Table 1.
Yield of different protocols used for microRNA quantification.
Fig 1.
Flowchart displaying selection of informative miRNA targets.
From 377 possible target miRNA, 144 are eliminated due to low expression in ≥80% of samples. After exclusion of fifteen target miRNA with a CV ≥4%, 181 target miRNA with a CV<4% are retained (low technical variability). The remaining target miRNA with unknown CV are also retained. The rationale behind the cutoff values of Cq ≤35 and >20% of samples is explained in the Discussion. Cq = quantification cycle, CV = coefficient of variation, miRNA = microRNA, TLDA = TaqMan Low Density Array.
Table 2.
Evaluation of different normalization techniques.
Fig 2.
Box and whisker plots of measures of data dispersion: CV (left) and SD (right).
First, ΔCq was calculated using the mentioned normalizing method for all 218 miRNA in each sample, then mean ΔCq values and SD were calculated for each miRNA across the 18 samples, finally CV was calculated for each miRNA by dividing SD by mean. Each normalization strategy consistently resulted in an increase in CV compared to the raw data CV. SD decreased in all normalization strategies except for U6 compared to the raw data SD. Ath = ath-miR-159a spike-in normalization, CV = coefficient of variation, Cq = quantification cycle, ΔCq = relative expression of target miRNA (calculated by Cq of target miRNA—Cq of reference gene), GM = global mean normalization NF = NormFinder algorithm normalization, QB = Qbase normalization using geNorm algorithm, Raw = raw (not normalized) data, SD = standard deviation, U6 = U6 endogenous reference RNA normalization.
Fig 3.
Minus average (MA) plots for each normalization method, with LOESS lines plotted for each sample.
Each colored line represents the LOESS line for one sample. NF and QB normalization methods display the straightest LOESS lines closest to zero, i.e. less dispersion compared to Ath, U6 and GM methods and raw data. Ath = ath-miR-159a spike-in, LOESS = local regression, GM = global mean, NF = NormFinder algorithm, QB = geNorm algorithm/Qbase software, U6 = endogenous control RNA U6.
Fig 4.
Agreement between different normalization methods.
Intra-class correlation coefficient closer to 1 indicates better agreement. Good agreement is seen between U6, GM and NF methods and between Ath and QB methods. Poor agreement is seen between other methods. Ath = ath-miR-159a spike-in, GM = global mean, NF = NormFinder algorithm, QB = geNorm algorithm/Qbase software, U6 = endogenous control RNA U6.
Fig 5.
Relationship between number of remaining informative miRNA and the needed percentage of samples with a Cq ≤35.
If a miRNA needs to be expressed in all samples (100%) with a Cq ≤35 to be retained, only 62 informative miRNA would remain. If a miRNA does not need to be expressed in any sample (0%) with a Cq ≤35 to be retained, all 377 target miRNA would remain. The elbow of the curve is situated at 20%, which means a miRNA needs to be expressed in 20% of samples with a Cq ≤35. Cq = quantification cycle, miRNA = microRNA.