Introduction

The genome-wide association studies (GWAS) era has transformed our understanding of bipolar disorder (BD). Ten years ago, BD was considered a distinct, highly heritable disorder for which genes of major effect had eluded detection by linkage studies but were expected to be found eventually. Now, numerous common genetic markers have been found by GWAS, none of which confers major risk for disease, and many of which overlap with markers associated with schizophrenia or major depression. A few higher-risk associations have also been identified, involving rare copy number variants (CNVs) that are usually not inherited. Now, BD can be regarded as a point on a spectrum of risk, ranging from major depression to schizophrenia. Despite this substantial progress, most of the inherited risk for BD remains unexplained, suggesting that there is still much to learn about the genetics of BD. In this review, we will summarize the key developments in BD genetics over the past decade and frame some open questions that will need to be addressed by future studies before we can fully realize the promise of “genomic medicine” in the diagnosis and treatment of BD.

The phenotype

Common

BD is among the most common of major mental illnesses, with prevalence estimates in the range of 1–4% [1]. However, since the diagnosis rests on reports of subjective symptoms that can be subtle, diagnosed cases probably represent the tip of an iceberg of very common disturbances in mood and behavior that blend imperceptibly into the clinical realm. Genetic studies have focused almost entirely on individuals who can be easily diagnosed by interview or are already in treatment, which undoubtedly provides an incomplete picture. Imagine trying to describe the genetics of hypertension by studying only stroke patients.

Varied clinical features

The genetic complexity of BD is belied by its complex and varied clinical presentation [2]. Although the first episode of major depression or mania typically begins between ages 18 and 24 [3], earlier or later onset cases are not rare. Episodes can be frequent or separated by many years, and some patients experience rapid cycling with a period of hours or days [4]. Comorbid anxiety [5, 6] and substance abuse [7, 8] are common, and psychotic features are often a component of mood episodes, particularly manias. Interepisode periods can be completely symptom-free or beset with chronic depressive or manic symptoms. Some people suffer only from manias, although this is uncommon [9]. Mixed states are frequent, as are periods of prolonged, treatment-resistant depression [2]. With such protean manifestations, it seems likely that what we now call BD may ultimately be resolved into dozens of biologically distinguishable disease entities.

Many studies have examined the familiality of clinical features in BD. Age at onset [10], psychotic symptoms [11, 12], frequency of manic and depressive episodes [13], and polarity (mania or depression) at onset [14] are all highly familial, while comorbid anxiety and substance abuse are less so [15]. Below we will address some of the genetic signals that may help explain these patterns.

High risk of suicide

Many studies have pointed to a high risk of suicide in BD [16,17,18,19,20]. On average, about 15% of people diagnosed with BD die of suicide [21], a number that has remained discouragingly stable for decades. Several small studies have reported that suicide may be especially common in some families with BD [18, 22, 23], suggesting specific genetic or shared environmental factors, but these have so far remained elusive.

Cycling as a distinct trait

Signs and symptoms of BD are so wide-ranging that they can be seen, in part, in just about every major psychiatric disorder. This makes for challenging differential diagnosis, one of the reasons that it has proven more difficult to accumulate very large samples of BD than schizophrenia, autism, or major depression. The one very distinctive trait seen in everyone with BD is cycling: episodic elevations and depressions of mood and behavior, separated by periods of relative or complete euthymia [4]. This is such a core feature of BD as currently conceived that we will probably not consider the genetics of BD to be solved until the genetic mechanism of cycling itself has been elucidated.

Response to lithium

Another relatively distinctive clinical feature of some people with BD is the response to lithium. Indeed about one-third of people diagnosed with BD will experience a dramatic improvement in the frequency and severity of mood episodes while receiving lithium, and another third with be at least somewhat improved [24]. Lithium is also the only drug shown to exert a protective effect against suicide in BD [17, 19, 20, 25]. No other major mental illness shows this kind of specific response to lithium, suggesting that genetic risk factors unique to BD are in some way related to the pharmacodynamics of lithium and that biologically meaningful subtypes of BD may be identifiable, at least in part, by response to lithium therapy. A few GWAS of lithium response have been published, but the results so far are divergent [26,27,28,29]. Some recent studies using cellular models lend support to the view that lithium-responsive BD carries a distinct neurobiological signature [30,31,32].

