Introduction

The parable of the blind men and the elephant has often been used as a metaphor for a reductionist cognitive process. Similar to the blind men trying to reconstruct the complexity of the elephant through fragmented experience, scientists aim at modeling complex human disease through functional studies of simpler, more approachable systems. In this perspective, availability of relevant experimental models is critical to further our understanding of any human disease, as well as to facilitate drug discovery. Advantages as well as potential limitations of employing human induced pluripotent stem cell (iPSC)-derived systems as an advanced technological tool to apply a reductionist approach to the study of autism spectrum disorder (ASD), have been extensively reviewed elsewhere [1,2,3,4]. Here, we focus on recent milestones in this space that are slowly enabling a more holistic outlook, and review efforts that apply iPSC-derived models to reconstruct, quantitate, and predict the complexity of the human brain. Specifically, we extensively discuss which cell types may be most vulnerable to the genetic variation underlying ASD and how their altered function might underlie behavioral changes in people with ASD. In this perspective, we critically evaluate neuronal iPSC-derived models used thus far to obtain such cell types in culture, and summarize reported cellular phenotypes that can represent partial readouts for ASD-like features. Then, we consider the few studies where iPSC technology has been coupled with human genetics to advance our understanding of the molecular makeup of neurodevelopmental complex genetic disease, including ASD. Finally, we contemplate ways that iPSC models can enhance the power of genetic studies to identify ASD risk factors and fuel discoveries of key pathways that are altered across many individuals with ASD and might serve as shared points for therapeutic intervention.

Background: the advantages of modeling disease with induced Pluripotent Stem Cells (iPSCs)

Animal models, especially mice, have historically been a key tool in basic research and therapeutics. However, because of existing species-specific differences in pathways implicated in health and disease [5], human cell-based models have always been regarded as appealing complementary systems [6]. Human primary cells are however generally unavailable for study due to their limited availability from patients [7, 8].

In the attempt to overcome this impasse, a landmark in the field of cellular disease modelling was the employment of in vitro grown embryonic stem cells (ESC) derived from human blastocysts, that have the ability to indefinitely self-renew and can give rise to any type of somatic cells [9, 10]. However, despite providing many human cell types for research and therapeutics, their employment immediately raised societal concerns over their early embryonic origin [11], as well as practical limitations due to lack of information about the donors, and their familial and medical history (including their propensity to develop ASD) [12].

Takahashi and Yamanaka’s pioneering studies in the 2000s [13] led to rapid expansion of induced pluripotent stem cell (iPSC) technologies, and opened unprecedented opportunities for disease modeling. The last decade is testimony of how iPSC-based studies can enhance biomedical research and personalized regenerative medicine [12,13,14]: somatic cells from easily accessible tissues of patients can now be routinely reprogrammed into an embryonic stem cell-like state, and subsequently differentiated into cell types that are relevant for the same patient’s disease. iPSC technology also yields virtually unlimited amounts of human tissue carrying a genetic variant of interest, that becomes easily available for manipulation and therapeutic endeavors. Furthermore, genetic variants can be examined on a genetic background sensitive to the disorder and that may account for unpredicted secondary effects [15, 16].

The considerations above are all especially relevant in the context of neuropsychiatric and neurodevelopmental diseases—including ASD—which are for the most part quintessentially human and often polygenic [17, 18]. Primary cultures of patient-derived cells are largely unavailable, because brain biopsies for establishing an in vitro neuronal cell line, are considered unethical [19], while human post mortem samples, despite being desirable alternatives, typically do not represent the developmental stage when the disease is firstly manifested, and can be confounded by other factors, including treatment for the disease of study or for some of its symptoms [20]. Additionally, neither genetically engineered animal models, nor post mortem samples, have thus far had the capacity to predict patient-specific clinical outcomes to candidate ASD therapeutics [21].

iPSCs meet all requirements to address these issues, as effectively unlimited quantities of patient-derived cells can be used to model cellular components of the human brain, to identify therapeutic targets, and to investigate said targets and design candidate therapies [2, 19,20,21] (Fig. 1). iPSCs are therefore an optimal resource to study various aspects of ASD in vitro, under the assumption that specific cell types are vulnerable to ASD, and that such cell types can be reliably derived from iPSCs using currently available protocols.

