Results and discussion

Specificity tests

Sixty-seven body fluid samples were amplified using multiplex M and U primer sets. The selection of ASIC4, FAM43A, and FOXO3 was guided by previous published studies [27, 28], which reported body fluid specific methylation markers for these loci, thereby reducing the potential for false classification in forensic applications. The representative samples from each body fluid are shown in Fig. 1. The M and U patterns of bisulfite-converted DNA from semen, saliva, and blood demonstrate successful amplification of the target markers. In semen, amplification with M primers yielded over 30,000 relative fluorescence units (RFUs) for the ASIC4 marker (Fig. 1a). In contrast, amplification with U primers (Fig. 1d) in the semen samples produced the lowest RFU value for ASIC4 (average = 3,119.6), whereas the average RFU values for FAM43A and FOXO3 were 21,311.8 and 14,842.7, respectively. Saliva and blood samples amplified with M primers also exhibited the highest RFU values for each specific marker (over 15,000 RFU), but not as much as those for semen (Fig. 1b, c).

Fig. 1: Electropherogram results of single body fluid samples amplified with tri-plex primer sets. Panels a and d show semen, panels b and e show saliva, and panels c and f show blood. M and U indicate methylated and unmethylated alleles, respectively

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. As shown in Fig. 2, semen samples exhibited significantly higher methylation levels at the ASIC4 marker compared to the other markers. Saliva and blood samples also displayed elevated methylation levels across their respective markers. However, all samples showed low methylation signals at the non-target markers (below 0.26), which may indicate cross-reactivity. Although we selected three markers with high average methylation percentage (β-scores) based on the study by Park et al. [28], these markers did not show 0% methylation in other body fluids. For the ASIC4 gene, the β-score for blood and saliva was 0.10 ± 0.08 and 0.06 ± 0.05, respectively. In FAM43A, β-scores for semen and blood were 0.01 ± 0.01 and 0.04 ± 0.04, respectively. For the FOXO3 gene, semen and saliva β-scores were 0.12 ± 0.04 and 0.17 ± 0.10, respectively. This indicates that methylation does not exhibit a strictly binary on/off pattern [28, 29]. The β-scores using the single-source samples in this study were calculated and presented in the Table 3. To reduce the risk of misinterpretation about false positives, the threshold values were determined as the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overline{x}}{(neg)}+{{k}{m}\cdot SD}_{(neg)}$$\end{document}.

Fig. 2: Box plots of methylation levels in semen, saliva, and blood samples. The methylation level for each marker was calculated as M/(M + U). The average methylation level of each box plot is indicated by an “x”

Marker name | Mean β-scores ± SD | Threshold: Mean β-scores + \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}
Semen(n = 10) | Saliva(n = 30) | Blood(n = 27)
ASIC4 | 0.88 ± 0.04 | 0.06 ± 0.05 | 0.10 ± 0.08 | 0.431 (k = 5)
FAM43A | 0.01 ± 0.01 | 0.61 ± 0.19 | 0.04 ± 0.04 | 0.237 (k = 5)
FOXO3 | 0.12 ± 0.04 | 0.17 ± 0.10 | 0.46 ± 0.10 | 0.341 (k = 2)

Analytical specificity for the developed MSP multiplex assay

To evaluate the specificity of the developed MSP multiplex assay, cross-reactivity was assessed using nine non-target body fluids and genomic DNA from four non-human species. Among the non-target body fluids tested, vaginal fluid (n = 6), urine (n = 1), and nasal secretion (n = 2) all yielded beta score values below the established cut-off threshold, indicating no cross-reactivity with the target methylation markers (Fig. 3). These results confirm that the assay does not produce false-positive signals in the presence of these matrices. Sweat and skin/contact DNA samples were also subjected to the identical experimental procedure; however, no amplification was obtained following bisulfite conversion. This outcome is attributable to the inherently low DNA concentration characteristic of these sample types, which is a well-recognized limitation in forensic casework. As bisulfite treatment induces additional DNA degradation, the already limited template in sweat and skin/contact DNA was insufficient for downstream amplification. Nevertheless, this result is forensically relevant, as it indicates that these matrices would not generate false-positive interpretations when encountered as trace evidence at a crime scene.

