neuroDEVELOPMENTS: Neuroscience Research to Clinical Relevance

New techniques to decipher autism’s complexity

Ron McKay, PhD, Chief Editor
Lieber Institute for Brain Development
Maltz Research Laboratories

Following an initial focus on genetic risk for psychiatric disorders widely distributed in the human population, contemporary studies increasingly emphasize the role of rare de novo somatic mutation in both coding and regulatory regions of the genome. In this issue, our newsletter highlights the promise and challenges of these advances. De novo mutation in autism spectrum disorder (ASD) was first convincingly shown using exome- and other-directed sequencing strategies. Here, in our first paper, technical advance using ultra-deep whole genome sequencing reveals the prevalence of mosaic somatic mutation in the cortex of ASD patients and controls (Rodin, Dou et al 2021). Alongside the advances in sequencing chemistry and data analytics that better define mutation, the consequential altered phenotype is being directly assessed by recombineering the human risk genes in a growing number of models. To illustrate the speed of advance in phenotyping, we focus secondly on a recent study where the growing power of CRISPR-based gene engineering is employed to systematically alter multiple risk loci and assess phenotypic mechanisms in mouse embryos (Jin et al 2020). In this pioneering study the authors focused on de novo mutation occurring within coding regions predicted to be highly penetrant. These papers and the comments of our Board members emphasize the continued rapid advance linking both germ line and de novo mutation to phenotype in human neurodevelopmental disorders.

--RM

neuroDEVELOPMENTS Editorial Board

Fred 'Rusty' Gage, PhD
President, The Salk Institute for Biological Studies

Daniel Geschwind, MD, PhD
Professor, UCLA School of Medicine

Elizabeth Grove, PhD
Professor, University of Chicago

Jürgen Knoblich, PhD

Director, Institute of Molecular Biotechnology, Austrian Academy of Sciences

Arnold Kriegstein, MD
Professor, UCSF

Pat Levitt, PhD
Professor, Keck School of Medicine of USC

Mu-Ming Poo, PhD
Director, Institute of Neuroscience, Chinese Academy of Sciences

John Rubenstein, MD, PhD
Professor of Psychiatry, UCSF

Nenad Sestan, MD, PhD
Professor, Yale University

Flora Vaccarino, MD
Professor, Yale University

Chris Walsh, MD, PhD
Chief, Division of Genetics & Genomics, Boston Children's Hospital

______________________________________

Venkata S. Mattay, MD
Managing Editor

Michele Solis, PhD
Science Writer

neuroDEVELOPMENTS

Starting to make sense

Though the question of finding genetic risk for autism is straightforward enough, the answers are not. Genetic studies of autism have detected many, many loci associated with risk for the disorder, and at last count, these pointed to at least 102 genes, implicated by rare de novo mutations (Satterstrom et al 2020) and inherited variants as well (Ruzzo et al 2019). A steady progression of genomic studies has identified different kinds of risk — common or rare, copy number variants or point mutations, exonic or regulatory — at work. Identifying the full catalog of genomic risk for autism or other neurodevelopmental disorders requires repeated forays into the genome, with new techniques in hand.

A new level of genetic complexity has entered the mix. Though textbooks teach us that all cells in our bodies contain the same DNA sequence, it turns out this is not necessarily the case. Researchers are finding that, in the brain, some cells carry mutations that set them apart from the rest. Though these discoveries don’t conform with dogma, they are hardly a surprise. The genome is replicated and split, over and over, during development, as a zygote develops into a many-celled organism. The flurry of cell divisions during embryogenesis gives rise to many billion neurons and non-neuronal cells. A cell carrying a mutation can pass it onto the daughter cells it creates upon division; this can leave a trail of cells carrying the same mutation, like related family members in a family tree.

Thus, the brain can contain cells that do not have identical genomes. This state of affairs is termed “somatic mosaicism” because it arises in non-germline somatic cells, and because it affects some but not all cells. The potential for this kind of DNA patchwork has been appreciated for some time, and now with new techniques, researchers are making headway toward detecting these DNA deviations and understanding whether they matter for brain function.

