Brain disorders are a greater cause of disability than any other class of diseases in modern societies. In Section 12.19 of Principles of Neurobiology, we saw how powerfully genes shape this burden: twin studies revealed up to 80% heritability for schizophrenia and bipolar disorder, and 30–40% for major depression and anxiety. Compared to environmental factors, which are multifaceted and hard to track, hope has been placed on identifying genes that contribute to psychiatric disorders as a means to gain new insight into their origins and identify new targets for drug development.
However, getting from genetic variations to mechanistic understanding and rational drug design requires answers to several questions. What is the gene’s role in regulating cell state and intra- and inter-cellular signaling? Which life stage and in which cell type does the gene function? How does disruption of the gene contribute to disease? Technical advances in recent years have enabled scientists to answer these questions with increasing resolution and scale.
Perturb-seq is a powerful technique for understanding the genotype-phenotype map—how perturbing individual genes changes a cell’s gene expression—in a high throughput manner. Using CRISPR-based systems (Box 14-1 in Principles of Neurobiology), one can introduce thousands of perturbations across many cells in a mosaic manner, sort the cells that receive at least one perturbation, and read out both perturbation identity and transcriptome-wide phenotype (mostly through single-nucleus and single-cell sequencing). Perturb-seq as a technique grew up in a dish, where cultured cells offer speed and throughput. However, applying it to study how disease risk factors work requires in vivo context: a cell in a dish is not a cell in a brain. In an intact circuit, the way cells interact in time and space matters a lot.
In a preprint posted in March 2026, Shi et al. generated a genome-scale, in vivo perturbation atlas targeting 1,947 disease-related genes across the whole mouse brain, profiling their cell-autonomous phenotypes with snRNA-seq across 7.7 million cells spanning major brain regions and neuron types (Figure 1). They found that a gene’s role depends strongly on where it lives. Some genes, like those involved in mRNA export and proteostasis, deplete neurons everywhere. Others are needed only by specific populations: non-GABAergic midbrain neurons (MB-HB-Glut-Sero-Dopa) and glutamatergic thalamic neurons (TH-EPI-Glut) turn out to be most susceptible, while others can shrug off the same hits.
Figure 1. In vivo whole brain Perturb-seq workflow (adapted from Shi et al., 2026). Cas9 mice (postnatal day 16) were injected with a library targeting 1,947 disease-related genes. After 3–4 weeks, perturbed nuclei were enriched via GFP (from the vector that produces pooled gene-specific guide RNA, or gRNA) and a neuronal marker NeuN, and over 7.7 million cells were profiled for transcriptome and gRNA identity.
More surprisingly, they found closely related genes expressed in the same cell type can regulate distinct pathways. Take Grin2a and Grin2b: both encode NMDA GluN2 subunits, with partially overlapping expression but distinct developmental trajectories and disease associations. GRIN2A variants are primarily linked to schizophrenia, whereas GRIN2B variants are strongly associated with early-onset conditions including autism and other neurodevelopmental disorders. A longstanding view attributes this divergence largely to their developmental dynamics: Grin2b dominates early cortical development, and Grin2a takes over postnatally as excitatory circuits mature. But Shi et al. found that even in the same cell type (L2/3 cortical excitatory neurons) and at the same developmental stage (early adolescence), knocking out each subunit drives opposing transcriptional programs: Grin2a loss pushes up activity-dependent and presynaptic genes, whereas Grin2b loss elevates genes related to structural remodeling and membrane regulation. This suggests the divergence probably isn't just about when these genes act, but what they do.
The disease-tied perturbation list also validates itself against human genetics: genes with stronger disease evidence and dosage-sensitive (dominant) genes produce the largest transcriptional disruption—the suspects humans most worried about turn out to have the longest rap sheets. Most hopefully, genetically unrelated neurodevelopmental-disorder genes tend to funnel their effects onto shared programs dominated by synaptic, NMDA-centered signaling. If hundreds of distinct mutations route through a few common nodes, targeting those nodes could possibly address many disorders at once.
In short, Shi et al. presented the first brain-wide, cell-type-resolved map of how disease risk genes reshape neuronal gene expression. A map this rich raises more questions than it answers. If two closely related subunits can diverge this much, how many other gene family members are hiding opposing functions? Why are thalamic and midbrain neurons so sensitive? And because this screen is neuron-focused during adolescence, could the next step be to tailor it to a specific question—combining conditional perturbations across developmental time or behavioral context, and adjusting cell-type coverage to fit? For instance, one might extend the readout to non-neuronal populations and into aging to study Alzheimer’s. And you might be thinking about another question.
Reference:
1. Shi, T., Korshunova, M., Kim, S., DeTomaso, D., Zheng, X., Vishvanath, L., Nyasulu, T., Huynh, N., Sun, A., Thompson, P. C., Zhang, Y., Wigdor, E. M., Rohani, N., Ali, S., Qiu, H., Geralt, M., Zhao, Z., Rabhi, S., Yao, Z., . . . Jin, X. (2026). Genome-scale functional mapping of the mammalian whole brain with in vivo Perturb-seq [Preprint]. bioRxiv. https://doi.org/10.64898/2026.03.16.711480