Fig. 3: Methylation levels of each marker (ASIC4, FAM43A, and FOXO3) for six vaginal fluid, two nasal secretion, and one urine samples. Each symbol represents the following non-target body fluids: (●) Nasal secretion, (▲) Urine, and (✕) Vaginal fluid. The dashed line represents the established cut-off threshold

For the assessment of species specificity, genomic DNA isolated from dog (Canis lupus familiaris), cow (Bos taurus), mouse (Mus musculus), and cat (Felis catus) was analyzed using the MSP multiplex assay. No amplification products were detected in any of the non-human DNA samples, demonstrating that the developed primers are strictly human-specific (Online Resource 2). This finding is particularly important in forensic investigations where biological evidence may be contaminated with or co-deposited alongside non-human biological material. Along with this, we also tested control DNA samples to confirm the absence of non-specific peaks for each marker; the data are shown in Online Resource 3. These results verify that our tri-plex primer sets correctly targeted semen, saliva, and blood samples, confirming their specificity. Collectively, these results confirm the high analytical specificity of the proposed MSP multiplex system, supporting its applicability in complex forensic scenarios involving mixed or potentially contaminated biological evidence.

Sensitivity tests

Serial dilution experiments were conducted in each triplicate using two distinct approaches. First, DNA samples were diluted before bisulfite-conversion and subjected to bisulfite-conversion to assess the minimum DNA amount required for biological evidence in crime scene samples (Fig. 4). This approach is hereafter referred to as the Dilution Before Bisulfite-conversion (DBB) method. In the DBB method, the total DNA input was adjusted from 20 ng down to 1.25 ng. In the other approach, bisulfite-conversion was performed using 20 ng of DNA, and then the bisulfite-converted DNA was serially diluted two-fold to 16-fold. The diluted, bisulfite-converted DNA was subsequently used for PCR amplification (Fig. 5). This approach is hereafter referred to as the Dilution After Bisulfite-conversion (DAB) method. In the DBB method, almost of samples were fully detected using as little as 1.25 ng of DNA for conversion. The DAB method demonstrated that all body fluids could be detected even with a small amount of bisulfite-converted DNA (Fig. 5). The observed sensitivity is comparable to that reported in previous SNaPshot-based methylation studies [12, 27, 30], supporting the reliability of our method in producing results consistent with established standards.

Fig. 4: Methylation levels of each marker (ASIC4, FAM43A, and FOXO3) for serially diluted DNA samples from (a) semen, (b) saliva, and (c) blood using the DBB (Dilution Before Bisulfite-conversion) method. DNA input amounts ranged from 20 ng to 1.25 ng. The gray horizontal lines represent the threshold values for each marker

Fig. 5: Methylation levels of each marker (ASIC4, FAM43A, and FOXO3) for serially diluted bisulfite-converted DNA from (a) semen, (b) saliva, and (c) blood using the DAB (Dilution After Bisulfite-conversion) method. Bisulfite-converted DNA prepared from 20 ng of input DNA was serially diluted two-fold up to 16-fold. Gray horizontal lines indicate the threshold values for each marker

LOOCV validation of body fluid markers

Using LOOCV with training-only one-sided cutoffs (cutoff = \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overline{x}}{(neg)}+{{k}{m}\cdot SD}_{(neg)}$$\end{document}), all three methylation markers showed strong discrimination for their target body fluids (Online resource 4). ASIC4, the semen-targeting marker, achieved perfect classification in this dataset with sensitivity 1.00 and specificity 1.00. FAM43A, the saliva-targeting marker, achieved sensitivity 0.967 with one false negative and specificity 1.000. FOXO3, the blood-targeting marker, achieved sensitivity 0.926 and specificity 0.950 with two false-positive non-target samples close to the cutoff. While the ROC curves for both ASIC4 and FAM43A markers approached the upper-left corner and AUC values were 1.00, AUC value of FOXO3 marker was shown by 0.973 (Online Resource 5), consistent with strong body-fluid specificity. Given the small sample size and near-complete separation, we emphasize binomial confidence intervals for sensitivity and specificity as primary uncertainty estimates.

Mixed sample tests

We also prepared mixed samples by combining 20 ng of each DNA from two or three body fluids and used the mixtures in the bisulfite-conversion procedure. All the mixture tests were conducted in triplicate. The methylation levels in all mixtures showed good specificity, but the sensitivity differed slightly according to sample type (Fig. 6). As shown in the mixtures of saliva and blood, the methylation level for the FAM43A marker (0.539) was slightly different from that for the FOXO3 marker (0.429). Also, in the mixtures of three body fluids, the methylation levels for the three markers were not displayed equally. Though we used the same DNA quantity, differences in the methylation status of each gene marker could affect this pattern. In this study, ASIC4 in semen showed the highest average β-score (0.88), whereas FAM43A and FOXO3 in saliva and blood exhibited lower scores (0.61 and 0.46). In addition, the reduced detection rate observed for longer target sequences commonly suggests that amplicon size significantly could influence amplification efficiency [31, 32]. The intermediate methylation levels observed in mixture samples are an expected consequence of combining body fluids with distinct methylation profiles and do not indicate assay instability [33]. Furthermore, differential amplification efficiency may arise from unequal binding of primers to methylated versus unmethylated templates and the inclusion of CpG sites, leading to deviation from expected mean methylation ratios in mixture samples [34, 35]. As shown in Table 4, all mixed samples were correctly classified as mixtures, and none were misclassified as single-source samples. Although a subset of single-source samples was conservatively classified as mixtures, particularly in samples exhibiting borderline marker positivity. This method sacrifices single-source sensitivity to ensure zero false single-source calls from mixtures.

