If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Institute of Legal Medicine, Medical University of Innsbruck, Innsbruck, AustriaForensic Genetics Unit, Institute of Forensic Sciences, University of Santiago de Compostela, Spain
National Forensic Centre (NFC), Swedish Police Authority, Linköping, SwedenApplied Microbiology, Department of Chemistry, Lund University, Lund, Sweden
Biological Traces, Netherlands Forensic Institute, Laan van Ypenburg 6, 2497 GB The Hague, The NetherlandsUniversity of Amsterdam, Swammerdam Institute of Life Sciences, Science Park 904, 1098XH Amsterdam, The Netherlands
Institute of Legal Medicine, Medical University of Innsbruck, Innsbruck, AustriaForensic Science Program, The Pennsylvania State University, State College, PA, USA
Development of new targeted bisulfite MPS assays for age estimation from semen.
•
Bisulfite multiplex PCR design targeting 13 or 5 age markers.
•
Inter-laboratory validation showed stable and accurate DNA methylation quantification.
•
Robust results with 50 ng DNA input at bisulfite conversion level.
Abstract
The analysis of DNA methylation has become an established method for chronological age estimation. This has triggered interest in the forensic community to develop new methods for age estimation from biological crime scene material. Various assays are available for age estimation from somatic tissues, the majority from blood. Age prediction from semen requires different DNA methylation markers and the only assays currently developed for forensic analysis are based on SNaPshot or pyrosequencing. Here, we describe a new assay using massively parallel sequencing to analyse 13 candidate CpG sites targeted in two multiplex PCRs. The assay has been validated by five consortium laboratories of the VISible Attributes through GEnomics (VISAGE) project within a collaborative exercise and was tested for reproducible quantification of DNA methylation levels and sensitivity with DNA methylation controls. Furthermore, DNA extracts and stains on Whatman FTA cards from two semen samples were used to evaluate concordance and mimic casework samples. Overall, the assay yielded high read depths (> 1000 reads) at all 13 marker positions. The methylation values obtained indicated robust quantification with an average standard deviation of 2.8% at the expected methylation level of 50% across the 13 markers and a good performance with 50 ng DNA input into bisulfite conversion. The absolute difference of quantifications from one participating laboratory to the mean quantifications of concordance and semen stains of remaining laboratories was approximately 1%. These results demonstrated the assay to be robust and suitable for age estimation from semen in forensic investigations. In addition to the 13-marker assay, a more streamlined protocol combining only five age markers in one multiplex PCR was developed. Preliminary results showed no substantial differences in DNA methylation quantification between the two assays, indicating its applicability with the VISAGE age model for semen developed with data from the complete 13-marker tool.
The discovery of the correlation between DNA hypo- and hypermethylation at cytosine-guanine dinucleotides (CpG sites) with the human lifespan led to the wide usage of DNA methylation as a biomarker of aging [
]. The quantitative analysis of DNA methylation at such CpG sites provides accurate chronological age estimation from biological material, which makes it a valuable tool for forensic genetic investigations [
A prerequisite for forensic assays is a maximum possible sensitivity due to the limited DNA amount that can be recovered from crime scene samples. This impedes the usage of assays targeting a high number of markers with microarray-based DNA methylation analysis, which require approximately 250 ng of DNA [
]. Therefore, forensic age estimation assays have focused on fewer markers capable of age prediction from a specific tissue, and showing reliable results with reported DNA inputs of 50 ng to 10 ng for bisulfite conversion (5 ng at PCR level [
]. Nevertheless, some of the known blood age markers also show age-associated DNA methylation in other forensically relevant tissues like saliva and bones [
] and although semen stains are commonly analyzed in forensic laboratories, only a handful of studies established age estimation models for this cell type. Lee et al. [
] demonstrated for the first time that age estimation from semen is possible by developing an age prediction model based on three CpG sites (cg06304190 in TTC7B, cg06979108 in NOX4/FOLH1B and cg12837463 in LOC401324) targeted by SNaPshot. The mean absolute error (MAE) from the chronological age of this assay was approximately 5 years [
] recently reported a smaller MAE of 2.4 years, however, their prediction was based on 51 microarray-typed genomic regions, a number which is currently difficult to incorporate into a viable forensic assay.
In this study, we aimed to further advance chronological age prediction from semen for routine forensic applications by developing a new assay in the course of the VISible Attributes through GEnomics (VISAGE) project. The VISAGE consortium has aimed to develop and validate tools based on targeted massively parallel sequencing (MPS) for predicting appearance, ancestry and age to provide information on an unknown sample donor for police investigations.
