I recently wrote a twitter thread for Works In Progress summarizing their article on gene drives as a tool to end malaria. There’s a section in that piece that I didn’t get to highlight in the thread which I want to talk about here: the offense-defense balance of gene drives.

What Are Gene Drives and How do They Work?

Gene drives are a form of biotechnology that allows DNA mutations to guarantee inheritance in offspring and rapidly spread through a population, regardless of their evolutionary fitness. Nature discovered this technology long before humans did, and around half of our DNA is currently made up of these gene-drive mutations.

Humans have recently discovered how to engineer their own gene drives by copying another answer off of Nature’s homework. The CRISPR/Cas9 protein complex is a piece of biological machinery that can cut, copy, and paste arbitrary sequences of DNA. It was originally developed by prokaryotic bacteria as a defense against viruses inserting DNA instructions to take over and turn the cell into a virus factory.

These Cas9 proteins are fully general DNA text editors. They can search for any spot on existing DNA using a guide RNA sequence, make a cut, and insert any sequence of edited payload genes to fix the break.

The instructions for making proteins are also encoded in DNA. That means Cas9 proteins can be programmed, not only to insert whatever mutation you want, but also to insert the instructions for making more Cas9 proteins that will copy the mutation, and themselves, all over again.

This recursive gene editing is sufficient for making a gene drive.

Each organism has two copies of their DNA in two chromosomes: one from their father and one from their mother. If you mutate both copies in a fruit fly, then when the fly breeds with an un-edited mate, their child is guaranteed to inherit one mutated copy. But when the child mates with an un-edited fly, the grandchildren only have a 50% chance to inherit the mutation. Unless the mutation is evolutionarily fit, it won’t spread.

However, if the mutation contains DNA instructions for manufacturing a Cas9 protein with orders to copy it over, the mutation will always be inherited. The children of mutated and un-edited fruit flies who only inherit one mutated chromosome will quickly become fully mutated. Their children are now guaranteed to get one mutated copy, which quickly turns into two and so on.

The Offense-Defense Balance

If you have practice looking at new technologies and considering the risks as access to them spreads, you may reasonably be terrified of gene drives.

Setting up a self-replicating gene edit with CRISPR/Cas9 is not expensive, maybe a few thousand bucks or maybe as little as $100. You’d need some pretty advanced scientific knowledge and equipment, but nothing that isn’t posted online and commercially available.

Bad actors could plant gene drives in livestock or crops that rapidly spread, devastating agricultural yields and potentially wiping out entire species. Or they could plant edits in bacteria or insects that make them more harmful to humans. They could even try to spread genetic modifications of their choosing through human populations.

Usually when there is a story about a technology that is cheap to use offensively with the potential for as much damage as gene drives, the conclusion is extreme caution and close regulation. But gene drives are different, and some important facts about the science make it much more defense-secure than it may seem on first glance.

The first defensive advantage is detection. If an attacker is trying to make all of your livestock infertile with gene drives, that will be easy to notice. But even more subtle edits can be screened for because the Cas9 protein which executes gene drives is only found naturally in bacteria. If you sample cells and find Cas9, you can be confident that they’ve been tampered with.

The second and more decisive advantage comes from using gene drives against themselves. You can just as easily code a Cas9 DNA package that targets and removes any unwanted edits and this will rapidly spread through a population, fixing any edited genes along the way.

Responses to hostile gene drives can be even more effective and precisely targeted than this too. Rather than creating an edit that contains both the targeting guide RNA and the instructions for creating Cas9, you can edit in just some guide RNA. Program this RNA to copy itself and to neutralize the instructions for creating Cas9.

Now, when this edit is in an unedited wild-type organism, it won’t do anything. It creates guide RNA but there’s no Cas9 protein to guide anywhere. However, when this organism mates with a gene-drive mutated organism, their offspring will have the gene drive on one chromosome and the neutralizer on the other. The previously dormant guide RNA now has a bunch of Cas9 floating around thanks to the gene drive. These Cas9 proteins are now turned against the original drive, halting the production of more Cas9 protein and copying the guide RNA so that further offspring will also inherit this passive protection.

So malicious attacks with gene drives aren't too expensive to counter. We can also avoid well-intentioned mistakes. The simple gene drive I explained above can be modified to be more precise and targeted.

Daisy chain gene drives add a timer to their spread, so we can pre-plan exactly how many generations we want to inherit the self-replicating drive before switching it back to standard Mendelian inheritance. We know how to make a gene drive that lasts only one generation. Edit in Cas9 protein code, the payload, and the guide RNA telling Cas9 to copy the payload, but not any RNA telling Cas9 to copy itself. The edit will spread to both chromosomes and thus be inherited in the next generation, but from then on it acts like any other gene since Cas9 isn’t copying itself.

We can extend this to an arbitrary number of generations. If we put the code to copy Cas9 instructions in a guide RNA that has no code to copy itself over, the gene drive lasts for two generations. The first generation has all the pieces. Therefore, the children get Cas9 and the mutation, so it will copy to both chromosomes. But, the guide RNA that says to copy Cas9 doesn't get copied. So the next generation inherits one copy of the payload, but no mechanism for copying it over and it’s back to standard inheritance.

There are other important risks and costs to consider, like off-target effects that can happen when Cas9 fails to target the right spot, but these are the standard risks of any new technology we are still learning about.

The point is that the prophesies of massive upsets to the offense-defense balance that are invoked to support extreme caution and tight regulation of e.g AI models or other biotechnology do not apply here.

This is important because gene drives have massive potential for good in the world. The most prominent and quantifiable example of this potential is ending malaria. Spreading a gene drive through the few populations of mosquitos that spread the malaria parasite is a cheap and feasible way to end this disease and save upwards of 600,000 lives every year. That’s about the population of Washington D.C or Boston, every year.

So we should move forwards with this technology ASAP. The risks are real and important, but not catastrophic. The risks of inaction are.

Open Philanthropy is admirably funding and trialing this technology. This is a huge endorsement to their intellectual honesty and consistency since it would have been easy for the EA org to fall back on well-worn scary-stories of potential misuse that are common in EA commentary on many other areas of technology.