pcrEfficiency: a Web tool for PCR amplification efficiency prediction
© Mallona et al; licensee BioMed Central Ltd. 2011
Received: 12 May 2011
Accepted: 20 October 2011
Published: 20 October 2011
Skip to main content
© Mallona et al; licensee BioMed Central Ltd. 2011
Received: 12 May 2011
Accepted: 20 October 2011
Published: 20 October 2011
Relative calculation of differential gene expression in quantitative PCR reactions requires comparison between amplification experiments that include reference genes and genes under study. Ignoring the differences between their efficiencies may lead to miscalculation of gene expression even with the same starting amount of template. Although there are several tools performing PCR primer design, there is no tool available that predicts PCR efficiency for a given amplicon and primer pair.
We have used a statistical approach based on 90 primer pair combinations amplifying templates from bacteria, yeast, plants and humans, ranging in size between 74 and 907 bp to identify the parameters that affect PCR efficiency. We developed a generalized additive model fitting the data and constructed an open source Web interface that allows the obtention of oligonucleotides optimized for PCR with predicted amplification efficiencies starting from a given sequence.
pcrEfficiency provides an easy-to-use web interface allowing the prediction of PCR efficiencies prior to web lab experiments thus easing quantitative real-time PCR set-up. A web-based service as well the source code are provided freely at http://srvgen.upct.es/efficiency.html under the GPL v2 license.
Since the development of quantitative PCR (Q-PCR) in the early nineties , it has become an increasingly important method for gene expression quantification. Its aim is to amplify a specific DNA sequence under monitoring and measuring conditions that allow stepwise quantification of product accumulation. Product quantification has fostered the development of analysis techniques and tools. These data mining strategies focus on the cycle in which fluorescence reaches a defined threshold (value called quantification cycle or Cq) [2, 3]; with the Cq parameter, quantification could be addressed following two approaches: (i) the standard curve method  and (ii) the ΔΔCq method .
It is worth noting that these classical quantification methods assume that amplification efficiency is constant or even equal to 100%. An efficiency value of 100% implies that during the exponential phase of the Q-PCR reaction, two copies are generated from every available template. But it has been shown that these assumptions are not supported by experimental evidences . With the aim of estimating PCR efficiency, and thus to include it in further analysis procedures, two strategies have been developed: (i) kinetics-based calculation and (ii) standard curve assessment.
Taking into account the reaction kinetics, which is basically equivalent to the bacterial growth formulae , amplification efficiency could be visualized in a half-logarithmic plot in which log transformed fluorescence values are plotted against the time (cycle number). In these type of graphic representations, the phase of exponential amplification is linear and the slope of this line is the reaction efficiency . Empirical determinations of amplification efficiencies show that ranges lay between 1.65 and 1.90 (65% and 90%) . Standard curve-based calculation method relies on repeating the PCR reaction with known amounts of template. Cq values versus template (i.e. reverse transcribed total RNA) concentration input are plotted to calculate the slope. Laboratories where few genes are analyzed for diagnostic may develop standard curves but they are in most cases out of scope for research projects where tens-hundreds of genes will be tested for changes in gene expression.
Several aspects influence PCR yield and specificity: reagents concentration, primer and amplicon length, template and primer secondary structure, or G+C content . The goals of a PCR assay design are: (i) obtaining the desired product without mispriming and (ii) rising yield towards optimum. In most cases, the sequence to amplify is a fixed entity, so setting up an efficient reaction involves changes in reagents concentrations (salts, primers, enzyme) and specifically an optimal primer design. Thus a plethora of primer designing tools have been published, regarding as little as G+C content for T m calculation [11, 12], evaluating salt composition  or even employing Nearest Neighbor modules, which consider primer and salt concentrations .
Efficiency values are essential elements in the ΔΔCq method and its variants: relative quantities are calculated using the efficiency value as the base in an exponential equation in which the exponent depends on the Cq. Thus efficiency strongly influences the relative quantities calculation, which are required to estimate gene expression ratios .
In this work, we analysed Q-PCR efficiency values from roughly 4,000 single PCR runs with the aim of elucidating the major variables involved in PCR efficiency. With this data we developed a generalized additive model (GAM), which relies on nonlinear regression analysis, and implemented it in a open, free online web tool allowing efficiency prediction.