The VISAGE enhanced tool for age estimation from semen (ET-13; Enhanced Tool-13) combines ten novel markers (regions) identified by Pisarek et al. [
]. The ET-13 was validated by five VISAGE consortium laboratories in a collaborative exercise to assess the assay’s reproducibility and sensitivity. Subsequent to the completed validation phase, the assay was used for data generation to establish the VISAGE age model for semen, which is presented in a separate publication that also provides details regarding marker selection [
]. As MPS allows for the analysis of CpG sites adjacent to the initially selected 13 candidate CpG sites (Table 1), 23 additional CpG sites located within the 13 marker regions were considered by Pisarek et al. [
]. The final VISAGE model for semen consists of six CpG sites from five age markers (model CpG sites, Table 2). Aiming to provide a more streamlined option for routine forensic laboratories, we optimized an additional five-marker assay (ET-5) based on only one multiplex PCR. A first evaluation of the ET-5 was performed to compare DNA methylation quantification between the two assays.
Table 1Age markers, genomic locations of candidate CpG sites used for initial assay development, primer sequences and final assay concentrations.
]. The combined marker set, termed ET-13, was designed to target the 13 candidate CpG sites listed in Table 1. Development of an age prediction model based on data generated with the ET-13 allowed for the simultaneous analysis of CpG sites adjacent to the selected candidate CpG sites (in total 36 CpG sites). Marker selection, which was conducted using multivariable stepwise linear regression, showed that only six CpG sites located in five of the 13 candidate markers were optimal for accurate age prediction (Table 2) [
]. These CpG sites were included in the final prediction model and used to develop a second assay (ET-5).
2.2 Concordance samples, stains and DNA methylation standards
Frozen semen samples of one 33-year-old (Sample 1; Cat No: 991–04-AS2) and one 42-year-old (Sample 2; Cat No: 991–04-AS2) donor were purchased from Lee Biosolutions (Maryland Heights, MO, USA). DNA was extracted with the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) following the user-developed “Isolation of genomic DNA from sperm” protocol 2 (QA04 Jul-10) available on the Qiagen website [
]. In brief, 100 µl semen were incubated with 100 µl Buffer X2 (20 mM Tris-Cl (pH 8.0), 20 mM EDTA, 200 mM NaCl, 80 mM DTT, 4% SDS and 250 µg/ml Proteinase K) for 1 h at 55 °C. After adding 200 µl Buffer AL and 200 µl ethanol, the protocol followed the “Tissue Protocol” of the QIAamp DNA Mini Kit. Final DNA extracts were quantified on a Qubit 2.0 Fluorometer using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA, hereafter referred to as TFS) and were used as concordance samples. Stains were prepared by applying 50 µl semen on Whatman FTA cards (Merck, Darmstadt, Germany). DNA was extracted from stains by the five participating consortium laboratories according to their in-house DNA extraction and quantification protocols (Table S1). DNA methylation standards were prepared using the WGA methylated & non-methylated DNA Set (Zymo Research, Irvine, CA, USA) diluted to 20 ng/µl in low TE (10 mM Tris, 0.1 mM EDTA, pH 8). The unmethylated DNA control sample was mixed with the fully methylated DNA control sample at different volume ratios to achieve 0%, 5%, 10%, 25%, 50%, 75% and 100% methylated DNA standards.
2.3 Assay development and protocols
Assay development was based on the previously published VISAGE basic tool for age estimation from blood [
Bisulfite conversion was performed with the Premium Bisulfite kit (Diagenode, Liège, Belgium), using 200 ng DNA if not otherwise specified. One extra dry spin was added before elution in 10 µl elution buffer to avoid ethanol carryover into the subsequent PCR step. Eluates were used for amplification on the same day.
2.3.2 ET-13: two multiplex PCR approach
Primers were designed using the PyroMark Assay Design Software 2.0 (Qiagen), except for primer sequences for NOX4/FOLH1B, which were described in [
]. The SYT7 reverse and the PALM forward primers each contained a single CpG dinucleotide in their binding sites. Therefore, we introduced an intentional base mismatch into these two primer sequences (Table 1) at the positions of the CpG cytosines. All primer sequences (Table 1) were tested in silico for specificity using BiSearch [
]. In vitro, all primers were tested in singleplex reactions using the Multiplex PCR Kit (Qiagen) and correct amplification was verified using Sanger sequencing. The BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) was used according to the manufacturer's instructions. Sequencing electrophoresis was performed on an Applied Biosystems 3500 Series Genetic Analyzer (TFS). Temperature gradient PCR was utilized for optimizing annealing conditions. Amplicon size and yield of multiplex reactions were evaluated on a Bioanalyzer instrument using the DNA 1000 Kit (both Agilent Technologies, Santa Clara, CA, USA) and the results were used to adjust primer concentrations. Different primer combinations were tested to obtain optimal multiplex PCRs. The final ET-13 is based on two multiplex PCRs carried out with 4 µl eluate from bisulfite conversion in 25 µl total volume using the Multiplex PCR kit (Qiagen). Cycling conditions were as follows: initial denaturation at 95 °C for 15 min, 40 cycles consisting of 94 °C for 30 s, 59 °C for 30 s, 72 °C for 1 min and a final extension step at 72 °C for 10 min. Allocation of primers to the two multiplex PCR assays of the ET-13, as well as all final primer concentrations are listed in Table 1. Clean up of PCR products was performed using 1.5X ratio Agencourt AMPure XP (Beckman Coulter Brea, CA, USA) according to the manufacturer’s protocol and eluted in 15 µl low TE buffer.