This issue of neuroDEVELOPMENTS highlights two recent papers that outline important approaches for making sense of the genetic variation detected in autism. One analyzes many post-mortem brain samples to estimate how often somatic mosaicism arises, what parts of the genome are susceptible, and whether brains from people with autism show a different pattern from neurotypical controls (Rodin, Dou et al 2021). As the catalog for mutation in autism swells, another study describes a method for understanding the action of many risk genes acting together to perturb brain development in intact mouse embryos (Jin et al 2020). While it is one thing to identify the many points of risk in the genome, it is quite another to link these to phenotype in a functioning brain.

“The timing is therefore important, with earlier mutations potentially having a larger functional impact.” – Flora Vaccarino

“The sheer number of risk genes being identified in psychiatric disorders invites the use of high throughput methods for testing function in dozens or hundreds of loci at a time, rather than the typical in-depth study of one gene at a time.” – Daniel Geschwind

“[I]f the mutations are present in multiple lineages, especially blood cells, then they can be detected – with the caveat that one can’t be certain that the same mutation is also in brain cells, let alone what part of the brain and in what types of brain cells.” –John Rubenstein

“One of the most interesting aspects of this paper for us was seeing that such enhancer mutations occur in normal people in the mosaic state, as well as in ASD brain, as though nature is constantly tinkering with different gene dosages in different parts of the brain.” – Christopher Walsh

Mosaics that matter

For the most part, the brain cells we’re born with are the ones we live with. So any somatic mosaicism that develops along the way will be with us for the rest of our lives. To understand how much somatic mutation occurs in the developing human brain, the first study extracted DNA from prefrontal cortex samples from 59 donors with autism spectrum disorder (ASD) and 15 control donors. Led by Christopher Walsh at Boston Children’s Hospital (and neuroDEVELOPMENTS editorial board member), the researchers applied ultra-deep whole-genome sequencing to these DNA samples in order to detect true point mutations anywhere in the genome. The massive amount of sequencing revealed a startling amount of somatic mutation: 26 single-nucleotide variants (SNVs) per brain, present in at least 4% of cells, and found in coding and regulatory regions alike.

First authors Rachel Rodin and Yanmei Dou and colleagues estimated 2-3 mutations occurred per cell division, making this a rather regular occurrence. With this rate, the whole body is projected to contain about 80 somatic single nucleotide variants in over 2% of cells. Half of people are predicted to have at least one somatic mutation in the brain that can potentially alter function. These parameters of prevalence match closely to those found in embryonic brain (Bae et al 2018) that we have discussed previously, and show that somatic mosaicism is a long-lasting part of the human genomic scenery.

As we have seen in previous newsletters, defining the role of mutation in psychiatric disorders has been the focus of major multi-center initiatives. The NIH Brain Somatic Mosaicism Network (McConnell et al 2017) has been central to the current study. Comparisons between autism and control samples found that, in autism, somatic mutations tended to land in enhancer regions of the genome. Enhancers are sequences that regulate gene expression, and so play a role in specifying cell types, and thus, specific tissues. Disruptions to enhancer function, in a mosaic subset of cells, could help explain brain-specific changes to gene expression that are not seen in other tissues. They may also help account for some of the phenotypic variability found in autism and other neurodevelopmental disorders.

Figure 1 Risk in Brain-Active Enhancers A) ASD brains show a propensity for somatic mutations that hit enhancers active in the brain. The x-axis lists the different types of regulatory elements, and the y-axis shows the odds ratio of finding a somatic mutation in one of these in ASD versus neurotypical control brains. B) Enhancer regions hit by somatic mutation are more frequently close to transcription start sites (TSSs) of genes with brain-specific expression (red) than any somatic mutation found near a TSS (blue). Figure from Rodin, Dou et al., 2021.

