A collection of primer pairs from PrimerBank covering most known mouse genes was tested by QPCR, agarose gel electrophoresis, sequencing and BLAST. An overview of the procedure used for primer validation can be seen in Figure 2. Universal mouse total RNA was reverse transcribed using random hexamers and the cDNA was used as a template. 26855 primer pairs, corresponding to 27684 transcripts, were tested by QPCR and the amplification plots and dissociation curves were analyzed. The same PCR conditions were used for all reactions. PCR amplification plots indicate SYBR Green I fluorescence which is proportional to PCR product formation. Dissociation curves indicate the loss of SYBR Green I fluorescence as the PCR product duplex dissociates. T_m and the shape of the dissociation curve are a function of GC content, sequence and length 231. From the amplification plots, PCR products appeared typically between 19 and 27 cycles of PCR, with a small variation of 1 or 2 cycles depending on the length of the PCR product and thus the amount of SYBR Green I bound to it. As a general observation, most shorter length products (from 60 bp) appeared between 20 and 27 cycles and their T_ms were between 75°C and 85°C, and most longer length products (>200 bp) appeared between 17 and 27 cycles and their T_ms were between 80°C and 90°C.

Agarose gel electrophoresis was used to confirm the correct size of the PCR product, and sequencing and BLAST were used to confirm that the expected transcript had been amplified. All successfully sequenced samples (24476) were BLAST analyzed. From the primer validation procedure, primer pairs were grouped into successful or failed, according to the analysis criteria. From 26855 primer pairs tested 17483 (65.1%) primer pairs, corresponding to 18324 transcripts, were found to be successful by QPCR, agarose gel, sequencing and BLAST analysis. 22189 (82.6%) primer pairs were successful based on agarose gel electrophoresis analysis and 19453 (72.4%) primer pairs were successful based on BLAST analysis. Primer pairs which failed based on the experimental validation procedure can be grouped into various types. Table 3 presents a classification of the types of failures. In a few cases (less than 0.8%), primer pairs were found to be successful based on the gel or BLAST analysis criteria, but no amplification could be detected with SYBR Green I. Sequencing can be very sensitive and a low abundance amplicon can thus be sequenced successfully despite low amounts. Also, in many cases where PCR products were short (~60–80 bp) it was not possible to obtain sequencing information for these samples.

A few representative examples of primer pairs are described [see Additional files 1, 2, 3, 4, 5], to demonstrate in detail the analysis of the results generated from the high-throughput primer validation procedure. Data are shown for five successful primer pairs, five primer pairs that failed based on agarose gel electrophoresis analysis and five primer pairs that failed based on BLAST analysis. Information on these primer pairs, such as PrimerBank IDs, primer sequences and amplicon lengths, is shown here [see Additional file 4]. More information on these primers, such as their T_m and location on the gene, can be found in PrimerBank, as well as alternative primer pairs designed for these transcripts.

All data generated from the high-throughput primer validation procedure can be freely accessed from PrimerBank http://pga.mgh.harvard.edu/primerbank/. See Figure 1 for the PrimerBank homepage. Users can search the PrimerBank database for primers for their gene of interest using several search terms such as: GenBank accession number, NCBI protein accession number, NCBI gene ID, PrimerBank ID, NCBI gene symbol or gene description (keyword). Search results include primer sequences together with some information about the primers, such as expected amplicon size and T_m. cDNA and amplicon sequences, and validation data can be viewed by clicking on the appropriate links. All validation data can be accessed from PrimerBank, since the validation criteria may be different from the criteria of the users. Also, users can use a BLAST tool found on the PrimerBank homepage (see Figure 1), to find any primers contained in the PrimerBank database that would amplify their sequence of interest. A BLAST tool for the PCR product sequence obtained from the validation procedure can be used to query the NCBI database and this can be found on the validation data webpage. The QPCR and reverse transcription protocols can be found on PrimerBank, as well as a troubleshooting guide.

