By Meredith W. Salisbury
Kevin FitzGerald has been using RNA interference technology longer than many scientists in the field. As group leader for emerging technologies in the department of applied genomics at Bristol-Myers Squibb, his team geared up six years ago and immediately took advantage of RNAi with its model systems unit. By the time papers started coming out showing that the concept could be used in mammalian cells, FitzGerald’s crew was well poised to dive right in and study loss of function genetics. As the story has gone so many times since then, it worked like a charm: “We had [had] a hell of a time coming up with drugs to target tumor suppressors,” FitzGerald says. An RNAi-based search for genes that would kill tumor cells and leave other cells alive turned up promising candidates.
A year ago, this story might’ve ended there; such was the wide-eyed excitement about this technology that even hard-core scientists were known to blithely overlook that nagging feeling that perhaps RNAi might be somehow too good to be true. In the past year, RNAi researchers have come to the conclusion that the technology — not surprisingly, of course — has been implicated in nonspecific binding events and off-target effects that may not show up in run-of-the-mill experiments.
But unlike predecessor technologies including antisense — “a series of what ended up being corpses along the road of disrupting genes,” as Stephen Friend, senior vice president of molecular profiling and cancer research at Merck, puts it — RNAi came of age at a time when other tools could be enlisted to help balance out the problems. Most notably, RNAi scientists have called on that old genomics standby, the microarray, to sort out the challenges.
The technologies can be paired in a number of ways. The most prevalent appears to be using gene expression studies to examine potential off-target effects of a completed RNAi experiment. Companies like Affymetrix and Agilent, both offering whole-genome arrays, are among the vendors urging scientists to look at the effects of RNA interference events across the entire genome. In another coupling, scientists use arrays to whittle down the field of genes of interest, and then perform their RNAi studies on just those genes to help save money. Perhaps the most innovative way to join these technologies was first demonstrated by David Sabatini, who conceived of the idea to spot cDNAs on an array and then plate cells over them to actually induce RNAi in parallel directly on the chip.
Using the technologies in conjunction with one another appears to be a rapidly growing trend in the field. In a recent Genome Technology survey, 151 respondents were asked about their interest and experience with this kind of work. Close to 11 percent of them said they regularly use arrays to study off-target effects of RNAi experiments, and nearly 19 percent said while it wasn’t regular practice, they had used arrays for that purpose. More than 45 percent of respondents said they hadn’t tried it yet but were interested in doing so. “Only a few thought leaders in the field … have been doing this on a regular basis,” says Bill Marshall, executive vice president of research and operations and site manager at Dharmacon.
Meanwhile, performing RNAi on microarrays seems to have more technological hurdles to face, and is growing at a slightly slower pace, according to the Genome Technology poll. Almost seven percent of respondents said they had successfully induced RNAi on an array, and some four percent said they had tried to do so but hadn’t gotten it to work. More than 58 percent of respondents said they hadn’t tried it but found the idea interesting — but close to 18 percent of respondents had never heard of the concept. For an explanation of this idea and a look at who’s been involved in it, check out the sidebar on p. 30.
Some experts say validation tools like microarrays could conceivably take RNAi as far as the therapeutic arena, assuming other hurdles like delivery can be overcome. Others warn that there are still challenges ahead: together, arrays and RNAi represent “a fantastic technology, but not without its limitations,” says Peter Scacheri, a postdoc fellow at NHGRI.
In the following pages, Genome Technology delves into the pairing of microarrays and RNAi to explore who’s using them, why, and where potential problems still lurk.
At first, it wasn’t clear that arrays would be needed to validate experiments. Peter Linsley, executive director of cancer biology at Merck subsidiary Rosetta Inpharmatics, was lead author of one of the first papers that outlined the off-target downside of RNAi. His goal when he began the project, however, was quite the opposite. “We were going to use clean genetics to show how dirty some of the compounds we were working with were,” he recalls. “We got a big surprise. … Every time we compared RNAi to a particular compound that hit that target, the compound was a lot cleaner.” His paper showed up next to two other RNAi papers in Nature Biotech last year, and “both of the other papers actually said there were no problems,” Linsley says. Since then, though, “I think the field’s come around to our point of view.”