Fig. 6: Methylation levels of each marker (ASIC4, FAM43A, and FOXO3) for mixed body fluid samples. Blue dots indicate mixtures of semen and saliva, orange squares indicate mixtures of saliva and blood, and green rhombuses indicate mixtures of semen and blood. The mixture of all three body fluids (semen, saliva, and blood) is shown in purple. Gray horizontal lines indicate the threshold values for each marker

True status | Called as single | Called as mixture | Inconclusive
Single-source | 65 | 2 | 0
Mixture | 0 | 4 | 0

Mixture inference under the predefined rule-based composite decision rule

Applying the predefined rule-based composite decision rule with an inconclusive zone to the mixture dataset (three technical replicates per mixture type) enabled explicit mixture calling and error quantification (Online Resource 6). Overall, 8/12 mixture replicates were correctly classified (66.7%), 2/12 were reported as inconclusive (16.7%), and 2/12 were misclassified (16.7%). Misclassifications were observed in semen + blood mixtures, where the saliva marker occasionally exceeded its threshold, leading to over-calling a three-fluid mixture.

Composite-score model for combined single-source and mixture classification

The LDA composite-score model provided a multivariate alternative that formally integrates all three markers (Online Resource 7). Under LOOCV across all samples (single-source + mixture), the model achieved 76/79 correct classifications (96.2%) with one inconclusive calls at the prespecified posterior threshold. The single misclassification involved two saliva samples classified as a two-fluid mixture (Blood + Saliva), reflecting a trade-off between sensitivity to low-level components and conservative multivariate class boundaries.

Performance of validation test from evidentiary samples

To assess the robustness of our MSP primer set, we evaluated its performance using DNA extracted from evidentiary samples collected from various crime scenes. A total of nine samples were analyzed using the same protocol as in the single-source test. It should be noted that the evidentiary samples used in this study had already undergone preliminary testing using the chemical or immunological methods during routine forensic examination, in which only one presumptive test was applied per sample based on the anticipated body fluid type. The results of the preliminary tests, MSP result interpretation, and the information about the nine evidentiary samples such as sample description and DNA concentration were shown in the Table 5.

Sample Namea | Sample Description | DNA Concentration(ng/μL) | Preliminary Test | MSP Result Interpretationa
SE1 | Victim's vaginal fluid | 101.3 | SM positive | SE only
SE2 | Stain on a bed | 8.4 | SM positive | SE only
SE3 | Panties | 1.6 | SM positive | SE only
SA1 | Swab collected from a drinking surface | 1.6 | SALIgAE® positive | SA only
SA2 | cigarette butt | 2.4 | SALIgAE® positive | SA only
SA3 | cigarette butt | 5.7 | SALIgAE® positive | SA only
BL1 | Swab collected from blood stains on a tire | 5 | LMG positive | BL only
BL2 | Swab collected from blood stains on the floor | 1.1 | LMG positive | BL only
BL3 | Swab collected from kitchen knife | 35.4 | LMG positive | BL only

The semen samples (SE1–SE3) demonstrated the most consistent performance, with successful detection. In the Fig. 7a, the methylation levels at the ASIC4 marker were greater than 0.89 for all semen samples. In evidentiary samples presumed to originate from semen, differential lysis is typically performed to enrich sperm cell fractions by removing female epithelial cells. This process may increase the relative proportion of sperm-derived DNA, which can enhance the detectability of semen-associated methylation markers such as ASIC4.

Fig. 7: Methylation levels at each marker (ASIC4, FAM43A, and FOXO3) from the nine evidentiary samples. Panel a shows three semen samples. Panel b and c show three saliva and blood samples, respectively. Gray horizontal lines indicate the threshold values for each marker

The methylation levels detected in all saliva samples were above the predefined threshold, suggesting that the DNA most likely originated from saliva (Fig. 7b). In the Fig. 7c, blood markers exhibited methylation levels above 0.5 in the blood samples (BL1 and BL2), which were collected from visible bloodstains on surfaces using swabs. Sample BL3 obtained from the kitchen knife has a low methylation level as 0.44. Methylation-based interpretations should be considered within a broader forensic and biological context rather than as absolute indicators.