Genetic epidemiology

Before the era of molecular genetics, much of our etiologic understanding of BD rested upon the methods of genetic epidemiology. Family studies demonstrated that BD runs in families, with a 10–15% risk of mood disorder among first-degree relatives of people with BD, but could not distinguish the effects of shared environment from those of shared genes [33]. Twin studies showed that much of the shared familial risk could indeed be explained by shared genes, with heritability estimates on the order of 70–90% [33]. Adoption studies lent further support to a largely genetic etiology, since BD was elevated only in the biological parents of adult adoptees with the illness [33]. Despite the strong and consistent evidence in favor of a genetic etiology; however, segregation analyses could not find a clear, Mendelian pattern of transmission, tending instead to favor more complex models of inheritance [34].

Assortative mating

Assortative mating refers to nonrandom mating among individuals in a population [35]. People with similar phenotypes may be more likely to mate or may selectively avoid potential mates with other phenotypes. A number of studies over the past decades have demonstrated varying degrees of assortative mating in BD, with an increased rate of matings between individuals with BD and those with BD, major depression, alcoholism, or other phenotypes [35,36,37,38,39,40,41,42,43]. Recent, large population-based studies have found similar patterns of assortative mating across psychiatric and other traits, including height [44], activity level [45], emotional intelligence [46], and educational and social status [47].

Such substantial rates of assortative mating are likely to have a major impact on the genetic landscape of BD but are often not considered in studies of the disorder. Theoretically, assortative mating can lead to accumulation of risk alleles in subsequent generations, with consequent increases in rates or severity of illness across generations of a family, a phenomenon known as anticipation [48]. Assortative mating across traits can also induce genetic correlations and comorbidity between the traits in offspring, but these are not likely to persist in the face of random mating by subsequent generations [49]. Assortative mating does not appear to effect heritability estimates by twin studies but may contribute to underestimates of heritability by empirical relationship methods based on SNP arrays [50]. This is because individuals drawn from populations with nonrandom mating will tend to share more risk alleles than would be expected based on their overall genetic relatedness.

Risk loci

Initial searches for risk loci depended on a very limited set of genetic methods, chiefly genetic linkage analysis [14, 51, 52]. However, since linkage methods do not work well in the face of complex patterns of inheritance, linkage studies of BD failed to produce definitive, replicable findings [53]. A similar problem faced linkage studies of most other common, complex traits.

Candidate genes

In an attempt to overcome the limitations of linkage methods, many researchers tried to find genetic markers that were chosen on the basis of their proximity to genes that encoded proteins of known neurobiological importance, such as the serotonin transporter [54]. Unfortunately, this candidate gene strategy was largely unsuccessful. This is because the selection of candidate genes with a high-prior probability of involvement in BD proved to be quite difficult. Most candidate gene studies of BD also suffered from the same biases due to small sample size and undetected genetic mismatch between cases and controls that bedeviled other such studies of a variety of common traits [55]. While meta-analyses do tend to support a small contribution from at least a few well-studied candidates, including the serotonin transporter, SLC6A4 [56,57,58,59], d-amino acid oxidase, DAOA [58, 60,61,62], and brain-derived neurotrophic factor [58, 63,64,65,66,67,68,69,70], the most reliable association evidence has come from GWAS.

GWAS

Genome-wide association studies, wherein large numbers of genetic markers spanning the genome are tested for association with a trait, typically in large, case–control samples, have so far been the most successful strategy for identifying genetic variants associated with BD. Since the first BD GWAS appeared in 2007 [71], almost 20 such studies have been published. Most have focused on typical case definitions of bipolar I disorder [26, 72,73,74,75,76,77,78,79,80,81,82,83], but some have examined clinical subtypes such as schizoaffective disorder [84], bipolar II [85], or BD in the context of personality [86] or other traits. The most recent published GWAS, based on ~50 K cases, detected 30 genome-wide significant loci, of which 20 were newly identified [87].

