Translation is Tricky

I read a paper that is well written and a surprising development in the field of gene/DNA therapeutics, so I wanted to share and summarize it, with a special focus on molecular biology. The title says it all, but you gotta be careful not to jump to conclusions: it’s not the AAV that seems to be the problem, nor is it the miRNA targeting the disease-causing gene. It turns out that the way the construct is designed causes toxic responses in animals. Cover art is a collage of the new constructs tested by Megan S. Keiser et al.

Toxicity after AAV delivery of RNAi expression
constructs into nonhuman primate brain

Spinocerebellar ataxia type 1 (SCA1) is a genetic disorder caused by repeat expansion in the ATXN1 gene. To treat this disease, you want a therapeutic that can silence the gene, since RNAi studies have shown success as a targeting strategy. This is a tall order for a small molecule, but maybe an anti-sense oligonucleotide could do the trick. For a gene disorder, it makes the most sense to consider a gene therapy , which is where adeno-associated viruses (AAVs) come in. AAV development is super popular right now, boosted in part by the the J&J vaccine which is built on this kind of platform, although AAV-based trials in the clinic are many years in the making (see this article for a recent metareview of the clinical landscape.) In SCA1, AAV gene therapies have been shown to be well tolerated and effective in mice mouse models, so you can imagine peoples surprise when an IND-enabling study in non-human primates showed unprecedented toxicity. So, what’s going on?

Well, maybe the AAV capsid is causing neuroimmune responses? Its possible, since the vehicle-treated animals had only minor issues while the animals dosed with AAV.miS1 showed cell loss, but there was no correlation between viral dose and severity. That’s a little weird (shouldn’t more AAV and/or miS1 dose cause worse severity/toxicity?), though the authors initially suggest its due to the variable time of vector expression in vivo. AAV.miS1 doesn’t seem to be messing up endogenous small RNA levels, which, according to the authors, can be a source of toxicity, so maybe the problem is miS1 itself. RNA sequencing of the tissue suggests that toxicity is related to an immune response, so we’re still not off of the capsid as the culpirt. What’s interesting is that, as the authors looked more closely at the RNAseq data for the tissue treated with AAV.miS1, they found that most of it mapped to the correctly packaged construct, which is good. However, they also detected a small number of reads that mapped to the backbone, which suggests cross-packaging.

What is cross-packaging, when does it occur, and why? It happens during AAV production process when plasmid or host-cell DNA gets packaged into the capsid, which you don’t want to happen. Addgene has a great introductory article on viral vectors, but to summarize, when making an AAV, you need to transfect 3 plasmids into mammalian cells: a packaging plasmid, an adenoviral helper plasmid, and the transfer plasmid. Think of the first two plasmids as standard, so the design you’re most concerned about is the transfer plasmid. In this case since they wanted to silence ATXN1, so it has to contain miS1. In order for miS1 to be packaged into an AAV, you have to have inverted terminal repeats (ITRs) flank it 5’ and 3’. The construct also has to be an optimal size, so an inactive “stuffer” sequence was also included. Last but not least, you want a promoter sequence that will promote transcription (in this case, U6.) In the most common case of cross-packaging, you’d expect to see empty vector, almost as if you only had 2 plasmids transfected (i.e. no transfer plasmid.) For quantitative analysis of AAV packaging efficiency, you would do droplet digital PCR, which lets you separate the individual DNA molecules present in a sample and PCR each one.

Now, back to the RNAseq data from the tissue treated with AAV.miS1. You should really only see virus-related reads map to the region between the ITRs, and at first glance, that IS what you see. However, as the authors note, while you don’t see many reads map to the ITRs when compared to the number of reads that map to the “drug” (>7000 reads), they’re still there (32-386 reads.) Looking at long-read sequencing of AAV.miS1 itself showed that yeah, most of it is cargo, and that cross-packaging reads accounted for a small % of the sequence data. However, when they quantified cross-packaging with ddPCR, it became clear that the right/3’ ITR was most abundant, so maybe cross-packaging, despite its low abundance, is toxic!

At this point, there are technically 3 sources to test as the cause of toxicity: the AAV capsid itself, off-target activity of miS1, or ITR activity. To get a better understanding of how the construct can influence toxicity, the authors redesigned it, replacing the U6 promoter with one from EIF1α (the authors state that “Pol II promoters have been shown to impede ITR-driven transcription”.) They also replaced the “inactive” stuffer region with an “active” ATXN1 homolog (“ATXN1L overexpression alone is therapeutic in SCA1 mice.”) In mice, this new construct (AAV.IntmiS1) did produce a dose-dependent response for miS1 expression and ATXN1 reduction, though miS1 expression wasn’t as robust as AAV.miS1. Still, the 3’ ITR was way less abundant, and the authors still achieved 30% target silencing required to prevent disease onset in mice; all promising signs.

Remember though, the problem that this paper brings up is that the mouse data didn’t predict the toxicity seen in NHPs. This is a big deal because in order to submit an IND application and move into humans, you need to study the effects of your clinical candidate in animals (at least two mammals, where one species is non-rodent) and depending on your indication (think CNS), NHP studies could be crucial.  Data from Figure 5 really drive home the point: Transcripts derived from stuffer and/or packaged backbone are sufficient to induce MRI abnormalities/toxicity in NHPs, while the new AAV.IntmiS1 showed similar results to an empty capsid. It’s not clear whether the transcripts themselves are toxic, or if they’re being translated into a toxic peptide/protein, but it’s still a major finding.

Don’t think that this an indictment against AAVs or gene therapies! The FDA had a meeting back in September 2021 on Cellular, Tissue, and Gene Therapies (CTGT) to discuss AAV toxicities (you can find the sessions here) and back in August of 2021, the American Society of Gene and Cell Therapies (ASGCT) had a rountable discussion on AAV integration (catch it on YouTube here.) The take home message in Keiser et al’s study is that, in gene therapies, context matters. The benefits of translational science are obvious, but surprises like these are a constant reminder that biology is hard.