Fig. 1
figure 1

Overview of all available model systems currently employed to model disease. iPSC-based models represent a source of unlimited patient-specific material, able to recapitulate neuronal development without ethical concerns linked to use of embryonic material or patient biopsies

Full size image

ASD is a complex, polygenic, and heritable disorder

Under the broad diagnosis of ASD is a variety of neurodevelopmental disorders marked by impaired social skills and restrictive-repetitive behavior [3]. Individuals diagnosed with ASD exhibit a variety of phenotypes depending on a complex interplay between genetic and environmental factors and often manifest other comorbidities, both neurological and non-neurological.

The phenotypic complexity of ASD reflects its underlying genetic architecture, made of contributions from rare variants of large effect, either CNV (e.g., 16p11.2 or 22q11.2 duplication and deletion) or point mutations (e.g., CHD8, SCN2A), and common variants each conveying small effect but collectively sha** most of its risk [4, 22,23,24,25,26].

Recently, an unprecedented expansion of genome-wide association studies (GWAS) have led to the identification of common variants associated with ASD [22, 23, 27], while large-scale exome sequencing studies of ASD have now identified over 100 high-confidence autism risk genes [24, 25, 28]. However, how disruption of such genes results in altered neurodevelopment and neurophysiology in individuals with ASD, is still largely unclear.

Nevertheless, granular understanding of ASD genetic architecture has provided a tool in determining the dynamics of ASD onset during development at the cellular level, using analysis of concerted expression of ASD risk genes [29], and has been pivotal in defining the identity of cell types most relevant to ASD physiopathology. Identifying cell types that are vulnerable to ASD can subsequently guide efforts in perfecting protocols to derive such cell types from iPSC models [30], providing a promising avenue to translate genetic information into cell modeling.

Cell types of both develo** and adult brain are vulnerable to ASD and can be modelled in vitro

The phenotypic complexity of ASD suggests that there might be multiple cell types vulnerable to ASD both during development and adulthood (Table 1).

Table 1 ASD-vulnerable cell types (selected studies)
Full size table

According to co-expression studies of ASD-relevant genes, a critical window for the onset of ASD, coincides with early fetal development, as specific molecular and cellular programs depending on coordinated expression of ASD risk genes, do not persist in the mature cells of the adult brain [29, 31, 32]. This argument nominates early time-points in neuronal maturation, such as Neural Progenitor Cells (NPCs), as attractive cell models. On the other hand, mature neurons have been strongly implicated in ASD physiopathology by a number of bulk gene expression studies showing changes in the neocortex of ASD patients, and indicating functional convergence of risk-gene expression in the adult brain [33].

Similarly, Satterstrom and colleagues, who applied co-expression network analysis to BrainSpan datasets, showed that ASD genes are indeed expressed at high levels not just in the develo** brain, but also in the adult cortex [28]. The authors were able to pinpoint specific cell types, based on expression modules of 4,261 cells from the prenatal human forebrain [34]. In accordance with previous evidence that identified excitatory glutamatergic neurons [29, 31, 32], they found neuronal cell types as being prevalently recapitulating the transcriptional signature of ASD, with most genes being expressed in excitatory and, to a lesser-extent, inhibitory lineage cells [28]. Other neuronal types found to be enriched for ASD signal were striatal interneurons. Interestingly, Cogill et al. demonstrated that also lncRNAs may play a role in ASD due to their convergence on shared pathways with ASD-associated coding-genes. It will be important to follow up on this finding and include lncRNAs in future re-analyses based on co-expression modules [35]. Despite the general lack of statistical power in molecular studies involving scarcely available post-mortem tissue, Velmeshev and colleagues recently conducted a single-nucleus RNA sequencing study on cortical tissue from patients, and found that expression of synaptic and neurodevelopmental genes is especially affected in cortical neurons [28].

In conclusion, although some non-neuronal cell types were found to be vulnerable to ASD (including microglia and OPCs), most of the transcriptional modules seem to converge on neuronal lineages, and notably on maturating neurons and excitatory neurons of the adult neocortex [36].

It should also be noted that abnormal pain sensitivity is commonly reported in ASD patients [37] and developmental disorders linked to monogenic forms of ASD are also associated with defects in somato-sensation [38]. Many studies conducted in rodent models of ASD with highly penetrant monogenic mutations, indicate that abnormalities in sensory reactivity correlate with ASD-related phenotypes, in line with the hypothesis that impaired sensory perception may impact brain development and function, and results in disparate symptoms associated with ASD [39]. This hypothesis provides a potential mechanistic link between otherwise heterogeneous ASD-related phenotypes, and implicates peripheral sensory neurons, in addition to the other neuronal types discussed above, in ASD etiology [40].

