ASD overview and 2D/3D modeling

Autism spectrum disorder (ASD) is a highly prevalent neurodevelopmental disorder (NDD) that impacts as many as 1/44 of children in the United States [1]. Clinical presentations of ASD vary widely among individuals but must include repetitive, restricted behaviors and social deficits [2]. To add to this complexity, comorbidities often include epilepsy and seizure disorders (up to 30%), intellectual disability (>30%), ADHD, gastrointestinal disorders (up to 70%), anxiety, and depression [3,4,5,6,7,8,9]. Risk factors for ASD can occur prenatally, perinatally, and postnatally and include genetic disruptions and environmental insults, of which there is likely a combinatorial or synergistic effect.

Twin studies were one of the first indicators of the genetic component to ASD, and while heritability estimates can range from 45 to 90%, it is more broadly thought to be 70–80% [10, 11]. More recently, cohort-based sequencing studies have provided a genetic framework to studying ASD and have identified several hundred implicated genes. Genetic disruptions include inherited rare variants and less common de novo mutations that exist as single nucleotide polymorphisms (SNPs), copy number variants (CNVs), and chromosomal abnormalities [12,13,14,15,16,17,18,19] Despite the immense progress in identifying ASD-risk genes, the encoded proteins and resulting pathobiology remains elusive. Scientists have turned to genetic modeling to better understand the molecular, cellular, and functional (circuit-based) consequences to disruption in these ASD-risk genes [20].

Genetic models of ASD have most commonly included mouse and human cell lines, which provide biologically and clinically relevant opportunities for study [21, 22]. These models are not without their caveats, however, as the development of a mouse brain excludes human-specific processes such as brain gyrification, the protracted development and formation of particular neural cell types, and human-specific gene transcriptional programs. In addition, mouse behavioral assays are often not translatable to the complex clinical presentations of those with ASD (reviewed in refs. [23, 24]). Conversely, human cell lines produce reliable, replicable environments for testing simple pathways, but are reductive and lack the ability to mimic complex developmental brain processes. Further, they can lack specific cell types and structures that play a crucial role in development, such as brain vascularization.

Induced pluripotent stem cells (iPSCs), which can be generated from human blood or skin-derived fibroblasts, have transformed the use of human cellular models [25]. iPSCs retain the unique genetic background from the individual, which is important given that idiopathic ASD represents roughly 80–85% of the ASD population [26]. Through directed differentiation, iPSCs can produce any germ layer cell types to model complex and inaccessible tissue such as the developing brain, allowing for a putatively unlimited supply of patient-specific tissues to study disease processes or drug screening [27] (reviewed in ref. [28], summarized in Table 1).

Table 1 Summary of current ASD model classes with relevant advantages and disadvantages.
Full size table

An essential process during brain development is the genesis and differentiation of neural stem cells (neurogenesis), which can be captured using hPSCs. Neurogenesis describes the emergence of various differentiated brain cell types from neural stem cells and neural progenitor cells (NPCs). Both precursor cell types are important for the formation of the neurons and glia that populate the cerebral cortex, and can be classified based on their mitotic state, location, and polarity (apical or basal). The polarity of an NPC or neural stem cell reflects the positioning of essential proteins and organelles such as the Golgi apparatus and can influence the cell fate and diversity of daughter cells. Disruptions in cell polarity have been associated with a number of NDDs including Fragile X, SCZ, and ASD [29,30,31,32].

NPCs differ from neural stem cells in that their pluripotent fate is more restricted; they have limited proliferation and are capable of producing most neural and glial cell types in the CNS. Given the frequent presentations of macrocephaly amongst ASD individuals, it is possible that excessive neural growth is an underlying factor that may contribute towards ASD pathogenesis, which has been examined using hPSC-derived NPCs [33,34,35,36].

NPC proliferation has been characterized using patient-derived hPSCs and can even be used to stratify subpopulations amongst patients with ASD. In a recent study, hPSCs derived from an ASD cohort that had either idiopathic ASD or a 16p11.2 microdeletion were used to generate NPCs to examine proliferating pathways. The team found that hPSCs derived from macrocephalic individuals with either the 16p11.2 deletion or idiopathic ASD showed increased proliferation and DNA synthesis and proliferation, whereas the remaining probands displayed the opposite trend [37]. The lines were categorized as hyperproliferative and hypoproliferative NPCs and were then treated with basic fibroblast growth factor (bFGF), a mitogen response element that can prime cells for cortical progenitor proliferation. Interestingly, NPCs from the hyperproliferative group displayed a blunted response to bFGF, whereas the hypoproliferative group showed an increase in DNA synthesis sensitivity and response to stimulation. This work highlights the complexities of disease modeling and how patient-derived NPCs can be used to identify subpopulations amongst heterogeneous clinical datasets.