A schematic representation of the agarose gel fail distribution can be seen in Figure 3. This analysis was based on determining whether one PCR product of the correct size could be visualized from agarose gel electrophoresis data. Most primer pairs were successful based on at least one step of the primer validation procedure. Two major types of failed primer pairs that comprise most of the failures are primer pairs that failed on agarose gels but were successful by BLAST and primer pairs that failed on BLAST but were successful on agarose gels. 3695 primer pairs failed based on BLAST analysis alone and another 1864 primer pairs failed based on agarose gel analysis alone. In most cases a primer pair failed in one of the analysis steps based on the criteria, but was successful in other analysis steps. The failed samples did not overlap in many cases and this could have been in some cases due to strict BLAST analysis criteria and new splice isoforms seen on the agarose gels. Also, some primer pairs failed by both BLAST and agarose gel analysis, although these are numerically minor. For a detailed description of the analysis criteria see Table 3. The criteria for success or fail may be different from the criteria users might apply and for this reason all validation data can be accessed from PrimerBank.

From the total agarose gel failed reactions, 46.7% were due to multiple amplification products apparent by gel electrophoresis. 13.8% of the total failed reactions were due to undesired amplification, seen as the wrong size band on the gel. 4.8% of the total failed reactions were due to poor amplification, and 34.7% of the total failed reactions were due to no amplification taking place. Multiple or undesired amplifications accounted for the majority (60.5%) of the agarose gel failed reactions. These may represent undocumented transcripts or splice isoforms that could have been amplified in addition to or instead of the expected transcripts. For the reactions that failed because no amplification had taken place, the template sequences may not have been present or present in very low copy number.

From the high-throughput PrimerBank mouse primer pair validation, 1745 samples (6.5%) failed because of no amplification, as seen from the QPCR amplification plots. From the gene description information we found several to belong to olfactory receptors, vomeronasal receptors, transcription factors and low abundance transcripts while others were of unknown function or RIKEN sequences (data not shown). In order to investigate the possibility that the templates for the failed amplification primer pairs were not expressed in the cDNA sample used, we repeated these reactions using genomic DNA as a template. It can be difficult to achieve amplification using genomic DNA as template in general, due to its complexity. However, it can be used successfully if technical difficulties are overcome and can be useful as a universal template as it contains a copy of all genes, and the same amount of template is present for all single-copy genes 32. We have found that enzymatic digestion (such as EcoRI/BamHI digestion used here) can be used for reduction of the complexity of the DNA and thus higher amplification rates. We matched 864 primer pairs to mouse genome sequences obtained from the UCSC genome browser. The remainder of the sequences could not be matched, probably because they were located on exon junctions. 640 of these primer pairs have no EcoRI/BamHI restriction sites in their expected PCR amplicons, and were used with EcoRI/BamHI digested DNA template to prepare the validation reactions. We tested 192 representative samples, from the 1745 total number of failed primer pair samples, whose expected PCR amplicon lengths range from 60 bp to 123 bp and whose amplicons have no EcoRI/BamHI restriction sites. 50 ng EcoRI/BamHI digested 129 mouse ES cell genomic DNA was used per 25 μl PCR reaction.

The amplification plots of all 192 samples (2 × 96 well plates) are shown here [see Additional files 6, 7]. The success rate of QPCR based on the amplification plots was high: 88.5% for the first plate [see Additional file 6] and 90.6% for the second plate [see Additional file 7]. However, Ct values differed significantly, from roughly 23 to 40 [see Additional files 6, 7]. The location of the reactions on the plate did not explain this variation. The samples were also analyzed by agarose gel electrophoresis and sequenced (data not shown). Sequences obtained were BLAST analyzed and matched to the expected sequences, confirming that the correct templates had been amplified (data not shown). Therefore, these primer pairs had originally failed because their respective templates were not present in the cDNA sample used and not because of poor primer design, in general.

