Worm Breeder's Gazette 16(5): 22 (February 1, 2001)

These abstracts should not be cited in bibliographies. Material contained herein should be treated as personal communication and should be cited as such only with the consent of the author.

On the reproducibility of large-scale RNAi screens

Fabio Piano1, Aaron Schetter1, Diane Morton1, Kris Gunsalus1, Valerie Reinke2, Stuart Kim2, Ken Kemphues1

1 Department of Molecular Biology and Genetics, Cornell University
2 Department of Developmental Biology, Stanford University Medical Center

RNAi is being used routinely to determine loss-of-function phenotypes and recently large-scale RNAi analyses have been reported (1,2,3). Although there is no question about the value of this approach in functional genomics, there has been little opportunity to evaluate reproducibility of these results.

We are engaged in RNAi analysis of a set of 762 genes that are differentially expressed in the germline as compared to the soma (4 -- "Germline"), and have reached a point in our analysis that allows us to look at the issue of reproducibility. We have compared the RNAi results of genes in our set that were also analyzed by either Fraser et al. (1 -- Chromosome 1 set "C1") or Gonczy et al. (2 -- Chromosome 3 set "C3").

In making the comparison we have taken into account the different operational definition of "embryonic lethal" used by the three groups. In the C3 study, lethal was scored only if there were fewer than 10 surviving larva on the test plate, or roughly 90% lethal. In our screen and the C1 screen the percent survival was determined for each test. To minimize the contribution of false positives from our set, in our comparison with the C1 set we defined our genes as "embryonic lethal" if at least 30% of the embryos did not hatch, but included all lethals defined by Fraser et al. (> 10%). For our comparison with the C3 set, we used a more restrictive definition of "embryonic lethal" that required that 90% of the embryos did not hatch. (This means that in Table 1, five genes from our screen that gave lethality between 30-90% were included in the not lethal category; one of these was scored as lethal by Gonczy et al.).

We have analyzed 149 genes from the germline set that overlap with the C1 set and 132 genes that overlap with the C3 set. The table below shows the number of genes scored as embryonic lethal (EL) or not embryonic lethal (NL) in each study. (Note that these comparisons do not include data from our published collection of ovary-expressed cDNAs.)

Table 1. Comparing RNAi analysis of the same genes in different studies.

GermlineChromosome 1GermlineChromosome 3
NL (117)EL (32)NL (97)EL (35)
NL (104)1004NL (89)872
EL (45)1728EL (43)1033

Overall, the degree of reproducibility is high. The concordance between our results and the published results was 86% with C1 (128/149 genes) and 90% with C3 (120/132).

However, we scored a larger number of genes as giving rise to embryonic lethal phenotypes than the other studies did. What does this mean? One possibility is that we are generating a large number of false positives (God forbid!). The other interpretation is that there is a fairly high frequency of false negatives in each screen (4-8% in our screen (2/45; 4/49); 22% in the C3 screen (10/45); and 35% (17/49) in the C1 screen).

It is no surprise that the different methods used by the three groups resulted in slightly different outcomes and we can only speculate on which methodological variation contributed most. In comparing our methods to those used in the C3 study we note that our two groups used different primer pairs for each gene; that we tested genes individually while they tested genes in pairs; and that the operational definition of "embryonic lethal" differed. Considering the latter two differences, we speculate that even with pools of two, the competition noted by Gonczy et al. in dsRNA pools could reduce levels of lethality below the 90% cutoff.

The major difference between our approach and the C1 approach is feeding vs. injection, raising the possibility that for some genes feeding may be a less effective means of administering dsRNA.

Whatever the basis for the difference, these comparisons indicate that genes scored as "non-lethal" in any single study may show an embryonic lethal RNAi phenotype when reanalyzed. It therefore seems useful to have more than one pass at analyzing C. elegans genes via RNAi.

We are indebted to P. Gonczy for very useful comments.

  1. Fraser, A. G., Kamath, R. S., Zipperlen, P., Martinez-Campos, M., Sohrmann, M. and Ahringer, J. (2000). Functional genomic analysis of C. elegans chromosome I by systematic RNA interference. Nature408, 325-330.
  2. Gonczy, P., Echeverri, G., Oegema, K., Coulson, A., Jones, S. J., Copley, R. R., Duperon, J., Oegema, J., Brehm, M., Cassin, E. et al. (2000). Functional genomic analysis of cell division in C. elegans using RNAi of genes on chromosome III. Nature 408, 331-336.
  3. Piano, F., Schetter, A. J., Mangone, M., Stein, L. and Kemphues, K. J. (2000). RNAi analysis of genes expressed in the ovary of Caenorhabditis elegans. Curr Biol 10, 1619-1622.
  4. Reinke, V., Smith, H. E., Nance, J., Wang, J., Van Doren, C., Begley, R., Jones, S. J., Davis, E. B., Scherer, S., Ward, S. et al. (2000). A global profile of germline gene expression in C. elegans. Mol Cell 6, 605-616.