hPSC-derived NPCs enable researchers to examine complex biological processes relating to proliferation and neuronal differentiation. Directed differentiation can produce robust cultures that can be subjected to high-throughput screening, drug testing and phenotyping [38]. Terminal differentiation can be achieved through the addition of various compounds and transcription factors to broaden the window of development that is examined [39,40,41,42], an example being the Ngn2 system that produces glutamatergic-like excitatory neurons. Two major caveats of this system are the reductive and overly simplistic 2D nature of the cultures, and their short lifespans. Unlike hPSCs, NPCs can only be passaged a discreet number of times, limiting the scalability of the model.

Although hPSC-derived monolayer cultures have deepened our understanding of CNS development, function, and pathology, 2D spatial-organizational constraints limit their ability to model three-dimensional (3D) tissue architecture with complex cell-cell and cell-extracellular matrix (ECM) interactions [43, 44]. Advances in stem cell technology over the past decade have led to the emergence of 3D self-organizing brain organoid models that recapitulate key cellular, structural, and circuital features of human development and disease (summarized in Table 2) [45]. These models were first pioneered with the use of hPSCs by Dr. Yoshiki Sasai’s group with the generation of cortical tissues and 3D optic cup structures in the early 2000s, which has since broadened towards established protocols that exploit either intrinsic or extrinsic signaling pathways to coax the differentiation of hPSCs toward cellular lineages reminiscent of whole or region-specific brain development, respectively [46] (Fig. 1).

Table 2 Overview of hPSC-derived ASD genetic modeling.
Full size table
Fig. 1: 2D and 3D Neural modeling opportunities from hPSC cultures.
figure 1

Schematic representation of human pluripotent stem cell (hPSC) differentiation into neural organoids. The cell-type diversity is highlighted in the right panel, and the organoid and assembloid opportunities are shown below.

Full size image

Unguided approaches rely on the spontaneous differentiation of hPSCs to ECM-embedded heterogeneous cerebral tissue [47]. The resultant unguided neural organoids (UNOs), formerly known as cerebral organoids [48], exhibit discrete regionalization reminiscent of in vivo human whole-brain development, such as markers of forebrain, midbrain, hindbrain, dorsal cortex, prefrontal cortex, hippocampal, occipital lobe, ventral forebrain, choroid plexus, meningeal, and retinal identity [47, 49]. Furthermore, they demonstrate cellular and structural features unique to human cortical progenitor zone organization, such as layers resemblant of ventricular and subventricular zones [49]. Epigenomic and single-cell transcriptomic analyses of UNO tissue have revealed a remarkable similarity to the early developing fetal cortex [50, 51]. However, undirected hPSC differentiation inherently results in stochastic organoid cellular composition that can hinders batch reproducibility [49]. Alternatively, guided organoid approaches incorporate exogenous signaling factors to direct hPSC differentiation towards region-specific lineages, such as those present in the cerebral cortex [52,53,54,55], forebrain [56,57,58,59], medial ganglionic eminence [54], midbrain [56, 60], thalamus [61], striatum [62], pituitary [63], hypothalamus [56, 64, 65], choroid plexus [66], cerebellum [67], brainstem [68], and spinal cord [69, 70]. These organoids generally display less batch-to-batch heterogeneity than their undirected counterparts and therefore may be more conducive to quantitative analyses [46]. Guided organoids have even been combined to generate assembloids comprised of different brain regions, which provide incredible promise to study pathophysiology within affected circuits.

Brain structure, assembloids, and circuitry

Broad structural and circuit abnormalities have been identified in multiple brain regions of individuals with ASD. In addition to generalized macrocephaly, ASD brains can have structural abnormalities within areas of higher order cognitive processing such as the cerebellum, frontal lobe, and limbic system [71, 72], and even manifest in enlargement of the ventricular cavities where newborn neurons originate [73, 74]. Due to a lack of standardized clinical imaging and EEG recordings, it is impossible to know how pervasive these brain abnormalities are within ASD populations alone, but recent population studies have estimated a frequency of 30–50% [75]. These structural changes are often subtle and variable across individuals with ASD, suggesting that dysregulated circuitry between affected regions and altered molecular pathways may be the underlying cause to this presentation [76,77,78].

