Exploring STR sequencing for forensic DNA intelligence databasing using the Austrian National DNA Database as an example

All samples were diluted accordingly in molecular grade water to amplify 0.5 ng of template DNA according to [22]. Multiplex PCR was performed targeting 0.5 ng DNA using the PowerSeq 46GY kit [22] on an Applied Biosystems GeneAmp 9700 thermal cycler (TFS) according to the manufacturer’s recommendations [23]. Each amplification reaction was treated with 5 µL proteinase K solution [504 µg/mL] (Roche Diagnostics, Mannheim, Germany) and purified using AMPure XP beads (Beckman Coulter, CA, USA) following [22]. The concentrations of amplification products were estimated spectrophotometrically prior to library preparation by measuring the absorbance at 260 nm according to [25] and as recommended in [23] with a NanoDrop ND-1000 spectrophotometer and analyzed with software version 3.8.1 (both Peqlab Biotechnologie GmbH, Erlangen, Germany).

End-repair, A-tailing, adaptor ligation, initial, and second purification were performed using the TruSeq DNA PCR-Free HT Library Prep Kit (Illumina, San Diego, USA; [24]) according to the manufacturer’s recommendations [22, 23], with the exception that supplied sample purification beads were replaced by AMPure XP beads (Beckman Coulter, CA, USA). To ensure balanced pooling, each library sample was quantified in duplicate by means of qPCR using the KAPA SYBR FAST Universal qPCR Kit (Roche) following [26] on an Applied Biosystems 7500 Fast Real-Time PCR instrument. Data analysis was performed using the HID Real-Time PCR Software v 2.3. Based on qPCR results, samples were diluted and normalized to 4 nM and equally pooled according to [22]. Sequencing was performed on the MiSeq FGx instrument (Verogen, [27]) using a 500 cycles MiSeq Reagent Kit v2 for 2 × 250 paired-end sequencing (Illumina, [28]) according to the manufacturer’s recommendations. The final library concentration was 12 pM with approx. 6.6% spiked PhiX control library (Illumina) following [22].

Size-based analysis of STR fragments was conducted using the GeneMapper ID-X software, version 1.2 (TFS) by applying in-house validated dye thresholds: blue — 50 relative fluorescence units (RFU), green — 80 RFU, yellow — 100 RFU, and red — 100 RFU.

Sequencing results were monitored using the Sequencing Analysis Viewer software (Illumina; [29]) to review relevant quality metrics. Compressed FASTQ files were manually extracted for data analysis using the open-source STRait Razor v2s software tool [30, 31]. Sequences were aligned to human genome assembly GRCh38 and genomic coordinates for STR markers were determined by post processing of the mpileup output from SAMtools [32] as described in [30, 31]. STR genotypes were analyzed by applying an analytical threshold (AT) of 50 reads, referring to [33]. Alleles were called above the in-house defined interpretation threshold (IT) of 500 or 100 reads for homozygous or heterozygous genotypes, respectively. Sequence-based allele frequencies are shown in Table S1 considering all available sequence information.

Stutter analysis was restricted to a subset of 50 samples (app. 20% of the entire sample set) selected according to the total number of reads that was calculated by summarizing the intensity (read count) of a given STR profile for all 22 aSTRs. Stutter sequences were determined as one repeat unit smaller than the parental allele and calculated by dividing the intensity of stutter sequence by the intensity of the corresponding allele. The selected samples were divided in two equally sized groups comprising low (category I) and high performing samples (category II), respectively. Selection criteria were established to investigate the effect of sample performance (total number of reads per genotype) on the formation of stutter height and defined as follows: category I comprised only samples that fell below 63,500 reads, while category II included solely samples above 199,000 reads (Table S2).

Microsoft Excel workbooks, IBM SPSS software, version 24 [34], and GraphPad Prism, version 8 for Windows [35], were applied.

Forensic and population genetic parameters, including allele frequency, observed and expected heterozygosity, expected homozygosity, power of exclusion, power of discrimination, matching probability, typical paternity index, and exact Hardy–Weinberg equilibrium (HWE) tests for the population, were calculated from CE data using in-house software according to formulae listed in Table S3 and the STRAF software package [36].