2.3.3 ET-5: one multiplex PCR assay
For development of the 5-plex PCR assay targeting the five age markers included in the final VISAGE age model (based on 6 CpG sites; Table 2), primer concentrations and thermocycler conditions were re-optimized with the same primer sequences used in the ET-13 set (Table 1). The final 5-plex PCR assay was performed using 8 µl eluate from bisulfite conversion in 25 µl total PCR volume and the following thermocycler protocol: Initial denaturation step at 95 °C for 15 min, 35 cycles consisting of 94 °C for 30 s, 59 °C for 30 s, 72 °C for 1 min and a final extension step at 72 °C for 10 min
2.3.4 Library preparation
All PCR products were quantified using the Qubit dsDNA HS Assay Kit (TFS). Library preparation was performed using the KAPA Hyper Prep (96rxn) kit with the KAPA Library Amplification Primer Mix (Roche, Basel, Switzerland) in half volumes following the manufacturer’s instructions. For each sample, 25 ng PCR products from the two multiplex reactions (ET-13) per sample were mixed and adjusted to 25 µl total volume for end repair and A tailing. Adapter ligation was performed with KAPA SI Adapter Kit Set A+B (Roche) diluted to 15 µM and library amplifications were carried out with five cycles. Agencourt AMPure XP (Beckman Coulter) were used for post-ligation and post-amplification clean-up steps, elution was performed in 12 µl and 20 µl low TE, respectively. Library quantification was carried out with the KAPA Library Quantification Kit - Complete Universal for Illumina Platforms (Roche) following the manufacturer’s instructions. Library preparation with the ET-5 was performed with 50 ng purified PCR products.
2.3.5 Massively parallel sequencing and data analysis
Libraries were diluted according to MiSeq System Denature and Dilute Libraries Guide, Protocol A (Document #15039740 v10; Illumina, San Diego, CA, USA). In brief, libraries were pooled equimolarly at 4 nM and further diluted to 7 pM with 2 µl of 20 pM PhiX Control v3 (Illumina) spiked into 598 µl total volume. The maximum number of samples pooled in one sequencing run was 24. Sequencing was performed with paired end sequencing and 201 cycles in both directions using the MiSeq FGx ForenSeq Reagent Kit (Verogen, San Diego, CA, USA). Generated fastq files were aligned with the Burrows-Wheeler alignment for bisulfite converted DNA sequences using the bwa-meth software [
] was used for bam file creation, sorting, filtering and indexing. All sample alignments were analysed using the Integrative Genomics Viewer (IGV) v2.4.10 [
]. Read counts at candidate CpG sites, model CpG sites (Table 2) and all non-CpG-Cs were extracted using bam-readcount (https://github.com/genome/bam-readcount) with a minimum mapping quality and minimum base quality set to 30. Methylation values were calculated as percentage beta-values (C reads were divided by the sum of C reads and T reads). Bisulfite conversion efficiency was estimated as the mean T read percentage at all non-CpG positions per sample. Sample coverage was calculated as the sum of reads at candidate CpG sites and normalized read depth represents the number of reads at a candidate CpG position divided by sample coverage. Base misincorporation rates at CpG sites were calculated by dividing read depth by the sum of unexpected bases at the respective positions, which would refer to the sum of A and G reads for CpG sites targeted on the sense strand and the sum of C and T reads for CpG sites targeted on the antisense strand. For methylation analysis, only CpG sites with a minimum number of 1000 reads were considered. Age prediction analysis was performed using the VISAGE model for semen, a multivariate linear regression model using percentage beta values obtained from model CpG sites (Table 2; [
]. The standard deviation was used to measure variation around mean values.
2.4 Validation framework
Five VISAGE consortium laboratories participated in a collaborative exercise and performed one MiSeq FGx run per laboratory for the developmental validation of the ET-13. Each laboratory analysed seven differentially methylated DNA standards at optimum DNA input (200 ng) to evaluate the assay’s reproducibility. One DNA methylation standard was assigned to each participant to test the assay’s sensitivity with duplicates of 50 ng, 20 ng, 10 ng and 1 ng DNA input into bisulfite conversion. One concordance sample (200 ng DNA input) and one stain were processed in duplicate by each laboratory, except for Laboratory 1, which analysed both samples (1 and 2). Duplicates were prepared after DNA extraction and quantification. Sample assignment to laboratories and sample numbers per MiSeq FGx run are indicated in Table S3. Overall, 99 samples were analysed for assay validation.
2.5 Species specificity testing
DNA extracts from the ten animal samples used for species specificity testing (Table S4) derived from an in-house sample collection [
], bisulfite converted (100 ng DNA input) and amplified following the ET-13 protocol. All samples were processed with human gDNA as an amplification control and PCR-grade water as a negative template control. Purified PCR products were loaded on a DNA 1000 chip for electrophoretic separation using the Bioanalyzer instrument. BiSearch's Primer search ePCR function [
] was used to scan animal databases for primers that may have created a PCR product in the ten animal samples. Primer pairs were tested in silico against bisulfite converted animal genomes using default parameters except for the “PCR product to display” option, which was increased to 500. The sequence of PCR products retrieved from BiSearch were further analysed using the BLAST [
] program “Needleman-Wunsch alignment of two sequences” to compare sequence homology. BiSearch results that hits on the antisense strand were reverse complemented prior to sequence comparison.
