Materials and methods

Marker selection and multiplex primer design

Target CpG sites with a high β-score, indicating a high methylation percentage, in the Illumina Human Methylation 450 K bead array, based on a previous study, were selected for the three body fluids (Table 1). The β-scores for the CpG sites targeting semen (cg17610929), blood (cg08792630), and saliva (cg09652652) were 0.92 ± 0.06, 0.50 ± 0.06, and 0.49 ± 0.17, respectively [27, 28]. Primers were designed to target CpG sites showing clear differential methylation between target and non-target body fluids, to be compatible with bisulfite-converted DNA, and to produce amplicons of distinct sizes suitable for multiplex capillary electrophoresis. Two sets of primers (M and U primers) were developed for the two separate amplifications. The 3' end of reverse primers was designed to be complementary to the CpG sites of each marker. The amplicons derived from semen, saliva, and blood using the M primers were the 71 bp, 95 bp, and 145 bp in length, respectively. Only forward primers were labeled with FAM dye at 5′ end, and the PCR products were distinguished based on their amplicon sizes during capillary electrophoresis analysis (Table 2). Degenerate bases were introduced at non-target CpG positions to accommodate sequence ambiguity resulting from bisulfite-conversion, where the methylation status of flanking CpGs is unknown. This approach was used to prevent preferential amplification of methylated or unmethylated alleles, while the target CpG site for discrimination was kept non-degenerate.

Target body fluid | Gene symbol | Target ID | CpG sites (hg 19) | Mean β-scores ± SD | References
Semen | Saliva | Blood | Vaginal fluid
Semen | ASIC4 | cg17610929 | Chr2:220,379,044 | 0.95 ± 0.06 | 0.01 ± 0.01 | 0.01 ± 0.01 | 0.01 ± 0.01 | [27]
Saliva | FAM43A | cg09652652 | Chr3:194,408,845 | 0.02 ± 0.01 | 0.49 ± 0.17 | 0.02 ± 0.00 | 0.03 ± 0.00 | [27]
Blood | FOXO3 | cg08792630 | Chr6:108,883,909 | 0.08 ± 0.02 | 0.09 ± 0.01 | 0.50 ± 0.06 | 0.10 ± 0.02 | [28]

Target body fluid | Primer sequence (5′–3′)a, b | Conc. (μM) | Amplicon size(bp)
Semen | M | F: GTTTTTGAYGTTYGTGTTGTYG | 0.3 | 71
R: GAAACCCTCCCCACG
U | F: GTTTGGTTTTTGATGTTTGTGTTGT | 0.15 | 76
R: AAAACCCTCCCCACA
Saliva | M | F: TTTYGTTAGGTYGAYGGY | 1.2 | 95
R: CCACGAATAAATAACCACGATAAAACG
U | F: ATTTGTTTTGTTAGGTTGATGGTGT | 0.09 | 100
R: CCACAAATAAATAACCACAATAAAACA
Blood | M | F: GGGTTTTGGTGTGGG | 0.24 | 145
R: CAAAAAAATAATAAAAAACGATAAAAAATCTCTCTTCG
U | F: GGGATGTGGGGTTTG | 0.3 | 131
R: CAAAAAAATAATAAAAAACAATAAAAAATCTCTCTTCA

Sample collection

Sixty-seven body fluid samples (ten semen, thirty saliva, twenty-seven blood) were collected from healthy Korean 15 male and 15 female volunteers using analytical procedures approved by the Institutional Review Board of the National Forensic Service. All volunteers were informed about the goal and procedure of this study. Volunteers were asked to collect saliva into a 50-mL conical tube, and male volunteers were also asked fresh semen into a collection cup. Whole blood samples were collected using a syringe and used without the addition of any anticoagulants or preservatives (e.g., EDTA or citrate). All collected samples were aliquoted into microcentrifuge tubes of 200 μL each and stored at—20 °C before the following experiment. For the collection of vaginal fluid, cotton swabs were used, and it stored at room temperature in dry conditions. Amplified human genomic DNA (completely unmethylated), as well as completely methylated and unmethylated human control DNA (bisulfite-converted) from the EpiTect® PCR Control DNA kit (Qiagen, Hilden, Germany), were also used in this study. Presumptive testing was not repeated for the evidentiary samples that had been previously examined using the SM, LMG, and SALIgAE® tests, because these specimens were supplied as pre-screened forensic materials and additional chemical testing required more sample volume.

DNA extraction and bisulfite-conversion

DNA was isolated from each sample using a QIAamp DNA Micro Kit (Qiagen). Quantification was conducted using a Quantifiler™ Trio DNA Quantification Kit (Thermo Fisher Scientific, Waltham, MA, USA) and a 7500 Real-Time PCR Instrument System (Thermo Fisher Scientific) according to the manufacturer's instructions. Bisulfite-conversion was performed using an EZ DNA Methylation-Lightning™ Kit (Zymo Research, Irvine, CA, USA) according to the manufacturer's instructions. To ensure consistency across samples, 20 ng of input DNA was used, with an elution volume of 10 μL. The bisulfite-converted DNA was stored at –80 °C until further use. In the mixture tests, all samples were used in the same quantity (20 ng) to prepare two- and three-sample mixtures. In addition, the input DNA for bisulfite-conversion was diluted two-fold for serial dilution tests from 20 to 1.25 ng.

