This post was originally posted on the MRC Weatherall Institute of Molecular Medicine’s blog
Dominic Owens, PhD student in the De Bruijn group at MRC MHU, recounts how puzzling results and a fortuitous lab meeting uncovered unexpected outcomes of CRISPR editing and changed the direction of his research.
It’s May 2018 and I’m feeling nervous. I was presenting the latest results in my PhD research project at a large joint lab meeting involving several research groups. The last few weeks of experiments hadn’t gone the way I expected, and I had gotten some odd results that were puzzling. Unknown to me, from this lab meeting, I would be embarking on a yearlong journey of despair and eventual discovery which became central to my first publication, published in Nucleic Acids Research (Owens et al. 2019, https://doi.org/10.1093/nar/gkz459).
I had been using CRISPR-Cas9 to examine the function of ‘junk DNA’. So-called ‘junk DNA’ contains no genes but instead is increasingly understood to play important roles in regulating gene expression. The enhancers I study are believed to regulate blood stem cell generation, by activating a gene called Runx1 during embryonic development. This gene is required for blood stem cell generation and is frequently mutated in human leukaemia.
In mouse cells, I had been using CRISPR to remove specific enhancers, and wanted to see what the impact on blood cell generation would be. If removing these enhancers made a difference, then I could reveal unknown factors important for blood stem cell generation.
After using CRISPR to cut out this DNA, I needed to sequence short pieces around the cut sites to determine what kind of cut took place. As is common practice, I used PCR to amplify the DNA and Sanger sequencing to read it, over a region around 200 base pairs on either side of the target sites. After doing this, I often saw cells with what looked like both copies of the region deleted, right where we expected. However, when I tested the function of the cells, I saw inconsistent results. Even though the sequencing told me that the region was missing, some of the cells were behaving like it was still there.
Back to the lab meeting: a colleague who is very experienced with CRISPR from a different lab suggested that I could be seeing larger than expected deletions which meant that one of the PCR primer binding sites I relied on was no longer available. This would in effect prevent one copy of the region from being amplified and sequenced, and thus render PCR blind to such larger deletions. I can still remember that moment—the significance of what she said truly dawning on me. So, after the lab meeting, my lab mates and I discussed what to do next. We decided that I should look for larger deletions in all the cell lines we had made. To do this I used longer range PCR, with primers that bound to the DNA much further away from the CRISPR target sites. This way I would be able to detect any larger deletions, if they were there. And when I looked, I found them. In fact, no matter how far I looked, larger deletions were always there.
This left me desperately wanting answers to three questions: why are these larger deletions happening? Just how large can the deletions be? And can they be predicted? The only problem was: I was working in a lab of exceptional developmental biologists, not genome engineers! So, my supervisor and I reached out to two local experts – Lydia Teboul at MRC Harwell, an expert in CRISPR editing, and Peter McHugh at MRC WIMM, an expert in DNA repair. Both were generous with their time, helping me plan and perform the necessary experiments to answer the questions I was puzzled about.
By sequencing many examples of larger deletions, I found that short regions of homology (microhomologies) were commonly found in places where the cut DNA was repaired back together. This suggested that a DNA repair pathway called microhomology-mediated end-joining was a likely culprit.
Next, I used a technique called droplet digital PCR that is incredibly useful for interpreting the outcomes of CRISPR experiments. The method uses an oil in water emulsion of droplets to perform 15,000 individual PCR reactions in a single tube. This is advantageous because it allows quantification of DNA with much greater accuracy and precision than comparable techniques. Taking advantage of this accuracy, my colleagues and I mapped deletions across thousands of base pairs of DNA in populations of cells targeted with CRISPR, without biases arising from the locations of longer-range PCR primers.
Modelling the effects of various potential predictors on the distribution of deletion sizes revealed the driving forces determining how large a deletion is likely to be. I found that the frequency of larger deletions was highly dependent on CRISPR’s cutting efficiency at a target cut site and the distance from it. This means that the closer to a cut site, the more likely a region is to be lost in a deletion. This makes sense of course, but this distance-dependency shows that the microhomology sequences often associated with the repaired breakpoints themselves cannot be used to predict the size of the deletions. This is in stark contrast to shorter CRISPR deletions, which have recently (Shen et al. 2018, Nature, and Allen et al. 2018, Nature Biotechnology) been shown to be highly predictable based on sequence characteristics such as microhomologies.
The work highlights important caveats about CRISPR that I wished I had known when embarking on this project. Larger deletions occur frequently after CRISPR editing and they will be easily missed if you aren’t looking out for them. Deletions can be any size when looking at individual cells or animals, but in general are much more common within 3 kilobases from a cut site. Even though extremely large deletions are rare, given a large enough population of cells, they should be expected to occur. In the case of gene-edited somatic cells for gene therapy, for example, a single rare oncogenic larger deletion could have disastrous health outcomes for recipient patients. In the future, inhibiting microhomology-mediated end-joining could help reduce this problem. Alternatively, gene-editing approaches that avoid DNA double-strand breaks altogether may be preferable.
I must admit, I was slightly disappointed when the questions I had set out to answer during my PhD didn’t quite work out. But I have also learned valuable lessons. It is often tempting to ignore outliers or repeat experiments that just look weird. But by staying curious and motivated to understand the things that don’t quite add up, you will understand the science more deeply, and maybe find your next big breakthrough!