Unguided neural organoids (UNOs) have been used to model both microcephaly and macrocephaly in disease contexts [49, 56]. A primary example includes studies of PTEN variants in ASD populations that are comorbid for macrocephaly [79]. Loss of PTEN function was investigated in UNOs by use of isogenic hPSC homozygous mutant lines; concordant with the loss of function mutations found in NDD populations with macrocephaly, the UNOs displayed an increase in size across development, in addition to aberrant tissue folding identified through light sheet microscopy [79]. More recently, PTEN gene-dosage sensitivity was assessed by comparing the isogenic KO to a lentiviral overexpression hPSC line, to model the 10q23.31 microduplication associated with patients with autosomal dominant primary microcephaly. Here it was found that UNO size was inversely proportional to PTEN expression, and could be rescued by an AKT inhibitor that acts on a known PTEN pathway [80]. This demonstrates the use of UNOs to model whole-brain structural abnormalities in ASD populations, and how they can be mined for pharmacological rescue.

Due to the developmental nature of the disorder, as well as the multiple brain regions affected, ASD may arise from miswiring amongst neural circuits during fetal development, with an enrichment in the developing cortex. Advances in guided neural organoids (GNOs, or brain-region-specific organoids) have enabled investigation into how different areas of the brain interact in a disease model. When merged, the resulting assembloids provide the necessary environment for cell-cell interactions and complex developmental processes including integration into circuitry. Assembloids can be examined to assess gross structural abnormalities, the migration of neuronal subpopulations, as well as inter-assembloid circuitry. Assembloids have included the combination of cortical (dorsal) and ventral forebrain organoids [58], cortical-thalamus [61], cortical-striatum [62], cortical-subpial spheroids [81], and even tri-part assembloids consisting of cortical-spinal fused to skeletal muscle assembloids [70]. These new model systems allow for the de novo generation of synthetic circuits in the lab, which have been shown to generate spontaneous neural oscillations comparable to that of the developing human brain [82].

Migration and circuit-based disruptions have been described in multiple ASD models, and in patients are often identified through MRI of gross structural abnormalities in the brain or inferred from EEG recordings of epileptic or paroxysmal activity [75, 83, 84] (reviewed in refs. [85,86,87]). EEG abnormalities include an increased frequency of focal spikes, or localized activity to a particular area of the brain [84, 88, 89]. Several wavelength frequency abnormalities have been characterized within ASD cohorts, including an increase in low-frequency (delta and theta) and high frequency (beta and gamma) wavelengths which is contrasted by a reduction in mid-range alpha frequencies, producing a characteristic U-shaped electrophysiological profile, in which the extremities of the power spectrum are enhanced in ASD populations and the mid-range values are reduced. Organoids generate many of the neural stem cell populations and mature cell types in the brain, and are capable of producing many of the EEG wavelengths mentioned above, in addition to increased firing rate, burst frequency, synchronicity, and population spiking across several months of development [45]. These qualities make organoids a promising model to examine functional aspects of ASD in a developing model. Assembloid systems are likely capable of complex neural activity and ossciations [90], and importantly can be used to probe for the innervation and migration of specific cell types in order to assess cellular circuitry between distinct brain regions.

Due to the nature of these tools and an inability to examine ASD pathology at a cellular resolution, the causative cell populations remain unknown. One potential cell population that may drive these global abnormalities are GABAergic interneurons, which are known to regulate the power of upper and lower- frequencies in the developing brain [91]. It is possible that disruptions in the connectivity of these and other cell types in the fetal brain are what produce the epileptiform changes, which can occur through local miswiring or the failure of a cell population to migrate to its intended destination. It should be noted that these processes arise in early fetal development and occur well before the postnatal time point of clinical assessments such as MRI and EEG in ASD populations.

Neuronal migration is an essential process in the developing brain, where excitatory cells emerge from the ventricular zone to create laminar structures in a well-defined, spatiotemporal manner. Recently, it has been found that a subpopulation of inhibitory neurons is also born from cortical progenitor cells, a phenomenon that appears to be human-specific [92, 93]. The remaining inhibitory neurons follow later in development to emerge from proliferative zones in the ventral telencephalon to migrate into the cortex [94,95,96]. This migratory process is well characterized in the human brain and known to be disrupted in NDDs such as ASD, Tourette Syndrome, and epilepsy [97]. Interneuron migration was recently investigated in human 3D organoid models of Timothy syndrome (TS), a severe neurodevelopmental disorder caused by mutations in the calcium channel, LTCC. Using patient-derived forebrain assembloids composed of cortical (dorsal) and subpallium (ventral) organoids, the researchers were able to identify disruptions in GABAergic interneuron migration originating from the ventral organoid; specifically, their saltatory movements were more frequent but less efficient, moving a lesser distance than control lines. Abnormal calcium signaling was thought to underlie the migration defects, and targeted pharmacological activation of the mutated calcium channel was found to rescue the migration phenotype. Importantly, this abnormality was found exclusively in assembloid-derived ventral organoids, and not in ventral organoids alone, demonstrating the utility of this system in modeling complex circuitry.

