Spatial CRISPR Screening Just Made Your Preclinical Models Look Like Guesswork

Picture a pathologist holding a tumor biopsy under a microscope. She can see exactly where the malignant cells cluster, which immune cells are crowding the margin, and how the stroma is reorganizing around an invasive front. Now picture a genomicist in the next room grinding that same biopsy into a single-cell suspension, running it through a CRISPR pooled screen, and producing a ranked list of gene hits. The pathologist’s spatial context — the who-is-next-to-whom that determines how a drug actually behaves — has been thrown in the trash. That tension between spatial biology and dissociation-based screening has haunted preclinical oncology for two decades.

A paper published this month in Nature Biotechnology closes that gap in a way that should force a rethink across every oncology program moving toward IND-enabling studies.

The method is called Perturb-DBiT. It co-sequences single guide RNAs and total RNA transcriptomes within the same intact tissue section, allowing researchers to observe what a CRISPR perturbation actually does in a native tissue environment — not in a dissociated cell pellet, not in a 2D culture well, but in the architectural context where tumor biology actually happens. The team deployed libraries of over 80,000 sgRNAs across mouse and human tissue models, including immune-competent and metastatic cancer models. The scale alone redraws what we thought was technically feasible.

The Problem Dissociation Was Always Hiding

To understand why this matters, start with what traditional CRISPR pooled screens systematically destroy. Tissue dissociation — the step where you enzymatically disaggregate a biopsy into individual cells before sequencing — severs every extracellular and cell-cell interaction that was operating in the original tissue. As documented in spatial CRISPR genomics research from the UK Health Security Agency Research Portal, this process prevents identification of gene functions that operate extracellularly or depend on specific tissue organization. In plain terms: you find out what a gene does to a cell in isolation. You do not find out what it does to the tissue.

That distinction has enormous downstream consequences for trial design.

Oncology’s Phase II attrition rate has remained stubbornly above 60% for fifteen years, and a significant fraction of that failure traces back to preclinical models that predicted single-cell responses without accounting for the microenvironmental signals that determine whether a tumor cell actually dies when a drug hits it. A target that looks compelling in a dissociated screen may be suppressed, compensated for, or irrelevant when neighboring stromal cells are present. Sponsors have been designing first-in-human doses, selecting patient biomarker populations, and writing adaptive randomization rules based on biology observed outside of its native context. Perturb-DBiT does not fix that retroactively — but it makes the excuse for continuing that practice disappear.

The Broad Institute’s development of Perturb-FISH, which combines imaging-based spatial transcriptomics with large-scale detection of CRISPR guide RNAs, signaled the direction the field was moving. What the Nature Biotechnology paper achieves is a sequencing-based implementation that handles total RNA — not just a targeted gene panel — and operates at a scale of over 80,000 sgRNAs simultaneously. That is the difference between a spatial hint and a spatial screen.

What the Tour of the Tissue Reveals

Walk through what Perturb-DBiT actually shows a sponsor’s preclinical team. In immune-competent cancer models, the method maps not only which perturbations alter tumor cell transcriptomes, but where in the tissue those effects concentrate, which cell populations are responding, and whether the perturbation is creating a local intercellular signal that travels to neighboring cells. That last capability — detecting functional transcriptional circuits that operate between cells — was previously invisible to sequencing-based screens. You had to infer it from imaging experiments conducted separately, in a different model, on a different day, by a different team.

Spatial co-registration of perturbation and response in a single experiment collapses a multi-step inference chain into a single observation.

For metastatic models specifically, this matters in ways that connect directly to clinical endpoints. Metastatic lesions are architecturally distinct from primary tumors. The stromal composition differs. The immune infiltrate differs. A gene that functions as a vulnerability in the primary site may be upregulated by the metastatic microenvironment as a resistance mechanism. Sponsors designing biomarker strategies for metastatic indications have been selecting targets based on primary tumor screens, then wondering why the response rates in their Phase II metastatic cohorts disappoint. Perturb-DBiT applied in metastatic tissue models does not guarantee a better answer — but it asks the right question in the right place for the first time.

The technology confirmed by Perturb-FISH research published in PubMed relies on in situ amplification of guide RNAs combined with imaging spatial transcriptomics, establishing that optical detection of perturbations in intact tissue is technically tractable. What the Nature Biotechnology paper adds is the total RNA dimension — every expressed gene in the cell, not just a targeted readout — which is what allows unbiased discovery rather than hypothesis confirmation. The distinction between confirming a target and discovering one is, operationally, the difference between a biomarker enrichment trial and a broad first-in-class program.

The Regulatory Implication Nobody Is Discussing

Here is the counterintuitive read that the field is not yet having. The conventional assumption is that spatial biology tools are research instruments — interesting for publication, useful for understanding mechanism, but one step removed from the regulatory package. The IND enables the trial; the preclinical data support the IND; and the preclinical data come from whatever models were available when the program was advancing. Spatial CRISPR screens are therefore treated as a future refinement, not a current obligation.

That logic holds until you examine what the FDA actually asks for in a Type B meeting when a first-in-human oncology program has a biomarker-selected population. The agency wants to understand the mechanistic basis for the selection criteria. If the sponsor’s target validation rests entirely on dissociated-cell screens, and a reviewer asks why the target was assumed to function identically in the tumor microenvironment, the answer has historically been “that is the best data available.” After June 2026, that answer is no longer accurate.

Perturb-DBiT was validated in human tissue models. It is a sequencing-based method, not an exotic imaging platform requiring specialized infrastructure that only three labs in the world can operate. The barrier to adoption is not technical inaccessibility — it is the inertia of preclinical workflows that were designed around dissociation because dissociation was all that existed.

Sponsors advancing immuno-oncology programs with tumor microenvironment hypotheses, cell therapy developers designing for solid tumor penetration, and rare disease programs targeting tissue-specific gene function all face the same asymmetric risk: the cost of incorporating Perturb-DBiT into IND-enabling work is a workflow addition; the cost of not incorporating it is a Phase II cohort that was selected on biology that was never actually measured in context. The FDA has not yet issued guidance requiring spatial validation — but the agency has issued guidance on the sufficiency of mechanistic evidence, and the threshold for “sufficient” just moved.

The question for every oncology program director reviewing their preclinical package this quarter is not whether spatial CRISPR screening will eventually become standard. It already is. The question is which programs submitted their INDs before the standard changed — and which ones will have to explain that gap to a reviewer who just read the same Nature Biotechnology paper.

References

1. Nature Biotechnology — “Spatial CRISPR screens map total RNA in tissue”
2. UKHSA Research Portal — “Spatial CRISPR Genomics Identifies Regulators of the Tumor Microenvironment”
3. Broad Institute — “New technology puts a spatial lens on CRISPR screening” (Perturb-FISH)
4. PubMed — “Simultaneous CRISPR screening and spatial transcriptomics reveals intracellular, intercellular, and functional transcriptional circuits”