While recent studies provide evidence for mechanisms that may contribute to disease pathology, excitement must be tempered by experimental knowledge addressing the caveats of in vitro disease modeling (Fig. 1). A downside of iPSC technology is a significant loss of epigenetic modifications after reprogramming, which poses a challenge for studying the impact of environmental factors on psychiatric disorders. However, it is possible that some epigenetic modifications are recapitulated following neuronal maturation in vitro [6]. Furthermore, iPSC-derived neurons are immature and their transcriptional profile is comparable to fetal neurons. Therefore, in vitro phenotypes may represent developmental phenomena preceding disease manifestation, presenting an opportunity for studying psychiatric disorders during development.
Another issue is that of variability between cell lines and across experimental batches, possibly due to somatic mosaicism in donor cells prior to reprogramming, accumulation of de novo mutations with selective advantages, stochastic events during differentiation, and heterogeneous patient genetics [6]. However, iPSC models which capture patient heterogeneity may provide a system for individually tailored assays for diagnostics and drug testing. As a complex picture of the variables at play emerges, complementary approaches and study designs addressing these caveats will be necessary to glean biologically meaningful information (Fig. 1).
One such approach is the stratification of large patient cohorts based on factors such as genetic risk, pharmacological response profiles, distinctive and quantitative endophenotypes, or comorbidities with other diseases. Modeling genetic risk includes rare variants conferring large genetic risk, such as copy number variation, or higher frequency common variants, such as single nucleotide polymorphisms, conferring relatively lower risk [4]. Cellular phenotypes associated with highly penetrant mutations can be studied using genome-edited isogenic iPSC lines or patient-derived iPSC lines. Experiments with the latter would ideally entail one-to-one comparisons between diseased and healthy individuals from the same family, controlling for genetic background. However, for idiopathic patient cohorts, genetic contributors are often unknown, and patient cohort segregation using drug responsiveness has proven to be a successful strategy for uncovering cellular phenotypes in, for example, schizophrenia and bipolar disorder [5]. Additionally, exploring the effects of pharmacological agents on human neural cells in vitro has indicated which molecular pathways and phenotypes may be therapeutically relevant. Collective data from such studies could provide deeper understanding of how diverse genetic risk factors converge on common biological processes and cellular phenotypes.
Another strategy is to study iPSC-derived neurons from a subset of well-characterized patients from a larger cohort. Here, in vitro phenotypes can be correlated with multiple continuous variables such as clinical severity, behavioral/biological measures, brain activity, and blood metabolites. Obtaining such multidimensional data from even small patient cohorts could inform the predictive value of individual variables and lead to the discovery of biomarkers. The explosion of rich neuropsychiatric disease sequence databases coincides with the emergence of powerful and accessible predictive machine learning tools. In conjunction with large-scale genetic data, deep-learning models may enjoy improved performance by using intermediate cellular phenotypes from patient-derived cells to bridge the gap between molecular and circuit or clinical features [7].
In addition to careful study design, picking appropriate in vitro models will be critical for discovering clinically relevant in vitro phenotypes. Three-dimensional iPSC-derived organoids may be able to recapitulate maturation-related signatures in developing circuits, as has been successfully done with autism spectrum disorder [8]. Similarly, transdifferentiation of adult somatic cells directly to neurons may partially conserve non-cell autonomous disease- and age-related epigenetic signatures that may be lost during reprogramming. Interestingly, processes such as inflammation have been implicated in psychiatric disorders, and microglia and astrocytes are emerging as central players in this process. Generating inflammation-sensitive glial cells from patient-derived iPSCs and co-culture experiments with neurons may prove useful for studying disease-relevant cellular interactions [9].
It is increasingly clear that gaining fresh insights into the biology of psychiatric disorders demands a multipronged approach, including but not limited to patient iPSC-based diseased modeling. Furthermore, concerted efforts across laboratories for tackling the inherent variability of in vitro systems may pave the way for establishing standardized in vitro parameters, which would be immensely helpful for moving toward high-throughput profiling and screening in the future [10]. Despite the gap in our knowledge of the biological causes underlying mental illness, iPSC technology—situated at the intersection of molecular biology and higher-order circuit properties—is well positioned to play an important role in disease study and biomarker discovery. We anticipate that, in the future, it may be possible to use patient iPSCs for predictive diagnoses and precision medicine.