Circuitry-based disruptions have also been described more broadly in copy number variant (CNV) models of ASD (15q11.3, 15q13.3, 22q11.2, 22q13.3, 1q21.1) [98,99,100,101,102], which frequently include epileptic comorbidities. These functional deficits in ASD have been explained as an imbalance in the ratio of excitatory: inhibitory cells and have been explored in a cohort of NDD patients with the known ASD-related CNV, 22q13.3. Importantly, this deletion encompasses a lead ASD-risk gene, SHANK3, which is highly expressed in human striatal tissue and has been implicated in corticostriatal circuitry disruptions in ASD individuals [62, 103]. Using an assembloid model of cortical organoids fused to striatal organoids, the group examined the axonal innervation from glutamatergic excitatory neurons into the striatum, which functionally connected to medium spiny neurons in the striatum organoid, similar to a developing brain [62]. Patient-derived assembloids were sliced or dissociated for single-cell patch clamp and calcium signaling, respectively, and both assays showed a hyperexcitable phenotype. Interestingly, this change was not present in individual striatal organoids, demonstrating the importance of assembloid modeling to capture complex circuit-based abnormalities in ASD models.

Migration defects can result in a failure for cellular integration and may contribute to downstream disruptions in cell signaling and activity that is present in ASD models. Disruptions in cell circuitry can be probed on a functional level using tools such as single-cell patch electrophysiology, multielectrode array (MEA) recordings of large neuronal populations, and [two-photon] calcium imaging to record abnormalities in firing patterns. These techniques have been used in the aforementioned migration studies as a means of complimenting the findings, as well as in epileptic studies [90, 104,105,106] to profile synaptic activity in a more sophisticated manner to include recordings of neural oscillations in both single organoids and assembloid systems [82, 90, 107, 108]. Using MEA recordings paired with traditional fMRI and calcium imaging, researchers have identified epileptiform changes that are unique to the brains of those with ASD and epilepsy [109,110,111], and have even used the models to explore unconventional pharmaceutical rescue of these abnormalities [90]. This method has also been used to show dosage-dependent responses of brain organoids to convulsant and antiepileptic compounds in seizure liability drug screens [108], demonstrating the utility of this system as a putative translational medicine tool. It is with these tools that researchers and clinicians can begin to understand circuitry pathogenesis in ASD to better provide targeted therapeutics (Fig. 2). hPSC modeling has truly revolutionized the field of ASD research and has enabled scientists to examine the pathophysiology from the top (phenotyping broad structural and growth components) to the integrated circuitry between different cell types, all the way down to the core mechanistic components to the disorder (Table 2).

Fig. 2: Applications of hPSC modeling in ASD populations: from the benchtop to the clinic.
figure 2

Schematic representation of human pluripotent stem cell (hPSC) modeling of autism spectrum disorder (ASD). The top panel describes modeling avenues as genetic, clinical, or environmental, which can be assessed on the molecular, biological, and functional level (bottom left panel). Therapeutic outcomes from each of these assessments are described in the bottom right panel.

Full size image

Despite the versatility of recording neural activity in 3D models, they are not without caveats. Techniques such as MEA and electrophysiology are difficult to scale up, and often record superficial neuronal populations on the direct surface of the organoid. Furthermore, a clear caveat comes from the assembloid system itself, as it includes the merging of regionalized systems that are created independently (and artificially) rather than together as with a truly developing brain. Despite being able to generate innervation and achieve neuronal migration in a biologically relevant manner, the order in which these processes occur does not represent that of the fetal brain and would require a more sophisticated approach (concurrently guiding merged organoids/co-cultures) to better represent the complex development of neural circuitry in a fetal brain. It is possible that the artificial timing is what hinders production of rare neural cell types and more complex circuitry [112]. There is also considerable variability in the cell types produced that later participate in organoid circuitry; select long-term organoid cultures have been shown to produce cortical progenitor-derived interneurons [82] however this is not the case for all protocols [113]. The absence of these cell types may result from a failure to reach maturation, or a lack of guidance factors provided by neighboring cell types or directed in the neural medium [114]. Current organoid protocols differ in terms of media components, extracellular matrix use, embedding, shaking, and even so far as nomenclature itself. The latter of which has recently been addressed in a joint call for a standardized naming system within the larger organoid community [48], and will likely expand to include more universal protocols in years to come.