State-of-the art iPSC-derived differentiation protocols that model cell types vulnerable to ASD

iPSC-based differentiation protocols offer a valuable resource to generate cell types relevant to virtually any disease of interest, with the caveat that such cell types can be reliably derived from iPSCs using currently available protocols.

A selected overview of the otherwise large number of published protocols for neural differentiation of cell types relevant to ASD, is reported in Table 2. They are all based on the premise that it is possible to mimic embryonic differentiation in a dish, with the distinction that some recapitulate intermediate NPC states, while others achieve direct differentiation to the terminal neural cell type of choice. Either way, many protocols start with dual SMAD inhibition [41]. Subsequent differences in concentration of patterning factors or in the timing of their addition can yield a variety of neural cell types and impact largely the homogeneity and nature of neurons. In order to overcome culturing heterogeneity, many protocols are now based on inducible transcription factor expression, via virus transduction or integration into a safe harbor locus [42,43,44,45,46]. These protocols generally produce highly differentiated and homogeneous cells in a shorter time frame and on a larger scale when compared to others. However, it is still essential to share detailed experimental guidelines to guarantee reproducibility of each new protocol.

Table 2 Protocols for fast generation of specialized neurons from iPSC cells (selected studies)
Full size table

High degree of reproducibility was reported by Nehme and colleagues, who combined small-molecule with transcriptional patterning, to generate cortical excitatory glutamatergic neurons [47]. After three weeks of maturation, transcriptomic analysis confirmed homogenous maturation of cultured iPSCs into upper layer cortical projection neurons. Microelectrode array and patch clamp electrophysiology also showed AMPA and NMDA-mediated synaptic transmission, which are hallmarks of postnatal cortical neurons. Of particular interest for NPC in vitro modelling, Wells and colleagues adapted the protocol developed by Nehme [47], allowing rapid (48 h) generation and maintenance of human stem cell-derived progenitor cells (SNaPs) [48].

Finally, the association between ASD and altered somato-sensation, suggests that generating iPSC models of sensory neurons from ASD patients will be a valuable system for testing the ability to reverse some ASD-related cellular phenotypes, as previously done in mouse models [49], and spinal cord [50]. Although few protocols have been previously reported (Table 2), none has been utilized in ASD research thus far.

Complex cell culturing systems to study ASD

The translational potential of iPSC-derived models can be further enhanced by complementing cell culture with the inclusion of additional components of the in vivo niche of the cell type of interest, or mimicking cell–cell and cell–matrix interactions that occur within organs and tissues [51] (Fig. 2). Complex culturing systems, including co-cultures and three-dimensional (3D) cultures, may also account for non-cell-autonomous effects on differentiation, and help modulate neuronal activity and drug response, while also promoting neuronal maturity [52].

Fig. 2
figure 2

Summary of various iPSC-based culturing systems. Bi-dimensional cultures can be adapted for co-culturing of more than one cell type at a time. Three-dimensional cultures can be supported by microfluidic devices

Full size image

One of the most ASD-relevant examples of co-cultures consists of iPSC-derived neurons and glial cells, an abundant cell population in the human brain that have critical supporting roles for neurons in both health and disease [53, 54]. Co-culturing neurons with microglia has been key in studying chronic inflammation correlated with ASD and neurodegeneration [55]. Other examples of cell types that have been co-cultured with neurons in ASD modeling are oligodendrocytes (impacting neuronal myelination [56]) and astrocytes (impacting viability, synaptic function, and neurite outgrowth [57]).

3D culturing conditions are also an important development for improving physiologically-relevant in vitro disease models. For instance, it is now possible to model specific regions of the brain, allowing a more holistic comparison between ASD-derived and control cultures [58]. Although 3D cultures can be achieved through microfluidics and bioprinting, one of the most promising technologies is that of iPSC-derived brain organoids. Brain organoids consist of multicellular aggregates that differentiate and self-organize, mimicking its in vivo development [59], and offer new models for assessing the pathogenesis of ASD, especially in the context of monogenic syndromes [60,61,62,63]. Additionally, implementations of organoid protocols allow production of specific regions of the brain, including hippocampus and cerebellum, as well as cortical folding, enabling a holistic study of the human brain in development and disease [58]. Attempts have also been made to use this technology to recreate early stages of corticogenesis, particularly relevant to the study of prenatal brain organization and function [64].