3. Results
3.1 Development and inter-laboratory validation of the 13-marker enhanced tool for age estimation from semen (ET-13)
The ET-13 was designed to target 13 candidate CpG sites in two multiplex PCRs (Table 1) that can be run in parallel on one thermocycler and are combined for library preparation. The minimum and maximum amplicon sizes were 75 bp and 242 bp (mean = 141.8 bp), respectively. In silico testing showed alternative PCR products for TUBB3 (size 733, antisense strand) and NOX4/FOLH1B (size 194, antisense strand). Due to three mismatches within the primer binding sites in each of the two TUBB3 primers, the amplification of the alternative larger amplicon was considered as negligible. In contrast, primer sequences of NOX4/FOLH1B match the binding sites of the alternative PCR product with the same amplicon size. Further investigation of the NOX4 gene showed that a pseudogene (LOC729960) is located on the other arm of chromosome 11, which impedes the design of primer specific only for the candidate CpG site. In fact, a previous study has identified a potential cross hybridization for probes targeting the NOX4/FOLH1B CpG site (cg06979108) on the Illumina Infinium MethylationEPIC BeadChip [
] and measurement of the combined DNA methylation level from the NOX4/FOLH1B candidate CpG site and the CpG site of the co-amplified pseudogene (chr11:49,251,356) showed statistically significant age correlation in the study by Pisarek et al. [
]. Therefore, the candidate CpG site was retained in the marker set and is hereafter referred to as NOX4/FOLH1B.
A first test run was performed using differentially methylated DNA standards, which showed a promising assay performance and robust methylation quantification at all target CpG sites (data not shown). The five participating VISAGE consortium laboratories received detailed instructions for validation experiments. The data presented here is a combined analysis of the 99 samples to validate the technical performance of the assay prior to data generation for the age model development.
3.1.1 Performance evaluation
General assay performance was evaluated using methylated DNA standards at optimum DNA input for bisulfite conversion (N = 35). In a first step, sample coverage was calculated and compared among laboratories (Fig. 1A, Table S5). Whereas Laboratory 1 (median = 388,165 paired reads), 2 (median = 199,483 paired reads) and 5 (median = 574,037 paired reads) showed uniform coverage values, Laboratories 3 (median = 801,010 paired reads) and 4 (median = 1,816,451 paired reads) yielded wider coverage variability and higher sample coverages compared to the overall results (median = 540,564 paired reads). The higher read depth obtained by Laboratory 4 could be due to a higher cluster density during sequencing (1568 K/mm2) compared to the overall average cluster density (674.4 K/mm2, Table S6). Furthermore, variable library inputs during preparations of the library pool might explain high read depth outliers within one run. Consequently, median values were considered for evaluation of the sequencing depths. The overall median read depth at the 13 candidate CpG sites was 29,815 reads. Dropouts (target position < 1000 paired reads) were only observed for samples from Laboratory 4 at candidate CpG positions on SYT7 and TBX4. These CpG sites were excluded from downstream analysis in all samples of Laboratory 4. The read distribution between the 13 amplicons was evaluated by calculating the normalized read depth at candidate CpG sites for each sample (Fig. 1B). Four markers showed higher than expected read depth values, with NOX4/FOLH1B outperforming the remaining markers. In contrast, SH2B2, TBX4 and ARHGEF17 showed lower read depths in all laboratories. However, median read depth at underperforming markers can still be considered sufficient with 7874; 9555; 13,923 paired reads at the three candidate CpG sites, respectively.
Fig. 1(A) Sample coverage of the seven DNA methylation standards used per laboratory. The median is shown for each MiSeq FGx run. (B) Normalized read depth per candidate CpG sites of DNA methylation standards (N = 35) across laboratories. The dashed line indicates the expected normalized read depth per marker (= 0.08; 1 divided by 13 markers).
Sequence quality was evaluated by calculating base misincorporation rates, bisulfite conversion efficiency and sequencing baseline noise. Base misincorporations rates refer to the percentage of wrongly incorporated bases and help to estimate the intrinsic precision of the assay. Additionally, higher than expected frequencies could indicate misaligned reads. The obtained mean misincorporation rate at the 13 candidate CpG sites was low with a mean of 0.1% ( ± 0.3) wrongly incorporated bases. Base misincorporations above 1% were only observed at the SYT7 candidate CpG site (six DNA methylation standards, Laboratory 2) and the ARGHEF17 candidate CpG (four DNA methylation standards, Laboratory 1) and carefully inspected using the IGV software. Misalignments were found at a low percentage, not reaching the full amplicon. This may result from the reduced sequence complexity of bisulfite converted DNA. Examples can be found in Supplementary File 1A–B. The minimum obtained bisulfite conversion efficiency at non-CpG C bases per sample was 99.4%, indicating that non-converted cytosines were unlikely to compromise the final calculated methylation levels. Sequencing baseline noise was assessed for each run by calculating the percentage of total reads from NTCs to the respective median sample coverage. Only one NTC showed a percentage of aligned reads above 1% (1.2%). Visual inspection of alignments using the IGV software showed incorrect read alignment indicated by high base misincorporation rates (mean = 8.9% ± 19.9) and incomplete reads (Supplementary File 1C – D).