Modeling environmental insults in ASD using brain organoids

Prenatal environmental insults, comprising either acquired (e.g., infection, substance use, heavy metal exposure, etc.) or inherent (e.g., vitamin deficiencies, stress, diabetes, etc.) pregnancy and birth complications, have been increasingly linked to NDDs including ASD [115]. However, evidence and specificity of these associations is mostly observational. Organoid systems offer unrestricted temporal access to early human neurodevelopmental milestones to examine these epidemiological associations in vitro through perturbation studies (summarized in Table 3) [115].

Table 3 Overview of known and associated environmental insults to ASD modeled using neural organoids.
Full size table

Since 2016, 3D hPSC models have been used to study the effect of infectious agents on early brain development. Maternal infection during pregnancy has been associated with an increased risk of ASD in offspring [116], and strong pathogen-host specificity has previously hindered the potential of traditional animal models to reliably recapitulate human transplacental and intrauterine infection [115]. The advantage of brain organoids to model environmental insults first became apparent in 2016, when the cellular basis of Zika virus (ZIKV)-associated microencephaly was investigated in human UNOs [117,118,119]. ZIKVBR-infected organoids exhibited a reduced growth rate and average growth area compared to mock-infected controls [119], which allowed investigators to provide supporting evidence of a causal link between the 2015 Brazilian ZIKV outbreak and increased incidences of congenital brain malformations in the surrounding population. More recently, organoid studies have provided critical insight into the virulence and putative cellular tropisms of SARS-CoV-2 infection in the developing brain [120,121,122,123], as well as potential therapeutic strategies [124]. Other groups have used UNOs or GNOs to explore the consequences of ToRCH infections (e.g., toxoplasmosis [125], cytomegalovirus [126], herpes simplex virus [127], human immunodeficiency virus [128]) on early neurodevelopment [122].

Growing epidemiological evidence implicates substance use and in utero chemical exposure with an increased risk for ASD [129]. To date, various groups have modeled the developmental effect of early exposure to chemical substances using brain organoids, including alcohol [130], nicotine [131], cocaine [132], heavy metals [133], valproic acid [134], and diesel particulate matter [135]. For instance, UNOs continuously exposed to ethanol from day 10–30 of differentiation exhibited increased apoptosis, impaired neurogenesis, and attenuated neurite outgrowth [136]. Furthermore, 2-month UNOs exposed to alcohol levels comparable to binge drinking displayed apoptosis in a cell-type-specific manner, increased metabolic stress, and altered gene expression in key pathways implicated in various neurological diseases [130]. Scalable organoid-based toxicological screens have also shown promise in identifying and assessing the cellular basis of species-specific neurotoxicity [137].

Parental factors and pregnancy complications have also been associated with ASD diagnosis [138]. Organoids allow researchers to investigate the influence of discrete environmental stressors in a controlled environment and precise genetic background, such as maternal stress (e.g., induced by glucocorticoid hormones [139]) and birth complications (e.g., hypoxia [140]). However, despite their clear advantages, organoid modeling of environmental programming is constrained by their inherent limitations. Groups should thoroughly consider the biological implications of missing cell diversity and circuitry during project design and interpretation. For example, 3D cellular models lack intrinsic maternal protective barriers (e.g., placenta, blood-brain separation, xenobiotic inactivation, etc.) that play an essential role in preserving neurodevelopment against environmental insults and may themselves be influenced during environmental programming [115]. Likewise, current environmental perturbation studies using a direct application of a given insult (e.g., toxin) to organoids is not physiologically representative and likely causes exacerbated effects. However, future studies could improve this using organoid transplanted into rodent models where physiological concentrations and drug metabolic processing may be better obtained.

