Salk Institute scientists have published Superb-seq, a laboratory method that, they say, reads CRISPR genome edits — including unwanted off-target cuts — and the resulting changes in gene activity, one cell at a time. The work appears in Nature Methods on 16 November 2026, under the title “Joint single-cell capture of Cas9 edits and transcriptomes reveals on- and off-target effects on gene expression.” The Salk Institute described the method the same day.
It is a method tested in cultured cell lines, not a therapy and not a clinical result.
What Superb-seq does, and what the Salk team measured
Current tools, Salk says, can show where unwanted CRISPR edits sit in a sample, but not which individual cells carry those edits, or how the edits change the cell’s behaviour. Superb-seq is designed to do both. The Nature Methods paper says the method “captures single-cell edits and transcriptome effects by integrating Cas9 editing, T7 in situ transcription and single-cell RNA sequencing.”
According to the Salk Institute, “the Salk team tested it on 34,000 cells from three different cell lines. In one group of 10,000 human leukemia cells grown in the lab, Superb-seq detected 43 edit sites—36 of which were off-target—and 5,761 edited cells.” The paper identifies that 10,000-cell experiment as K562 cells, targeted with seven guide RNAs against four chromatin-remodeller genes, and reports the same 5,761 edited cells and 36 off-target sites.
Graham McVicker, an associate professor at Salk and senior author of the study, said: “I knew off-targets were an issue, but they were far more frequent than I was expecting.”
The paper also says Superb-seq improved estimates of how edits change gene expression compared with Perturb-seq, a predecessor that reads gene activity and guide RNAs but not the resulting edits. Nineteen off-target edits were associated with differential gene expression, nine of them in a cell-type-specific way. Companion analysis software, Sheriff, is available on GitHub.
Suggested uses, and what the authors have disclosed
Salk presents four possible uses: basic research, drug development, personalised medicine, and training machine-learning models. Co-first author Mickey Lorenzini said the tool “could be immensely helpful in establishing the safety of genome therapy candidates early on in development.” McVicker also suggested it could generate better single-cell data for models that predict how cells respond to perturbations. Those are the institute’s suggestions, not results from a treated patient or a trained clinical model.
The authors have filed a patent. The Salk release says McVicker, co-first author Brad Balderson and Lorenzini “have filed a patent for Superb-seq.” The paper’s competing-interests statement adds that the same three authors “are inventors on a patent application related to Superb-seq” and “are founders of Cascadal Inc., which is developing” the technology. Other co-authors include Madelyn Light, Olalekan H. Usman and Gavin Kurgan of Integrated DNA Technologies in Coralville, Iowa. The work was supported by National Institutes of Health grants and by private philanthropy, Salk said.
Superb-seq is a laboratory readout, not a treatment. That places it alongside other recent TSN science files rather than beside a therapy approval. Nature Medicine has just published OpenScientist, an open AI co-scientist tested on existing datasets, another methods paper rather than a clinical deployment. Closer to the genome-instability theme is NYU’s work linking ESCO2 loss to PARP inhibitor sensitivity, and, further back, a study finding that a share of fetal neurons carry large DNA copy-number changes. Those are different experiments. None of them turns Superb-seq into a bedside test.
A sharper map of off-target cuts is not a gene therapy
The firm facts are that a Nature Methods paper has described Superb-seq, that Salk reports tests on 34,000 cells from three cell lines, and that in 10,000 laboratory-grown human leukaemia cells the method found 43 edit sites, 36 of them off-target, and 5,761 edited cells. McVicker’s comment is that off-targets were “far more frequent than I was expecting.” The paper supplies a method — Cas9 editing, T7 in situ transcription and single-cell RNA sequencing — and some further comparisons, including 19 off-target edits linked to changes in gene expression. The authors have a patent application, and three of them have founded a company. Three co-authors are at Integrated DNA Technologies.
None of that is a demonstration that a CRISPR therapy is safer, or that a machine-learning model trained on Superb-seq data works in the clinic. Salk is entitled to say where it thinks the method could help. Those uses remain prospective. The experiments were done in cell lines, including K562 leukaemia cells grown in the laboratory, not in patients. Off-target frequency in that setting is a finding about the method and the guides tested, not a general verdict on every CRISPR application.
What the paper does offer is a way to see, cell by cell, both the intended and the unintended edits and whether they change gene activity. That is a laboratory advance, and it is the right scale at which to read this result. The next official facts to watch are ordinary ones: whether other labs adopt Sheriff and Superb-seq, whether later papers test primary cells or therapeutic candidate guides, and whether Cascadal or Salk describes a product rather than a method. Until then, the news is that a single-cell CRISPR readout has been published, not that genome editing has been made safe.
Sources
- Lorenzini, Balderson, Lavalle et al., “Joint single-cell capture of Cas9 edits and transcriptomes reveals on- and off-target effects on gene expression,” Nature Methods, 16 November 2026
- Salk Institute, “Can genome editing be safer and more effective?,” 16 November 2026
- Sheriff software on GitHub
Get TSN’s biggest tech stories, explained.
News and insights across AI, tech, crypto and cybersecurity. Every source linked. Unsubscribe anytime.
Get the newsletterRelated stories
- Nature Medicine Publishes OpenScientist, an Open AI Co-Scientist is the closest recent TSN account of a laboratory methods paper, and the right next stop for another tool that assists research rather than treating patients.
- NYU Links ESCO2 Loss to PARP Inhibitor Sensitivity
- 26% of Fetal Neurons Carry Big DNA Errors, Falling to 6% After Birth
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