Genome-wide significant loci reported to date are summarized in Table 1. As with most other common traits, risk loci are numerous, most of the lead SNPs are noncoding, and odds ratios are small (1.1–1.3). Although many of the loci have been implicated by several studies, only a few loci can be resolved to single genes [88, 89] based on current information, so it is still too early to make firm conclusions about specific risk genes underlying most GWAS loci. As functional genomic data accumulates, convergent findings are expected to point toward specific risk genes and pathways.

Table 1 Genetic loci associated with BD.
Full size table

Convergent data so far highlight at least three genes. ANK3, located on chromosome 10q21.2, was one of the earliest genes to be implicated in BD by GWAS [72, 90,91,92,93]. Significant association has now been found between BD and SNPs near ANK3 by several studies, and several of those SNPs affect expression of ANK3 [90, 91, 94,95,96]. ANK3 encodes ankyrin B, a protein involved in axonal myelination, with expression in multiple tissues, especially brain [97]. Numerous alternative transcripts exist, suggesting a potential role for alternative splicing [98]. A conditional knock-out mouse displays cyclic changes in behavior that resemble BD and respond to treatment with lithium [99]. CACNA1C, located on chromosome 12p13, has also been implicated by genome-wide significant SNP associations in several studies of BD, along with schizophrenia and major depression; some of the associated SNPs are also associated with expression of CACNA1C in multiple tissues, including brain [73, 74, 87, 100,101,102,103]. The gene encodes an L-type voltage-gated ion channel with well-established roles in neuronal development and synaptic signaling. Heterozygous knockdown of the gene in mice alters a variety of behaviors thought to reflect mood, but without a clear syndromic resemblance to BD [102]. TRANK1, which resides on chromosome 3p22, has been implicated by genome-wide significant association with nearby SNPs in studies of BD and schizophrenia [75,76,77, 104, 105]. TRANK1 encodes a large, mostly uncharacterized protein, highly expressed in multiple tissues, especially brain, and may play a role in maintenance of the blood–brain barrier [106]. The expression of TRANK1 is increased by treatment with the mood stabilizer valproic acid, and cells carrying the risk allele show decreased expression of the gene and its protein [104]. Recent transcriptomic studies suggest that DCLK3 may be another gene in the same 3p22 GWAS locus that contributes to risk for both BD and schizophrenia [88, 107].

While each individual GWAS “hit” has only a small effect on risk, polygenic risk scores that combine the additive effects of many risk alleles (often hundreds or thousands) can index substantially more genetic risk by including variants that have so far escaped detection individually at genome-wide significance [108]. Recent studies that use the PRS strategy have shown that common variation accounts for about 25% of the total genetic risk for BD (less of the phenotypic variance), that PRS overlap substantially between BD and schizophrenia, and that PRS derived from large schizophrenia samples are associated with increased rates of psychotic symptoms and decreased response to lithium in BD [101, 105, 109].

Copy number variants (CNVs)

CNVs are stretches of DNA that occur in one (deleted), three (duplicated) or more copies on a chromosome, rather than the typical two copies expected in the diploid human genome. Initially discovered by use of hybridization or SNP array methods that could detect deletions and duplications too small to be found reliably by cytogenetic methods, large (30–1000 kb) CNVs have since been shown to play a major role in neurodevelopmental disorders [110,111,112,113,114,115,116] and some cases of schizophrenia [110, 117,118,119,120,121,122,123].

CNVs seem to play a smaller role in BD [124], but at least two CNVs have been associated with BD in large, case–control samples. The 650 kb duplication on chromosome 16p11.2 was initially described in a de novo study of schizophrenia [125] and was later detected as a de novo event in a proband with early-onset BD [126]. Genome-wide significant evidence of association with BD is based on a large meta-analysis of SNP array data, in which the duplication conferred an OR of 4.37 (95% CI: 2.12–9.00) [127]. This same study also found evidence of association with a deletion on 3q29, but this fell short of genome-wide significance [127]. Both of these CNVs have also been associated with schizophrenia, autism, and intellectual disability [128]. A reciprocal deletion in the 16p11.2 region is associated with autism and ID [129, 130]. One recent study found enrichment of genic CNVs in schizoaffective BD [131]. Taken together, these findings suggest that the genetic overlap between BD and schizophrenia extends beyond common, low-risk alleles to rare alleles of larger effect.