For the six model CpG sites, additional quality control was performed by calculating read depth and base misincorporation rates. Since model CpG sites of each marker are located on the same amplicons as their initial candidate CpG sites, only minor differences (<1.5%) in read depth were observed between them. Lower read depths were only observed at the IFITM2 model CpG site, with a difference of 12,665.8 (40.2%) paired reads between overall means. Furthermore, the average base misincorporation rates were comparable to rates at candidate CpG sites with an elevated frequency of 1.8% ( ± 1.5) only at the model CpG site of IFITM2. Visual inspection of alignments showed misaligned reads in the reverse strand after a poly-G tract causing the incorporation of Gs at the IFITM2 model CpG site (Supplementary File 1E). Notwithstanding, methylation quantification at the model CpG site was robust. When comparing standard deviations of the mean methylation quantifications (N = 5) of the seven DNA methylation standards, no statistically significant differences were observed (paired t-test, P-value > 0.05) between the IFITM2 model and the candidate CpG site.
3.1.2 Reproducibility
Seven premixed DNA methylation standards were sent to the five participating laboratories to assess methylation quantification (Fig. 2). A tendency to underestimate the expected methylation levels was observed for most markers, appearing as a uniform bias from all laboratories. Variability between laboratories was assessed by calculating the standard deviation for each DNA methylation standard. Notably, the highest average standard deviation across markers was found at 75% expected DNA methylation level (mean SD = 3.1%), followed by 50% (mean SD = 2.8%) and 25% (mean SD = 2.2%). Whereas almost no variability was observed at 0% methylation (mean SD = 0.4%), the fully methylated control DNA showed an average standard deviation of 1.6%, which was higher than at 10% (mean SD = 1.3%) and 5% (mean SD = 1.0%) expected methylation level. When comparing markers, the highest standard deviations were observed at SH2B2 (SD = 7.4% at 50% methylation level), which also showed lower read depths compared to other markers (Fig. 1B). Overall, the methylation levels obtained by the five laboratories demonstrate that the assay generates reproducible results, with small standard deviations for the majority of markers.
Fig. 2Mean methylation quantifications versus the expected methylation of seven differentially methylated DNA standards analysed by five laboratories using the ET-13 and quantifications obtained for the two replicates with the ET-5. Error bars of the ET-13 represent the standard deviation, the dashed line represents the line of identity (intercept = 0, slope = 1).
Sensitivity was tested by lowering the DNA input into bisulfite conversion of five methylated DNA standards to 50 ng, 20 ng, 10 ng and 1 ng. The 1000 reads threshold was met for almost all markers and samples down to 10 ng DNA input. Dropouts only occurred at SYT7 and TBX4 candidate CpG sites of samples prepared by Laboratory 4 and one additional dropout was registered at the candidate CpG site of EXOC3 at 10 ng DNA input. In contrast, all 1 ng samples yielded read depths below the threshold at two or more candidate CpG sites (in total 36 dropouts). As read depths between MiSeq FGx runs showed large differences (Fig. 1-A), normalized read depth was considered for comparing the assay performance between 200 ng and lower DNA inputs (Fig. S1-A). Overall, the 200 ng samples showed no statistically significant differences (Pairwise Wilcoxon test, Bonferroni corrected P-value > 0.05) to samples with 50 ng DNA input. At lower DNA inputs, two markers from 20 ng SD samples (NOX4/FOLH1B and ARHGEF17), three markers of 10 ng samples (NOX4/FOLH1B, ARHGEF17 and IFITIM2) and six markers of 1 ng samples (NOX4/FOLH1B, ARHGEF17, IFITIM2, PALM, GALR2 and EXOC3) showed significant differences in normalized read depths to the optimum DNA input samples. Assessment of sequence quality showed base misincorporation rates similar to the optimum DNA input samples. The average bisulfite conversion efficiency decreased at 1 ng DNA input (mean = 97.2%), however, comparison between amplicons showed that almost all incomplete conversions of non-CpG C bases derive from the EXOC3-amplicon (Supplementary File 1F). Misaligned reads at non-CpG C bases of this amplicon reached the 1000 read threshold, which decreased the average bisulfite conversion rates, whereas the candidate CpG sites of EXOC3 were excluded due to insufficient read depth.