Perturbing the path to function

With thousands of genetic loci implicated in different human diseases, and the occurrence of many mutations in some affected individuals, means that the one-at-a-time approach of developing knockout models to understand their function will fall short. Researchers need systematic methods for studying the functions of many genes in intact brains. A new paper from Paola Arlotta of Harvard University and colleagues have provided a starting place to ask questions of groups of genes with their new platform, called Perturb-Seq. Combining CRISPR-Cas9 genome editing with early brain development in mice, the researchers evaluated 35 risk genes for autism or other neurodevelopmental disorders.

First author Xin Jin and colleagues introduced mutations to the 35 risk genes into the brains of embryonic mice. They did this by injecting lentiviruses loaded with guide RNAs into the embryonic mouse brain in utero, where the cargo was taken up by stem cells lining the ventricular space. This produced different brain cells carrying a genetic perturbation to one or more of these 35 genes. At postnatal day 7, the researchers examined the brain cells with single cell RNA sequencing to give a read out of mutation and phenotype in specific cell types. To get a high-level view of changes in gene expression, the researchers looked at 14 modules of genes whose expression co-varied. They found that nine of the perturbed genes had significant effects on five of these modules in cortical neurons, astrocytes, and oligodendrocytes.

This distillation of consequences on gene expression in mice may well translate to humans. Eight of the gene modules found in the mice were conserved in human brain tissue, and overlap was detected between gene expression changes in the Perturb-Seq mouse data and that found in cells from autism brain donors. This suggests that the many risk loci identified for autism converge onto a limited set of biological pathways that can now be systematically explored in mouse and human models.

“In many ways, the ‘hit’ on multiple cell types by a single gene is both disturbing and revealing – the developmental impacts are complex.” – Pat Levitt

“Progress in understanding the etiology of a disease as complex as autism is likely to depend on identifying convergent signaling pathways affected by multiple gene perturbations associated with ASD.” – Arnold Kriegstein

Figure 2 Of mice and men Fourteen human genes show differential expression (DE) in autism brains compared to controls (listed on the y-axis; the extent of DE in human brain is shown on the x-axis, with the red dots indicating the log Fold-change (FC)). Perturb-seq data from mice show that perturbing mouse versions of ASD genes sometimes resulted in similar changes in expression of some of the fourteen gene mouse orthologs (black dots). Note the pronounced decreases in expression in blue for NRN1 and SST. Figure from Jin et al., 2020.

Piecing it together

Understanding the biological basis of autism and other neurodevelopmental disorders is not easy, but these papers make advances on two fronts. One expands the catalog of genetic variation operating in autism by adding somatic mosaicism, which makes a small but potentially important contribution to risk; this may help explain some of the heterogeneity of symptoms surrounding this and other neurodevelopmental disorders. The other offers a way to understand the function of many genes at a time in many cells at a time, and future work will more fully connect changes in gene expression to other aspects of brain function. Together the two approaches tighten the connections between genotype and phenotype, improve the biological understanding of neurodevelopmental disorders, and inform ideas about how to better treat them.

###

The full commentary from our Editorial Board on the two papers highlighted in this issue of neuroDEVELOPMENTS.

1. From Flora Vaccarino, MD, Yale University:
Using the most sensitive techniques—single cell sequencing and cloning—both the Walsh and our group have shown that the total number of somatic SNV per cells is around 500 to 1,000 per cell already at birth. Due to limited discovery power, bulk RNA sequencing will only reveal the most frequent mutations—those present in more that 1-2% of the cells and which arise early in development. These are only the tip of the iceberg--however, they are often the most important functionally. Somatic mutations arising earlier in development are going to be present in a larger fraction of cells, and may also be shared by several tissues. The timing is therefore important, with earlier mutations potentially having a larger functional impact. The finding that in brains of individuals with an ASD diagnosis these early mutations are more often located in enhancer elements is intriguing. Enhancers regulate cell type and region specific gene expression through the binding of transcription factors. Depending on their location within these enhancers, somatic mutations may change the affinity of transcription factor binding and alter gene expression in a specific cell type with potential consequences on cell fate or cellular function. This points to the importance of mapping brain-specific enhancers in human brain development and understanding their role in governing gene expression.

