However, a forensic laboratory often has to deal with DNA samples that are less than ideal. The technical capacity to analyze samples expected to contain very few cells is often limited by the quality and quantity of DNA. The main limitation of existing forensic DNA protocols lies in the minimum size and/or amount of DNA fragments that can be typed by a capillary electrophoresis assay (ranging from 100–400 bases in length and from 500–1250 pg) [4]. Sometimes forensic samples may contain only small amounts of genomic DNA (less than 100 pg), referred to as LCN DNA (low copy number DNA), and/or may also be degraded into fragments smaller than 100 bp, resulting in failure of amplification of the STR loci and an incomplete DNA profile with lower discriminatory power [5]. To overcome this problem, several protocols for whole genome amplification (WGA) have been developed. The aim of WGA is to amplify a limited DNA sample non-specifically in order to generate a new sample that is indistinguishable from the original but has a higher DNA concentration. This method was specifically designed to increase the quantities of DNA in clinically relevant samples such as single cell analysis, pre-implantation genetic diagnosis and forensic typing [6]. However, the application of WGA methods to forensic casework is demanding because of technical artefacts that are particularly relevant when STR markers are analyzed. Such artefacts include contamination, PCR failure, preferential allele amplification, the complete absence of one allele (allele drop-out, ADO) in heterozygous loci and the non-specific generation of extra alleles (allele drop-in) [7,8]. Recently, an isothermal WGA method was introduced using bacteriophage Φ29 DNA polymerase and random hexamer primers [9,10]. This method, termed MDA (Multiple Displacement Amplification), takes advantage of the bacteriophage Φ29 DNA polymerase, a proofreading enzyme with high fidelity and potent strand displacement activities. MDA is based on the isothermal strand displacement mechanism and leads to hyperbranching amplification of both circular and linear DNA templates [10]. Its advantages include limited bias in the amplification of loci (maximally 6-fold compared to 10³- to 10⁶-fold in PCR-based methods), higher average length of amplification products (12 kb compared to 100–1000 bp in PCR-based methods), lower error rate (<10^-6 compared to 3 × 10^-4 to 10^-5 for alternative methods) and higher DNA yield [11]. Recent data have shown that the distance to telomere predicts failure only in WGA samples while percentage-GC is positively correlated with PCR failure [12,13]. MDA has been recognized as the most effective WGA method but its applicability to forensic casework has not yet been considered because technical artefacts have not been eliminated [14,15]. Significantly more alleles are revealed when STRs are analyzed by application of MDA to LCN samples than to non-amplified control samples although false allele generation (allele drop-in) was reported. The cause of such artifacts are not yet completely explained although it appears more frequent in STRs analysis [7,16]. Moreover STR analysis in MDA samples also shows an increased number and height of stutter peaks together with a high level of allelic imbalance, affecting the interpretation of results, especially in cases where two or more individuals may have contributed to the DNA evidence [16,17]. Stutter peaks result from the PCR process when STR loci are copied by a DNA polymerase. Because stutter products are the same size as actual allele PCR products, it can be challenging to determine whether a small peak is a real allele from a minor contributor or a stutter product of an adjacent allele especially when the height of the stutter is higher than that normally expected [18].

Moreover, comparative analysis of genomic and MDA samples starting from 6 ng of DNA reported only a 90–95% concordance rate for STRs [19,20]. The amount of template gDNA required for the MDA reaction is also a critical determinant of genotyping performance because it is generally unacceptable for forensic purposes. Thus, since current forensic profiling is based on STRs and the concordance rate with LCN DNA has not proved acceptable for forensic profiling, MDA is not regarded as valuable for typical forensic casework. However, better concordance rates were observed when MDA was applied to SNPs. Until now, the STR concordance rate has been worse for amplified than for genomic DNA, even with an input of 200 ng of gDNA into the MDA reaction; whereas optimal TaqMan^® SNP genotyping is expected for MDA DNA inputs greater than 4 ng [21]. With less than 5 ng genomic DNA in the MDA reaction, the amplification bias and the number of genotyping errors increases [17,22]. Similar technical limitations were also reported when MDA was considered for pre-implation genetic diagnosis, with up to 32% ADO for single-cell samples [6].

Recently, we developed and validated an SNP panel for human identification, rigorously selected to have very constant frequencies throughout the population tested and showing high specificity for the human genome [5]. The average size of our amplicons (77 bp) is much less than current forensic markers, making them more sensitive for LCN DNA and less prone to PCR bias.