We next set out to determine the uniformity of amplification using fully validated PrimerBank primer pairs ie. primer pairs that had been successful in all steps of the validation procedure. 96 primer pairs were chosen with expected PCR amplicon length ranging from 80 bp to 120 bp and containing no EcoRI/BamHI restriction sites in their sequences. Both forward and reverse primers were chosen to be on the same exon in order to amplify the same template on genomic DNA. EcoRI/BamHI digested 129 mouse ES cell genomic DNA was used as template. After digestion the DNA was purified for PCR by phenol extraction and ethanol/salt precipitation. 50 ng of DNA template was used per 25 μl PCR reaction, which was found by optimization experiments to give a reasonable Ct value.

See Figure 4 for the amplification plots and dissociation curves. As can be seen from Figure 4A, the Ct values for each sample are not exactly the same. This is expected since there will be some stochastic variation. Also, different primer pairs were used for each sample. However, the Ct values are similar, so amplification using PrimerBank primers appears to be relatively uniform. The statistical significance of the difference in Cts observed was determined by plotting a frequency distribution of the number of samples versus the Ct (Figure 5A). A statistical normality test was also used for the analysis of these Ct values, but the data did not pass this test. The effect of primer length and primer GC% on the Ct was studied, by plotting these values against the Ct, and no correlation between these parameters was found (see Figure 5B,C). The effect of the PCR product T_m on the Ct was also studied, by plotting the T_m values against the Ct, and again no correlation was found (see Figure 5D). Since the expected PCR product size varies from 80 bp to 120 bp, some small variation in T_m is expected, and this can be seen from the dissociation curve data (see Figure 4B). The T_m data (obtained from the dissociation curves) was also plotted as a frequency distribution and did not pass the statistical normality test (data not shown).

In order to determine the technical reproducibility of the QPCRs, five 96 well plate assays were prepared using the same technical procedure. Reactions were set up using the same 96 primer pairs and DNA template (129 mouse ES cell EcoRI/BamHI digested genomic DNA) that were used for the uniformity of amplification test. The coefficients of variation for each 96 well plate assay are all < 0.1 and the average coefficient of variation for all assays is 0.07 [see Additional file 8]. The individual primer pair Cts for each 96 well plate assay and coefficients of variation are shown here [see Additional file 9]. Ct data from each assay initially did not pass the statistical normality test. The Ct values were normalized, using the formula:

(LnCt - LnCt_av)/SD,

where LnCt is the natural logarithm of the Ct value used, LnCt_av is the natural logarithm of the average Ct value of the assay and SD is the standard deviation of the LnCt values for each assay, and outliers were removed. The normalized data passed the normality test, so the data appear to be log normal. The plots of the frequency distributions of the log normal data are shown here [see Additional file 10].

Analysis of pipetting variation during liquid transfer of the fluid handling system was carried out and the transfer efficiency of the robot was found to be 97.3% [see Additional file 11]. The data from the liquid transfer test passed the statistical normality test only after the 9 lowest value outliers were removed (data not shown), but the coefficients of variation are low (less than 0.03) [see Additional file 11]. Variation in liquid transfer can only account for a small amount of the variation observed in QPCR reactions, and hence other factors must be responsible for the differences observed in Ct values.

The SYBR Green I dye has been widely used as a non-sequence specific dye for fluorescence detection of QPCR products 20. Studies of SYBR Green I-DNA binding showing some sequence specificity of the dye have been reported but these have not been conclusive 203334. We investigated whether SYBR Green I is sequence specific by adding the dye to a series of amplicons and taking fluorescence readings. 8 amplicons of increasing length and 7 amplicons of increasing AT% [see Additional file 12] were used, whose concentrations were accurately determined (see methods). From these experiments, we did not observe any length dependent or AT/GC dependent sequence specificity of SYBR Green I [see Additional file 13]. However, we cannot exclude the possibility that SYBR Green I can show specificity to sequences such as homopolymer regions of DNA 20 or specific sequences. We also investigated whether SYBR Green I dye binding is sequence specific by estimating the number of PCR product molecules at threshold using the ABI PRISM 7000 Sequence Detection System (Applied Biosystems) 3536. For this, the same 14 amplicons as above were used and a template titration series of reactions was prepared for each amplicon. SYBR Green I threshold cycle (Ct) fluorescence will be the same for all amplicons (and all reactions), since the same threshold was used to compare all reactions. However, if SYBR Green I is sequence specific, this fluorescence will correspond to a different number of molecules at threshold for each amplicon. These experiments were inconclusive, as the stochastic error was too large to be able to accurately determine the molecules detected at the threshold (data not shown).

