Deconvolution of the confounding variations for reverse transcription quantitative real-time polymerase chain reaction by separate analysis of biological replicate data

Abstract

Reverse transcription quantitative real-time polymerase chain reaction (RT-qPCR) uses threshold cycles (Ct values) for measuring relative gene expression. Ct values are signal-to-noise data composed of target gene expression and multiple sources of confounding variations. Data analysis is to minimize technical noises, evaluate biological variances, and estimate treatment-attributable expression changes of particular genes. However, this function is not sufficiently fulfilled in current analytic methods. An important but unrecognizable problem is that Ct values from all biological replicates and technical repeats are pooled across genes and treatment types. This violates the sample-specific association between target and reference genes, leading to inefficient removal of technical noises. To resolve this problem, here we propose to separate Ct values into replicate-specific data subsets and iteratively analyze expression ratios for individual data subsets. The individual expression ratios, rather than the raw Ct values, are pooled to determine the final expression change. The variances of all biological replicates and technical repeats across all target and reference genes are summed up. Our results from example data demonstrate that this separated method can substantially minimize RT-qPCR variance compared with the traditional methods using pooled Ct profiles. This analytic strategy is more effective in control of technical noises and improves the fidelity of RT-qPCR quantification.

Figure 1. Variance components produced in RT-qPCR experiments

RT-qPCR usually includes 5 experimental steps, each of which inevitably produces either technical variance (green bars) or biological variance (blue bars). The variance from all steps accumulates stepwise to final Ct data. Thus the Ct values among experimental replicates set up from the first to last RT-qPCR steps exhibit decreased variance. Each step is suitable to set up experimental replicates; however, if triplicates are used for each step, they will generate a large volume of Ct values (3⁵ = 243/gene/treatment type). For realistic sampling, most often researchers design RT-qPCR replicates at two experimental steps (red arrows).

Figure 2. Ct distribution among independent replicates

A total of 153 Ct means of PCR duplicates from 3 independent replicates are plotted across 17 genes (3 target plus 14 validated reference genes; 5 non-selected reference candidates are not included). Ct values for a particular gene are pooled together without distinguishing treatment types. Note that the Ct values in replicate 1 (blue circles) are constantly lower than those in replicate 3 (red triangles). The Ct values in replicate 2 (green crosses) are constantly between replicates 1 and 3. The gene names marked in red are the target genes, and the rest are the validated reference genes. Note that all replicates in these data were designed in the same PCR plate to avoid run-to-run variations and using exactly the same threshold setting for Ct readout [14]. Otherwise, the difference among replicates would be larger.

Figure 3. The expression variances of target genes are sensitive to analytic methods

The normalized expression ratios (R, in blue) of the target genes and their standard errors (SE, in red) are calculated using two different methods. The traditional method combines all replicate Ct (3 biological replicates × 2 PCR repeats) into an average value across genes for data normalization and calculation of R (R combined, blue circles) and SE (SE combined, red circles). The separated method is based on sample-specific data normalization. The R and SE are labeled as “R separated” (blue crosses) and “SE separated” (red crosses) respectively. Note that most “R combined” and “R separated” overlap (except for Hsp70 in T05); whereas all “SE combined” and “SE separated” have substantial differences (except Tor in A05). Normalization details are described in Table S2 and S3 in the supplementary material.