Methylation quantification at low DNA inputs was evaluated at two levels. First, we assessed differences between duplicates. At low DNA inputs, large differences in DNA methylation rate estimates are expected due to stochastic effects. The mean difference in DNA methylation percentage (DNAm%) between duplicates was 4.0% (median = 1.9%) for 50 ng samples and increased to 8.8% (median = 4.3%) for 20 ng, 11.7% (median = 4.9%) for 10 ng and 19.9% (median = 8.4%) at 1 ng DNA input. Second, we calculated the mean of duplicates and compared it to the mean value obtained with 200 ng DNA input (Fig. 3) for all five methylation ratios. At 50 ng DNA input, the assay could retain accurate methylation quantification. Although some markers showed stable methylation quantification down to 10 ng DNA input (e.g. CpG sites at NOX4/FOLH1B, TBX4, SYT7, LOC401324), the number of outliers with high differences started to increase below 50 ng DNA input.
Fig. 3Absolute difference between mean methylation of sensitivity samples to optimum DNA input samples (200 ng) per candidate CpG site.
The ET-13 set was tested with two concordance samples (DNA extracts from semen) and two “mock” casework samples (semen stains). The assay demonstrated a good overall performance with semen samples, showing low base misincorporation rates (mean = 0.07% ± 0.3) and bisulfite conversion efficiencies per sample exceeding 99.5%. High read depth values at all candidate CpG sites were observed for all laboratories (median = 38,615 reads, excluding SYT7 and TBX4 candidate CpG sites of Laboratory 4). Moreover, normalized read depth of semen samples exhibited a different read distribution between the 13 markers than observed for DNA methylation standards (Fig. S1-B).
Methylation quantifications were evaluated based on DNAm% differences between duplicates and the comparison of the mean methylation quantifications from duplicates among laboratories (Fig. 4). One outlier (Laboratory 3, stain 2) was detected, resulting in a large difference in observed methylation for these duplicates (mean = 26.7%). When inspecting the respective quantification values from library preparation, the laboratory reported a library concentration of only 0.3 pM (mean = 281.3 nM), which indicates sample loss during the laboratory workflow probably leading to distorted methylation quantification results although normal read depth values were obtained. The respective sample was removed from further analysis. Methylation quantifications of the concordance and “mock" casework samples showed only small differences between technical duplicates with a mean difference of 0.9% ( ± 0.8) and 0.7% ( ± 0.8) across the 13 markers for concordance and ”mock” casework samples, respectively.
Fig. 4Observed DNA methylation values at candidate CpG sites for concordance and “mock” casework samples 1 and 2. Each sample was analyzed by three participating laboratories in duplicates (one outlier removed) using the ET-13: Sample 1 by Laboratory 1, 2 and 5; Sample 2 by Laboratory 1, 3 and 4. DNA methylation values obtained by Laboratory 1 using the ET-5 are indicated for the respective markers at SH2B2, EXOC3, GALR2, IFITM2 and NOX4/FOLH1B CpG sites only.
Laboratory comparisons revealed an absolute difference in DNAm% from one laboratory (mean of duplicates) to the mean quantifications obtained by the other laboratories of 0.8% and 0.9% for concordance sample 1 and 2 and 0.8% and 1.4% for stain 1 and 2 across candidate CpG sites (Fig. S2). The highest differences between laboratories were seen for SYT7 with a mean absolute difference of 3.8%. Additionally, the absolute differences in the overall mean methylation quantification results between concordance and “mock” casework samples was low for all candidate CpG sites with a mean difference of 0.7% ( ± 0.5). This suggests robust methylation quantification throughout all laboratories with low differences between the analyses of pre-extracted DNA samples and semen stains. Furthermore, we compared the final age prediction outcome among laboratories (Table 3). DNA methylation percentages at the six model CpG sites were extracted for each laboratory and used for age prediction with the VISAGE age model for semen. The mean predicted age varied from 35.7 to 41.2 years (5.5 years difference) for sample 1 and 38.6–42.7 years (4.1 years difference) for sample 2. The difference from the chronological age was on average 5.4 years for sample 1 and 2.0 years for sample 2 across concordance and “mock” casework samples.
Table 3Mean predicted age (N = 2) based on DNAm levels obtained with the ET-13 or the ET-5 by laboratory.
Comparison between electropherogram profiles (Agilent Bioanalyzer) produced by animal DNA samples and the human control showed clear differences in both multiplex PCR assays, except for the 7-plex PCR products of the Orangutan sample (Supplementary File 2). This can be explained by the genetic similarities amongst primates. However, some animal samples yielded PCR products, potentially deriving from conserved genetic regions that could interfere with human methylation quantification. Even though most of the PCR products were detected at much lower concentration than in the human control sample, all primers were tested in an in-silico PCR using BiSearch with the respective animal genomes. No bisulfite converted genomes were available for Goat and Orangutan. BiSearch results showed that only the primer pairs targeting TUBB3, ARHGEF17 and LOC401324 were able to produce in silico amplicons in samples from Cattle, Rat, Sheep, Dog and Horse. These sequences were further compared to the human amplicons using BLAST (Needleman-Wunsch alignment of two sequences). Amplicon sequences retrieved with primers targeting TUBB3 and LOC401324 obtained only a low percentage of identity (16–54%) when compared to the human sequence, which would most likely lead to unmapped reads which would not interfere with the final methylation quantification results. For the primer pair targeting ARHGEF17, one amplicon with a high sequence identity (91%) was obtained for the converted Dog genome. Although this may lead to mapped sequence reads, ARHGEF17 primers show five base mismatches within the primer binding sequence, which likely reduces PCR efficiency. A careful inspection of alignments should be performed, in particular if animal-human mixtures may be present.