Methylation-specific PCR

Multiplex PCR amplification was conducted using an EpiScope® MSP kit (TAKARA Bio, Shiga, Japan) and a GeneAmp PCR System 9700 Thermal cycler (Applied Biosystems, Foster City, CA, USA) under the following conditions: initial activation at 95 °C for 30 s; 37 cycles of 98 °C for 5 s, 58 °C for 40 s, 60 °C for 40 s, and 72 °C for 1 min; and final extension at 72 °C for 20 min. Amplification was conducted using a 50-μL reaction volume comprising 5 μL of bisulfite-converted DNA, 25 μL of 2 × MSP Buffer, 0.5 μL of 100 × TB Green Solution, 1.2 μL of MSP enzyme, 18.3 μL of primers and distilled water. The optimized concentration for each primer is shown in Table 2. The PCR products were stored at 4 °C until the next procedure.

CE and data analysis

The analysis included a 10-μL mixture of Hi-Di™ Formamide and GeneScan™ 600 LIZ™ dye Size Standard v2.0 (Applied Biosystems), along with 1 μL of PCR products. The mixtures were denatured at 95 °C for 3 min and cooled on a PCR cooler. Subsequently, the products were analyzed on a 3500 Genetic Analyzer (Applied Biosystems) using a 36-cm capillary array and POP-4™ Polymer (Applied Biosystems). The injection was conducted under the following conditions: 1.2 kV for 24 s, and electrophoresis was performed for 1,550 s at 13 kV and 60 °C. The data were analyzed using GeneMapper™ ID-X Software (Applied Biosystems).

Set threshold of methylation level

The methylation level, defined as M/(M + U), was used to evaluate the methylation status, where M and U represent methylated and unmethylated RFUs, respectively. For each marker, a one-sided upper cutoff was defined from the distribution of negative (non-target) samples:

\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{cutoff }={\overline{x}}_{(neg)}+{{k}_{m}\cdot SD}_{(neg)}$$\end{document}

Here, x denotes the methylation level M/(M + U); \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overline{x}}{(neg)}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{k}{m}\cdot SD}_{(neg)}$$\end{document} are the mean and standard deviation of x calculated from non-target (negative) samples. k__m is a marker-specific constant that controls the stringency of the one-sided cutoff for each target marker. A sample was classified as positive for the target body fluid when its methylation level was ≥ cutoff. To determine an appropriate constant k__m (m = [ASIC4, FAM43A, FOXO3]) for each target marker, we evaluated _k_m = 1 to 5 using the negative (non-target) β-score distribution for each marker. We summarized the false-positive/false-negative trade-off across k__m values (Online Resource 1) and selected marker-specific k__m values to balance specificity and sensitivity. Based on this trade-off assessment, we used k = 5 for ASIC4 and FAM43A to eliminate non-target false positives, and k = 2 for FOXO3 to avoid excessive loss of blood sensitivity.

Validation and performance metrics

To avoid optimism bias, cutoff estimation and performance evaluation were separated using leave-one-out cross-validation (LOOCV). In each LOOCV iteration, one single-source sample was held out as an independent test sample. The cutoff was calculated using only negative samples within the training set and then applied unchanged to the held-out sample. Predictions were aggregated over all iterations. Performance was evaluated in a one-vs-rest manner: ASIC4 (semen vs saliva + blood), FAM43A (saliva vs semen + blood), and FOXO3 (blood vs semen + saliva). Sensitivity, specificity, and accuracy were calculated from the aggregated confusion matrix and reported with 95% confidence intervals using Wilson binomial intervals. ROC curves and AUC were computed from continuous methylation levels.

Composite decision rules for single-source and mixture calling

Composite decision rule for single-source and mixture inference (rule-based) with an inconclusive zone

For each marker, a one-sided upper cutoff was defined from the distribution of negative (non-target) samples:

\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{c}={\overline{x}}{(neg)}+{{k}{m}\cdot SD}_{(neg)}$$\end{document}

To reduce borderline calls, a marker-specific inconclusive (gray) zone was defined as [c-Δ, c + Δ], where Δ = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${SD}_{(neg)}$$\end{document}. Marker status was assigned as Positive if x \ge c + Δ, Negative if x łe c-Δ, and Inconclusive otherwise. Specimen-level interpretation used the following composite rule:

1. Single-source call: exactly one marker is Positive and the other two are Negative.

2. Mixture call: two or more markers are Positive (mixture composition reported as the set of Positive markers).

3. Inconclusive: any Inconclusive marker status, or no Positive markers.

Mixture performance was evaluated using replicate mixture experiments (three technical replicates per mixture type), and misclassification and inconclusive rates were summarized.

Composite-score classifier (LDA) with a probabilistic inconclusive rule

As an alternative composite rule, a linear discriminant analysis (LDA) classifier was trained using [x(ASIC4), x(FAM43A), x(FOXO3)] to classify seven classes (semen, saliva, blood, semen + saliva, semen + blood, saliva + blood, semen + saliva + blood). The predicted class was assigned as the class with the highest posterior probability, and samples were reported as Inconclusive if the maximum posterior probability was < 0.8. Model performance was evaluated using leave-one-out cross-validation (LOOCV) across all samples (single-source + mixture).