Perturb-seq in skin organoids exposes a biphasic NFKB2 differentiation switch

What the work claims

Squiers et al. report the first Perturb-seq screen carried out in epidermal organoids, a three-dimensional, air-liquid-interface culture that develops the stratified architecture of human skin. The central claim is that this organoid setting captures physiologically relevant differentiation programs that two-dimensional calcium differentiation misses, and that screening transcription-factor knockdowns at single-cell resolution can expose regulators with stage-specific, and sometimes opposing, effects. The headline finding is that NFKB2, a subunit of the non-canonical NF-κB pathway, acts as a biphasic switch: it suppresses the basal-to-spinous transition while later promoting cornification. The authors also argue that gene-expression programs derived from the organoid screen line up with the genetic drivers of human skin diseases, including cutaneous squamous cell carcinoma and ichthyosis.¹

How it works

The model is an air-liquid-interface epidermal organoid grown from Ker-CT keratinocytes, an hTERT- and CDK4-immortalized human foreskin line. Cells are seeded at confluence on transwell inserts, switched to differentiation medium three days later, and airlifted ten days after seeding. Cultures are harvested at 21 days. Histology and immunohistochemistry for KRT5 and KRT10 confirm stratification with basal and differentiated layers.

For the screen, the authors generated a Ker-CT line expressing dCas9-KRAB, infected it with a pooled lentiviral sgRNA library targeting transcription factors with known or suspected roles in epidermal differentiation, and sorted cells by BFP expression. After organoid differentiation, they dissociated the cultures and ran single-cell RNA sequencing using the 10X Chromium Single Cell Gene Expression Flex platform with feature-barcode capture of the sgRNAs. The final dataset contains 236,112 cells. Leiden clustering resolved five transcriptomic states, including basal, spinous/granular-like, and late differentiated populations. A pseudotime trajectory rooted in basal cells recapitulates the expected rise of KRT10 and KRTDAP along differentiation.

To quantify perturbation effects, the authors computed Energy Distance between each transcription-factor knockdown and non-targeting controls. At a p-value threshold of 0.001, they identified 21 significant perturbations. They then decomposed the data into eight gene-expression programs using consensus non-negative matrix factorization. The programs map onto proliferation, cell migration, cell-cycle regulation, translation, cell adhesion, antimicrobial function, and cornification. NFKB2 knockdown produced one of the largest transcriptome shifts. In the organoid, NFKB2 expression is biphasic: high in basal cells, low in spinous cells, and high again in granular cells. This pattern is mirrored in normal human skin immunohistochemistry from the Human Protein Atlas, where NFKB2-positive nuclei are reported at roughly 55% in the basal layer, 12% in the spinous layer, and 97% in the granular layer.

The functional split is the interesting part. NFKB2 knockdown increases Cell Adhesion program scores, consistent with accelerated exit from the basal state. But the same knockdown lowers Cornification program scores, suggesting that NFKB2 is required for the final differentiation step. The authors validated this dual role in an orthogonal two-dimensional calcium-differentiation model: NFKB2 knockdown increased KRT10 expression but the cells did not complete cornification. The screen also links disease gene sets to specific programs: cutaneous squamous cell carcinoma drivers are enriched in the Cell Cycle program, while ichthyosis-associated genes are enriched in late differentiation programs such as Cell Adhesion and Cornification.¹

Where a skeptic should push

The most load-bearing assumption is that the Ker-CT organoid faithfully represents the genetic regulation of native human epidermis. Ker-CT cells are immortalized with hTERT and CDK4, which removes some replicative limits but also alters cell-cycle and stress pathways. The authors compare their organoid expression patterns to published human skin data and find agreement, yet the model remains a single engineered line. A result that holds in one immortalized keratinocyte line is a proof of principle, not proof that the same hierarchy operates in primary patient keratinocytes.

Sample size is another concern, though it takes a different form here. The single-cell dataset is large at 236,112 cells, but those cells come from pooled knockdown cultures whose replicate structure is not described in the preprint. It is unclear how many independent organoid differentiations were performed for each perturbation, or how batch effects were controlled across sequencing runs. Without that information, it is hard to judge whether the Energy Distance clusters reflect biology or technical variation.

The biphasic NFKB2 claim is supported by orthogonal two-dimensional differentiation and by public immunohistochemistry, but the mechanism is still correlational in the organoid. The authors show that NFKB2 levels track with differentiation state and that loss of NFKB2 changes program scores, yet they do not demonstrate which downstream targets mediate the early versus late effects, or whether pharmacologic activation of non-canonical NF-κB can rescue the cornification defect. The preprint also does not report whether the findings replicate in primary human keratinocyte organoids or in organoids from ichthyosis or squamous-cell-carcinoma patients.

Implication for organoid models of human organs and drug discovery

For organoid-based drug discovery, the practical message is that a three-dimensional, stratified skin model can support large-scale genetic screens and yield program-level readouts that are closer to organ physiology than two-dimensional calcium differentiation. That matters because many epidermal disease genes, including those underlying ichthyoses and genodermatoses, act during specific differentiation stages. A screen that only scores live versus dead cells would miss stage-specific regulators such as NFKB2, whose loss both accelerates one transition and blocks another. The ability to map a perturbation onto a pseudotime trajectory and a set of gene-expression programs gives compound and gene-therapy developers a more nuanced target-evaluation framework.

The opportunity is a standardized human epidermal platform for screening modulators of barrier formation, cornification, and squamous differentiation. Toxicology programs could use the KRT10 and Cornification programs as early safety readouts for compounds that inadvertently perturb epidermal maturation. Inherited-skin-disease programs could test whether a candidate therapy restores the late-differentiation programs that are disrupted in ichthyosis gene sets. The authors make the disease relevance explicit by showing that known cSCC drivers load onto the Cell Cycle program and ichthyosis genes load onto Cell Adhesion and Cornification programs.

The threat is the same one that haunts every organoid screen: generalization failure. The finding is demonstrated in one immortalized male foreskin keratinocyte line under one differentiation protocol. If the field rushes to build screens on this exact platform without validating in primary keratinocytes, female lines, different ancestry backgrounds, and patient-derived disease organoids, it risks optimizing compounds for a model-specific phenotype. The second threat is over-interpreting the disease associations. Enrichment of cSCC or ichthyosis genes in organoid programs is suggestive, but it does not show that NFKB2 modulation will help either condition. A drug that pushes cells out of the basal state could promote differentiation in ichthyosis or, conversely, accelerate tumor progression in cSCC, depending on context.

A deeper, less obvious implication is methodological. The authors pair Perturb-seq with organoids, a combination that has been far more common in cancer and neural systems than in epithelial barrier tissues. If this approach works robustly in skin, it is portable to other stratified epithelia, including esophageal, cervical, and corneal organoids. That could expand the range of human organ models that can be screened genetically at single-cell resolution, turning organoids from descriptive models into causal-discovery platforms.

The bottom line

This is a method-development preprint with a strong mechanistic finding rather than a definitive target-validation study. The demonstration that epidermal organoids can support Perturb-seq, and that NFKB2 has a biphasic role in differentiation, is well supported by the single-cell data and by the orthogonal two-dimensional validation. What would confirm the finding is replication in primary human keratinocyte organoids, rescue experiments with non-canonical NF-κB activators, and identification of the direct NFKB2 target genes that mediate early versus late effects. What would weaken it is evidence that the biphasic expression pattern is a culture artefact of the Ker-CT line or that the Cornification defect is caused by an off-target sgRNA effect. For now, the work is best read as a blueprint for moving genetic screens into stratified epithelial organoids and as a reminder that differentiation regulators can have opposite effects at different stages.