We first assessed the performance of multiple displacement amplification (MDA) in terms of genotype concordance between gDNA (50 ng) used as reference genotype, genomic and amplified LCN DNA (≤ 1 ng). Initially, a human-specific real-time PCR assay, a Quantifiler™ Human DNA Quantification Kit (Applied Biosystems, Foster City, CA), was used to quantify the genomic DNA in 100 selected samples. All the samples were re-sequenced to assess the genotype at each locus tested. We then prepared serial dilutions of the DNA to obtain concentrations ranging from 1 to 0.01 ng/μl. MDA was performed on 1 μl of the three serial dilutions (1 ng, 0.1 ng, 0.01 ng) for each sample. The MDA products were quantified using the Quantifiler™ Human DNA Quantification Kit; 14.5, 1.5 and 0.27 μg of DNA were produced from the 1 ng, 0.1 ng and 0.01 ng dilutions, respectively. These quantities are only indicative, because Quantifiler amplifies a small fraction of the hTERT gene, which is located adjacent to the 5p telomere, and previous studies have shown that telomeric regions may be under-represented [16]. We also quantified the MDA products using Nanodrop spectrophotometer. The mean final yields of MDA obtained from the 1 ng, 0.1 ng and 0.01 ng dilutions were 19.5 μg, 8.5 μg and 6.5 μg, respectively. As previously reported [23] a considerable part of DNA obtained after MDA was a nonspecific DNA products. The RNAse P real-time assay was performed to estimate the amount of specific DNA obtained after MDA (10.0 μg, 2.6 μg and 0.3 μg for each dilutions respectively). TaqMan assays were performed on 1 μl of MDA DNA and gDNA. To evaluate the genotype concordance, all the TaqMan assay results from serial dilutions of gDNA and MDA DNA (1 ng, 0.1 ng and 0.01 ng) were compared to genotypes previously confirmed by direct re-sequencing of the same samples (50 ng). In parallel, TaqMan SNP assay results were examined on the same dilutions of gDNA as technical controls (Additional file 1). Optimal genotype concordance (100%) was observed between the sequence and TaqMan. The TaqMan assay performed on 1 ng of genomic DNA revealed no failure of amplification (100% call rate) and optimum genotype concordance (Additional file 1). The TaqMan assay performed on an input of 1 ng in MDA revealed PCR failure for the marker rs11881170 (99.960% call rate) but perfect genotype concordance. When the TaqMan assay was performed on 0.1 ng of genomic DNA, 100% genotype concordance was confirmed but there was failure of amplification in two samples for the markers rs3130315 and rs10866988, in six samples for marker rs585070 and in four samples for rs873289, resulting in a call rate of 99.461%. The TaqMan assay performed on a 0.1 ng input in MDA revealed a single PCR failure for the same marker, rs11881170 (99.961% call rate) and a single genotype discrepancy for rs1981752 (concordance rate 99.961%). When the TaqMan assay was performed on 0.01 ng of genomic DNA, 100% genotype concordance was again confirmed but there were more failures of amplification: in four samples for marker rs3130315, in two for marker rs585070, in 14 for rs1506981, in four for rs478347 and in six for rs873289. The resulting average call rate was 98.846%. In contrast, the TaqMan assay performed on a 0.01 ng input in MDA revealed a non-absolute concordance rate (99.379%) with two genotype discrepancies observed for marker rs2278741, two for rs7740233, two for rs10866988, two for rs1533800, two for rs154659 and three for rs873289. It is nevertheless remarkable that more positive calls were observed with 0.01 ng input of MDA-DNA than with unamplified DNA at same dilution. In particular, 20 samples failed to be typed (99.230% call rate) compared to 30 PCR failures with unamplified DNA at 0.01 ng dilution.

In order to determine whether a different reaction mixture might improve both concordance and call rates, we tested a recently-released master mix, the TaqMan^® Genotyping Master Mix (Applied Biosystems). This is able to optimize the preferential binding of the allele-specific probe, providing better separation and clustering of alleles and consistently strong fluorescence signals. Better results were obtained by repeating the same serial experiments under identical conditions using this new reaction mix (Additional file 2). In particular, all three dilutions of genomic DNA gave 100% results for both concordance and call rates. Also, we observed 100% concordance and call rates for MDA inputs of 1 ng and 0.1 ng DNA and 99.923% concordance and call rates for an MDA input of 0.01 ng (Additional file 2). We failed to assess the correct genotype in two out of 2598 samples, resulting in a genotyping error rate of 7.7 × 10^-4. The two genotype discrepancies observed in MDA were ADO (false homozygous), and we also failed to find evidence that the typing results correlated with the GC content and/or the distance to telomeres. We note that this rate of genotype discrepancy is much lower than those reported for current STR forensic protocols [23,24]. We also never observed positive amplification in negative controls.

On the basis of these findings we selected the TaqMan^® Genotyping Master Mix as the default reaction mixture for our panel and for the further experiments presented here.