The most common method for the calculation of the amplification efficiency of a QPCR reaction requires preparation of a series of serial dilutions of the sample and creation of a standard curve, whereby efficiency is estimated from the slope of the standard curve 3637. However, this method does not provide an accurate value of the efficiency, as the efficiency can vary between different reactions and as input concentration changes. A number of analytical methods have been described for the calculation of the amplification efficiency of a reaction from single reaction kinetics 38 (for a correction in equation 3 of this paper see: 39), 404142. These methods can be more accurate and, when automated, less laborious compared to the standard curve method 43. Using the following analytical method, we estimated the amplification efficiency values for 13 QPCRs using PrimerBank primer pairs that had been previously used. The log2 fluorescence data was plotted versus the Ct number and the slope of the linear regression was taken to be equal to the efficiency of each reaction [see Additional file 14]. Cycle values closest to the Ct were used, as this region will be the most accurate. The efficiency values ranged from 79% to 96% [see Additional file 14]. Replicates can be used to improve accuracy when using either the standard curve or analytical single reaction kinetics methods 3944.

We compared amplification efficiency estimation using the standard curve and analytical methods in order to determine the accuracy of each method using the same 13 PrimerBank primer pairs as above [see Additional file 15]. Either the log2 of pg of input template DNA data, for the standard curve method, or the log2 fluorescence data, for the analytical method, was plotted versus the Ct number. Ct was the independent variable and log2 of pg of input template DNA/fluorescence was the dependent variable. The slope of the linear regression was taken to be equal to the efficiency of each reaction. From these results the analytical method shows a smaller variance of efficiency values and the range is smaller compared to the standard curve method [see Additional file 15]. One-way ANOVA analysis was done to determine if amplification efficiency varied significantly between different PrimerBank primer pairs, using each primer pair in a series of titration reactions of template DNA [see Additional file 16]. The average efficiency, standard deviation and coefficient of variation for each group of primer pairs are shown here [see Additional file 17]. The P value is > 0.05 (0.7338) therefore the amplification efficiency is similar between these groups.

In order to account for sample effects, it is useful to provide a model of the experimental measurement of fluorescent PCR product accumulation 4546474849. The following equations can be used:

Log2pgDNA = β₀ + β_Ctx_Ct + ε,

where Log2pgDNA is the dependent variable, β₀ is the intercept, β_Ct is the regression coefficient for the x independent variable, and ε is the error. Equation 1 can be used for the standard curve method.

Log2Fluorescence = β₀ + β_xx_c + ε,

where Log2Fluorescence is the dependent variable, β₀ is the intercept, β_x is the regression coefficient for the x independent variable of cycle c, and ε is the error. If β_x = 1, amplification efficiency is 100%. Equation 2 can be used for the analytical methods.

PrimerBank primer pairs have been designed irrespective of their location on exons. Data from the UCSC genome browser were downloaded and used to find the location of 26854 mouse primer pairs with respect to exons (see Table 4). 19668 primer pairs matched to sequences from the genome browser. Most of the matched primer pairs (16356) are located within exons and at least one primer from the rest of the primer pairs is located on an exon boundary. Primers can be designed to be located on exon boundaries, in order to avoid non-specific amplification of genomic DNA during PCR, but in many cases it was not possible to design primers located on exon boundaries that fulfilled all of the criteria for primer design, most trivially because some transcripts consist of a single exon.

