Skip to content

Xin Jin (keynote ★) — in vivo Perturb-seq, from a pilot screen to a whole-brain genome-scale CRISPR screen

Speaker: Xin Jin, Scripps Research (confirmed via SCGD26 speaker list) Talk title (official): "Towards a functional brain genome" Bio (as introduced): grad student finishing in Cory Bargmann's lab (Rockefeller) → Harvard Junior Fellow → co-developed in vivo Perturb-seq with Feng Zhang and Paola Arlotta (first prototype: Jin et al., Science 2020). Co-founder of PerturbAI. Slot: SCGD26, Jun 12, 2026 (11:00–11:40 AM EDT) Recording: SCGD26 livestream (YouTube) — verify archived link Status: Captured from session transcript

Thesis

Human genetics has handed neuroscience a long list of large-effect, haploinsufficient (loss-of-function) risk genes for autism / neurodevelopmental disorders / schizophrenia — far more than can be studied one gene at a time. The answer is to scale functional genomics directly in the mammalian brain at single-cell resolution: in vivo Perturb-seq pairs pooled CRISPR perturbation (delivered in utero) with single-cell read-out postnatally, with built-in internal controls that regress out animal-level and library-prep confounds.

Methods & technical advances

  • AAV-CH9 for fast embryonic expression. Lentivirus was the in vivo bottleneck. A repurposed-screen across 86 barcoded AAV serotypes (collaboration, UC Irvine) found AAV-CH9 (originally a Schaffer-lab DNA-shuffling variant from UC Berkeley) expresses fast and early enough for corticogenesis — ~10× less work than the 2020 approach, more cells recovered.
  • Single-nucleus multiomic guide capture. Moving from single-cell to single-nucleus (for ATAC + RNA multiome, adult tissue) broke guide-RNA recovery; ~2 years of optimization (student "Xinhe"). Fix: barcoded, polyadenylated guide transcript + MS2 or PP7 RNA-hairpin capture by a nuclear-anchored binding-protein fusion — both anchoring vectors improved gRNA assignment ~3.8× vs. non-anchoring, without significantly compromising editing. Preprint: In vivo multiomic Perturb-seq with enhanced nuclear gRNA capture, bioRxiv 2026 (vector deposited to Addgene — specific plasmid ID not confirmed, see below).
  • Whole-brain genome-scale screen. With Broad / Allen / "Ox" / Pertrip AI collaborators (leads incl. "Tuo," Maria Kushchenova, Soo Young Kim): 1,947 disease-associated genes, 8,588 sgRNAs, >7.7M cells after QC (per-hemisphere/mouse/litter barcoding). Preprint: Genome-scale functional mapping of the mammalian whole brain with in vivo Perturb-seq, bioRxiv 2026 (PMC). Atlas openly on Hugging Face: perturbai/wholebrain_crispr_atlas (CC-BY-4.0), also released via PerturbAI + the NVIDIA AI Blueprint for single-cell analysis. MOI/cell-number titration keeps ≈1 guide/cell; CRISPR knockout (not CRISPRi) often yields in-frame deletions → nonfunctional protein even when transcript is detectable.

Key results

  • Pilot (Fox g1 / Foxg1): (cortical-development pilot screen — Jin et al., Cell 2024) two independent guides give highly correlated log-fold-change → two-allele LoF support. In layer-6 corticothalamic neurons, knockout de-represses a transcription-factor network (ectopic Bcl11b, Rorb, etc.) → failed fate maintenance / hybrid fate — and this is layer-6-specific (layer-5 IT neurons, same lineage, respond completely differently). Bulk would have missed it.
  • Whole-brain: ~20–30% of perturbation×subtype pairs have a significant effect. Top-vulnerable programs are proteostasis/proteasome (post-mitotic neurons can't dilute proteotoxic stress), RNA splicing, epigenetic regulation. The four top PSMC hits are physically interacting 26S proteasome components — biochemical partners rediscovered genetically.
  • GRIN2A vs GRIN2B vignette: Grin2b perturbs more cell types (cortical + subcortical; severe/early NDD) than Grin2a (cortical, milder/later schizophrenia) — consistent with clinical onset. "Phenocopy" sharing with clathrin-mediated endocytosis genes; some "phen-opposite" cell types where Grin2a and Grin2b drive opposite DE.
  • Perturb-Clear (postdoc "Boli"): tissue clearing + light-sheet + sparse perturbation → manual reconstruction of dendritic cytoarchitecture + long-range connectivity, paired with Perturb-seq. ADNP knockout → basal-branch-specific complexity loss in layer-4 (not oblique/tuft, not layer-2/3), with concordant large DE only in layer-4. Morphology + RNA are concordant overall and more interpretable jointly.

Q&A highlights

  • "Sequencing cost is no longer the blocker." The next bottleneck is which cell-type × perturbation combinations to sample — the space is sparse (some cell types matter far more), so the move is computation-guided iterative sampling of high-effect-size genes, not brute force.
  • On polygenic disease: start with large-effect de novo (haploinsufficient) variants because CRISPR knockout models them well; polygenic effects may still converge on a smaller set of conserved large-effect nodes (cites Jonathan Pritchard).

Connections to the corpus

Sources

  • SCGD26 speaker list (confirms Xin Jin, Scripps Research; title "Towards a functional brain genome"): https://satijalab.org/scgd26/
  • Jin et al., Science 2020 (first in vivo Perturb-seq / ASD genes): https://www.science.org/doi/10.1126/science.aaz6063
  • Jin et al., Cell 2024 (massively parallel in vivo Perturb-seq, cortical development pilot): https://www.cell.com/cell/fulltext/S0092-8674(24)00476-8
  • Whole-brain atlas preprint, bioRxiv 2026: https://www.biorxiv.org/content/10.64898/2026.03.16.711480v1 — https://pmc.ncbi.nlm.nih.gov/articles/PMC13108593/
  • Whole-brain CRISPR atlas on Hugging Face: https://huggingface.co/datasets/perturbai/wholebrain_crispr_atlas
  • Nuclear guide-capture (MS2/PP7) preprint, bioRxiv 2026: https://www.biorxiv.org/content/10.64898/2026.03.15.711739v1

Still unverified

  • Exact lab-member name spellings ("Xinhe," "Tuo," "Maria Kushchenova," "Soo Young Kim," postdoc "Boli"): not confirmed against an authoritative author list (bioRxiv full-text returned 403; searched Google Scholar/bioRxiv abstracts, names not individually verified).
  • Specific Addgene plasmid ID for the nuclear guide-capture/anchoring vector: searched Addgene + the preprint listing; candidate Multiome Perturb-seq sgRNA vectors exist (e.g., pEM040) but could not be confirmed as the exact deposit from this preprint.