Accurate age estimation from semen has the potential to greatly narrow the pool of unidentified suspects in sexual assault investigations. However, previous efforts utilizing semen age-related CpG (AR-CpG) markers have shown lower accuracy compared to blood AR-CpG-based methods. This discrepancy may be attributed to DNA methylation (DNAm) interferences from “round cells” such as leukocytes and immature sperm cells in semen. This study aimed to develop age calculators based on sperm-specific AR-CpG markers and to achieve performance-improved age estimates from sperm DNA. Through an analysis of publicly available MethylationEPIC microarray data from 90 sperm samples of healthy males aged 22–51 years, we identified 31 sperm-specific AR-CpG markers with absolute Pearson’s R values > 0.5 and Benjamini-Hochberg adjusted p values < 0.013. The top 19 AR-CpG markers with the largest absolute R values and beta ranges > 0.10, along with 3 reported semen AR-CpG markers (cg06304190, cg06979108, and cg12837463), were integrated into two methylation SNaPshot panels (Ⅰ and Ⅱ), each containing 11 markers. The 21 qualified AR-CpG markers showed absolute R values ≥ 0.427 in an independent validation cohort of 253 sperm DNA samples (22–67 years), with cg21843517 exhibiting the strongest age correlation (R = 0.853). The optimal models, constructed using sperm DNAm data of the training set (n = 214, 22–67 years) and markers from panel Ⅰ (n = 11), panel Ⅱ (n = 10), or both panels, achieved mean absolute errors (MAEs) of 2.526–4.746, 3.890–5.715, and > 9.800 years on the test sets of sperm (n = 39, 23–64 years), semen (same donors as the sperm test set), and whole blood (n = 40, 22–65 years), respectively. The simplified models incorporating 3, 5, 9, or 14 AR-CpG markers (MAE = 2.918–4.139 years for sperm) still outperformed the Lee et al. original model (MAE = 6.444 years for semen) and the reconstructed panel Lee model (MAE = 6.011 years for sperm). The final models, utilizing all sperm DNAm data (n = 253) and markers from panel Ⅰ, panel Ⅱ, or both panels, yielded mean MAEs of 2.587, 2.766, and 2.200 years, respectively, on the 50 test sets generated by 5 repeats of 10-fold cross-validations. Additionally, multiple markers in both panels demonstrated the ability to discern sperm or semen from blood with 100% accuracy. In summary, our study substantiates the potential of sperm-specific AR-CpG markers for precise age estimation from sperm DNA, providing an improved toolset for forensic investigations.

从精液中准确估计年龄有可能大大缩小性侵犯调查中身份不明嫌疑人的范围。然而，与基于血液 AR-CpG 的方法相比，之前利用精液年龄相关 CpG (AR-CpG) 标记的研究显示准确性较低。这种差异可能归因于精液中白细胞和未成熟精子细胞等“圆形细胞”的 DNA 甲基化 (DNAm) 干扰。本研究旨在开发基于精子特异性 AR-CpG 标记的年龄计算器，并根据精子 DNA 实现性能改进的年龄估计。通过对来自 22-51 岁健康男性的 90 个精子样本的公开 MmethylationEPIC 微阵列数据进行分析，我们确定了 31 个精子特异性 AR-CpG 标记，其绝对 Pearson R 值 > 0.5，Benjamini-Hochberg 调整后的 p 值 < 0.013。将绝对 R 值最大且 beta 范围 > 0.10 的前 19 个 AR-CpG 标记与 3 个报道的精液 AR-CpG 标记（cg06304190、cg06979108 和 cg12837463）一起整合到两个甲基化 SNaPshot 组（Ⅰ 和 Ⅱ）中，每个包含 11 个标记。在 253 个精子 DNA 样本（22-67 岁）的独立验证队列中，21 个合格的 AR-CpG 标记物显示绝对 R 值≥ 0.427，其中 cg21843517 表现出最强的年龄相关性（R = 0.853）。使用训练集（ n  = 214，22-67 岁）的精子 DNAm 数据和第一组（ n  = 11）、第二组（ n  = 10）或两个组的标记物构建的最佳模型在精子测试集（ n  = 39，23-64 岁）、精液（与精子测试集相同的供体）和全血（ n  =分别为 40 岁、22-65 岁）。包含 3、5、9 或 14 个 AR-CpG 标记的简化模型（精子的 MAE = 2.918–4.139 年）仍然优于 Lee 等人。原始模型（精液的 MAE = 6.444 年）和重建的面板 Lee 模型（精子的 MAE = 6.011 年）。最终模型利用所有精子 DNAm 数据 ( n  = 253) 和第一组、第二组或两个组中的标记，在 5 次重复生成的 50 个测试集上分别产生了 2.587、2.766 和 2.200 年的平均 MAE。 10 倍交叉验证。此外，两个面板中的多个标记都证明能够以 100% 的准确度从血液中辨别精子或精液。总之，我们的研究证实了精子特异性 AR-CpG 标记物在根据精子 DNA 精确估计年龄方面的潜力，为法医调查提供了改进的工具集。