Co-culture and three-dimensional (3D) cultures are therefore a promising development for iPSC modeling of complex tissues (reviewed in [65, 66]), although currently still challenging to establish and reproduce, mostly due to batch-to-batch or organoid-to-organoid heterogeneity and long differentiation periods (60–120 days to reach differentiation levels similar to mid-gestation, compared to 14–30 days of NGN2-based patterning protocols).

iPSC-based models manifest ASD-related, measurable phenotypes

NPCs and neurons derived from syndromic and idiopathic individuals diagnosed with ASD, display a wide range of phenotypes [30, 67,68,69,70]. The phenotypic diversity observed in many individuals with ASD is representative of the underlying heterogeneity of their genetic background and is also reflected on the diversity of reported cellular phenotypes observed across iPSC models. Comprehensive catalogues of ASD-relevant cellular phenotypes as well as detailed description of current Biobanks of deposited patient-derived cell material, have been compiled in a number of excellent reviews [71, 72]. However, it is often challenging to compare qualitative observations, and the field would certainly benefit form adapting standardized quantitative measures to evaluate the impact of genetic background on cellular phenotypes.

Modelling ASD with iPSC technology and classifying each model based on a well-defined subset of quantitative qualifiers or “phenotypic classes”, has proven crucial in revealing novel cellular and molecular mechanisms underlying it. Thus far, the most robust quantitative measures utilized to stratify cellular phenotypes have focused on cell proliferation and brain growth; RNA-processing; synapse density and dendritic arborization; electrophysiology; and calcium signaling (Fig. 3). Here we provide a summary of the most reproducible, robust, and representative cellular phenotypes according to these metrics (Table 3), and focus on few representative examples to discuss how they relate to ASD symptoms, and whether they can be used for translational endeavors (Fig. 4).

Fig. 3
figure 3

Overview of measurable phenotypes observed in cell types either derived from ASD patients or obtained via gene-editing. Neuronal phenotypes can be cell autonomous or mediated by interaction with co-cultured non-neuronal cells

Full size image
Fig. 4
figure 4

iPSC-based models as fundamental tools to bridge human genetics and functional studies in ASD, through the employment of large-scale assays, including PPI networks, CRISPR-screens and “villages”

Full size image
Table 3 Summary of published studies using iPSC-based models to study ASD (alphabetical order)
Full size table

Cell proliferation

Several clinical studies of ASD have reported accelerated brain growth in the first three years of life of patients [73]. This translates in macrocephaly, that is in fact a characteristic phenotype of some genetic subtypes of ASD. Conversely, microcephaly, the inverse phenotype, is also associated with autism. For example, deletions and duplications at two loci, 1q21.1 and 16p11.2, have opposing brain growth phenotypes [74], as well as individual gene mutations: CHD8′s [124], and in animal models. Notably, ** and colleagues used in vivo Perturb-seq to introduce frame shift mutations in 35 genes strongly associated with ASD in humans, and studied their effect on mouse postnatal brain [125]. Given the recent advances in adapting CRISPR-screen to human neurons [126], it is easy to imagine how a similar experimental design could be soon translated to iPSC-derived human models.

Additionally, a number of studies have recently focused on employing genetic heterogeneity within populations, to exploit the multiplexing potential of single cell RNA-seq [127]. Specifically, genetic information offers a natural identifier or barcode to demultiplex pooled samples, allowing complex combinatorial experimental designs of single cell RNA-seq experiments. This approach has been successfully combined to CRISPR-screens, and iPSC-derived neurons represent an ideal cell model to analyze effects of ASD-risk genes on cellular phenotypes [128]. Another interesting application of recent advances in pooled RNA-seq technology is represented by systematic sequencing of villages of neurons obtained from patient-derived iPSCs [129, 130]. For instance, Cederquist and colleagues, pooled in a single dish 30 isogenic iPSC lines harboring de novo ASD mutations to disentangle ASD genetic heterogeneity [131]. The same approach can be used to explore polygenic risk or to assess the impact of genetic background on ASD-relevant phenotypes and eQTLs (expression quantitative trait loci). Lines carrying monogenic ASD mutations of patients that have been extensively genotyped certainly represents an attractive proof-of-principle for further developments of this approach.