2. From John Rubenstein, MD, PhD, University of California, San Francisco:
Rodin et al highlights the need to consider de novo somatic mutation as a mechanism in causing a broad spectrum of neurodevelopmental disorders, beyond what is known that somatic mutations causing malignancies (e.g. retinoblastomas), and hyperplasia resulting from somatic mutations at the PTEN, AKT and TSC loci. Their findings show that standard genotyping methods only reveal germline genetic information, which sets the stage for genetic risk, but is only part of the story for some affected individuals. An individual who is a germline heterozygote for an ASD allele may be at increased risk for ASD, if somatic mutation of the wild-type allele is present in a sufficient number of brain cells, in which case it would cause a homozygosity phenotype in part of the brain. Unfortunately, if the somatic mutations are restricted to the CNS lineages, it seems unlikely that diagnosticians can uncover these mutations in living individuals; however, if the mutations are present in multiple lineages, especially blood cells, then they can be detected – with the caveat that one can’t be certain that the same mutation is also in brain cells, let alone what part of the brain and in what types of brain cells. Importantly, this advance opens the door to greater understanding of the genetic mechanisms that work together to increase risk for neuropsychiatric disorders. Further, it implies that we need to understand both the effect of heterozygosity and homozygosity of ASD risk genes on brain development and function.

Jin et al likewise presents an important conceptual change in the way scientists approach genetic disorders that are caused by many genes. The Perturb-seq method screens the effects of mutating tens of genes in a single developing brain – in this case in a mouse. The approach can be applied to other systems, such as human cortical organoids. The Perturb-seq yields high quality data and insights into transcriptomic phenotypes. On the other hand, I don’t think that it replaces standard genetic analyses on single genes, as these are more likely to give a deeper understanding of molecular, developmental, cellular, neurophysiological and behavioral phenotypes.

3. From Pat Levitt, PhD, University of Southern California:
With 20+ years and multiple laboratories contributing to the concept of somatic mutation as a normal component of non-heritable DNA modification, the paper by Rodin et al makes contributions that create new directions for ongoing research efforts regarding the genetic etiology of neurodevelopmental disorders. They demonstrate clear advantages of applying ultra-deep sequencing methods to address what are, in totality, events that impact a very small percentage of neural cells. The ‘so what’ question is moving closer to being addressable directly. The strategies of such sequencing approaches with interrogating clonal somatic mutations brings greater resolution to issue of whether there are specific, and perhaps even unique tissue/cell type patterns of genomic somatic mutations that arise early in development. As if biology didn’t have sufficient options for enhancing diversity at the cellular level, the experimental data reported in Rodin et al demonstrate that these mutations indeed can arise early, and eventually may lead to studies in experimental models that provide a sense of the degree to which this activity influences typical neurodevelopmental processes. Additionally, while the prevalence of somatic mutations in control and autism spectrum disorder (ASD)-derived dorsolateral prefrontal cortex did not differ, the enrichment in enhancer regions of the genome create additional opportunities to determine the impact of potentially altered gene regulation in both developmental and functional processes. Given that ~75-80%+ of ASD diagnoses are likely to be due to variant combinations in non-coding, regulatory regions, this finding is encouraging. The authors note that in some sense, because of the limited sample size, this is an exploration in providing greater definition to future strategies that can be used to understand the origins of cellular diversity developmentally, and the contributions of the somatic mutation landscape to pathophysiological processes. Application in other neuropsychiatric disorders of developmental origin will be an exciting initiative to undertake.

The excitement around the methodology applied in the Jin et al paper, using a novel ‘Perturb-seq’ strategy, may overshadow another advance that has broad implications for experimental approaches to determining the role that specific genes play in neurodevelopment. We tend to utilize animal models with germline transmission of gene-targeted deletions as a way to screen, somewhat broadly and superficially, for biological impact. We oftentimes look under the lamp post for phenotypic disruptions, focusing on a particular cellular or functional component, limited by time and our own technical capabilities. Perturb-seq data reported by Jin et al demonstrates the potential for taking a different approach, identifying specific cellular elements that exhibit distinct changes in transcriptome modules based on single cell RNA sequencing. In many ways, the ‘hit’ on multiple cell types by a single gene is both disturbing and revealing – the developmental impacts are complex. But Perturb-seq will provide important guidance for probing deeply the outcome of specific cellular elements (more than one or two). With the use of gene editing and single cell analytics, as described in the paper, investigators can more rapidly focus their efforts on testing what they believe are the most relevant cell populations for deciphering the biological impact of human genomic variants on brain development. This will be exciting and likely reveal new neurodevelopmental relations in the regulation of histogenesis, differentiation and circuit formation and maturation.

