Jennifer Smith, PhD, deputy director of the ICCB-Longwood Screening Facility, and Yanhui (Claire) Hu, PhD, senior bioinformatician at the Drosophila RNAi Screening Center in the Department of Genetics at Harvard Medical School, share with contributing editor Tanuja Koppal, PhD, the excitement as well as the apprehension working with the CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)/Cas technology for gene editing. Dr. Smith talks about the challenges on the experimental side, finding the right reagents, and libraries for developing the best CRISPR assays, while Dr. Hu elaborates on the issues underlying design rules, off-target effects, and data analysis.
Q: Can you provide some background on how and why you started thinking about CRISPR after having done RNA interference (RNAi) screening for such a long time?
Jennifer Smith, PhD: The introduction of CRISPR is very exciting because it is another tool available to us to answer the biological questions that we are interested in. As a core facility, we are most interested in using the best technology available to help scientists answer their research questions. I think what has become evident over the past couple of years is that utilizing the CRISPR technology is actually addressing a different question than using RNAi, and they are both complementary. There is a clear difference between knocking something out with CRISPR and knocking something down with RNAi.
Q: What are some of the things you are interested in doing with CRISPR screening?
JS: I am happy to share with you what we are considering, because we have just ordered our first arrayed CRISPR test set from Dharmacon (part of GE Healthcare). In [the interest of] full disclosure, ICCB-Longwood has a ten-year relationship with Dharmacon. Over the years, their reagents have worked well for us, and I found [the company] very willing to work with us as well. We have certainly been thinking about CRISPR screening and talking with a lot of people about how to do it. But we have not yet run a CRISPR screen at ICCB-Longwood. In the Boston area, a number of institutions, including The Broad [Institute of MIT and Harvard], have led the effort in pooled CRISPR screening and are making their reagents available to the community. The pooled CRISPR screening approach works very well in a forward genetics approach, depending upon the readout. Our background in functional genomics is really in arrayed siRNA (small interfering RNA) screening. A lot of the screens that we have supported have been image-based readouts done in a one-gene/well format. We were really excited when companies began discussing having arrayed constructs for CRISPR screening. It is possible to have arrayed viral libraries too, but it is challenging to do [that] in a whole genome format. We have developed a great workflow for siRNA screening, and if this arrayed CRISPR screening works with RNA constructs, it could be really powerful.
Q: What were some of the things you needed to do as you started thinking about CRISPR-based functional screening?
JS: It is amazing to see so many CRISPR- related publications [appear] in such a short time. What we have been doing is just keeping our ears and eyes open. The functional genomics community has formed over the past 15 years, and it is the same people who are now testing the CRISPR strategies. We found that people have been really open to talking with us about what is working and what isn’t working. We are using the lessons of our peers to help guide us. Since the pooled screening strategy is handled well, individual labs can purchase lentiviral libraries from a lot of different vendors. What we want to do is create a setting where we can also enable people to do CRISPR screens in a focused arrayed format.
One of the first things to look at is whether you want the Cas9 protein stably expressed in your cells or stably expressed but under an inducible promoter, or whether you want to transiently introduce it just for the editing event. We are going to pilot arrayed CRISPR screening in an assay we have that is well-defined from a full genome siRNA screen. We know the genes that when knocked down lead to a certain phenotype, and we have done the follow-up analysis to eliminate off-target effects. We are going to test that system using stable Cas9 in the cells and also try different transfection protocols to see if we can introduce the Cas9 protein or transcript transiently.
So, initially we are going to look at the classic CRISPR knockout in an arrayed format to see how that goes, and then we can expand to CRISPRa or CRISPRi, if there is interest. One of the things that will be interesting, which is different from siRNA screening, [will be to observe] the cell-specific differences in the DNA sequence. The guide RNAs make such a difference that you often sequence the gene you are modifying when doing single CRISPR knockouts for gene editing events. You need to know [whether] the gene is knocked out but not leading to a phenotype, or if that guide RNA did not work, and hence there is no change in expression. However, is this scalable? I guess this is why we continually advise investigators to follow the positive results from screens.
Q: Do you think that you will have to invest in some additional infrastructure or hardware as you transition from RNAi to CRISPR screening?
JS: Since we are trying an arrayed format, it actually will be very similar to RNAi screening. Colleagues who have used the arrayed strategy are using the CRISPR libraries at similar concentrations to the siRNA libraries. There may be more consumables involved if there are multiple transfections needed, but I don’t think the hardware will need any modifications. From a software perspective I think additions will be needed, since people are still figuring out what the caveats of the analysis are. For CRISPR screening, what will be most important is how people are choosing to validate the phenotypes and determine whether the editing took place. We will just have to work through those challenges and determine at what stage of the screening process those validations are done.
Q: Any advice to lab managers as they continue to grapple with the pros and cons of investing in and working with a new technology?
JS: It is fun to get really excited about a new technology, but I think with CRISPR it can also get really overwhelming because there is so much information available. I think you should just look at the question that you are trying to answer and determine how to best answer it. We have the ability to utilize multiple approaches, and not pigeonhole ourselves into doing only one thing. If we don’t have the resources to do something, I would much rather refer people to another screening core that would meet their needs and help answer the biological question. The resources are there, so even if you have to meet someone new or collaborate with a different group, do it.
Design and Data Challenges
Q: From an assay design perspective, what challenges did you run into?
Yanhui (Claire) Hu, PhD: You really need to know the sequence for the guide RNA. Since CRISPR works on the genomic DNA and not mRNA, you really have to think from a genomic perspective to design it. For knockout, you have to think about both efficiency and off-target effects, and the rules required are quite different from RNAi. The rules were pretty simple with RNAi. With CRISPR, there are a lot of algorithms, e.g., more than eight different scores predicting efficiency. People really have to think about which is the best score they want to use and which one is the closest to the species and system they are using. The scores are more context-specific.
One thing that is important for CRISPR library design is [that] people have to design enough controls. The controls used for RNAi and CRISPR are different. If you do not have negative and positive controls, that is a problem. If you get the controls right, it will make your analysis easier and get your probability scores right. Especially in cancer cells, a lot of chromosomal regions are duplicated and high copy number correlates to gene-independent cell response to CRISPR/Cas9 targeting (http://www.ncbi.nlm.nih.gov/pubmed/27260156). Therefore, you want a lot of negative controls spread out across those regions. Something like this is unique to CRISPR. However, in general, most people seem to think that CRISPR data is much cleaner than RNAi, with no microRNA- type off-target effects associated with short hairpin RNA (shRNA) or siRNA reagents.Q: Do you have any advice for downstream data analysis?
YH: I haven’t done any analysis of data from CRISPR screens yet, so I can’t talk from experience, but only from my understanding of biological data sets. After a CRISPR screen is done and the raw data is analyzed, a lot of gene hits are identified. There are many tools for enrichment analysis of the gene hits. The enrichment step offers proof that the assay works and helps generate more hypotheses for the next step, such as for secondary screening. After the enrichment analysis, if say, five of the top ten hits [are] from one gene family or one set, those are very likely to be real.
I also suggest looking at a few CRISPR papers to see how they visualize the data. Each data set is different and how you display and visualize the data is important. It also helps cross-reference the data to other publications or to a 3-D structure database, to find which gene is easier to target and which genes are being worked on, which helps prioritize the hits.