The mean DNA quantity of the 247 samples was 7.29 ng/µL (SD: 5.20, median: 6.23) with a minimum of 0.71 ng/µL and a maximum of 33.04 ng/µL (Table S5). The mean storage period for DNA samples used within the current study was 2.96 years (SD: 1.90; median: 3.12) and varied between 1 and 13 years (Table S5). As there were no changes in sampling and DNA extraction over the entire time period, the latter is the only remaining variable that could influence the DNA quantity and quality. As shown in Figure S1, storage time did not affect sample performance and was comparable for the majority of samples included in each storage time group (Figure S1). Although relative marker performance decreased slightly with increasing amplicon length, we did not observe differences between storage time groups I–III (Figure S2). The decline in relative marker performance, in relation to amplicon size, did not adversely affect downstream data analysis.

Relative marker performance was evaluated (for each sample and STR locus) by dividing the intensity of marker reads by the total number of reads for all markers of a single STR profile. Assuming equally performing STR markers, the expected relative marker performance value for each of the 23 markers (incl. amelogenin) of the PowerSeq 46GY panel would equal 4.35% of all reads. The effective mean relative marker performance values were found to lie close to the expected value and were comparable for the majority of markers included in the present MPS panel, except for D1S1656, D2S1338, TH01, D12S391, D16S539, Penta D, D22S1045, and amelogenin (Fig. 1). The average performances of the latter eight markers were outside the well-balanced range, which we defined as mean relative marker performance + / − one SD_{(marker mean)} (SD_{[marker mean]}: 1.23%; Table S5). These results indicated that the PowerSeq 46GY prototype library kit consists of a more balanced marker set compared to other prototype/early access MPS-based STR multiplexes, when only aSTRs are considered [45].

Heterozygote balance (HB) is expressed (for each sample and STR locus) as ratio between the minor and the major allele intensity for heterozygote genotypes. On average, all markers showed HB ratios ≥ 0.80, except D12S391 (mean: 0.79; SD: 0.13), D19S433 (0.78; 0.14), Penta E (0.77; 0.15), and D2S1338 (0.70; 0.19) (Figure S3, Table S4). In seven samples (0.13% of the sample set), we observed highly imbalanced genotype calls (HB ratios ≤ 0.30) at D2S1338 (5 ×), D19S433 (1 ×), and D21S11 (1 ×) (Figure S3, Table S4). Generally speaking, HB ratios were similar to those obtained with CE-based STR kits [46, 47] or reported in earlier MPS studies [48, 49]. D2S1338 was found to be most susceptible to heterozygote imbalance, which has been described by [49] before. Since, high imbalances potentially lead to marker drop-out, further optimization steps, especially for D2S1338, are required as already indicated by [49]. We note that known imbalances at D22S1045 [2, 50‐52] and D5S818 [51, 53‐55] were not observed for genotypes amplified with the PowerSeq 46GY library kit. In one sample, heterozygote imbalance was found at D5S818 after CE analysis but not with MPS (typed alleles — CE: 10/11, HB: 20%; MPS: 10/11, HB: 72.1%). Sequence analysis of the latter sample unveiled the presence of an A > G transition (rs182073376), which was located 36 nucleotides upstream the repeat region of D5S818 [56] and was found only once within our dataset. To understand the reason for this imbalance, we examined the SNP location with respect to the previously published PowerPlex 16 (Promega) primer sequences [57]. Indeed, the observed A > G transition was located ten nucleotides upstream of the 3′ end within the forward primer’s binding site, which apparently decreased thermal stability of the primer-template complex and, therefore, reduced PCR performance (Figure S4) [58, 59].

Stutter products are commonly known artifacts in STR typing that are caused by strand slippage of the DNA polymerase during the extension phase of PCR. Strand slippage results in the addition or deletion of typically one repeat unit in the nascent DNA strand [60]. Besides of rarely seen stutter products in n + 4 and n − 8 positions, the most prominent observed stutters are one repeat unit shorter, i.e., n − 4 for tetranucleotide STR markers.

Note, by applying MPS for STR analysis, each repeating motif may result in the formation of a, i.e., n − 4, stutter product and is therefore multivariate [61]. Furthermore, the formation of stutter seems to be also influenced by adjacent nucleotides [61]. As only single source samples were analyzed, we decided to reduce the complexity of sequence-based stutter analysis by considering stutter products as “defined by length” (univariate; recalculated from MPS results) instead of as “defined by sequence” (multivariate). As expected, stutter ratios increased with the growing number of repeat units (allele size) [60, 62]. Locus-specific stutter analysis showed that stutter ratios for integer alleles were higher than those for intermediate alleles (Figure S5), which is in line with earlier reports [61]. We found no evidence of increased stochastic variation of stutter height for samples belonging to category I (samples with reads < 63,500; Table S2), which is in contrast to the findings described by [62].

