Hi all. New to Crispr. I’m trying to put GFP at the 3’end of my favorite gene. I have two sgRNAs selected, one at 95% and the other at 98%. Both have off-targets in genes (~10 genes each). That seems pretty bad to me, though I have no context to compare to. Also, one of the sgRNAs overlaps my stop codon (the 95% option) and the other one is upstream ~10bp. Are these essentially equivalent options given the plan to repair these with a second contruct? I guess I’m not quite clear yet on the repair process and the need to cut close to the site I’d like to alter.
To add a small piece, you can now in the UCSC genome browser ask for a sgRNA crispr track.
It gives you all the possible sgRNA possible for a given region, and these sgRNA have a color code according to their quality (calculated from 3 different scoring approaches)
So maybe you can check if your sgRNA are really good or not.
Hope this helps.
Cheers.
This is a great point from ThomasD. I’ve just started using it, and it makes choosing sgRNAs SO much easier. Instead of using two or three different programs, it lets you look at a region of interest and scan through sgRNAs by hovering over them with a the cursor. Nicely, the site gives scores on specificity (off-target sites) from the MIT site, and activity from the Doench and Morena-Mateos algorithms (plus I think it lists a ton of other scores). Highly recommended. I actually just met the developer at a UCSC CRISPR club last week, so he’s really interested in improving it. If any ideas or issues be sure to send him a message and I bet it will be sorted.
A few other points:
- I participated in a CRISPR workshop here in the TMC where the people surveying guide efficiency at the scale of many thousands, in cell culture, insist that there is no benefit for guides that have “better scores” based on various algorithms. Naturally, nobody is assessing chromatin state when doing these measurements, and we know that chromatin state matters, at least in certain contexts. If you figure out how to do that in the proper region of the distal arm of the worm gonad, let us know! In the meantime, we just guess. Also, Dickinson (2015) claims that score doesn’t matter. But they don’t control for how far away from the cut site you are inserting your protein (see below).
- That said, we still parse our guides really carefully. Since things seem to work, why mess with it? The Xu score has been good to us. I believe it is mostly based on Wang, Sabatini, Lander (2014). In a nutshell, “GCGG” is the best sequence at positions -4 through -1 before the PAM. Ts are bad here. This fits with Farboud and Meyer’s observations. Our work is not systematic, so we cannot derive any efficiency claims.
- Re: proximity. There is evidence that inserting close to your cut site matters. A lot. Apparently efficiency drops off precipitously as you move away from the insert site.
- Because of that, we are always using two cuts sites and simply replacing a chunk of the gene with our insertion in the middle. That way we are always inserting right on top of the cut site. A little more work at the cloning end, but we hope that the knockin efficiency is higher. Plus, we can use whatever guide we want without constraints.
- You have to inject a lot of animals really well. Look at Dickinson (2015) and see how many they inject. And that is with positive/negative selection. The Paix and Seydoux papers with short inserts and short homology arms claim higher effiiciency. We’ve tried that once and it didn’t work. Could have been the guide. But we’ve had a lot of luck with Dickinson’s positive/negative selection.
- Isaiah Neve, a student with Meng Wang at BCM, actually tested his guides to see if he could isolate deletions in his gene, detected by PCR. Then he went on to knock in. Seemed like a clever approach, since some of his guides didn’t work. Companies sell kits to test your guides, but I am dubious: that’s not chromatin, so does it reflect reality? I doubt it.
- Seydoux’s ssODN to direct deletions and missense mutations works really well. But, we tried it on a larger scale with a gBlock of dsDNA. Didn’t work. Maybe it is the dsDNA that strand invades poorly compared to the ssODN. Or just another anecdote: the whole field is built on them!
