Two members of Memorial Sloan-Kettering Cancer’s high-throughput drug screening facility this month published a study highlighting major shortcomings of RNAi screens, including the failure of such studies to yield reproducible hits.
The publication also showed that screens based on pooled shRNAs led to “unprecendented enrichment” of hits that do not show up in screens conducted with individual siRNAs or shRNAs, raising questions about the scientific merit of the approach, Hakim Djaballah, director of the MSKCC screening facility and co-author of the study, told Gene Silencing News.
According to Djaballah, his study stems from previous efforts to publish the results of an RNAi screen as a educational tool that can help guide other investigators in setting up their own screening experiments.
In that paper, he and his colleagues used an arrayed, genome-scale, lentiviral-enabled shRNA screen to identify lethal and rescuer genes, and described a methodology for performing such large-scale screens and analyzing high-confidence hits.
But the group ran into trouble with the journal’s reviewers, who raised questions about the failure of the screen to home in on genes that are well-established as essential for cell survival.
“That prompted us to go into the literature and try to figure out if everyone who has ever run a lethality screen managed to identify such hits,” Djaballah explained.
“Each time you perform [such a] screen, you should identify the same signatures of genes that are essential for the survival of any cell,” he added. “When we put the story together, we were very surprised and dismayed that there are zero hits and commonality across everything that has been published.”
To conduct their study, Djaballah and Bhavneet Bhinder, a computation analyst at MSKCC, surveyed the literature for published RNAi screens and collected a set of 64 gene lists from 30 studies comprising both siRNA and shRNA technologies, according to the latest paper, which appeared in Combinatorial Chemistry & High-Throughput Screening.
“We censored an initial hit list of 7,430 nominated genes resulting from an arsenal of different analysis methods utilized by the different groups to score hits in genome-wide, as well as focused screening data sets,” they wrote. “Considering a conservative estimate of at least 25,000 genes in the human genome, it is rather surprising that 30 percent of them seem to be required for cellular viability and survival.”
Gene lists were segregated into groups based on whether they were derived from siRNA screens or shRNA screens, with genome-wide and focused screens being analyzed separately. Curiously, none of the gene hits were common to all of the studies, and some of the major players of cellular viability were missing entirely.
For instance, PLK1 was identified as a “prominent gene candidate” in a systematic analysis of siRNA duplex gene lists, yet it was only marginally present in shRNA gene lists. Meanwhile, EIF5B emerged as the most common hit, but only in shRNA screens.
The most surprising finding came when Djaballah and Bhinder focused on just pooled shRNA screens and discovered that 5,269 out of 6,664 gene candidates — roughly 80 percent — culled from all shRNA screens were exclusive to the pooled format.
To Djaballah, the fact that shRNA screens identified so many genes as hits at all was a red flag. “It’s almost 7,000 genes, and the genome is [around] 25,000” genes, he said. “The sheer percentage is just ridiculous.”
Overall, the paper emphasizes the “dismal reproducibility of RNAi screening data outputs with zero commonality across [around] 147 distinct lethality gene lists, while showcasing a compelling prominence of gene candidates resulting from relative hairpin depletions in pooled shRNA screens, questioning the merits of this technology in gene discovery and validation,” Djaballah and Bhinder wrote.
Djaballah said that there remains much value in RNAi screens, but urged the research community to pay close attention to data-analysis practices especially given the technology’s combinatorial nature.
However, he cautioned against the use of pooled shRNAs — a viewpoint he conceded was controversial.
“For shRNAs, you have to remember that all the sequences are theoretical … [and] there is no evidence that the cell itself will generate the particular sequence you think it will,” he said, citing work from other groups in which the same shRNA library was used to run screens in two different cell lines and gave different results.
“That speaks to the fact that every cell line will have different Dicer activity [and] different processing activity,” he said. “You may have a sequence you think is targeting gene X, but … you are actually targeting six other genes.”
Djaballah added that, in the roughly 13 years RNAi screening has been around, “there hasn’t been anything that came out of [pooled RNAi screens] that enabled a drug-discovery program anywhere in the world. What has come out of it is a large catalog of targets that no one is going to be bothered looking [at] because there are so many of them and people have seen so many papers being retracted.
“Our advice has been not to do a pooled shRNA screen — it’s not really worth it,” he said. “And from my perspective, it doesn’t have any scientific merit because of the deconvolution aspect associated with it.”
Screens based on individual shRNAs, while slightly more expensive, “work just as well,” he said. Meanwhile, “siRNAs are much more affordable, [their use is] straightforward, some people have had very good success with” them.