Most published CNV studies to date have relied on technologies that cannot reliably detect CNVs much below ~30 kb. As WGS and other technologies come to the fore, we will doubtless find very large numbers of smaller CNVs in the human genome. Many such smaller CNVs may also be associated with various neurodevelopmental and adult psychiatric disorders and may well be found to play an important role in BD in the future.

Single nucleotide variants (SNVs) and and small insertions/deletions (indels)

Next-generation sequencing (NGS) technology has enabled a search for rare single nucleotide and small insertion/deletion variants that are not represented in SNP arrays [132, 133]. Such studies may uncover alleles conferring greater risk than the common alleles detectable by GWAS, but the lower allele frequencies and large number of potential variants usually demand very large sample sizes, often larger than those needed for GWAS [134].

A few early NGS studies have been published in BD and several others are underway [135,136,137,138]. While the early studies lacked statistical power to demonstrate significant evidence of association after correction for multiple testing, as sample sizes grow significant findings may emerge. Ongoing consortia efforts that aim to achieve larger sample sizes through meta-analysis of multiple independent samples have perhaps the best likelihood of success. Studies that leverage the increased frequencies of otherwise rare alleles sometimes seen in unusual populations [134, 139, 140] may also succeed as sample sizes grow and sequencing technology improves.

Other studies have used NGS to sequence RNA expressed in brain tissue obtained post-mortem from people diagnosed with BD [107, 141, 142]. Such studies can identify diagnosis-associated changes in gene expression, inform efforts to fine-map GWAS loci to individual genes [143], and potentially reveal other transcriptomic events (such as alternative splicing [144]) that mediate risk of inherited genetic variants.

Pathways

One way to deal with the substantial genetic heterogeneity of illnesses like BD is to group implicated genes across studies into pathways or networks of functionally related genes. In this way, increased power to detect association may follow if different alleles in different genes converge at the level of gene sets. Several such pathway studies have been published, with little apparent agreement so far [85, 93, 145,146,147,148,149,150]. The multiplicity of implicated pathways and probably reflects genetic heterogeneity, the relatively small number of robust genetic associations found so far for BD, and the still-challenging problem of assigning common genetic markers found by GWAS to the appropriate gene or genes. Calcium signaling is probably the most supported pathway in BD to date. Calcium signaling has been implicated by animal and ex vivo models of BD [90, 151, 152]. The most compelling genetic evidence for this pathway in BD follows from the known function of the risk gene, CACNA1C [73, 102, 103, 153]. Lithium is also theorized to act by decreasing intracellular calcium signaling [154].

Pathways related to chronobiology and circadian rhythm have long been suspected to play a role in BD. Sleep disturbance is often reported by patients suffering from BD, and changes in sleep schedule (as in transmeridian travel) can provoke episodes in susceptible people [155,156,157]. Genes that influence entrainment of circadian rhythm to the light/dark cycle have been widely studied in BD, with some nominally significant findings [141, 158, 159], but none of these genes have so far been directly implicated by GWAS. Mutations of the CLOCK gene, a canonical gene in the circadian pathway, have been associated with mood disturbance and sleep disorders [160].

Mitochondrial dysfunction, with resulting disturbance in energy metabolism, has also long been theorized to play a role in BD. Patients with some known mitochondrial disorders also show increased rates of mood disturbances consistent with depression or BD [161, 162]. There is also some evidence of mitochondrial dysfunction in induced pluripotent stem cell (iPSC)-derived neurons from BD patients [163]. However, GWAS have failed to detect any significant association between mitochondrial DNA polymorphisms and BD [164].

