[jfarlow]

Well, as CRISPR was originally found as a kind of immune system, there do exist a number of anti-anti-Cas9 systems against that evolved alongside it. There are a number of small inhibitors of Cas9 [1] (which themselves could be used to tune Cas9 in therapeutics). However up-taking such a defense is admittedly an unlike route for a virus like HIV to take to evolve resistance to a CRISPR-based therapy.

More practically, HIV has such a high mutation rate, that it's likely very difficult to target every HIV sequence with a sequence-specific Cas9 therapy. If the Cas9 guide sequence is too generic it'll take out stuff besides HIV (stuff you need). And if the guide sequence is too specific it won't get all the viral inserts because many are degenerate. As with all things though, 95% success with viral excision via CRISPR, in conjunction with 95% success via immunotherapy [2], and 95% from standard anti-retrovirals [3], get's you pretty good 99.9999% coverage.

That's the power of convergent technologies. It's an interesting slice through a number of modern therapeutic technologies all applied to one of the most challenging of tailored foes. You see convergence of small molecule biochemistry along with immunotherapy, along gene therapy, along with cutting edge synthetic biology - all approaching the problem from different angles.

[1] www.cell.com/cell/fulltext/S0092-8674(16)31683-X

[2] https://serotiny.bio/notes/proteins/ecd4ig/

[3] https://en.wikipedia.org/wiki/Category:Antiretroviral_drugs

[Turing_Machine]

"More practically, HIV has such a high mutation rate, that it's likely very difficult to target every HIV sequence with a sequence-specific Cas9 therapy. If the Cas9 guide sequence is too generic it'll take out stuff besides HIV (stuff you need). "

Unless I'm grossly misunderstanding how CRISPR works, there's no conceptual reason why you couldn't target multiple sequences at the same time, with a cocktail method. That is, rather than trying a single overly broad match, you could go for (say) ten highly-specific targets at the same time. A particular HIV virion would then have to differ in all ten regions to avoid getting chopped up.

Just spitballing here:

Google tells me that HIV has a mutation rate of about 4E-3 per base.

Let's say you choose target sequences ten bases long (I don't know what the maximum practical length for the technology is, nor the minimum length you'd need to reliably tell HIV from human, and Google isn't any immediate help there).

The probability that there will be a mutation in that sequence is then about 0.04. However, if you target ten sequences simultaneously, the probability that all ten would be mutated is (0.04)^10 ~= 1e-14.

That's likely more than good enough to assure that there weren't any resistant mutants around (if by some chance there are...lather, rinse, repeat).

This is hand-waving, to be sure. If you have better numbers, plug them in.

Edit: fixed fat-fingering the calculator.

[new299]

I think ∼3E−5 per base per replication. Might be a more useful number [1].

If you use 10mers, that only gives you 1048576. I'd be almost certain that >90% of those sequences also exist in the human genome. So your target isn't specific enough (take out stuff you need as the parent suggested).

So you need to use a longer sequence, perhaps 25bp. Maybe there's a stable region or set of regions you can target (in which case the high overall mutation rate doesn't matter). Or a cocktail of sequences, specific to the global HIV population (I doubt this, HIV mutates more in a single individual than Flu does in the global population).

But if not, then you first need to figure out what the viral population in this individual looks like. So you sequence a subset of population, and come up with a 25mer or set of 25mers that target this population.

That might be a lot of sequences (significant problem). Which you then need to get synthesized (will take weeks).

Now. It's taken days to run your sequencing experiment, and weeks to get your CRISPR stuff synthesized. In this time the viral population has been generating 10E11 new virions per day. You're population has moved on, and almost certainly contains members which don't have your previous cocktail of 25mers in them and will survive the treatment.

Because HIV mutates so much, there was some interesting work I saw a while back on guiding the evolution of the population. You'd use drugs which don't wipe out the infection, but push the population toward specific genotypes. Specifically those which you have good treatments for, in the hope that you can wipe out most of the population at once.

[1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3530041/

[Turing_Machine]

"Maybe there's a stable region or set of regions you can target (in which case the high overall mutation rate doesn't matter). Or a cocktail of sequences, specific to the global HIV population (I doubt this, HIV mutates more in a single individual than Flu does in the global population)."

Hmm... I would bet that there is a cocktail of sequences such that if they are not conserved, the virus effectively becomes no longer HIV (no longer infectious, no longer capable of producing symptoms...).

HIV is obviously not a human being, right? Find every sequence where it differs, target them all. :-)

[new299]

I did a quick literature search, but couldn't find anything. It should be easy to answer that question.

There appears to be at least one conserved protein. However there's a lot of scope for different underlying sequences due to synonymous codons.

Depending on how long a fragment you need to target, that could end up being a lot of sequences, and unpractical.

[new299]

It's also possible that HIV could stick introns into the sequence too to avoid CRISPR...

[Turing_Machine]

Now that I've read the abstract, it looks like that's exactly what they're doing. They tried both dual targets and quad targets.

[BWStearns]

> As with all things though, 95% success with viral excision via CRISPR, in conjunction with 95% success via immunotherapy [2], and 95% from standard anti-retrovirals [3], get's you pretty good 99.9999% coverage.

I wonder if several narrow CRISPR payloads simultaneously would achieve the same effect since any survivors would need n many mutations simultaneously.