4. From Arnold Kriegstein, MD, University of California, San Francisco:
The list of risk genes for ASD keeps growing, but determining how and where in the developing brain the presence of these mutations perturb normal function is also a growing challenge. A major step forward would be determining the cell type and the stage of brain development where dysfunction begins. The Jin et al study presents a novel experimental model that can help shed light on this problem, but how well does the mouse model match human brain development and how well does it highlight pathways affected in human disease?

The Perturb-Seq data suggest that perturbations in the Chd8 and sGatad2b gene, associated with ASD, significantly decreases the expression of a gene module, ODC1, which is related to the production of oligodendrocytes and would be predicted to accelerate oligodendrocyte maturation. Interestingly, myelinating oligodendrocytes were found to be increased in number in a Chd8 germline heterozygous mutant mouse model, validating the perturb-seq results. But given the obvious differences between cellular composition and organizational complexity between mouse and human brain, can the Perturb-Seq results really inform us about a human disease such as autism? Jin and colleagues addressed this question by comparing their results to a single nucleus RNA-seq dataset of postmortem ASD brain samples. They found that two genes, SST and NRN1 that were decreased in interneurons and excitatory neurons respectively from ASD patients also showed decreased expression in their dataset. This convergent result suggests that the Perturb-Seq data has the potential to shed light on human diseases. Progress in understanding the etiology of a disease as complex as autism is likely to depend on identifying convergent signaling pathways affected by multiple gene perturbations associated with ASD. The Perturb-Seq approach may help discover some of these pathways.

5. From Daniel Geschwind, MD, PhD, UCLA School of Medicine:
The sheer number of risk genes being identified in psychiatric disorders invites the use of high throughput methods for testing function in dozens or hundreds of loci at a time, rather than the typical in-depth study of one gene at a time. Gene editing approaches involving CRISPR provide a powerful approach (Kampmann 2020), but are challenging, especially when conducted in vivo. Jin et al.’s work is very important in providing the first proof of principle for the value and potential power of pooled genomic screening approaches to test ASD candidate genes in high throughput in vivo in mouse using Perturb-seq.

This is a technically demanding experimental approach. A key aspect of their experimental design is that because for some of the genes and cell types there are not many observations, so rather than focusing on individual gene perturbations, they work at the pathway level, using WGNCA to identify modules of co-expressed genes. WGCNA, which has been widely applied in bulk tissue (Parikshak et al 2015), is quite noisy in single cell data alone, requiring integration with bulk data to simulate the types of high-quality modules defined previously (Morabito et al 2021). So, it is surprising and potentially important that the authors are able to define viable modules across cell types in these data. Further, because single cell data is so noisy, often meta-cells are created by combining subsets of cells to reduce the noise and improve network reproducibility. Here, they use single cells without pooling as input, and apply non-standard parameters in the WGCNA, removing modules driven by outlier cells. Understanding more about how these steps were conducted, and how network power and other parameters were chosen would be very helpful in guiding future studies. Understanding the details of this analysis and others is important.

One particularly interesting result is the identification of a module related to Ndnf+ interneurons (and Ank2 perturbation). Ndnf+ cells are a type of neurogliaform inhibitory cell in layer 1 that modulates excitatory neurons across multiple layers, essentially regulating cortical column function and the sensory state of the animal (Cohen-Kashi Malina et al 2021). This cell type has not been widely studied, but is potentially very interesting with regards to modulation of sensory processing, a phenotype strongly connected to autism. Another interesting observation is the finding that CHD8 deletion impacts oligodendrocyte progenitor development, which is not predicted from human expression data (Polioudakis et al 2019), but is consistent with detailed functional data from mouse in vivo (Marie et al 2018). It will be important to validate these findings in human tissue or in vitro systems.

