For the two independent miR-HTS experiments, 115 and 116 candidates fulfilled this criterion in the first and second replicate screens, respectively (Figure 4A, B). Fifty-nine candidates were identified in both miR-HTS replicates; 58 candidate growth-inhibitory individual miRs, and 1 candidate growth-inhibitory miR cluster (see Supplementary Table S4 for the average fold changes of the two replicate screens and Supplementary Table S3 for fold changes of each miR-HTS). No growth-promoting miR candidates were reproducibly identified in both miR-HTS replicate screens (Figure 4B). This observation suggested that none of the miRs in the library strongly stimulated IMR90 cell growth under the screening conditions employed, since we detected both under- and overrepresented lenti-miRs during initial evaluation of GRE-qPCR assay performance (Figure 2). As expected, miR-34a was among the growth-inhibitory miRs identified. Cells containing the lenti-miR-34a cassette had a ≥8-fold decrease in abundance at both of the last two time points (t2 and t3). At t2, the representation of miR-34a virus–infected cells was 0.7% of that at t0, a >140-fold decrease in abundance (Supplementary Table S4 and Figure 4C). We did not observe a further decrease in barcode abundance of miR-34a fromt2 to t3, because the GRE-qPCR assay specific for miR-34a did not detect any miR-34a-infected cells at t2 (or at t3). Because the detection limit of a 20-μL qPCR reaction is Ct = 35 for detecting a single molecule, qPCR results of Ct > 35 were unreliable (39, 40). Thus, we assigned Ct = 35 when the raw Ct measurements were >35; this allowed us to calculate a minimal fold change when input templates were below our detection limit [as is commonly practiced in the literature (41, 42)].
We observed different growth-inhibitory kinetics among the 59 growth-inhibitory candidates (Figure 4C, Supplementary Table S4). Twenty-six of the 59 candidates caused ≥8-fold decreased in abundance at that final time point (i.e., candidates with t3/t0 ≤0.125 in Supplementary Table S4), illustrated by the miR-449a~449b cluster. Fifteen of the 59 candidates caused ≥8-fold decreased abundance at both of the last two time points (i.e., candidates with t2/t0 and t3/t0 both ≤0.125 in Supplementary Table S4), illustrated by miR-34a. Eight of the 59 candidates caused ≥8-fold decreased abundance at all three experimental time points (i.e., candidates with t1,2,3/t0 all ≤0.125 in Supplementary Table S4), illustrated by miR-550a-2. We also identified 10 growth-inhibitory candidates that were undetected at all four time points (i.e., candidates with t0,1,2,3/t0 = 0 in Supplementary Table S4), illustrated by miR-892b. The presence of these 10 lenti-miR viruses in the initial library was verified by infecting REH, a human acute leukemia cell line, using the same lot of lenti-miR library (not shown). Therefore, failure to detect these 10 lenti-miRs in IMR90 cells during miR-HTS was not due to a technical problem with this lentivector or GRE-qPCR assay, because REH cells infected with these 10 lenti-miRs were easily detected at t0 when a miR-HTS was conducted in the REH.
We selected 12 candidate miRs with different growth inhibitory kinetics to test individually for validation. For each selected miR, IMR90 cells were transfected with mature miR mimic, and cell growth was monitored (Figure 5A). Caenorhabditis elegans miR-67 (cel-miR-67), which has minimal sequence homology to any human miR, was used as a negative control. miR-937, which was found not to inhibit IMR90 growth in the miR-HTS assays, provided an additional predicted negative control. Nine of 12 (75%) growth-inhibitory miR candidates tested suppressed IMR90 growth as compared with cel-miR-67 mimic, whereas miR-937 mimic did not affect IMR90 growth (Figure 5A, Supplementary Figure S2). Transient overexpression of candidate miR mimics demonstrated their direct growth-inhibitory capacity, and ruled out the possibility that the effects observed in the miR-HTS were due either to serendipitous mutations that may have arisen during the month-long miR-HTS culture period or due to lentiviral insertional disruption of the host genome. Except for miR-513a, for which we assayed both mature miR strands that could be expressed from a single miR hairpin precursor, we assayed only the mature strand, which is currently thought to be expressed predominantly [most mimics used only had one strand commercially available; strand expression data was obtained from miRBase, www.mirbase.org (43, 44)]. Interestingly, both miR-513a-5p and miR-513a-3p mimics suppressed IMR90 growth (Figure 5A), suggesting that both strands contributed to the miR-513a-mediated growth inhibition observed in the miR-HTS. Since both mature miR strands are expected to have been overexpressed during the miR-HTS, it remains to be determined whether the mature strand mimics not assayed in this study could suppress IMR90 growth, especially for the three tested miRs (i.e., miR-150, miR-381, and miR-606) that appeared to be false-positive candidates. Thus, at least 75% (i.e., 9 of 12) of the candidate miRs examined were verified to inhibit IMR90 growth (Figure 5, Supplementary Figure S2).