Statistical results for univariate analysis
Z = -7.4398
Z = -5.423
Sequence G+C content
Z = -10.2664
Z = 2.1004
Z = 3.9818
Z = -5.294
Z = -7.1808
Primers T m
Z = 1.4653
Primers self complementarity
Z = 11.9002
Z = 4.4161
Primer GC imbalance
Z = 11.1367
Primers GC content
Z = 4.5921
Z = -3.4951
χ2 = 585.616
χ2 = 1241.562
Variety or line
χ2 = 585.8386
χ2 = 940.8915
χ2 = 727.4887
Primer's 3' last two nucleotides
χ2 = 237.911
Pr (> | t |)
Approximate significance of smooth terms
R-sq.(adj) = 0.41 Deviance explained = 42.1%
GCV score = 0.0094091 Scale est. = 0.0092293 n = 3944
In order to ease PCR efficiency prediction prior to wet-lab PCR set-up, we developed a user-friendly, freely available web tool for assessing primer suitability before and during primer design. For this purpose, we wrote a set of Python/Biopython scripts  requested through a Common Gateway Interface (CGI). These scripts were developed to call to the Primer3 software , working inside the EMBOSS package . The web tool called PCR efficiency calculator allows primer design starting from a DNA fragment producing a set of theoretical PCR efficiency values. It also predicts PCR efficiency values for preexisting primers and DNA template combinations.
Several tools have been developed to assess primer design procedures. Most of them consider hairpin structure formation avoidance, selection of nucleotides in 3' termini, primer melting temperature, etc. . However, intrinsic amplicon characteristics are not contemplated in primer design. The work we present includes this important parameter as amplification was found to be highly dependent on template structure (Tables 1 and 2). Indeed PCR specificity or PCR failure have been found to be dependent on sequence similarity between primers and template, lack of mismatches, or number of priming sites [29, 30]. Using logistic regression analysis, Benita et al. found that PCR success is highly dependent on regionalized G+C content in the template thus showing the importance of template structure as a second step in PCR optimization. Generally, PCR success is evaluated as a dichotomy by presence or absence of product. However in Q-PCR experiments amplification efficiency becomes an important parameter to perform proper statistical analysis that should yield the actual differences in expression between several transcripts. Thus PCR efficiency becomes as important as Cq values to determine differential gene expression. Our model showed that G+C content in the amplicon plays a key role in PCR efficiency confirming previous work and including it inside as a predictor of PCR efficiency.
Multiple parameters, such sequence palindrome abundance or the nucleotide at the 3' primer termini, were found to significantly contribute to the PCR efficiency when analyzed separately. However, they were not significant when included to a multiple component GAM. Unexpectedly, variables such the species or the line the template was extracted from, or the operator involved in the PCR set-up, were found to be significant. Nonetheless, the modelling procedure disregarded these variables, as GAM with and without them did not differ significantly. Thus we ascribe this result as covariation, because each primer pair-amplicon combination is normally used in a certain project which is limited to a species, operator and one or a few lines.
The model presented in this work estimates an efficiency value per PCR reaction regarding three parameters, each of one represents the interaction between two independent variables: the interaction between G+C content of the amplicon with its length; the interaction between G+C content of the primers with their length; and the interaction between G+C content imbalance between primers (gcImbalance; their difference in G+C) with their tendency to hybridize and thus to form primer dimers (primerDimers). Our model gives a high influence to the difference of G+C content between primers. Previous works noted that PCR using unequal primer concentrations have better efficiencies when their melting temperatures differ in ≥ 5°C . However, when our tool is piped to the Primer3 primer-design work flow, this difference is restricted by the Primer3 algorithms, thus avoiding design of highly unequal primers. Very high or very low amplicon G+C content affects amplification success [32, 33]. Specially, regionalized G+C-content has been shown to be relevant in PCR success prediction .
