Researchers have uncovered a fundamental flaw in how we design CRISPR gene-editing tools: they've been ignoring the physical shape of DNA inside living cells. A team at Imperial College London, the MRC Laboratory of Medical Sciences, and the University of Sheffield found that when DNA twists and coils under natural cellular stress, it becomes far more vulnerable to unintended cuts by CRISPR-Cas9, the molecular scissors used in gene therapy.

This discovery challenges decades of CRISPR design philosophy. Most high-fidelity CRISPR variants have been engineered based on how the enzyme behaves with straight, linear DNA in laboratory conditions. But inside cells, DNA is constantly being bent, stretched, and twisted as genes are transcribed and replicated. One common state is negative supercoiling, where the DNA helix becomes slightly underwound and buckles into loops, much like a spring that coils up when slightly untwisted.

Why Does DNA Shape Matter More Than We Thought?

The problem is expensive and dangerous. Off-target CRISPR edits, where the enzyme cuts DNA sequences it wasn't designed to target, cost the biotech industry between $300 million and $900 million annually in profiling, guide redesign, and treatment delays. For patients, off-target cuts can compromise the safety and efficacy of gene therapies meant to treat conditions like sickle cell anemia and certain cancers.

The Imperial College team made a breakthrough by creating tiny DNA circles called minicircles that could be deliberately supercoiled while remaining small enough to visualize using cryo-electron microscopy. This was a delicate engineering challenge: the circles had to be small enough to reveal atomic-level detail but large enough to maintain a supercoiled state. As one researcher noted, it was a "Goldilocks problem" of finding the sweet spot.

What they observed was striking. The same DNA sequence, when supercoiled, was cut by CRISPR-Cas9, but when the same sequence was in linear form, it remained completely intact. This suggests that many off-target cuts observed in cells occur not because of sequence mismatches alone, but because the DNA is supercoiled.

"The same DNA with the same sequence, in linear form, is not cut by Cas9. It remains completely intact. But if you supercoil it, now it's cut by Cas9," explained Professor David Rueda, Chair in Molecular and Cellular Biophysics at the MRC Laboratory of Medical Sciences.

Professor David Rueda, Chair in Molecular and Cellular Biophysics, MRC Laboratory of Medical Sciences

How Does Supercoiling Change CRISPR's Behavior?

At near-atomic resolution, the researchers observed that when Cas9 binds to supercoiled DNA, the enzyme physically changes its geometry to prepare for cutting. Specifically, the HNH domain, the part of the enzyme responsible for cutting the DNA strand, moves closer to the target site, priming the enzyme to cut. This repositioning doesn't happen with linear DNA.

The team's hypothesis is that as DNA twists and buckles, it reduces the energy required to unzip the double helix, lowering the energy barrier for Cas9 binding and cutting. This facilitates off-target activity. Additionally, the twisting of the DNA helix during supercoiling makes mismatches easier for Cas9 to tolerate. Imperfect sequence matches between the CRISPR guide and DNA would normally stop Cas9 from working, but the new structures revealed new types of mismatches being made, showing new mechanisms of off-target cutting.

When the researchers compared their newly discovered molecular structures of Cas9 bound to supercoiled DNA with previously published structures based on linear DNA, the differences were dramatic. Most earlier structures lacked the realistic mechanical stresses found in coiled DNA in living cells. The new versions reveal that supercoiling fundamentally changes how the enzyme behaves.

Steps to Designing Better CRISPR Tools

• Account for DNA topology: Future CRISPR variants must be engineered to account for how supercoiled DNA behaves, not just linear DNA, to reduce off-target errors in living cells.
• Test in realistic conditions: High-fidelity CRISPR designs should be validated against supercoiled DNA structures, not only linear DNA experiments, to ensure they work safely inside patients.
• Develop topology-sensitive variants: Create Cas9 enzymes that are sensitive to DNA topology and can distinguish between intended targets and off-target sequences even when DNA is supercoiled.

The implications for gene therapy are substantial. Current strategies for developing high-fidelity, low-error CRISPR precision rely heavily on structural information obtained from linear DNA experiments. But if Cas9 behaves differently on supercoiled DNA, those designs may not fully capture what happens inside living cells. Understanding the biomechanics of how supercoiling affects CRISPR accuracy could lead to CRISPR enzymes that dramatically reduce errors in the system.

This research also connects to a broader effort to harness genomic data for AI-driven medicine. Basecamp Research, a frontier AI lab for biological design, is partnering with PacBio to generate the Trillion Gene Atlas, a landmark initiative that will sequence approximately 100,000 deeply sequenced samples from over 31 countries across 5 continents. This massive dataset will train biological AI models like EDEN, which is designed to enable "prompt-to-medicine" capabilities and design programmable therapeutics across diverse modalities including cell and gene therapies.

Meanwhile, researchers at the Icahn School of Medicine at Mount Sinai have developed a complementary AI tool called Variant to Phenotype (V2P) that identifies disease-causing genetic mutations and predicts the types of diseases they are likely to trigger. Unlike current genetic testing tools that simply estimate whether a variant is harmful, V2P goes further by predicting the category of disease a mutation is most likely to cause, such as neurological disorders or cancer. When tested on real patient data, the tool ranked the true disease-causing mutation among the top ten candidates.

"Our approach allows us to pinpoint the genetic changes that are most relevant to a patient's condition, rather than sifting through thousands of possible variants," explained Dr. David Stein, first author of the V2P research.

Dr. David Stein, Icahn School of Medicine at Mount Sinai

The convergence of these advances, from understanding CRISPR's behavior on supercoiled DNA to building AI models that link mutations to disease outcomes, represents a fundamental shift in precision medicine. Rather than treating genetic information as static sequences, researchers are now recognizing that the physical properties of DNA, the context of genetic variants, and the biological diversity captured in large datasets all play crucial roles in designing safer, more effective therapies.

For patients waiting for gene therapies to treat rare and complex conditions, these discoveries offer hope that the next generation of CRISPR tools will be not only more powerful but also significantly safer.