Endocrine Abstracts

JOINT713

Background: The genetic basis of male fertility remains poorly understood. Previous studies, including those using ICD-10 codes and small-scale datasets, have failed to identify genetic loci associated with male fertility with consistent reproducibility. Testicular volume, a reliable proxy for quantitative spermatogenesis and male fertility, presents distinct challenges for population-level assessment. Traditional measurement methods are invasive, often introducing selection bias and restricting the scalability of studies. Despite its pivotal role in reproductive health, the genetic architecture of testicular volume has not been systematically explored at population level.

Methods: Using machine learning-based segmentation techniques, we analyzed bi-testicular volume in a total of 22,499 male UK Biobank participants with abdominal DIXON MRI data. A subset of 350 scans was manually segmented by a medical professional to train a fully convolutional network (U-Net model), generating 22,149 segmentations. Of these, 859 were excluded by an algorithm for incomplete segmentations, and 2,182 were excluded after manual review for not meeting quality standards. Additionally, 460 participants with male-specific ICD diagnoses that could potentially affect segmentation were excluded. By adding the manually segmented scans, this resulted in 18,998 participants remaining for population level analysis. The segmented volumes were subsequently used for a genome-wide association study (GWAS) of bi-testicular volume, conducted using linear regression in PLINK and adjusted for age and the first ten genetic principal components.

Results: The U-Net used for MRI segmentation achieved a median Dice score of 0.87, demonstrating human-level performance and ensuring accurate volume quantifications used in the analysis. The mean (SD) bi-testicular volume was 52 (16) mL. Our population-based GWAS identified 14 genome-wide significant loci (P <5×10⁻⁸), revealing key genetic determinants of testicular volume and male fertility. The strongest signal was observed at rs12271187 in the FSHB locus on chromosome 11 (P = 1.33×10⁻⁴¹), alongside a significant association at the FSHR locus on chromosome 2, highlighting the central role of the follicle-stimulating hormone, including its ligand and receptor, in male reproductive biology. Functional annotation revealed that many significant SNPs are enriched in regulatory regions, such as UTRs and intronic sequences, suggesting their involvement in transcriptional regulation.

Conclusion: Our study demonstrates the potential of machine learning for population-level analysis of testicular volume, addressing key challenges in male reproductive health research. It provides a foundation for future exploration of the genetic determinants of male fertility.