The comparison of the model performance in the ROC space discriminates 1.60 as the classifier threshold which leads to the worst model behaviour, but shows only minor differences for the other cut-offs. The analysis of PR curves allows further comparisons and highlights that 1.80 shows the highest degree of resolution. Tuomi and coworkers described 1.80 as boundary for optimized PCR reactions .
It is worth noting that the tool developed aids in primer design prior to the wet lab experiments. Since it remains clear that there are physical constraints which establish the maximum PCR efficiency of a given set of one amplicon and a pair of oligos, bias is introduced in many ways (pipetting, reactives, PCR machine, etc.). We would like to point out that our work does not intend to substitute the experimental efficiency calculation nor modify the quantification settings; its aim is to complement the existing primer design tools and thus minimize the need for primer combination testing.
Using a wide range of amplicons and PCR set-ups, we statistically modelized the response of the PCR efficiency value, a parameter affecting PCR success and involved in effective gene expression quantitation. In order to ease PCR primer design for Q-PCR experiments, the efficiency-predicting model was included in the Primer3 design pipeline and freely provided as a web tool. This tool should help to generate primer combinations with similar theoretical efficiencies to well established PCR primers or to ease multiplex PCR reactions where efficiencies should be similar among templates.
We used a variety of DNA templates to obtain data for efficiencies including genomic DNA from bacteria, yeasts, plants and humans, and plasmid DNA. Samples using cDNA as template were produced from isolated mRNA from different sources. Synthesis of first strand cDNA was performed from DNAase treated mRNA, using the Maxima kit from Fermentas as described in the protocol. Samples amplified by whole genome amplification using the ϕ29 DNA polymerase were performed with the Genomiphi kit (GE-Healthcare) according to manufacturers manual. A summary of the data is available as Additional File 1.
PCR reactions used were carried out with the SYBR Premix Ex Taq (TaKaRa Biotechnology, Dalian, Jiangsu, China) in a Rotor-Gene 2000 thermocycler (Corbett Research, Sydney, Australia) and analysed with Rotor-Gene analysis software v.6.0 as described before . A second set of reactions was performed with a Mx3000P machine (Stratagene, Amsterdam) and analyzed with the qpcR R package . Reaction profiles used were 40 cycles at 95°C for 30 s, an amplicon-specific annealing temperature for 20 s and amplification at 72°C. In order to ensure the specificity of the reaction, uniqueness of Q-PCR products were checked by melting analysis (data not shown). Fluorescence data acquisitions during the cycling steps were collected at 72°C step, temperature at which eventual primer-dimers should be melted, thus avoiding artifactual contribution to the fluorescence measure. Once finished, analysis was followed by a melting curve whose ramp was delimited between the annealing temperature and 95°C. Reaction volume was 15 μL and each primer was 240 nM.
Reaction efficiency was calculated using the amplification curve fluorescence, analyzing each PCR reaction (tube) separately as before . Efficiency value (E) was defined as , in which n is determined as the 20% value of the fluorescence at the maximum of the second derivative curve. Efficiency calculations were performed with the qpcR R package . Curves were formed by 40 points, each one representing a fluorescence measure in each amplification cycle. The Rotor-Gene 2000-based runs were baseline corrected either by standard normalization (substraction of the fluorescence present in the first five cycles of each sample) or by "dynamic tube" normalization (which uses the second derivative of each sample trace to determine the take-off, thus asigning a threshold separately to each reaction), whereas the Mx3000P were by "adaptive baseline" correction (which assigns a threshold independently to each sample).
Data mining was performed with the R statistical environment v2.7.1 and v2.10.1  with the following libraries: coin v1.0-4 , mgcv v1.7-5 , ROCR  v1.0-4, compute.es v0.2  and verification v1.31 . The final model was implemented in a CGI server-side set of Python/Biopython scripts interacting with the web browser requests. Source code of both modelling procedures and the server-side application are available at the website.
This work was funded by Fundación Séneca 11895/PI/09 and MICINN BFU-2010-15843. IM received a grant from Fundación Séneca. The authors would like to thank Luis Pedro García and SEDIC for aid in conducting the computational-intensive part of this work, as well the web server set-up. Luciana Delgado-Benarroch is acknowledged for web usability suggestions and Marta Pawluczyk for comments on the manuscript.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.