Indeed. From pharmas to academia, researchers all over are learning to back up their RNAi experiments with gene expression studies to test for off-target effects. For a community so well-versed in the complexity of response from perturbing any little part of the genome, what was more surprising than the discovery of cross-target repercussions was that somehow this seemed to sneak up on scientists. Marshall at Dharmacon explains, “Whenever there’s a new technology … everyone wants it to be the greatest thing since sliced bread, and they tend to look away from some of the potential issues with it.” But now, he says, people are starting “to look at the warts.”
Wilson Woo, director of strategic programs in the bioresearch solutions division of Agilent Technologies, says his company has had a research collaboration with Dharmacon for the past year to study the complexities of off-target effects. Most people are used to inducing RNAi and then following one mRNA to check for knockdown, he says. “That’s a quick and dirty way, but it doesn’t really tell the whole story.” Woo says his group’s studies have shown that even the delivery vehicle for the RNAi molecule, such as a lipid, “can cause some up- and down-regulation in the cells.” Understanding what actually happened requires separating out the real gene knockdown signature from related and unrelated off-target effects as well as the delivery vehicle effects, Woo says.
Of course, it remains to be seen how much of an impact these off-target effects actually have, says Bill Hahn, an assistant professor in the department of medical oncology at the Dana-Farber Cancer Institute and a PI for the RNAi Consortium (see box on p. 34). “Until you test hundreds of these reagents then it’s hard to know exactly how to evaluate the extent of the off-target effects,” he says. “It doesn’t seem like it’s going to be a huge problem, but we’d like to know is it a one percent of the time problem, a 10 percent of the time problem, or is it worse.”
FitzGerald at BMS notes that interpreting off-target effects is incredibly complex because of an organism’s ability to handle genetic changes. “Most heterozygous mice and people for that matter … [their] biological system can take a 50 percent reduction and usually adapt to it,” he says. “Chips are a very sensitive technology to detect change,” he adds, noting that a 50 percent change in gene expression on a chip would appear statistically relevant but would offer no guarantee that it’s actually biologically relevant.
Regardless of how important they turn out to be, vendors understand that it’s critical to be able to offer customers a way to check on downstream effects. Dharmacon’s Marshall predicts that RNAi as a procedure will become increasingly reliable with the release of new reagents and appropriately validated controls. “We’re developing easy-to-use methods to understand any … off-target events in general,” he says. In the next quarter, he expects to launch a new siRNA molecule that allows for transcriptome-wide screening, he says.
Research goes on
Concerns about off-target effects, however, don’t seem to be slowing the use of RNAi in research labs. Scientists are taking advantage of microarrays and RNAi together to better understand the biological implications of gene silencing. Stephen Friend at Merck says the thought for his team was, “Let’s use the genome as a sensor pad and take the patterns that come up using expression arrays” to help improve the specificity and potency of RNAi.
Merck isn’t the only one to see the benefit of the technologies working in concert. “I think this is going to be a standard technique going forward,” says Agilent’s Woo. As a vendor, it’s to be expected that he urges users to perform a whole-genome study of their RNAi experiment; but he adds, “I wouldn’t say in every single stage of the experiment you need to do a whole-genome [scan].” He predicts that researchers will start out looking broadly across the genome and then narrow down their array studies to particular genes that seem to be affected by RNAi.
That fits with what Amit Kumar, CEO of CombiMatrix, says he has observed in this field so far. “We’re finding that a lot of people don’t want catalog arrays” such as the whole-genome chips some vendors offer, he says. “They want custom arrays.”
That philosophy makes a lot of sense for people like David Uhlinger, principal scientist in the bioinformatics R&D department at Johnson & Johnson. The pharma’s “extensive in-house microarray effort” has been in place for the last seven or eight years, according to Uhlinger, and comes in handy for spotting custom arrays. Uhlinger works with in silico expert Xiwei Wang, project lead in bioinformatics, who supplies gene leads to the wet lab team and helps hone various technology platforms. The J&J scientists actually use their arrays and RNAi in reverse: Wang’s team culls down the genes of interest with microarray experiments, and then Uhlinger’s crew attacks those selected genes with RNAi studies, primarily for target identification and validation work, Uhlinger explains.
Another perk, Wang says, is that the technologies help each other combat background noise. Microarrays, of course, are inherently noisy. Of the ways to deal with that, most — such as increasing sample size — are unappealing to pharma researchers. But “what you can do is introduce RNAi,” Wang says. “The major theme of these combinations is you want to use RNAi to create mutant background” to help minimize noisy signals from the array studies.
