The Villain of CRISPR

Yang et al (2013): "To assess whether a marker transgene could be inserted into an endogenous locus, we coinjected Cas9 mRNA, sgRNA, and a double-stranded donor vector that was designed to fuse a p2AmCherry reporter with the last codon of the Nanog gene (Figure 2A). A circular donor vector was used to minimize random integrations. To assess toxicity and to optimize the concentration of donor DNA, we microinjected different amounts of Nanog-2A-mCherry vector. Injection with a high concentration of donor DNA (500 ng/ml) yielded mCherry-positive embryos with high efficiency, with most blastocysts being retarded, whereas injection with a lower donor DNA concentration (10 ng/ml) yielded mostly healthy blastocysts, most of which were mCherry-negative. When 200 ng/ml donor DNA was used, 75% (936/1,262) of the injected zygotes developed to blastocysts, 9% (86/936) of which were mCherry-positive (Figure 2C; Table S1)."

So efficiency is inversely proportional to toxicity, they treated many more than 5-25 cells (the selection would occur at the level of the embryo), and there was only 1-10% rate of mutation detection. Also, they used "Superovulated female B6D2F1 mice". This procedure leads to chromosomal abnormalities and probably genetic instability so we would expect elevated presence of mutations at any given site: http://jhered.oxfordjournals.org/content/77/1/39.full.pdf

I'll have to look closer at the HDR aspect though (primers used, etc). But what may be going on that usually they detect the insertion via PCR: there is one primer to a sequence unique to the cassette and another upstream or downstream that should only be in the cells. Then the segment spanning the junction is amplified which supposedly is conclusive evidence of insertion at the correct location. The problem is you can get single primer amplification and also the homology arms required for HDR are likely to contain similar sequences to the "cell-only" primer. Eg: http://link.springer.com/protocol/10.1385%2F0-89603-258-2%3A...

Hwang et al 2013: "On the next day, injected embryos were inspected under stereoscope and were classified as dead, deformed or normal phenotypes. Only embryos that developed normally were assayed for target site mutations"

They don't seem to tell us how many embryos were injected. And that study does not appear to use any type of control group at all. AFAICT, that is exactly the type of study that is consistent with a selection effect.

That could possibly explain some results, but not those involving transmission. If the knocked-in DNA is transmitted to the next generation then I'd think it must have gotten incorporated somewhere, however, this need not be at the intended site if the primers are amplifying the template.

Then again, supposedly shingles is caused by extragenomic Varicella-zoster DNA that is somehow stable for decades and can be passed on during pregnancy. I'm not sure I believe that though, and of course that is viral DNA.

Anyway, in that Ruan et al (2015) they claim to have detected exactly the expected sequence across the junction in at least a few cells. I can't think of any explanation for that data other than CRISPR working as advertised. However, they don't report in what percent of the cells this was observed.

Edit: I mean supposedly those exact sequences shown in figure S2 never physically existed before and now they do, exactly as predicted by the theory. That is strong evidence.

>"It sounds like you doubt the dsDNA nuclease activity of Cas9."

Not at all. This would be why the treatment is toxic and suppressive of proliferation.

At this point, I still think the presence of indels at the site (ie the proposed NHEJ mechanism) is just as easily explained by selection for pre-existing mutants. The experiments involving insertion of DNA (ie the proposed HDR mechanism) are better, but lack controls for "off-target" PCR amplification when showing the gels. IE we need to know how often the template itself will be amplified under their primers/conditions, both free and if it gets incorporated in some random location.

When segments across the insertion junction are amplified, sequenced, and reported, I find this convincing as it is a precise prediction that matches the data and I can think of no other explanation. The other experiments are pretty much redundant and add nothing. However, the reports I have seen contain little methodological or quantitative information regarding these sequences which does make me remain skeptical, especially about claims of efficiency. Those claims seem to always be determined using the former experiments that can be explained in other ways.