Future perspectives: iPSC models to drive therapeutic intervention

In this review, we summarized how iPSCs have been utilized to model certain aspects of ASD, and to quantitatively assess ASD-associated phenotypes. Looking into the future, it will be important to enhance the translational potential of current technologies. Given the heritability of ASD, gene therapy offers a complementary alternative to small molecule-based approaches, especially in the monogenic syndrome space. This approach has been already largely explored for treatment of several diseases and notably for some neurological monogenic disorders [132, 133]. Genetic correction could be tested on cell systems to optimize both optimal carriers and efficiency in rescuing specific phenotypes. AAVs have emerged as the principal delivery candidates, and have proven effective in mice carrying a null allele for Mecp2 [134, 135]. However cell toxicity, optimal time-window for transduction, and off-target effects have not yet been determined, and need further investigation [136]. iPSC models represent an immediate venue for these types of evaluations. An appealing alternative to genetic correction is modulation of gene expression by knockdown of mRNA transcripts through antisense oligonucleotides (ASOs) or short interfering RNAs (siRNAs). Both technologies are based on Watson–Crick base pairing to particular mRNA transcripts aimed at preventing their translation (detailed mechanisms of action, are reviewed elsewhere [137]). ASOs targeting Ube3a-ATS have been used to correct cognitive deficits in a mouse engineered to model features of Angelman syndrome [138]. In the same vein, ASOs have been used to normalize MeCP2 levels and rescue the neurological defects observed in mice carrying an extra copy of MECP2 [139]. This avenue appears to be broadly attractive for treatment of many syndromes caused by haploinsufficiency, where effective therapeutics should aim at restoring normal range of gene expression rather than editing the genome.

Although humanized mouse models, circumventing the issue of potential interspecies differences across vertebrates, may be utilized for testing these studies, patient-derived iPSC models represent a more expedited and scalable tool to test transability of these results in humans and vulnerable cell types, as in the case of other neurological and neurodegenerative conditions, including ALS/FTD and AD [140, 141].

Conclusions

ASD comprises a group of highly inheritable neurodevelopmental disorders characterized by impaired social interactions as well as the presentation of restrictive and repetitive behaviors [3]. The phenotypic complexity of ASD reflects its underlying genetic architecture, made of contributions from highly penetrant rare variants, and common variants each conveying small effects but collectively sha** most of its risk [4, 22,23,24,25,26].

In this review, we discussed how iPSC technology has become central to modelling various aspects of complex human disease, and can potentially allow researchers to advance our understanding of the pathophysiology of ASD, and to test personalized drug candidates. However, since the goal of iPSC-based ASD models is to reproduce and somehow functionally break down the complexity of the human brain, they need to be highly elaborate, yet reproducible. As a consequence, there are many challenges to face, including improving reliability and robustness of iPSC culturing and differentiation protocols. In this perspective, a good balance should be found in the current attempts on advancing the complexity of cellular models [142], and the vast batch effects, that might account for the most part of measured phenotypes [143].

Development of more robust protocols, employment of isogenic lines, and use of gene editing to compare variants within the same genetic background, will certainly contribute to overcome most technical roadblocks encountered in the past [118,119,120]. The rapid development of protocols to derive three-dimensional cultures and organoids, as well as to efficiently maintain co-cultures of mixed cell-types, coupled with technological advances in engineering culturing devices, all seem to be promising paths towards obtaining more complex and accurate disease models [60,61,62, 144]. Additionally, iPSC biobanks, providing access to a plethora of established, well-characterized and well-annotated iPSC, have significantly improved our understanding of the biological basis of natural genetic variation [117,118,119].

Generating faithful models is of utmost importance, but efforts in obtaining more reliable cell models, must be matched with advances in standardizing measurable cellular phenotypes related to certain aspects of ASD. In fact, the community would immensely benefit from a standardized assessment of in vitro phenotypes that reflect disease-relate mechanisms rather than generic experimental and/or genetic variance. Individual studies, mostly performed on iPSC-derived models of monogenic syndromes, have focused on assaying cell-proliferation, altered RNA-processing, electrophysiological properties, synaptic structure and calcium signaling [68, 69, 104, 106]. Assessing these phenotypes in larger-scale studies, comparing vast numbers of iPSC-derived cell lines, and linking them to dysfunction of discrete molecular pathways, is more than ever necessary. In parallel, continuous efforts in clinical sequencing of stratified patients and broadening biobank databases, will be key in advancing our understanding of complex genotype–phenotype correlations at the individual and cellular levels [48, 129].

In conclusion, recent advances in the field of human genetics, with tens of genes being identified as concentrating ASD risk [27], and hundreds of rare-variants with different degrees of penetrance [28], emphasized the need for a better understanding of the complexity of ASD. iPSC-based cell systems, while offering an unprecedented opportunity for modeling measurable ASD-related phenotypes, also provide a unique platform to rapidly validate and enhance genetic findings by nominating pathways that are disrupted across groups of ASD patients [48, 129]. These might represent as hot-spots for ASD vulnerability and desirable targets for therapeutic intervention.