Given that IMR90 yields ~0.002 ng genomic DNA per cell, >250 million IMR90 cells would be needed to produce the 578 μg genomic DNA needed to perform 578 GRE-qPCR assays (250million IMR90 cells would require ~62 confluent T175 flasks). To reduce cell-number requirements for the miR-HTS, we utilized a nested PCR strategy to produce sufficient DNA from fewer cells. Integrated lentiviral DNA segments containing the miR cassette and GFP transgene were first amplified from genomic DNA samples, using external common forward and reverse primers designed to minimize PCR amplification bias favoring shorter amplicons (Figure 1A). Approximately 70% of the miR cassettes were 500–600 bp, with a global size range of 182–951 bp (Supplementary Figure S1A). The external common primers included a total of 1850 bp flanking each miR cassette, which reduced the relative length ratio of the longest to the shortest amplicon from 5.2-fold to 1.4-fold, as compared with the miR cassette alone (Supplementary Figure S1B). GRE-qPCR assays and GFP loading control qPCR assays were all nested within the amplicon defined by the external common forward and reverse primers (Figure 1A).
After removal of the external common primers to prevent their unwanted further amplification, purified PCR amplicons were used as templates for GRE-qPCR assays to determine the relative abundance of each miR cassette (Figure 1B). Using PCR amplicons (obtained by 25 cycles of amplification of the same initial genomic DNA samples used in Figures 2A, B), GRE-qPCR assays accurately quantified the number of cells infected with either lenti-miR virus, ranging from 25 to 104 cells (Figure 2C, D). The fitted regression line showed a strong log-linear inverse relationship between corrected Ct values and numbers of cells (r2 = 0.99 for either miR-155 or miR-222, with only 0.2 ng PCR amplicon used per GRE-qPCR assay; Figure 2C, D). We could not reproducibly detect a signal when the infected cell number was 10. Similar detection sensitivity and accuracy was observed when either 1.5 or 1 μg genomic DNA was used to generate PCR amplicons, whereas the accuracy and sensitivity in cell number quantification became unreliable when only 0.5 μg genomic DNA sample was used (not shown). We found that 10–30 cycles of PCR amplification generated similar sensitivity and accuracy in cell number quantification (not shown). We opted to use 25 cycles in the miR-HTS to minimize cost; this required only 0.2 ng purified PCR amplicons per GRE-qPCR assay, instead of the 20 ng purified PCR amplicons, which would be required using 20 cycles of amplification. The yield of PCR amplicons was ~300 ng per 1.5 μg genomic DNA at 25 cycles of amplification, generating more than sufficient templates for all 578 different GRE-qPCR assays. While 1 or 1.5 μg genomic DNA was optimal to maintain accuracy and sensitivity using this nested PCR strategy for IMR90 cells, optimization should be performed as in Figure 2 if the genomic DNA yield per cell of a given cell line is significantly different from that of IMR90.
Once the GRE-qPCR assays were shown to accurately quantify the abundance of lenti-miR–infected cells, we performed the miR-HTS twice in IMR90 cells, as illustrated in Figure 1B. Genomic DNA was harvested from the lenti-miR library–infected cultures at the reference time point (t0, 3 days post-infection, to allow virus integration and GFP expression) and at three experimental time points (t1, t2, t3; days 11, 19 and 27 post-infection, respectively, corresponding to every 2 passages; 4–5 population doublings of IMR90 cells per passage). To determine fold changes in abundance of each lenti-miR over time, the abundance of each lenti-miR at each experimental time point was quantified and normalized to its abundance at the reference time point (i.e., tN/t0). Cells overexpressing a miR with no effect on growth would be expected to maintain their starting representation over time (tN/t0 = 1; e.g., miR-Z). In contrast, cells overexpressing a miR promoting growth should become overrepresented (tN/t0 > 1; e.g., miR-Y), and cells overexpressing a miR inhibiting growth should become under-represented (tN/t0 < 1; e.g., miR-X).
Technical replicates of GRE-qPCR assays were highly reproducible; 2 independent GRE-qPCR assays performed using the same templates (i.e., assaying a single batch of PCR amplicons) were nearly identical (r2~1; Figure 3A). Furthermore, when 2 batches of purified PCR amplicons were generated independently and then each used separately as templates for GRE-qPCR assays, only 4% of the GRE-qPCR assays had differences of ≥3 cycles in corrected Ct (Figure 3B). A 3-cycle difference in corrected Ct (i.e., ΔΔCt = 3) is equivalent to a 8-fold change in abundance (37). The 4% frequency of unexpected, ≥8-fold differences observed between the replicates might be explained by the Monte Carlo effect occurring during independent PCR amplifications (38); this further suggested that any observed change less than 8-fold might due to variations in PCR amplification instead of a true difference in abundance. Therefore, we set our candidate selection threshold at ≥8-fold change from t0; that is, only miRs that showed a ≥8-fold change in abundance in miR-HTS were considered as growth-inhibitory or growth-promoting candidates (e.g., an 8-fold increase would be tN/t0 = 8, and an 8-fold decrease would be tN/t0 = 0.125 in Figure 4, Supplementary Table S3, and Supplementary Table S4). As we gain additional experience with this assay, we should be able to estimate the variance of the method more precisely, and we may be able to reduce the 8-fold threshold by reducing the Monte Carlo effect, for example, by pooling multiple independent PCR amplifications prior to the GRE-qPCR.