Meanwhile, in the functional genomics department of Novartis, nucleic acid sciences unit head Jonathan Hall says these technologies are moving in the direction of high-throughput gene screening. “That’s probably where microarrays are best combined with RNAi,” he says. The challenge for really ramping that up now is not off-target effects but rather analysis methods to correlate the dual data streams rapidly and comprehensively, Hall adds. “The number of groups that actually have the resources to run such high-throughput gene screening approaches are relatively few,” he says.
Still ahead: proteomics
As RNAi pushes forward, experts agree that microarrays won’t prove sufficient validation of specificity. The consensus seems to be that proteomics will have to play a major role before off-target effects can really be understood and predicted.
“You do need to follow the protein, no doubt,” says Woo at Agilent. “A western blot is a good way to do it, a cheap way.”
“The microarray ap-proach underestimates what is happening at the protein level,” says Peter Scacheri at NHGRI. Studies his group has done show that at least some of the minor off-target effects on the mRNA level proved to be much more pronounced on the protein side. “There’s a significant need for a protein genome-wide approach for looking at nonspecific effects of siRNAs,” he adds.
Marshall at Dharmacon says these kinds of tools are already in the works. His company is “actively exploring traditional proteomics approaches,” he says, with a focus on antibody arrays. After that, he says, “of course the next level is cellular events — looking at it in the context of living cells. We’re looking at that with GE Healthcare.”
Ultimately, of course, the real question is: will all of this help get scientists closer to churning out some kind of siRNA therapeutic? There’s less agreement on this front, in part because of other barriers such as mode of delivery.
“Obviously the next step we and all pharmas want to do is in vivo [RNAi],” says Uhlinger at J&J. Getting there is a major obstacle, adds Peter Linsley at Rosetta. “There’s been a lot of talk about using RNAi for therapeutics,” he says. “Nobody really knows what happens in vivo. … Nobody’s done the experiments yet.”
Hall at Novartis says using arrays to help beef up confidence in RNAi experiments won’t make a difference in the long run for the therapeutic side. “I don’t think it’s [this] combination … that will make siRNAs into drugs,” he says. “It’s being able to get over the hurdles of using siRNAs in vivo. … If one could find the right indication areas, these could very well make it onto the market.”
An early impact in the therapeutic arena, and one that won’t interfere with delivery issues, could come in the form of disease-specific microarrays, says CombiMatrix’s Amit Kumar. “People will eventually want a diabetes chip, a cardiovascular chip, an Alzheimer’s chip,” he says. That would provide an obvious boost for RNAi experiments targeted at those areas. But getting to those chips will take more study: “Today we don’t know what those genes are,” Kumar adds.
“There’s great potential here,” says Merck’s Stephen Friend. “But the tipping point for therapeutic use is in need of a technological breakthrough.”
Sabatini’s RNAi Spots
A key joint use of the RNAi and array technology is still emerging from early-stage work, and that’s a cell array, or reverse transfection protocol. It was first dreamed up by the Broad Institute’s David Sabatini, who says, “The general idea … is very simple. … We print nucleic acids onto a microarray [and] culture cells on top of that array.” The cells that fall on the nucleic acid spots take up the RNAi reagent, allowing for localized transfection for as many as 5,000 spots on a single chip. The concept was originally demonstrated with cDNAs and has been proven to work with siRNAs as well.
It may be the first significant step toward high-throughput, multiplexed RNAi experiments to go beyond the gene-by-gene approach. It may also have arrived slightly before its time: Sabatini started a company called Akceli in 2001 to commercialize the technology, but the firm shut its doors early this year.
To be sure, while a handful of papers from various labs have come out indicating that the process works — Sabatini says researchers have had success with human and Drosophila — there are plenty of technical obstacles facing potential users. Cost is a major one: with limited access to affordable, comprehensive libraries, the technology won’t move forward anytime soon, Sabatini says.
David Uhlinger at Johnson & Johnson says his trouble with Sabatini’s RNAi arrays is that “for a lot of the cells that we’re interested in, they don’t transfect or they don’t take things up readily, or they don’t live very long and they don’t like being abused.” For more robust cell types, he says, the technology may be more applicable.
