age of onset. Also in this case, the small sample size may not have been powered for the low effect sizes observed in other studies.

A similar lack of effect was reported by another study, this time on neuropsychological measures (70). The study included a network analysis in 132 first-episode psychosis patients, assessing cognitive functioning and psychopathology at 2 months and 2 years, and found that no mood PGS were significantly associated with cognitive domains, but, again, the small sample size and the complex network analysis could suggest power issues.

A much larger study reported interesting findings on symptomatology (71). The study analyzed UK Biobank data from 409,630 participants for chronotype and 239,918 for insomnia, reporting that BD PGS (p = 4.8 × 10⁻³) and MDD PGS (p = 8.07 × 10⁻⁴) were both significantly associated with an evening chronotype, and that both BD PGS (p = 2.9 × 10⁻⁷) and MDD PGS (p < 2.2 × 10⁻¹⁶) were significantly associated with insomnia, suggesting that genetic liability for mood disorders contributes to symptomatology heterogeneity and circadian dysregulation. This finding is of relevant clinical interest given the known impact of circadian dysregulation in mood disorders outcome (72–75).

Harrington et al. (76) investigated peripartum depression in 178 parous female inpatients from an Italian sample, dividing them into MDD (n = 72) and BD (n = 106) subgroups and applying a multipolygenic risk framework with 341 PGSs, finding that both MDD and BD PGS were negatively associated with peripartum depression, though not consistently in the two subgroups. However, the relatively small sample size and the many comparisons in the study suggest caution in interpreting findings of this study.

Suicidality is another potentially interesting phenotype and it was the focus of another very recent study in 232 youth (mean age 16.7 years) with BD (n = 125) or at high risk for BD (n = 107) in Canada (77). Results suggest that MDD PGS was nominally associated with suicidal ideation (β = 0.36, SE = 0.16, p = 0.017; remaining significant when controlling for family history (β = 0.37, SE = 0.15, p = 0.016), whereas BD PGS did not significantly predict any suicidality outcomes, again suggesting a different effect of MDD genetic liability versus BD liability also on suicidal behaviors, an area that should be further investigated for its potential clinical benefit.

The different effect may reflect on the potential diagnostic power of the two PGS. In a very large study Panagiotaropoulou et al. (78) analyzed 51,149 individuals (15,532 BD cases, 12,920 MDD cases, and 22,697 controls) from the Psychiatric Genomics Consortium with replication in an independent iPSYCH cohort (n = 25,966, including 2524 BD and 23,442 MDD cases) to differentiate BD from MDD using genome-wide association analyses and PGS calculated with SBayesR, finding that BD PGS significantly differentiated BD from MDD (AUC = 0.62, Nagelkerke R² = 2.29%) and that combining BD PGS, MDD PGS, and a BD versus MDD GWAS-based PGS improved classification (AUC = 0.64, R² = 4.56%), with MDD PGS alone contributing little, thereby reinforcing that BD and MDD PGS, though correlated, have different effects. Though focusing only on diagnostic status and not on outcome, the study is interesting in further underlining the specific effects of MDD versus BD PGS.

In a similar study, Chen et al. (79) applied deep learning algorithms to genetic data from multiple large datasets (including MGS, SCCSS, CATIE, PGC, WTCCC, among others) to classify SCZ, BD, and MDD based on PGSs for 42 comorbid traits, and reported that for BD classification, the target BD PGS achieved an accuracy of 0.895 ± 0.020 and an AUC of 0.965 ± 0.003, while for MDD classification, the target MDD PGS achieved an AUC of 0.854 ± 0.010, with performance improving when PGSs for comorbid traits were added, suggesting that although disorder-specific polygenic risk is informative, genetic overlap with comorbid traits can further enhance diagnostic classification and possibly outcome.

Following on the possible pleiotropic effects of PGS, Segura et al. (80) examined the impact of PGSs on antipsychotic-induced metabolic dysregulation in a longitudinal study of 231 first-episode psychosis (FEP) patients over 6 months and found that MDD PGS, but not BD PGS were associated with total cholesterol levels (FDR = 0.006) and also at month 2 (FDR = 0.030). Though interesting, and in line with previous evidence of a correlation between metabolic and depressive backgrounds (81), also in this study the relatively small sample size suggests caution.

The complex interplay with the environment was further investigated in a study involving 573 FEP cases and 1005 controls from the EU-GEI study to investigate PGSs and environmental risk interactions (82). Results suggest that for affective psychosis, BD PGS was the strongest genetic predictor [OR = 1.50, 95% CI = (1.18–1.91), p = 0.001] and MDD PGS was also significant but with a smaller effect [OR = 1.34, 95% CI = (1.10–1.63), p = 0.004], though not interacting with environment. Therefore, whereas SCZ-spectrum disorder was primarily driven by SCZ PGS, affective psychosis may be influenced from a combination of mood disorder genetic liability and environmental factors.

Finally, Song et al. (83) examined 5180 BD cases from Sweden and 2577 BD cases from the UK to assess associations between BD PGS and MDD PGS with BD subphenotypes, finding that BD PGS was positively associated with full interepisode remission [OR = 1.16, 95% CI = (1.10–1.23), p = 1.05 × 10⁻⁷] and with higher Global Assessment of Functioning (GAF)-function scores [β = 0.78, 95% CI = (0.38–1.17), p = 1.06 × 10⁻⁴], and was negatively associated with comorbid anxiety disorders [OR = 0.88, 95% CI = (0.83–0.93), p = 1.60 × 10⁻⁵], whereas MDD PGS was negatively associated with remission [OR = 0.84, 95% CI = (0.80–0.89), p = 2.78 × 10⁻¹¹] and GAF-function [β = −0.70, 95% CI = (−1.00 to −0.40), p = 3.76 × 10⁻⁶] and positively associated with comorbid anxiety [OR = 1.15, 95% CI = (1.09–1.21), p = 8.73 × 10⁻⁷], thus providing further evidence that MDD and BD are quite distinct polygenic liabilities that underpin different aspects of BD heterogeneity. This study is interesting also because it focuses on more detailed outcome and severity phenotypes, a much needed line of investigation, and that results further confirm the detrimental effect of MDD PGS while a mixed effect of BD PGS.

Collectively, these studies (Table 2) suggest a comprehensive picture of how PGSs for MDD and BD may influence a wide range of phenotypes that may relate to outcome: environmental exposures, symptom expression, diagnostic transitions, developmental trajectories, cognitive and educational outcomes, help-seeking behavior, treatment response itself, and even pharmacokinetic parameters, suggesting that, while higher MDD PGS is frequently associated with increased exposure to adverse events and poorer functional outcomes, BD polygenic risk exhibits a more complex pattern that can sometimes be linked to better clinical outcomes (such as higher remission rates and functioning in BD subphenotypes) and cognitive advantages, yet also modulates other aspects in a detrimental direction such as psychosis; however, the overall predictive power of these PGSs remains modest, and their clinical utility would benefit from further refinement through larger, multiancestry, longitudinal studies that integrate genetic data with environmental, clinical, and biomarker information.