Morphological Remodeling Processes During Spermiogenesis in Three Closely-Related Mouse Species Revealed byMachine-Learning Analyses of Cell, Nuclei and Manchette Shape and Size

Introduction & Objective: Spermiogenesis is a complex morphophysiological process involving nuclear, cytoskeletal and cellular remodeling that transforms round spermatids into spermatozoa. Despite its central role this process has traditionally been described using qualitative criteria, limiting reproducibility and comparisons across species. There is a need for advanced, objective techniques to identify and quantify stages in isolated mixed germ cell populations. Methods: A quantitative and unsupervised machine-learning framework for the analysis of spermiogenesis was used to compare three mouse species (Mus musculus, M. spretus, M. spicilegus). Germ cells (n = 5 males of each species) were labeled with DAPI (nucleus) and anti-acetylated tubulin (manchette). Each cell was imaged using fluorescence microscopy (nucleus, manchette) and phase contrast (whole cell). Shape and size were analyzed for each structure using NMA software v1.20.0.Morphometrics of entire cell, nucleus and manchette of spermiogenic steps were combined with Principal Component Analysis (PCA), k-means clustering and Linear Discriminant Analysis (LDA) to objectively characterize progression within and between species. Results: Changes followed a continuous and conserved trajectory in all species. Yet, interspecific morphological divergence was not uniformly distributed along the process. Informative interspecific differences emerged during spermiogenic steps 11-12 in the three structures analyzed (entire cell, nucleus, manchette) both in overall angle profiles and specific segments 0 and 1, and in morphometry, particularly nuclear minimum diameter, entire cellmaximum Feret and bounding height, and manchette’s area and minimum diameter. Inclusion of manchette descriptors enhanced the discriminative capacity of models that included only entire cell and nuclear parameters, improving separation of steps and reducing overlap of the three species. k-means clustering after PCA achieved 91.7% total similarity score. The species best grouped within the same, distinct clusters were M. spicilegus and M. spretus (96.3% and 95.8%, respectively), followed by M. musculus(81.0%). These differences allowed for a robust discrimination between species, even with a limited number of morphometric variables. Conclusion: A quantitative and unsupervisedmachine-learning framework for the analysis of spermiogenesis was used to compare three mouse species (Mus musculus, M. spretus, M. spicilegus). Germ cells (n = 5 males of each species) were labeled with DAPI (nucleus) and anti-acetylated tubulin (manchette). Each cell was imaged using fluorescence microscopy (nucleus, manchette) and phase contrast (whole
cell). Shape and size were analyzed for each structure using NMA software v1.20.0.Morphometrics of entire cell, nucleus and manchette of spermiogenic steps were combined with Principal Component Analysis (PCA), k-means clustering and Linear Discriminant Analysis (LDA) to objectively characterize progression within and between species.