The pathway analyses of genes implicated in the most recent BD GWAS highlight ion transport, neurotransmitter receptors, insulin secretion, and endocannabinoid signaling, which may provide novel targets for therapeutic development [87].

Genetic architecture

Heritability

Twin studies have consistently demonstrated that most of the individual difference in risk for BD is explained by inherited genetic factors. Studies that compare monozygotic with dizygotic twins have estimated values for narrow-sense heritability of about 70% [165]. Some concern has been raised that the traditional twin design may overestimate heritability under specific circumstances that violate model assumptions [166]. These include assumptions about unbiased ascertainment, equivalence of environments shared by MZ as compared to DZ twins, and potential gene-environment correlations [165]. (Gene–gene and gene–environment interactions, however important they may be in BD, do not contribute to narrow-sense heritability estimates [167]). Recent, population-based studies that do not depend on the same assumptions as twin studies have found very similar heritability estimates [168]. Thus, any overestimation of heritability in the earlier twin studies is likely to be small.

Recent methods allow estimates of heritability based on distant kinds of relatedness that may exist in large, case–control samples [169]. These methods rely on empirical estimates of relatedness derived from sharing of common alleles genotyped by SNP arrays. As has been observed for most common, complex disorders, the SNP-based heritability estimates for BD tend to range from around 25–45% [78, 170]. This “heritability gap” or “missing heritability” is not fully understood, but may reflect imprecision in the method, overestimates of heritability in twin studies (noted above), or a contribution of rare variants not captured on SNP arrays.

Models of etiology and risk

We still lack good models that can bring together genetic and other data heuristically. Four possibilities broadly consistent with the available data come to mind, but others are hard to rule out: (1) Two-hit model. Under this model, we imagine that classes of risk factors interact nonadditively to determine outcome, with combinations accounting for phenotypic distinctions [171]. For example, given two individuals with similar polygenic risk burden, one might develop BD while the other, exposed to a second hit from maternal influenza, develops schizophrenia. (2) Multifactorial threshold model. Under this model, there is a large but finite set of nonspecific genetic and other risk factors, whose total dosage determines specific phenotypes [172]. Thus, BD would occupy some intermediate space, with more risk factors than depression but fewer than schizophrenia. This is a more general version of the two-hit model and fits best when each risk factor has a small, additive effect on outcome. (3) Risk-resilience model. Under this model, genetic differences might confer risk or resilience, with the phenotypic outcome reflecting a delicate balance of harmful and protective factors [173, 174]. Thus, BD might result from genetic risk factors conferring, say, unstable mood, nearly balanced by stable temperament, and advantageous life circumstances. (4) Omnigenic Model. Under this model, almost all genetic differences contribute in some small way to risk (or resilience), while phenotypic outcomes are determined largely by which genes are involved and their relative importance in relevant cells and tissues [175]. Thus, BD might result from genetic risk factors that happen to impact genes that play an important role in cells that underlie neural circuits involved in regulation of mood and behavior.

It has been said that all models are wrong, but some are useful. Each of these models has supporters and critics. The two-hit model resonates with long-held theories of gene × environment interaction, but robust evidence of such interactions has proven elusive [176,177,178,179,180]. The Omnigenic Model has generated much recent debate, since it would seem to imply that larger and larger GWAS cannot alone solve complex traits. In any case, we clearly need more and better ways to incorporate nongenetic risk factors into models of etiology and risk prediction.

Genetic correlations

Genetic correlation refers to the degree to which two distinct traits share genetic influences (or more formally, the proportion of additive genetic variance—heritability—that is shared [167]). Traditionally, estimated through laborious twin and family studies, genetic correlation can now be estimated much more easily from overlapping sets of common SNPs genotyped in existing samples [181]. Such studies have so far revealed many expected and some unexpected genetic correlations with BD. In addition to the substantial genetic overlap with schizophrenia that was already apparent early in the GWAS era, significant genetic correlations are observed between bipolar and major depressive disorder [87, 182, 183], attention deficit hyperactivity disorder [184], neuroticism [185], and borderline personality disorder [86]. Small but significant genetic correlations have also emerged between BD and educational attainment [87], creativity [186], and leadership [187]. These findings lend support to the view that BD represents a point on a spectrum of genetic risk, with quantitative rather than categorical genetic differences underlying a range of common disorders of mood, perception, and cognition (Fig. 1).