Lastly, the authors compare these results in mouse to expression profiles from human brain. One surprising result is that they find only 14 genes differentially expressed (DE) between ASD and controls in 3 major cell types at an FDR of 0.2, which is orders of magnitude less than the original paper (Velemeshev et al.) and in bulk tissue studies that preceded that work (Parikshak et al 2016, Voineagu et al 2011). Despite the small numbers of DE genes, the authors note that SST and NRN1 are reduced across their perturbations and in ASD post mortem brain. It should be emphasized that it is unclear whether any of the overlap observed is statistically significant, and these results are presented somewhat anecdotally. But, previous papers have shown substantial reduction in interneuron markers in ASD brain, including PARV and SST, which are among the largest changes observed in post mortem ASD cortex (e.g. Parikshak et al 2016, Voineagu et al 2011). These data in Jin et al. are quite interesting in that they provide a potential avenue for understanding how these in vivo expression profiles result and what biological states they represent in post mortem brain. For example, the authors can now use these perturbation models to understand how perturbation of ASD risk genes in excitatory neurons or other cell types, influences the activity and expression profiles of interneurons, which perhaps implicates local inhibitory-excitation alterations in ASD.

6. From Christopher Walsh, MD, PhD, Boston Children’s Hospital:
One of the most striking aspects of autism genetics is that so many of the mutations are heterozygous, such that the patient still has a remaining functional copy of the gene, suggesting that the disorder reflects exquisite sensitivity of the brain to the dosage, or expression level, of many different genes. While recessive mutations, affecting both gene copies, cause at least 3% of autism cases, heterozygous mutation of just one copy is an even more common scenario, and the same seems true for schizophrenia. This dosage sensitivity is also apparently reflected in the importance of copy number variants—deletions and duplications of multiple genes—which are the second most common identifiable mutation type in ASD after de novo heterozygous point mutations. And it seems like for many genes, too much of the gene can be as damaging to the brain as too little, suggesting that evolution has carefully selected just the right level of gene expression that it wants, within a very narrow range.

The evidence from Rodin et al, as well as from several other excellent recent papers, are starting to elucidate that mutations in gene enhancers, which control where and when a gene is expressed, are likely to be important in many cases of ASD that are as yet undiagnosed. This makes perfect intuitive sense given this broader dependence of the phenotype on gene expression level. Enhancer mutations, and other noncoding mutations, are devilishly difficult to identify with statistical evidence, because the noncoding genome is so vast, and each gene has multiple enhancers. Therefore, the study sample sizes needed to confidently identify specific noncoding elements important in ASD will be even larger than what has already been analyzed.

One of the most interesting aspects of this paper for us was seeing that such enhancer mutations occur in normal people in the mosaic state, as well as in ASD brain, as though nature is constantly tinkering with different gene dosages in different parts of the brain. These noncoding mutations, since they leave the coding sequence of the gene intact and so result in a normal protein, can either boost gene expression in the case of some mutations, or other mutations might impair gene expression. A common question is whether the “genomic diversity” of neurons created by mosaic mutations is ever positive, as opposed to only being associated with disease. In principle certain enhancer mutations could be positive in some people, while different enhancer mutations might be disease-causing in some cases of ASD.

Given the diversity of genetic mechanisms and causes implicated in ASD, the Perturb-seq paper by Paola Arlotta and colleagues could not come at a better time. It offers a hope and potential way forward to analyzing large numbers of genes at scale, both by presenting ways to disrupt them in vivo, and ways to study the results of that in vivo disruption. With the clear importance of levels or doses of gene expression revealed by the genetics, the idea of monitoring gene effects by measuring levels of gene expression makes a lot of sense. And the finding that many distinct genes may converge on a relatively smaller set of mechanistic pathways provides a hopeful sign for potentially finding therapies that might impact a pathway, rather than having to engineer a separate solution for hundreds of different genes.

Forward