Stutter ratios were found to be similar for category I and category II samples (Figure S5; Table 2). The majority of STRs included in the PowerSeq 46GY panel showed mean stutter ratios ranging from 10 to 15% for both categories (Table 2). Stutters appeared to be generally higher for MPS-based STR genotyping than for CE-based kits [46, 47, 63] (Table 2), which is in line with earlier reports [45, 64‐66]. Rarely, we observed stutter values exceeding 20% (Table 2). However, as D22S1045 consists of a trinucleotide repeat, which is more prone to stutter artifacts, higher stutter values were expected and also known from earlier studies [44, 45, 67].

Sequence-based analysis revealed an increase of heterozygosity relative to CE-based genotyping due to isometric alleles (alleles of identical size but different sequence). Using MPS, 181 of the 1075 (16.8%) homozygous allele pairs were unveiled as isometric heterozygotes at 13 aSTRs: D5S818 (34 ×), D3S1358 (25 ×), D13S317 (23 ×), D21S11 (16 ×), D7S820 (15 ×), D8S1179 (14 ×), D16S539 (14 ×), vWA (11 ×), D2S1338 (9 ×), D2S441 (8 ×), D12S391 (6 ×), D1S1656 (4 ×), and TPOX (2 ×) (Table 4). Our findings for increased heterozygosity were in line with an earlier report [42], except for TPOX. For example, Gettings et al. (2018) [69] observed a minimal increase in heterozygosity (< 1%) at TPOX, while Silva et al. (2020) [49] did not observe sequence variation at TPOX, D7S820, and D13S317. In two samples, MPS was able to identify isometric heterozygous genotypes at TPOX, allele 8 (repeat structure variants [AATG]8 vs. [AATG]8 containing a G > T transversion located in the flanking region [rs149212737]).

The primary intention of this study was to characterize the full sequence information of STRs with respect to forensic DNA intelligence databasing. Despite a high degree of concordance between MPS and conventional CE methodologies, we observed sequence variation that could potentially cause differences between the apparent CE length and allele calls based on counting repeat units in the MPS data as applied below. Furthermore, and as described earlier, sequence variation adjacent to the repeat region is known to affect allele nomenclature based on repeat unit counting [73]. In addition, Bodner and Parson (2020) [9] reported that SNP- and insertion/deletion-caused discrepancies between MPS and CE allele sizes were also a main reason of errors and discrepancies detected during STRidER quality control of MPS datasets.

Instances of flanking region deletions were observed at D2S441, D19S433, and Penta D, respectively (Table S6). At D2S441, we identified a single [T/-] deletion (rs888232687) of the first base downstream of the repeat block (CE: 9.3; MPS: 10) [56]. At D19S433, alignment revealed a single [CT/-] deletion (rs745607776) located in the 5′ flanking region (CE: 14.2; MPS: 15) that was previously described [42, 45, 56]. At Penta D, we identified a 13-nucleotide [AAGAAAGAAAAAA/-] deletion (rs1190908807) that results in the length-based allele 2.2 previously reported by [42, 56, 69]. Additionally, we observed a [A/-] deletion (rs536566765) located in the 3′ flanking region of Penta D (CE: 13.3; MPS: 14) that could cause discordance between MPS and CE as previously reported by [71, 74]. Considering sequence information, the two aforementioned samples revealed five and 14 full repeat units, respectively (Table S6). We note that all these described deletions would have caused seemingly shorter amplicons resulting in discrepant allele size estimations by mimicking intermediate alleles with CE-based detection methods (Table S6).

In addition, at D19S433, we identified 55 intermediate alleles (CE: X.2; representing 11.1% of all D19S433 alleles) that contained a well-characterized [TC/-] deletion (rs147936416; Table S6) [56]. This particular polymorphism is located inside the repeat region of D19S433, and spans the border of a counted repeat unit and an uncounted nucleotide block (for more information, see [56, 73]). This might lead to different micro-variant allele designation depending on the technology and counting method used. For example, an allele called 12.2 based on CE would result in an allele 12.3 based on repeat unit counting from MPS results (Table S6). Of note, alignment of these micro-variant alleles currently also differs in catalogues [10, 56]. As any kind of insertion/deletion can impact concordance between MPS and CE, it is inevitable to come up with a straight forward nomenclature system that maintains backward compatibility to the size-based STR profiles stored in national DNA databases. Based on this, it is important to collect and describe as many sequence variants as possible to increase sequence-based genotype accuracy [71].