Bill Hahn at DFCI, who calls Sabatini’s work “important,” says that the field may not be ready for high-throughput just yet. “There are significant issues related to how do you make viruses in quantities and in sufficient quality to make plated arrays,” he says. “Like most screens, they require a lot of commitment from the lab.”
Kevin FitzGerald and his team at Bristol-Myers Squibb have also worked with Sabatini and the array technology. “It does seem to work quite well,” FitzGerald says. “You can cover a lot of ground on an array.” But he has ventured away from the arrays and started using regular well plates, which can be used to array RNAi plates, and have the advantage of being able to hold more cells — ultimately giving data with a higher statistical significance. “You’re limited on a glass slide array by the size of the spot,” FitzGerald says.
But there are still plenty of people willing to tackle these challenges. John Kang, a senior applications scientist at Alpha Innotech, has been working with a company called Integral Molecular — together, the companies hope to be able to deliver a cell array and high-end imaging technology that could be used for RNAi arrays. (At this time, notes cell array firm Integral Molecular’s chief Ben Doranz, the company’s products are not yet directed at RNAi.) Kang notes that Alpha Innotech has been working on the imaging side of things, and says the company expects to introduce an instrument to look at the arrays at this month’s Chips to Hits conference with pricing expected to be “significantly less” than the seven-figure price tags on competitors’ instruments.
“Certainly this is going to require a lot of development,” says Marshall at Dharmacon, noting that his company is actively considering various approaches along these lines. “We think there’s great promise in this technology.” — MWS
‘Proceed with Caution’
Even advocates of using arrays to validate the effects of RNAi experiments agree that it’s cost-prohibitive to study the gene expression of every single RNAi test you perform. What’s a concerned scientist to do? Genome Technology rounded up the following tips to help boost your confidence in these experiments.
“If you just use one siRNA you can easily be misled,” says Peter Linsley at Rosetta. “If you see a consistent phenotype using multiple siRNAs you have a lot more confidence that what you’re seeing is due to what you think it is.” Bill Marshall at Dharmacon puts the magic number at three replicates for a sound study.
“There is nothing more important than having a good control to use in your experiments,” says Bill Marshall at Dharmacon. Adds Bill Hahn at DFCI, “If you introduce a cDNA that’s resistant to the effects of the RNAi, that can be very powerful to show that the effects are specific.”
“Everybody should proceed with caution,” notes Peter Scacheri at NHGRI, who recommends using “a real solid proteomics method to explore on a genome-wide level what happens when you silence a target gene.”
There are several ways to couple microarrays and RNAi to advance research. Here are a few our experts mentioned:
Iterative honing. After performing an RNAi experiment and evaluating the phenotype, says Kevin FitzGerald at Bristol-Myers Squibb, “you can do an array to try to figure out what is happening at the transcript level and use informatics to mine what pathways are changing. You can use the chip data to actually then go back and design further RNAi experiments.” That process can be repeated to make the RNAi experiments more and more specific and to gather additional data to help map gene function.
Making RNAi molecules. CombiMatrix has a proprietary technology it uses to synthesize up to 12,000 DNA sequences on one chip, and then transcribe that into RNA in preparation for an RNAi experiment, says CEO Amit Kumar. The method uses “electrochemical synthesis to synthesize different sequences at each spot,” he says.
Study the biological impact of knockdown. By combining “the siRNA knockdown of a specific gene product with genome-wide molecular phenotyping for functional assessment of the target gene,” says Tarif Awad, senior scientist in genomics collaborations for Affymetrix, researchers can “determine the impact of gene inhibition on the biology of the cell.”
In the Works: RNAi Consortium
Since early this year, details have been trickling out about the RNAi Consortium, centered at the Broad Institute. At press time, the Broad still hadn’t formally announced the program, but some information has come to light. The goal, according to PI Bill Hahn at the Dana-Farber Cancer Institute, is to “over a couple of years make short hairpin RNA vectors for every mouse and human gene.” A second goal, he adds, is “trying to develop methods to validate the degree of suppression for all the vectors we make.”
The labs working on this — other PIs include Eric Lander, David Sabatini, Sheila Stewart, and Nir Hacohen, among others — have production pipelines with quotas and classes of genes to work on, Hahn says. Participating institutions include Whitehead, the Broad, DFCI, and the Harvard Institute of Proteomics, according to Sabatini’s website and David Root, project leader.