A commercial composite mouse RNA preparation was chosen as the source of DNA template for QPCRs, which contains RNA from a panel of eleven different mouse cell types for a good representation of the majority of mouse genes. The composite mouse RNA is composed of total RNA from: whole embryo, embryonic fibroblasts, kidney, liver, lung, B-lymphocyte, T-lymphocyte, mammary gland, muscle, skin and testis. The success rate of the high-throughput PrimerBank primer validation experiments was high as seen both from agarose gel and BLAST analysis. We validated some of the failed reactions using genomic DNA as template, and found that most of the failures in which no PCR product had formed could be due to very little or no cDNA present in the source of DNA template. In order to increase amplification success, specific tissues may be used as sources of cDNA templates where expression of the genes of interest is known.

The PrimerBank primer design was based on a successful approach for the prediction of oligonucleotides for the interrogation of protein coding regions by microarrays 27. However the primer design differs by the addition of filters that are thought to be important for primer specificity 26. All primers have been designed to work using a relatively high annealing temperature of 60°C and this temperature was used throughout the primer validation experiments described here. High annealing temperatures help reduce non-specific amplification. A high percentage of the total failed samples were due to undesired or multiple amplification, however this may have been for other reasons such as new unidentified genes or splice isoforms. In 3.9% of the cases where multiple bands could be seen on the agarose gel and in 14.6% of the cases where bands of other than the expected size could be seen on the agarose gel, no sequencing information was obtained. Also, 29.7% and 55.2% respectively, did not match to the expected sequences by BLAST. So, sequence homology existed in most cases of undesired or multiple amplification. From the genome-wide primer validation experiments presented here, we have found a high success rate of primer pairs that amplify the transcripts for which they had been designed. For primer pairs that failed because no amplification could be detected, we found that the reason for which they had initially failed was because their target sequences were not present in the target cDNA used. Another reason for failure in the high-throughput validation procedure, may be that protein coding genes in the human genome are fewer than previously thought, and the same may apply to the mouse genome 50.

As mentioned previously, larger than expected or multiple bands were visible on the agarose gel for some samples, however, sequences for these matched confidently by BLAST to the expected sequences. Therefore, the template sequences amplified in these cases could be new genes or splice isoforms. These unrecognized genes or splice isoforms may contribute to primer cross reactivity which results in a lower success rate on the agarose gels. Good primer design depends on accurate genomic information about genes and splice isoforms and it is suggested that many unidentified genes and splice isoforms could exist. All primer pairs that failed because of non-specific amplification, but when BLAST analyzed matched to the expected sequence, could have amplified new non-identified isoforms. This information would be very useful for other researchers, in addition to other strategies for identifying new genes and splice isoforms 5152. PrimerBank primers could also be used for determining copy-number variation of a gene or splice isoform 5354.

Several online databases exist containing experimentally validated primers, however, only a few thousand primer pairs are currently present in these databases 555657. We have previously designed PCR primers for the human and mouse genomes, which are available from PrimerBank 26. The PrimerBank database currently contains 306800 primers for the mouse and human genomes and is tightly integrated with information from the NCBI databases. PrimerBank has been designed so that researchers can search for primers for their gene of interest using several search terms such as: GenBank accession number, NCBI protein accession number, NCBI gene ID, PrimerBank ID, NCBI gene symbol or gene description (keyword). Currently, all validated primers can be retrieved by searching PrimerBank. In many cases, alternative primer pairs for genes also exist in PrimerBank. NCBI sequences have been attached to the primer information page and NCBI LocusLink indices have been used internally for gene locus mapping. All primers have uniform properties such as T_m, length and GC content and can work using the same PCR conditions.