Profiling developmental trajectories across time

Two of the major barriers to ASD therapeutics are a lack of available biomarkers and a poor understanding of disease trajectory. Prior to the use of human-derived hPSC modeling, scientists were limited to postmortem brain tissue to identify neural biomarkers for ASD, which are subject to degradation and often depict less relevant developmental timepoints. Less invasive imaging techniques can be used at more pertinent timepoints, and even across a developmental continuum, however their low resolution fails to capture relevant biological pathways and the repeated measures across time are often between different individuals and underpowered to provide a conclusive understanding of the brain in a disease context. This has been addressed in recent longitudinal imaging studies [141, 142], but require more patient representation to capture the spectrum of the disorder. Capturing the disease trajectory is further complicated by the time at which ASD diagnoses occur; behavioral abnormalities are likely present well after the disorder and many important neurodevelopmental processes are established [2]. For example, deficits in neuron migration or differentiation may be identified at a later time point, but the causative cell population or biological pathways will remain unidentified using basic clinical assessments. Similarly, many critical synaptic pruning events that are disrupted in ASD and identified through MRI are undetectable by adulthood [143], indicating a critical developmental window that must be more thoroughly examined. The identification of vulnerable windows in development can better guide when populations should be assessed for biomarkers. Should biomarkers be identified during a pre-symptomatic period, at-risk children could be better supported, which is critical when considering the increased risk factor for neuropsychiatric illnesses for individuals with ASD diagnosis later in adulthood [144]. 3D organoids faithfully produce cell types in a spatiotemporal manner similar to that of a developing brain, and so disruptions in any of these processes can be assessed spatially, functionally, and through multi-omic approaches (Fig. 2).

Bulk RNA sequencing can be used to elucidate affected pathways in both pre and postnatal development to identify mechanisms underlying alterations in developmental trajectories [111, 145, 146]. In 1- to 3-month-old cortical GNOs, bulk RNA transcriptomic signatures demonstrated that 1-month-old organoids most closely resembled early mid-fetal (13–16 gestational weeks) through late mid-fetal (19–2 gestational weeks) periods, whereas 3-month-old organoids capture profiles of late mid-fetal (19–24 gestational weeks) through neonatal-early infancy (up to 6 months postnatal) developmental periods. The ability of this model to mirror developmental windows across time lends itself to studying early developmental processes, particularly prenatal neurodevelopment that were previously inaccessible using human samples. Downstream patient clinical information can then be combined with these models to guide core developmental questions surrounding neurodevelopment.

Bulk RNA sequencing was also used in patient-derived forebrain assembloids from individuals with Timothy Syndrome, a monogenic form of ASD. Using ventral tissue at multiple timepoints, the team identified alterations in GABAergic signaling at early differentiation stages where they had previously identified disruptions in interneuron migration [57]. They then used weighted gene co-expression network analysis (WGCNA) after gene set enrichment analysis to generate modules of highly correlated genes within the datasets. From these modules, they concluded that GABAergic signaling disruption was likely due to dysregulation of calcium signaling, which was rescued pharmacologically. Taken together, this transcriptomics approach identified early windows in development that are impaired in models of TS syndrome and further distilled the disruptions down to core mechanistic pathways amendable to pharmacological rescue.

Similarly, RNA sequencing in patient-derived brain organoids was used in the common 22q11.2 CNV to profile developmental processes across 100 days of development. This CNV presents highly variable clinical presentations, ranging from cardiac impairments to general developmental delays [147]. Multiple timepoints were used to capture disruptions in biologically relevant pathways such as pattern specification, NPC proliferation, membrane potential regulation, and glial differentiation [148]. The authors were able to identify biologically relevant windows sensitive to neuronal excitability, which they corroborated with functional assays such as single-cell electrophysiology and calcium imaging. Despite the high variability in clinical presentations, these cellular phenotypes remained consistent across multiple patient-derived lines and were even recapitulated with the heterozygous KO of a single gene, DGCR8. This demonstrates the versatility of using organoids to explore potential therapeutic avenues and driver genes within CNVs.

Bulk transcriptomics can identify unbiased biological pathways and biomarkers that may otherwise be missed with conventional phenotyping methods such as immunofluorescence or single gene expression. One notable disadvantage to the technique is the homogenization of highly heterogeneous tissue to capture the average global gene expression; in doing so, subtle intracellular signatures among heterogeneous populations are overlooked. Single-cell RNA sequencing (scRNA Seq) is an alternative transcriptomic approach to capture cell-type identity and individual transcriptomic profiles and trajectories over time. Downstream analyses have been aided with the release of publicly available databases, which include hPSCs, ESCs, embryoid bodies, and neural organoids at multiple developmental timepoints [149, 150]. Information from these databases can be combined with spatial anatomical tools such as the Allen Brain Atlas to provide reliable cell-type identification and pseudotemporal gene expression alignment for identification of developmental trajectories. More recently, scRNA Seq has been combined with lineage tracing inducible CRISPR technology, termed iTracer, to identify cell lineage dynamics and clonality across UNO development [151]. This technique introduces a barcoding library to identify cell types from an hPSC pool, which is retained in daughter cells throughout division and differentiation, and when paired with an inducible CRISPR scar can track lineage dynamics during a discrete window of time. This can putatively be used to identify small windows of changes to cellular fate during UNO development and can be complemented with techniques such as 4D light sheet microscopy to track migration of daughter cells and newly generated neurons. This platform identified lineage disruptions in a neurodevelopmental dysplasia KO model [151], which was paired with 4D spatial sequencing to show disruptions in brain regionalization consistent with these lineage disruptions. Using sophisticated lineage tracing in a heterogeneous human model enables us to ask questions about population-specific dynamics throughout space and time—formerly something that was restricted to animal models.