Fig. 1
figure 1

Genetic and symptomatic relationships between bipolar and some other psychiatric disorders. Shared heritability of bipolar disorder (BD) with schizophrenia (Scz), attention deficit disorder (ADD), and major depressive disorder (MDD). Genetic correlation values were extracted from Ref. [181].

Full size image

Pharmacogenetics

Pharmacogenetic studies aim to use genetic information to help match patients with the safest, most effective treatments. Several pharmacogenetic studies have been performed in patients with BD, but replicated findings have not yet emerged. This may reflect the fact that many past studies relied on a candidate gene design, while GWAS have not generally been able to achieve sample sizes large enough to detect variants of minor effect. The measurement of treatment response in BD brings additional challenges, since the episodic nature of the illness makes short-term assessments of outcome unreliable.

Some promising findings have nevertheless emerged from recent studies. The largest study to date, by the Consortium on Lithium Genetics, carried out a GWAS of lithium response in over 2000 individuals with BD who were treated with lithium and systematically rated for response. Significant association was detected with a set of genetic variants within a noncoding region on chromosome 21 [27]. Another recent GWAS compared lithium-responsive patients to healthy controls, revealing significant association with a SNP near SESTD1 [188]. The apparent lack of agreement between these two GWAS studies probably reflects limited power to detect small effects. One study in a highly selected set of Taiwanese claimed a locus of major effect [28], but several well-powered studies have failed to replicate this finding [29, 189,190,191]. As sample sizes grow, it seems likely that common loci influencing response to lithium or other drugs will be identified. Larger samples may also enable PRS derived from pharmacogenomic studies to illuminate pathways of drug response or help identify subgroups of patients most likely to respond to a specific treatment regimen.

In contrast to studies of treatment response, those focused on serious adverse events have detected strong and reproducible signals for drugs that are sometimes used in the treatment of BD. Patients exposed to carbamazepine occasionally develop serious adverse cutaneous reactions (ACR), such as Stevens–Johnson Syndrome. Genetic association studies initially carried out in people of Asian ancestry identified an HLA haplotype that conferred substantial risk of ACR after carbamazepine exposure [192]. Subsequent studies have confirmed this association also in patients of European ancestry [193], albeit with a different HLA haplotype. Other studies have identified additional, apparently independent HLA haplotypes that predispose to ACR after exposure to lamotrigine or phenytoin [194]. Based on these findings, HLA testing is advised in all patients being considered for carbamazepine and may also be informative for treatment decisions concerning other anticonvulsants [195].

Genetics of clinical subtypes

It has long been assumed that the clinical diversity of BD reflects, at least in part, differences in underlying risk alleles. Limited statistical power has so far forestalled a complete genetic dissection of the bipolar phenotype, but several studies have found suggestive evidence of genetic differences in bipolar cases with psychosis or catatonic features, and in cases with bipolar II disorder [84, 105, 196, 197]. One large study found a significant positive correlation between genetic risk for schizophrenia and psychotic episodes in patients with BD [84]. This same study detected significant heritability, as estimated from genome-wide SNP data, for psychotic features and suicide attempts in BD.

Ongoing studies aim to go beyond clinical symptoms to define subtypes of disease based on neuroimaging [198,199,200,201], neurocognitive tests [202, 203], and EEG patterns [201, 204, 205], as well as genetic markers. Such studies hold promise for a future nosology of bipolar (and other psychiatric) disorders that better reflects neurobiological disease entities.

Future directions

Cellular phenotyping

The generation of iPSCs from patients allows for in vitro evaluation of cell-autonomous traits that might be associated with clinical diagnosis [206, 207]. Cellular morphology, gene expression, and cellular functions are just some of the phenotypes that can be analyzed using iPSC-based cellular models. More complex models, such as 3D organoids, can explore more macroscopic interactions and might shed light on disorder-specific changes in brain circuitry. So far, only a few published studies have used iPSC derived from patients with BD [104, 151, 163, 208, 209], but several studies are underway. Initial results suggest some differences in neurons derived from patients with BD.

