PhD, biochemistry and molecular biology, Harvard University — 1992
AB, biochemistry and molecular biology, Harvard University —1986
This month, University of Massachusetts Medical School professor and RNAi pioneer Phillip Zamore and colleagues showed how they were able to design siRNAs specific enough to distinguish between targets that differ by a single nucleotide.
In the September issue of the journal PLoS Genetics, Zamore and his colleagues investigated whether such siRNAs could target mutant genes associated with Huntington’s disease and familial amyotrophic lateral sclerosis while leaving the wild-type unaffected.
Recently, RNAi News spoke with Zamore about this research.
Could you talk about how you started … designing siRNAs that are so specific?
It all started when my colleague Neil Aronin, chair of endocrinology [at the University of Massachusetts Medical School], and my colleague Zoushang Xu, [who] is a professor of biochemistry here … within six months of each other came to me … with the exact same question: Can one make siRNAs that can discriminate between a disease allele and a normal, functional wild-type copy of a gene? At the time, we strongly suspected that you could. But we weren’t sure if it would be ad hoc or if there were some strategies one could use.
Both Zoushang, with whom I first started the collaboration, and Neil, who joined us later, work on diseases where a dominant disease allele causes a debilitating neurodegenerative disease. Zoushang works on Lou Gehrig’s disease, or ALS, and Neil works on Huntington’s disease. In each case you have two copies of the gene: the normal wild-type gene and the mutant destructive gene that ultimately leads to some kind of cell death in the nervous system.
For Huntington’s disease, this is the only cause of the disease. For Lou Gehrig’s disease only about 10 percent of patients have a [form of] disease caused by a defective gain-of-function allele of the superoxide dismutase 1 gene — those are the familial Lou Gehrig’s disease cases, rather than the sporadic ALS cases, which are the majority.
In the case of familial ALS, it was much more obvious that this was a disease where, if you could distinguish between two alleles that differed by as little as a single nucleotide, you would potentially be on the road to a therapy.
For Huntington’s disease, it wasn’t quite so obvious, and the question that was put to me initially was: Could we design siRNAs that distinguish between the length of the polyglutamine expansion? The disease is caused by an increase in the number of CAG repeats in the first exon of the Huntingtin locus. The normal gene that you and I have has as many as 30 CAG triples — a reach of 90 nucleotides, approximately — whereas someone who has a very early onset form of the disease may have 100 CAG repeats — that is, 300 nucleotides. Both of those lengths are much longer than a single siRNA. I was extremely pessimistic that we could make an siRNA that would differentially down-regulate the disease allele versus the wild-type allele based on the length of the CAG repeat.
On the other hand, Huntington’s disease is inherited; there is no other way known to get it. … Because [of this], it seemed very likely to me that there would be polymorphic differences between the normal and mutant alleles — differences that might in no way correlate with the disease but that for every patient we might be able to find a single nucleotide difference between the two messenger RNAs. In fact, in preliminary work Neil and I can now show that [while] maybe we can’t find such a difference for every patient, for the overwhelming majority of patients you can find such a … single nucleotide polymorphism.
For both of these diseases it became imperative to understand if one could develop single nucleotide-specific siRNAs. I want to contrast this with the more standard strategy that companies that are trying to bring siRNA therapeutics to the market are taking, for the most part. The general strategy is to search the sequence of the target gene for the best siRNA in terms of activity and specificity, irrespective of which part of the sequence that siRNA falls into. This is exactly the opposite. If you have two genes, and the only difference between them is a single nucleotide, the siRNA has to target that region of the gene.
So we needed two kinds of tools. One, we needed to be able to discriminate between two alleles based on the difference of a single nucleotide. And two, we needed some tricks to make all possible siRNAs active without having to change their sequence because if the sequence of the siRNA that’s required for specificity produces an siRNA that has poor activity, there isn’t a lot of wiggle room for us — we can’t go to another part of the message.
So the second part of that problem was how we ended up getting into the whole idea of functional asymmetry, which you may recall my lab and Anastasia Khvorova’s lab discovered in 2003 — the idea that for every siRNA, one strand is generally preferred for assembly into RISC.
The other part was a little longer in coming. We published a paper … in which we had the first hints that you could develop these kinds of SNP-specific siRNAs. Then the three labs — mine, Zushang’s, and Neil’s — spent a huge amount of effort trying to understand the extent to which we can make siRNAs that discriminate well. And Peter Linsley at Rosetta [Inpharmatics] helped us with some really rigorous in vivo tests using microarrays for specificity and strand choice.
That’s what’s really in this [PLoS Genetics] paper — the idea that we’d do a comprehensive survey of all the possible places you could introduce a mismatch between the target and the siRNAs, and all the possible mismatches you could have.
So there’s a heavy bioinformatics component to this approach.
In Peter’s contribution, they are using tools that they had developed previously to look at off-target effects. Here what we’ve done is design cell-culture experiments to maximize off-target effects [using] a really cool tool that Peter developed that lets you know which strand of an siRNA is dominating RISC assembly. The nature of off-target effects is that they have seed matches to the strand of the siRNA that’s down regulating them. So you can determine which strand is dominating RISC assembly by the nature of the off-target effect.
Peter used that method to check whether our strategy for forcing one strand versus the other to go into RISC was working. And for the most part it did, but there were a few exceptions where we would probably, if we were going to try to carry those siRNAs forward in any clinically meaningful way, we’d have to inactivate the passenger strand.
Is that something you want to look into, or was this a proof-of-concept sort of thing for other people to move forward on?
In terms of developing SNP-specific siRNAs that have clinical utility, me, Neil Aronin, and Zuoshang Xu will be carrying these things forward into clinically relevant models. So, we certainly are going to take it through the preclinical stage. Neil and I are working in Huntington’s models in rodents. The hope is that if one can succeed there, you would try to test safety and delivery in primates. So we’re quite serious for Huntington’s disease, and Zoushang for ALS, in carrying this forward into some kinds of translational research where the goal is to get it into the clinic.
For the issue of getting the right strand in, my interest is much more in terms of understanding the fundamental biochemistry of strand choice. Others like Anastasia [who is now with Dharmacon] and Peter have done gorgeous work showing how you can chemically modify siRNAs to alter that process.
Can you touch on the tricks you use to make sure you get the right strand where you want it to go?
Our strategy has always been very simple, which is to introduce a mismatch at the 5’ end of the guide strand. Within the siRNA, there’s an unpaired base at the very first nucleotide of the guide strand. But once the two strands separate, the guide strand pairs perfectly with the target, although in fact the first base never has to pair with the target — under some circumstances it improves efficacy not to.
That’s not a central focus of the [PLoS Genetics] paper, but it is something that is there because one of the things we discovered about our earlier work was that it’s very easy to overestimate siRNA specificity if the siRNA itself is sub-optimal because you’re not loading very much RISC. So if you have a fairly inactive siRNA, it looks like it’s active against the perfectly matched allele and inactive against the mismatched allele. But if you take that same siRNA and increase its activity, you can start to see activity against the mismatched allele.
From our perspective, the lesson is in order to ask intelligent questions about discrimination between two alleles, you have to start with really good siRNAs.