Understanding transcriptomics on a single-cell level compliments the diversity of cell types that organoids can produce and has helped establish vulnerable cell-type populations to ASD [152, 153]. The exact localization of these cell types in an ASD model has yet to be fully established but can be aided with the use of spatial sequencing to determine the cytoarchitectural microenvironment within individuals with ASD. A recent study examined the adult human cortex using 10x Genomics’ barcoding platform, Visium, to generate spatial maps of gene expression within the human dorsolateral prefrontal cortex. When this dataset was integrated with other NDD datasets, including those of ASD patients, there was a profound layer-specific enrichment of known ASD genes, highlighting the need to understand ASD genetics both spatially and functionally [154]. Defining cellular transcriptomes with spatial resolution is especially relevant when using 3D models that establish brain laminar structure and distinct cytoarchitecture. To this end, spatial transcriptomics lends an unbiased perspective on cell-type-specific abnormalities through generation of spatial gene maps that, when paired with imaging techniques such as MRI, could help delineate structural abnormalities and underlying circuit defects not identified through scRNA Seq alone. Cell population microenvironments can even be probed for activity-dependent pathways to help establish the affected circuits and their regionalization. Spatial transcriptomics have been used in organoids to establish neural lineage dynamics with spatial resolution (iTracer), neurodevelopmental patterning factors [155], and can be paired with fluorescent tagging to isolate or identify specific regionalization within heterogeneous organoid or diseased tissue [156].

The use of transcriptomic profiling provides powerful information about cell identity, lineage, and localization. Analysis pipelines enable the user to infer cell trajectory, intercellular communication [157], and can even be used to predict drug response [158, 159]. Recent developments in electrophysiology can also allow a glimpse into the synaptic activity of a given cell via Patch-Seq, a modified version of whole-cell patch-clamp electrophysiology that enables transcriptomic capturing as well as morphological rendering of a given cell. This three-in-one platform provides comprehensive information about the functionality of a cell as well as its operative biological pathways [160] and could have powerful implications in understanding ASD pathophysiology. It is a low-throughput alternative, however, and should be used selectivity within cell populations that are known to be disrupted in the disorder.

Another tool to capture functionality across time includes multielectrode arrays, which are capable of recording neural populations in 2D and intact 3D cultures across development. Importantly, these recordings are done in an unbiased manner to capture population-wide recordings and synchronous activity of diverse neural populations [107, 108]. MEA recordings in cortical organoids have been shown to correlate with that of human preterm neonatal EEG signatures [82], and can therefore provide a glimpse into the network activity of ASD populations during critical developmental windows. Of note, these oscillations can even be captured in assembloid systems [90], offering the ability to capture inter-organoid circuitry, generalized EEG patterns between both organoids, as well as focal signatures to a particular brain region [after stimulating the other]. Knowledge of how particular brain regions are affected functionally can help clinicians decide which pharmaceutical approaches may be most beneficial to their patients [161,162,163]. They can provide screening opportunities for clinicians to modify pharmaceutical compounds in a controlled environment to target ASD-specific pathways [82, 164,165,166], and can even identify causative driver genes that can be targeted by AAV- or ASO- based gene therapy [166,167,168,169,170,171,172,173]. Following refinement and rigorous testing, patient-derived neural organoids can be used to enhance and personalize cell therapies, gene therapy, and drug discovery, thereby accelerating their transition from the benchtop to the clinic (Fig. 2).

Limitations to human modeling and future directions

2D and 3D human models have made enormous progress in the past decade; with the emergence of stem cell reprogramming, patient-derived skin and blood samples are now capable of producing hPSCs that can later go on to mimic general and brain-region-specific processes. These models have great potential for clinical applications and to understand the mechanisms of ASD pathology. They are not without their caveats, however, which include limitations to growth, tissue maturity, and an absence of vascularization and external input from the peripheral nervous system (summarized in Table 1).