Reverse phenotyping

As we begin to identify genes that have a substantial influence on risk (either collectively, as with PRS, or individually, as with certain CNVs or rare variants), it may be instructive to study individuals who carry substantial risk but do not present in a psychiatric clinic. This approach, dubbed “reverse phenotyping” [210] or “genetics-first” [211, 212] has begun to bear fruit in studies of CNVs and aneuploidies that confer high risk for ASD or schizophrenia [116, 213,214,215]. These kinds of studies are needed for accurate estimates of penetrance [110, 114, 216, 217] and may also reveal an unheralded range of phenotypes related to identified genetic risk factors [218, 219]. Longitudinal studies of genetically high-risk individuals may also shed light on protective or resilience factors and could provide the basis for assessing the impact of primary prevention strategies.

Drug development

The path from the identification of risk alleles to the development of new drugs is complex and beyond the scope of this review. Readers interested in exploring this topic further should consult some recent reviews [220,221,222].

Clinical genetic testing

Genetic testing with utility for the diagnosis of BD or its treatment is not on the horizon right now. Too little of the risk is explained by current polygenic risk scores [170], and known pathogenic CNVs are so far quite rare in BD [124, 127]. However, some models suggest that PRS may ultimately prove useful in psychiatric diagnosis as GWAS samples reach sizes on the order of one million, at least for those individuals with the highest risk allele burdens [223, 224].

Genome-wide approaches help us navigate through the complex genetic landscape in an unbiased manner. However, multiple testing means that GWAS can only detect robust associations in large samples. Increasing the number of samples through involvement of different sample collection sites may improve power but can also introduce substantial genetic heterogeneity. This could be due to the innate genetic variability present across different populations and differences in ascertainment or clinical diagnosis by different research groups. This challenge highlights the need for further global-scale collaborations, standard practices of clinical assessment and phenotype characterization across different groups, and genome-scale modeling that can elucidate the biological impact of the many different risk alleles that are detected in large, population-based studies.

Conclusions

What emerges most clearly from molecular genetic findings over the past decade is a concept of BD that includes several features: (1) BD is a heterogeneous set of illnesses united by the core clinical feature of cyclic elevation in mood and activity, with substantial individual variation in depressive and psychotic symptoms; (2) there is strong sharing of weak, common genetic risk factors with schizophrenia and major depression; (3) high-risk alleles also exist, but they are rare and nonspecific, and there is so far no evidence for monogenic forms of BD.

As a disease entity, BD may resemble stroke or type II diabetes in the sense that several subclinical states create a meta-stable condition that periodically erupts in symptoms. For stroke, we understand that hypertension and cerebrovascular disease create vulnerabilities that may present periodically with paralysis, language, or cognitive deficits. And while there are rare, high-risk alleles that cause stroke, most of the genetic risk resides in large numbers of common alleles that each have a small impact on blood pressure, vascular health, and coagulability [225]. This analogy suggests that we need to identify the fundamental neurobiological processes that are most directly influenced by common risk alleles and we should expect that these processes are underway long before the first manic episode. The analogy further suggests that secondary preventive strategies will need to take aim at these underlying processes, probably beginning at or around the time of the first manic symptoms.

It remains to be seen whether genetic findings to date will continue to coalesce into clear neurobiological pathways. If they do, identification of new drug targets may be possible. The advent of cellular modeling through iPSC technology offers a new platform for screening large numbers of potential new drug treatments, but the success of this approach will depend heavily on the identification of robust cellular phenotypes that reflect at least some of same the genetic risk factors that predispose to bipolar or related disorders. Meanwhile, even if single genes of large effect remain elusive, it seems likely that polygenic approaches incorporating numerous common risk alleles will continue to be useful for research and may ultimately find modest applications in some clinical settings. We have finally made it through the first era of molecular genetics of BD, but the road to new methods of diagnosis and treatment may well remain long and uncertain.