129 Embryonic Stem cell mouse genomic DNA (isolated by ethanol precipitation) was used. The DNA was digested completely using EcoRI and BamHI restriction enzymes. Digests were made by adding 20 μl EcoRI buffer (10×) (New England Biolabs), 20 μl 10× BSA, 4 μg DNA, 40 U BamHI (New England Biolabs), 40 U EcoRI (New England Biolabs) and water to 200 μl total volume. Digests were incubated at 37°C for 4 hours and 30 minutes and heat inactivated at 75°C for 10 minutes. The digested DNA was phenol extracted and ethanol/salt precipitated. DNA pellets were resuspended in TE pH 8.0.

QPCRs were performed in polypropylene 96 well plates on the ABI 7300 Real-Time PCR System (Applied Biosystems). For each reaction, 12.5 μl of Absolute Q-PCR SYBR Green ROX mix (ABgene) were added to 2.5 μl of 2 μM forward and reverse primer mix (final concentration of each primer is 200 nM), 1 μl of 50 ng/μl BamHI/EcoRI digested genomic DNA and made to 25 μl with water. The following PCR conditions were used: 50°C for 2 minutes (step 1), 95°C for 15 minutes (step 2), 95°C for 15 seconds, 60°C for 30 seconds, 72°C for 30 seconds (step 3 – repeated another 39 times ie. 40 cycles in total), 72°C for 10 minutes (step 4). Dissociation curves were obtained by heating and cooling the samples at: 95°C for 15 seconds, 60°C for 30 seconds, 95°C for 15 seconds.

Amplicons were prepared large-scale by PCR, in two steps. For the first step PCR, 75 μl PCR reactions were prepared for each sample. For each reaction, 37.5 μl Absolute Q-PCR SYBR Green ROX mix (ABgene) were added to 3 μl of 5 μM primer pair mix (final concentration of each primer is 200 nM), 3 μl universal mouse cDNA (see: 'preparation of cDNA sample' section in methods) and made up to 75 μl with water. The following PCR conditions were used: 95°C for 15 minutes (step1), 95°C for 15 seconds, 60°C for 30 seconds, 72°C for 30 seconds (step 2 – repeated another 39 times ie. 40 cycles in total), 72°C for 10 minutes (step 3). The PCR products were purified using the MinElute PCR purification kit (Qiagen). Purified amplicons were used as templates in large-scale 40× 100 μl PCRs, each reaction containing 50 μl 2× LC1v3 buffer (40 mM Tris-HCl pH8.8, 40 mM KCl, 40 mM ammonium sulfate, 4 mM MgCl₂, 200 μg/ml BSA, 0.2% Triton X-100, 400 μM dNTP mix, 2.5 M betaine), 4 μl of 5 μM forward and reverse primer mix, DNA template, 1 μl Taq polymerase and water to 100 μl. The PCR conditions used were the following: 95°C for 3 minutes (step 1), 95°C for 15 seconds, 60°C for 30 seconds, 72°C for 30 seconds (step 2 – repeated another 39 times ie. 40 cycles in total), 72°C for 10 minutes (step 3).

PCR reactions were phenol extracted and isopropanol precipitated. DNA pellets were resuspended in TE pH8.0. DNA was purified using Performa DTR Gel Filtration Cartridges (EDGE BioSystems), following the recommended procedure. Amplicon concentrations were determined by taking OD260 readings of each preparation using the ND-1000 Spectrophotometer (Nanodrop). The average value was taken and the OD260 reading from a no DNA template control was subtracted, in order to remove the contribution from primers and buffer components to the spectrophotometric absorption.

DNA samples in 1× Absolute Q-PCR SYBR Green ROX mix (ABgene) were pipetted into OptiPlate-96F black 96 well plates (Perkin Elmer). SYBR Green I fluorescence was detected using the Analyst AD fluorescence plate reader (Molecular Devices) by excitation at 485 nm and emission at 530 nm (505 nm dichroic mirror).

5 μl of 10 mM dNTP solution were added to 95 μl water and the OD260 readings were taken using the Spectra Max Plus Spectrophotometer (Molecular Devices).

Mouse genome sequences were downloaded from the UCSC genome browser 59 and the primer pair sequences were matched by BLASTn to the genome sequences, to identify the primer locations with respect to exons.