Integrating Polygenic Scores and Phenotypic Data to Understand Psychiatric Outcomes

Genetic predisposition plays an important role in the development and clinical management of most psychiatric conditions. Traditionally, it is evaluated through family history, but recent advancements have introduced polygenic scores (PGSs) as an alternative. PGSs, sometimes referred to as polygenic risk scores, measure genetic risk based on common genetic variants identified through genome-wide association studies (GWASs). By combining GWAS data from thousands to millions of individuals, researchers can readily estimate the genetic risk for discrete diagnostic entities and continuous measures, such as neuroticism or body mass index. This approach yields a single measure of polygenic liability, typically calculated as the weighted sum of risk alleles in a given individual. PGSs allow quantification of the genetic loading in situations when family history is uncertain because of small family size, unknown family background, or in cases where family history is impractical and difficult to measure. Thus, PGSs can be used to quantify genetic effects for which family or twin data are scarce.

In this issue of the Journal, Jonsson et al. (1) present the results of a study of bipolar disorder in which they analyzed the relationship between the genetic contribution to several psychiatric phenotypes and occupational disability, namely, periods of unemployment and long-term sick leave, as well as the severity of illness, approximated by the frequency of psychiatric hospitalizations. The sample, obtained from a large Swedish registry cohort, included individuals diagnosed with bipolar disorder (N=4,782) and control subjects (N=2,963). The authors derived PGSs for bipolar disorder, major depressive disorder, schizophrenia, attention deficit hyperactivity disorder (ADHD), Alcohol Use Disorders Identification Test (AUDIT) score, and educational attainment.

The study found that higher PGSs for major depressive disorder and ADHD and a lower PGS for educational attainment were significantly associated with both occupational outcomes. This was the case among individuals with bipolar disorder as well as among those in the control sample. The paper does not provide a comparison of the PGSs for these traits between the groups, but the relative risk is similar in the two groups. These findings argue for an effect that is not disease specific. The paired effects of PGSs for ADHD and educational attainment for both outcomes raise questions surrounding the relationship between the two factors. It may be that individuals in this sample with a high PGS for ADHD are not receiving the necessary supports to succeed academically or in the workplace. Additionally, higher PGSs for schizophrenia and major depressive disorder were associated with longer periods of unemployment, also in both groups.

As for psychiatric hospitalization, higher PGSs for bipolar disorder and schizophrenia were associated with more frequent admissions. These results were replicated in a separate sample of individuals from the United Kingdom with bipolar disorder (N=4,219), although it appears that the bipolar I disorder subgroup was driving this association. The U.K. sample also showed a negative relationship between the PGS for ADHD and frequency of psychiatric hospitalization, which was not observed in the main sample and is rather counterintuitive. Even so, the primary hospitalization results are consistent with several earlier studies. Similar findings were reported by Kalman et al. (2), where PGSs for bipolar disorder and schizophrenia and not major depression were associated with psychiatric hospitalizations in a sample of patients with bipolar disorder. Berutti et al. (3) found that a family history of mood disorders was associated with increased psychiatric hospitalization among individuals with bipolar disorder. Poorer outcomes have been indirectly connected with PGSs, such as in the sample investigated by the Consortium for Lithium Genetics (ConLiGen) (4), where the PGS for schizophrenia was associated with poor response to lithium (although in the ConLiGen data, the PGS for bipolar disorder was not associated with treatment response). As for older family studies, Grof et al. (5) found poor response to lithium prophylaxis in patients with first-degree relatives with schizophrenia, and Post et al. (6) reported worse functioning in patients with a family history of mood disorders. There is also a well-replicated association between family history of mood disorders and both early onset and poor outcome (7).

In summary, Jonsson and colleagues found that the contributors to negative occupational consequences and hospitalization as a direct result of psychiatric illness are divergent. This observation has interesting implications for measures of severity of illness in bipolar disorder samples, because these common measures may reflect distinct etiologies. An overview of the results is presented in Table 1. The study draws on a large genotyped sample of affected individuals and control subjects with accompanying data for psychiatric hospitalizations and occupational history. In contrast, the replication sample had only psychiatric hospitalization information available for comparison. The absence of detailed phenotypic information is a common occurrence, particularly in large-scale data sets such as biobanks or commercial genetic databases. This lack of detail is especially problematic when categorical diagnoses are based on self-reports rather than measures obtained through standardized instruments, for instance, anthropometric measures or validated questionnaire data, which are more reliable. Furthermore, diagnostic categories defined by the current psychiatric classification systems do not necessarily correspond to unique sets of predisposing genes. PGSs can address the overlap by quantifying genetic correlations between certain disorders and isolate genetic underpinnings of specific symptoms. Moreover, PGSs are proving useful in clarifying the genetic contribution to traits beyond categorical diagnosis, and Jonsson et al. illustrate this point by examining the link between two measures of occupational functioning and PGSs for nonaffective phenotypes. However, researchers should exercise caution and refrain from indiscriminately testing increasing numbers of hypotheses without fully understanding the current limitations of polygenic analyses.

When introduced in human genetics, PGSs were considered in the context of diagnostic improvements and risk prediction; however, their accuracy remains low. PGS-derived heritability for many traits is substantially lower than the heritability estimates obtained by, for instance, twin studies. The predictive power of PGSs is expected to increase with availability of larger samples for deriving the appropriate summary statistics. Nonetheless, the gap between the use of PGSs and both GWAS and twin heritability approaches will remain, because PGSs index only the heritability that results from common variants and do not account for rare variants and heterogeneity. Another challenge relates to the ethnic origin of samples used to derive summary statistics for PGS calculations. To date, the largest proportion of data comes from Caucasian populations, inherently limiting the degree to which these PGSs can be generalized to other ethnic groups. Fortunately, there has been a concerted effort to include more diverse ancestries in GWAS samples, in addition to the development of statistical methods that can accommodate multi-ancestry populations. More diverse and well-characterized samples are especially achievable with the decreasing cost of genotyping and the small amount of DNA required for sequencing, although it should be noted that the availability of summary statistics for a great number of traits requires a careful balance between the number of PGSs included in a study and the statistical penalty for multiple comparisons.

It is clear from GWASs that psychiatric disorders are highly polygenic and genetically heterogeneous, and GWASs are unable to fully explain the etiology and mechanisms that produce the varied clinical presentation between individuals. Similarly, the present classification systems do not adequately capture the phenotypic variance within psychiatric disorders. PGSs allow for more granular explorations of psychiatric illness, deconstructing clinical presentations to dissect the complex symptomatology. The use of a combination of PGSs and clinical data has been shown to increase predictive accuracy in studies that analyzed the risk of psychosis (8, 9) and treatment response to lithium (10). There is a need for more studies such as that of Jonsson et al., which incorporate both genetic and phenotypic information, to continue unraveling the dimensions of psychiatric disorders to improve diagnosis and obtain accurate risk prediction.