When melanoma patients on the targeted drug vemurafenib stop responding to treatment, the cancer has usually found a workaround. The problem is that dozens of different genetic changes can each independently fuel that resistance, making it nearly impossible to predict which escape route a tumor will take. A team at Rockefeller University has now built a screening platform, called PerturbFate, that tackles the problem head-on: it silences more than 140 resistance-linked genes one at a time across a population of over 300,000 individual cells, then watches what happens inside each cell at two molecular levels simultaneously.
The results, published in Nature and described in a Rockefeller University release in May 2026, reveal that many genetically distinct paths to drug resistance funnel through a small number of shared regulatory control points. The researchers say they plan to apply the same approach to other cancers and to neurodegenerative diseases including Alzheimer’s, though experimental work outside cultured melanoma cells has not yet been published.
Why existing screens fell short
PerturbFate belongs to a family of methods known as Perturb-seq, which pair CRISPR gene editing with single-cell sequencing to measure the effect of genetic changes one cell at a time. Earlier versions of Perturb-seq primarily captured a snapshot of which genes were active in a cell at a single moment. That is useful, but it misses a deeper layer: the physical packaging of DNA, called chromatin, which controls whether genes can be switched on or off in the first place.
A precursor technique called compressed Perturb-seq expanded the number of perturbations that could be tested in one experiment by applying compressed sensing to deconvolve pooled signals. A separate method, Multiome Perturb-seq, showed that chromatin accessibility and gene expression could be read at the same time in CRISPR screens. PerturbFate integrates both advances and adds something neither offered: trajectory inference, a computational strategy rooted in the Monocle framework that reconstructs the sequence of regulatory events a cell passes through on its way to a new state. Instead of a static picture, researchers get a movie of how a cell transitions from drug-sensitive to drug-resistant.
What the melanoma experiment found
The Rockefeller team, based in the Laboratory of Single-Cell Genomics and Population Dynamics, selected 143 genes previously linked to vemurafenib resistance in BRAF-mutant melanoma. Using CRISPRi, they silenced each gene individually across a large pool of cells, then captured two readouts from every cell: how open its chromatin was at regulatory regions, and which genes were actively being transcribed into fresh RNA.
By tracing how each knockdown altered chromatin and transcription over pseudotime, the team identified what they call convergent regulatory nodes. In practical terms, that means many genetically distinct resistance trajectories passed through a small number of shared bottlenecks. One gene highlighted in this analysis was VEGFC, a growth factor well known in vascular biology but now implicated as a hub in the resistance network. If multiple escape routes depend on the same regulatory factor, blocking that factor could, in principle, cut off several resistance paths at once.
For context, vemurafenib resistance is not a niche problem. BRAF mutations drive roughly half of all melanomas, and while vemurafenib and related drugs initially shrink tumors in most of those patients, resistance typically emerges within months. Current clinical strategies combine BRAF inhibitors with MEK inhibitors to delay resistance, but many patients still relapse. A tool that maps the shared wiring behind multiple resistance mechanisms could help identify new drug targets or combination strategies.
The Alzheimer’s angle, and its limits
The lab’s stated research agenda lists neurodegeneration alongside cancer as a primary focus, and the Rockefeller release frames PerturbFate as a platform that could be aimed at diseases like Alzheimer’s. The scientific logic is straightforward: neurodegenerative diseases involve gradual shifts in cell state driven by regulatory networks, exactly the kind of process that trajectory-aware perturbation screens are designed to dissect. If you could silence dozens of Alzheimer’s risk genes in neurons or brain organoids and watch how each one reshapes chromatin and transcription over time, you might find convergent nodes analogous to the ones identified in melanoma.
But that work has not been done yet. No published dataset shows PerturbFate applied to neurons, brain organoids, or animal models of neurodegeneration. The connection to Alzheimer’s rests on the lab’s research direction and on a reasonable mechanistic argument, not on completed experiments. Readers should treat it as an informed projection rather than a demonstrated result.
Separate studies have shown that perturbation screens can be extended to living animals using adeno-associated virus (AAV) vectors to deliver CRISPR perturbations in vivo. Those experiments used different designs and did not employ PerturbFate’s dual-readout approach. Bridging the chromatin-plus-RNA measurement into living brain tissue will require solving additional technical challenges around vector delivery, sequencing depth, and guide capture in complex tissues where many cell types are intermixed.
Open questions worth tracking
Several uncertainties remain. The Nature study does not include head-to-head benchmarks comparing PerturbFate’s accuracy and efficiency against Multiome Perturb-seq or compressed Perturb-seq. It is unclear how sensitive the trajectory inference is to noise in chromatin data, or how reliably convergent nodes can be detected when the effect of a given gene knockdown is subtle. The convergent-node finding is compelling within the melanoma dataset, but whether the same regulatory hubs appear in other cancer types is an open question that will require replication.
Scalability and cost also matter. Capturing both chromatin accessibility and nascent RNA at single-cell resolution while tracking CRISPR guides is sequencing-intensive. The Nature paper demonstrates feasibility at the scale of hundreds of thousands of cells, but it does not fully detail how the method would perform in larger, more heterogeneous samples such as primary patient tumors or mixed brain tissue. Practical deployment in translational settings will depend on whether similar insights can be obtained with smaller cell numbers or more targeted gene panels without losing the ability to resolve trajectories.
No independent group has yet replicated the PerturbFate workflow, which is typical for a newly published method but worth noting. Performance claims should be treated with appropriate caution until outside labs validate the approach in their own systems.
What this means for the field right now
The verified achievement is specific and significant: PerturbFate provides a way to connect genetic perturbations to dynamic regulatory changes in a melanoma model of drug resistance, revealing shared control points that could inform future therapies. It represents a genuine technical advance over earlier single-cell screening methods by integrating chromatin measurement, transcription measurement, and trajectory inference into a single workflow.
The broader promise, that similar strategies could illuminate the regulatory wiring behind other cancers and complex brain diseases like Alzheimer’s, is scientifically plausible but unproven. The gap between a successful in vitro screen in melanoma cells and actionable insights in neurodegeneration is wide, spanning different cell types, delivery challenges, and disease biology. Closing that gap will take years of additional work, and the results are not guaranteed.
For now, PerturbFate is best understood as a powerful new lens for studying how cells change state under genetic pressure. Where that lens gets pointed next, and what it reveals, will determine whether the tool lives up to its broadest ambitions.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