To address this issue, we tested two independent protocols using our recently-developed forensic panel of SNPs. The first protocol revealed dissimilar results when genomic and MDA DNA were analyzed. In particular, 14 and 30 genomic DNA samples were not amplified with 0.1 ng and 0.01 ng dilutions, respectively, but all genotypes were concordant with the controls. In contrast, only one, one and 20 MDA DNA samples failed to be amplified with the 1 ng, 0.1 ng and 0.01 ng input dilutions, respectively. However, one MDA DNA sample (for inputs of 1 ng and 0.1 ng) revealed a genotype discordant with the controls, and 36 MDA DNA samples (for input of 0.01 ng) gave results dissimilar to the sequenced controls.

All discordant genotypes were false homozygous samples resulting from the complete failure of amplification of one allele in MDA (allele dropout, ADO) [7]. We therefore decided to use a different master mix specifically designed to optimize the preferential binding of the allele-specific probe. We repeated the experiments and obtained better results. We observed 100% concordance and call rate for all dilutions of genomic DNA and for inputs of 1 ng and 0.1 ng in MDA. Furthermore, when inputs of 0.01 ng were analyzed, we observed a single PCR failure and only two samples gave results dissimilar to the controls. The resulting concordance and call rate were therefore increased to 99.923% for a 0.01 ng input in the MDA reaction, resulting in a probability of genotyping error of 7.7 × 10^-4, quite similar to that reported for STRs. Such results suggest that in combination with extra-short TaqMan assays and the specifically-designed master mix, MDA gave optimal findings in terms of sensitivity, reproducibility and robustness of genotyping. Independent studies have suggested that replicate MDA reactions should be pooled prior to genotyping in order to overcome the frequency of ADO and/or genotyping errors [25]. Tzvetkov observed that the concordance between MDA and gDNA genotypes shows a small but significant improvement (0.5%) in pooled samples compared to single MDA. In our opinion, this strategy could be applied when sufficient DNA template is available.

Nevertheless, the comparison between genomic and amplified DNA showed that MDA does not improve our ability to type LCN DNA for single SNPs, as demonstrated by the better concordance and PCR failure rates in genomic than in MDA DNA. The genotype concordance and PCR failure rates were 100% and 99.923% in MDA DNA and genomic DNA respectively. However, it should be noted that MDA amplification produces large amounts of DNA, so it is possible to type many SNPs; in contrast, genomic DNA is often present in limited quantities so fewer SNPs can be typed. Finally, MDA allowed us to type a large number of SNPs starting from LCN DNA as template. Moreover, our experiments on artificially-degraded DNA provide evidence that MDA can enhance our capacity to type severely-degraded samples. In our experiments, nine samples failed to be amplified in genomic DNA but were successfully typed in the amplified DNA. We observed 100% concordance and call rate for all SNPs using MDA on degraded DNA. The quality of our results in typing digested DNA by MDA was surprising, although this has also been reported to a lesser extent for STR typing [16]. Even though we typed completely-digested DNA (the average length of the degraded DNA (180s) was 80 bp down to less than 50 bp), it is likely that a small number of remaining intact fragments of template were amplified by MDA. In our opinion, these results depend on the marker types, the amplicon sizes and the genotyping technique. The results confirm and extend previous comparative analyses of SNP and STR typing in degraded forensic DNA (without MDA), demonstrating that good markers for degraded DNA depend on a small amplicon size [26]. Here, very small amplicons (77 bp on average) have been typed by the most sensitive genotyping assay.

Finally, we recommend using MDA in order to overcome limitations of DNA quantity (when the amount of DNA is limited) or when only poor-quality DNA samples are available (severely-degraded samples, to increase the number of typed loci).

We would also point out that 0.26 ng of genomic DNA (0.01 ng for each locus) is needed to type all 21 autosomal SNPs of our panel and 5 extra SNPs [5,27] successfully. It is also noteworthy that under these conditions, even though Real-Time PCR cannot allow multiplexing of the PCR, the amount of DNA needed to type a panel of SNPs in singleplex is comparable to current forensic STR methods and is much less than is needed to type forensic SNPs, as reported elsewhere [27]. As a result, its use in forensic practice should be seriously considered when traditional forensic kits fail.

If the amount of DNA proves insufficient to type all SNPs independently, it will be possible to apply MDA even in LCN templates. In this case, only 0.01 ng of input DNA is needed to type a large number of SNPs (even more than the 26 reported here). This amount of DNA is much less than needed in the forensic protocols reported to date, so it represents a significant improvement in current forensic DNA protocols. As mentioned above, the possibility of a technical artefact (i.e. ADO) or a mis-genotyping following MDA with 0.01 ng of input DNA is 7.7 × 10^-4. This is a lower error rate than that reported for the STRs used for forensic typing. Moreover, the mutation rate is higher in STRs than in SNPs: germline mutations at these STR loci lead to problems in the interpretation of genetic profile results.