Brain organoids have undergone extensive transcriptomic profiling to show the presence of many different brain-region cell types that emerge in a spatial-temporal manner [44]. Multiple studies have revealed the persistence of a stem cell niche alongside these mature cell types, which supports the use of brain organoids to model fetal development [111]. The presence of this niche is unique to organoids and makes late-stage developmental modeling difficult to achieve. In addition, long-term cultures are further hindered by a lack of vascularization and nutrient flow to the inner organoid core [174], which is compounded by the absence of the blood-brain barrier and its inclusion of immune cells such as microglia. This is especially a limitation to modeling autism, as microglia are a proposed vulnerable cell type within ASD and are thought to contribute towards its immunopathology [175, 176]. Luckily this caveat has been addressed with the introduction of blood vessel organoids that provide vascularization networks at the cellular level, which in turn increase NPC populations, and introduce microglia into the environment [177, 178]. This is incredibly important given the prenatal time point where neurovascularization occurs, the human-specific expression pattern of vascular cells, and its influence on brain structure and development [179].

Microglia populations have also been incorporated into growing organoids through direct co-culture or merging of NPCs and primitive macrophages, which are capable of synaptic pruning and phagocytic activity once mature. These models can be used to investigate the effects of the immune environment on brain pathology [180, 181]. The addition of microglia would provide critical developmental cues to all cell types in the organoids, while supplying a cellular substrate to understand how neuroinflammatory processes occur in NDDs. For example, over-pruning of synapses is one type of deleterious function of abnormal microglia that have yet to be modeled in organoids and would allow complex modeling in 3D.

Despite the enormous progress in modeling specific brain regions through guided differentiation, an element of the CNS that has been underrepresented in human ASD research is the eye. Multiple NDDs are associated with vision disorders, and there has been tremendous advancement of retinal organoid protocols. Retina morphogenesis is a highly regulated process both temporally and spatially, and much like the developing brain requires a stem cell niche that is present in early development [182,183,184]. Individuals who are blind are at least ten times as likely to have ASD, and clinical studies have shown comorbid vision impairment within ASD populations, although the underlying pathogenesis between these two conditions remains unexplained [185, 186]. Retinal organoids are capable of producing retinal pigment epithelia and functional photoreceptors, and their application to ASD modeling would provide novel insights into how retinal development may be impaired and later give rise to visual impairment and dysfunction. Further, the emergence of retinal-cortical assembloids [187], provides the necessary tool to study eye-brain connections in NDDs to understand how dysfunctional sensory input and function may arise.

In recent years, growing evidence suggests the importance of exploring ASD models beyond the CNS due to the high proportion of sensory dysfunctions reported amongst individuals with ASD (as high as 90% [188]). Mouse and fly studies highlight the role of the somatosensory nervous system in ASD sensory and behavioral deficits [189,190,191,192]. Building from the knowledge gained by 2D cultures, dorsal root ganglion-like organoids [193] and neuromuscular organoids [194, 195] have emerged and offer a promising opportunity to investigate the role of the PNS in ASD pathophysiology, such as altered sensory functioning. This approach would integrate external input into what has traditionally been a CNS-exclusive model, providing a more complete understanding of ASD pathogenesis.

An exciting model to examine developmental biology more comprehensively, and with theoretical inclusion of all the systems noted above, include synthetic embryos, or embryoids. These novel systems produce gastrulating embryo-like structures that are capable of undergoing organogenesis [196, 197]. While prototypes have been formed in mouse ESCs, and can only reach 8 days of development, optimizations in a human background may achieve month-long growth periods that would enable scientists a novel glimpse into human fetal development (reviewed in ref. [198]). An important consideration, however, would be the inclusion of stimuli to mimic true gastrulation both within and external to the womb. Notwithstanding, these exciting advancements also give rise to several important ethical considerations. While there have been some preliminary discussions concerning hPSC-based ethics in research [199,200,201,202], these discussions must be formalized, informed by science, and made jointly between experts in the field, policymakers, and activists in order to develop appropriate universal standards.

While each of the models discussed in this review provide novel insight into neural development and circuitry, they remain undoubtedly limited by the natural heterogeneity within both the model itself and the clinical pathophysiology of ASD. It is more likely that these models will provide a starting point for understanding ASD pathogenesis that, when coupled with a multitude of animal and clinical modeling, may ultimately result in a therapeutic breakthrough [203]. Importantly, we also acknowledge the essential role that members of the ASD community play in conducting thoughtful and meaningful research. Self-advocates have expressed their need for improved social support systems, and we hope that the incorporation of those personally affected into decision making will bolster the research done at the bench, and ultimately provide a more comprehensive and compassionate approach to addressing ASD therapeutics and clinical outcomes [204,205,206,207,208]. Given the accelerated pace of brain organoid research over the last few years, this human and patient-specific model system will undoubtedly play a critical role in helping to develop future therapies.