3.2 Development and performance evaluation of the five-marker enhanced tool for age estimation from semen (ET-5)
The ET-5 set combines only the final five VISAGE age model markers in one multiplex PCR and was evaluated with duplicates of the seven DNA methylation standards and the two concordance and “mock” casework samples by Laboratory 1. The assay achieved a median read depth of 113,235 paired reads across the five targeted regions (Fig. S3-A). When evaluating normalized read depth, a more balanced distribution was observed for concordance and casework type samples (N = 8) compared to DNA methylation standards (N = 14, Fig. S3-B). Sequence quality was assessed with the same parameters as with ET-13: by calculating base misincorporation rates at candidate (mean = 0.05% ± 0.02%, median = 0.04%) and model CpG sites (mean = 0.81% ± 1.99%, median = 0.04%), bisulfite conversion efficiency (mean > 99.7%) and analysis of the total reads percentage of two NTCs from the mean sample coverage (1.8% and 5.2%). Visual inspection of alignments from the NTCs showed that no uniform read depth across the amplicons was achieved. IGV captures of NTC amplicons with a read depth above 1000 reads at candidate CpG sites can be found in Supplementary File 1G-M. As with the ET-13, a higher frequency of base misincorporations was only detected at the model CpG site of IFITM2 (mean = 4.5 ± 2.8), caused by misalignments in the reverse strand. Likewise, these misalignments had no influence on the obtained DNA methylation level, which showed low differences between duplicates of DNA methylation standards (2.1% ± 1.3) and concordance and “mock” casework samples (0.9% ± 0.5) analysed with the ET-5.
Obtained methylation values at candidate CpG positions of DNA methylation standards were in accordance with the ET-13 findings (Fig. 2), indicating that no substantial differences in methylation quantification were introduced by changing the multiplex PCR conditions. Furthermore, the discrepancy between the ET-13 and the ET-5 methylation quantifications of concordance and “mock” casework samples were small (Fig. 4) with a mean absolute difference in DNAm% of 0.7% ( ± 0.6) across the five shared markers. To further evaluate the effect of differences in obtained DNA methylation values, we compared age prediction outcomes between ET-13 and ET-5. Age estimations obtained using the ET-5 are in accordance with ET-13 (Table 3).
4. Discussion
The presented protocols for the ET-13 and ET-5 marker sets are the first targeted bisulfite MPS assays intended for age estimation from forensic semen samples. Even though quantitative DNA methylation methods were compared among laboratories in previous studies [
], to the best of our knowledge, this current study presents the first inter-laboratory validation of a single targeted bisulfite MPS assay for forensic applications. Such validation studies are essential to fully understand technical assay variabilities and enable the implementation of the MPS-based age tools into routine forensic laboratories.
Our results show that the ET-13 set has an overall good performance and produced high quality sequences for all 13 amplicons. Read distributions between amplicons showed some variability between laboratories, however, the same tendencies for over- and underperforming markers were obtained. For optimum input samples (200 ng/bisulfite conversion), drops below the 1000 read threshold were observed in Laboratory 4 only, with candidate CpG sites on SYT7 and TBX4 being affected in all samples. The associated sequencing run featured a high level of imbalance. Here, the markers with very high read depths (Fig. 1A) may have outcompeted SYT7 and TBX4 target region amplification.
Reproducibility analysis showed that the quantification levels obtained were stable between laboratories, with an average standard deviation across markers of 2.8% for the 50% methylated DNA standard. As already documented for the previously established VISAGE tools for age estimation [
], higher variation was observed for the highly methylated DNA standards compared to lower methylated standards, which may be a result of the bisulfite conversion process or could be inherent to the in vitro methylated control DNA used.
A thorough characterization of the assay sensitivity is of major importance for forensic applications. We observed only minor differences in read distribution and methylation quantification results between the 200 ng (optimum as indicated by the manufacturer) and 50 ng DNA input into bisulfite conversion. Below 50 ng, differences in normalized read depth increased with decreasing DNA input, possibly resulting from imbalances at the multiplex PCR step, where certain markers seem to have outcompeted other markers. At 1 ng DNA for bisulfite conversion, the generally high sequencing depth could no longer overcome differences in marker performance. Variability in methylation quantification results increased below 50 ng DNA, affecting the difference between technical duplicates and the difference to the optimum input values (Fig. 3). It should be noted that the DNA amount at PCR level may be considerably lower due to DNA loss and degradation during bisulfite conversion. When assuming 55% [
Bisulfite conversion of DNA: performance comparison of different kits and methylation quantitation of epigenetic biomarkers that have the potential to be used in non-invasive prenatal testing.
] DNA recovery and 4 µl input into each multiplex PCR, a DNA input of 50 ng, 20 ng, 10 ng and 1 ng into bisulfite conversion would correspond to 11–14.8 ng, 4.4–6.1 ng, 2.2–3 ng and 0.2–0.3 ng at the PCR level, respectively. Further studies evaluating semen samples will be needed to pinpoint the assay’s sensitivity with regard to the final age prediction outcome.
Two semen samples were used to assess the assay performance and inter-laboratory variability with DNA extracts and “mock” casework samples (Fig. 4). For both sample types, the ET-13 showed robust methylation quantifications with small differences between duplicates performed by each laboratory. The results were consistent with previously recorded technical variabilities of a targeted bisulfite MPS assay reporting a maximum of 3% average difference of DNAm% for technical duplicates [
]. Additionally, inter-laboratory differences between mean quantifications remained small for DNA extracts and stains resulting in robust age prediction results. Differences between estimated and chronological age were within the expected range for the VISAGE model for semen (MAE = 5.1 years [
]. Therefore, not only were the 13 candidate CpG sites considered for age estimation, but also neighboring CpG sites. The final VISAGE age model for semen comprises six CpG sites on SH2B2, EXOC3, GALR2, IFITM2 and NOX4/FOLH1B (Table 2). Consequently, we developed the ET-5, targeting only these five model markers in one 5-plex PCR assay. The DNA methylation quantifications obtained from the ET-5 assay were in accordance with the ET-13, which is crucial for the age prediction model based on data from the whole ET-13 assay. The simultaneous amplification of all relevant age model markers is less labor-intensive and likely improves the assay sensitivity, since the eluate does not have to be split between two amplification reactions. The ET-5 therefore, represents a promising tool for forensic routine laboratories, which will be completed by a statistical software automating MPS data analysis and age prediction, currently developed by the VISAGE consortium.
Apart from validation studies that prepare the ground for forensic applications of age estimation assays, further investigations determining possible confounding factors like lifestyle-related influences (e.g., alcohol consumption [
DNA methylation signature in blood does not predict calendar age in patients with chronic lymphocytic leukemia but may alert to the presence of disease.
] provides an overview about the current literature on factors affecting epigenetic age, and thus will require systematic studies to evaluate their implications on forensic age estimation models. Furthermore, Jenkins et al. [
] have indicated the effect of environmental influences on sperm DNA methylation. Whereas the ET-5 protocol was re-optimized for routine applications, additional studies with the ET-13 may be useful to evaluate the impact of confounding factors on the DNA methylation levels of individual candidate CpG sites. This would enable further optimizations of the VISAGE model for semen. Of particular relevance for the VISAGE tools may be the effect of biogeographic ancestry on DNA methylation at age-associated CpG sites. By replicating an age estimation assay established for a Polish sample set [
] found slight differences in the predictive power of analyzed CpG sites. By retraining the model, they could improve the final age prediction outcome. Such differences in measured epigenetic ages of groups of different ancestry were reported by several studies [
]. An evaluation of the ET-13 with sample sets deriving from different biogeographic ancestries would allow for the development of fully optimized age estimation models for the respective populations. In combination with the VISAGE tools for appearance and ancestry [
], it would even become possible to choose the most accurate age model for an unknown sample donor.
Funding
The study received financial support from the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No. 740580 within the framework of the Visible Attributes through Genomics (VISAGE) Project and Consortium. MdlP is supported by a postdoctoral fellowship awarded by the Consellería de Cultura, Educación e Ordenación Universitaria and the Consellería de Economía, Emprego e Industria from Xunta de Galicia (Modalidade A, ED481B 2017/088).
Author contributions
AH developed assay design, performed laboratory work and data analysis. AP, AW, EP, WB, CX and HN contributed to assay development. MdP and CX contributed to laboratory work and data analysis. TS, FXL, JH, IB contributed to assay validation. NS, MU, MS, KJ, MVG, AU, JV performed laboratory work for assay validation. AH drafted the first version of the manuscript and WP and CX shaped the final version of the manuscript. WP conceived the study. All authors contributed to the article and approved the final version.
Data availability
The datasets generated for this study can be found in the European Nucleotide Archive (ENA) [https://www.ebi.ac.uk/ena/browser/home] under accession number PRJEB43859.
Acknowledgments
The authors would like to thank the Forensic Genetics and Casework section of the Institute of Legal Medicine in Innsbruck for DNA extraction of “mock” casework type samples and Leire Palencia Madrid (BIOMICs Research Group, Lascaray Research Center, University of the Basque Country, Vitoria-Gasteiz, Spain) for her laboratory support.
Bisulfite conversion of DNA: performance comparison of different kits and methylation quantitation of epigenetic biomarkers that have the potential to be used in non-invasive prenatal testing.
DNA methylation signature in blood does not predict calendar age in patients with chronic lymphocytic leukemia but may alert to the presence of disease.
00133-2&id=gr1.jpg){kind=link}
00133-2&id=gr2.jpg){kind=link}
00133-2&id=gr3.jpg){kind=link}
00133-2&id=gr4.jpg){kind=link}