A high-throughput method for unbiased quantitation and categorisation of nuclear morphology

The physical arrangement of chromatin in the nucleus is cell type and species specific. This is particularly evident in sperm, in which most of the cytoplasm has been lost; the shape of the nucleus reflects the shape of the cell. Mice have distinctive falciform (‘hook shaped’) sperm heads and nuclei. Quantification of the differences in shape variation between mouse species and lines often relies on manual measurement and classification that leads to subjective results, making comparisons within and between samples difficult. We have developed an analysis program for assessing the morphology of asymmetric nuclei, and characterised the sperm of mice from a range of inbred, outbred and wild-derived mouse lines. We find that laboratory lines have elevated sperm shape variability both within and between samples in comparison to wild-derived inbred lines, and that sperm shape in the F1 offspring of CBA and C57Bl6J lines is subtly affected by the direction of the cross. Hierarchical clustering can distinguish distinct sperm shapes with greater efficiency and reproducibility than even experienced manual assessors. We quantified the range of morphological defects in the inbred BALB/c line, demonstrating we can identify different morphological subgroups. This approach has applications for studies of sperm development, infertility and toxicology.

Nuclei are complex, dynamic structures within a cell. For many cell types, the nucleus is 3 generally spherical, but for other cell types the nucleus adopts a distinctive shape [1]. One of 4 the most profound changes to nuclear shape occurs during spermatogenesis: mammalian 5 sperm tend to have a spatulate, or 'paddle' shape, meaning the nucleus both condenses and 6 reshapes. The chromatin becomes wound ~4-6 times more tightly than in metaphase, 7 mediated via replacement of histones with smaller protamines [2], and various cytoskeletal 8 elements coordinating to shape the nucleus [3]. 9 10 In rodents, this process is even more elaborate: most rodents, including mice, have a 11 falciform 'hook-shaped' sperm, with varying degrees of hook length and body shape 12 between species (e.g. [4]). The mouse sperm head shape develops through a series of 13 interacting mechanical forces, reshaping the nucleus via the cytoskeleton and 14 nucleoskeleton. The sperm head is divided into developmental 'modules', each of which is 15 shaped by particular cytoskeletal components [5]. When these processes go awry, distinct 16 morphological abnormalities can result (e.g. [6]), linking phenotype with the underlying 17 genetic alterations. The reshaping of the nucleus is itself a distinct process from the 18 chromatin condensation and repackaging [3]. Reshaping precedes transition and protamine 19 replacement, and chromatin condensation then follows. 20 21 Analytical methods for categorising and quantifying sperm head shape variation have 22 developed markedly over the years, and the advent of computational processing of images 23 has dramatically increased the quality of data we can capture, and the sophistication of the 24 analyses. To date, morphometric approaches in sperm have fallen into three main groups; 25 the measurement of basic parameters such as lengths, widths, and areas of objects, the use 26 of elliptic fourier analysis to investigate differences in the two dimensional outline of the 1 object, and the use of Procrustes analyses to examine differences in fixed landmarks within 2 the sperm head. Each of these approaches has advantages and disadvantages. 3 4 Basic measures such as area and length were the first statistics recorded describing sperm 5 morphology (e.g. [7][8][9]. These still remain useful, especially in situations such as CASA 6 analysis for fertility screening, in which an assessment of semen quality must be made 7 rapidly across many different cells [10]. However, the parameters measured by these 8 analyses are dominated by the size of the object, not the shape, and can make it difficult to 9 consistently assess the number of normal sperm across populations [11]. 10 11 In contrast, elliptic Fourier descriptors [12] allow an arbitrary closed two dimensional shape 12 to be decomposed into harmonic amplitudes describing the curvature of the object 13 perimeter, thus allowing subtle variations in shape to be discovered [13]. This approach has 14 proved powerful for demonstrating differences between species, lines within a species, and 15 different treatments (e.g. [14][15][16]). However, the approach has the drawback that the shape 16 parameters and underlying mathematics are difficult for biologists to understand and relate 17 back to the biological structure that is affected [17]. Moreover, since Fourier analyses rely on 18 smooth harmonic deformations of an underlying elliptical outline, sharp points -such as 19 found at the tip of a mouse sperm -tend to be poorly fitted [18]. 20 21 The third major method, Procrustes-based geometric morphometric analysis, uses 22 landmarks and semilandmarks within the object to align individual samples to consistent 23 size, position and orientation (e.g. [4]). Principal component analysis (PCA) can then be 24 used to identify the major varying landmarks distinguishing samples [5]. This approach has 25 the advantage of tightly relating the variation to physical structures within the object: 26 however, since objects are aligned by a least-squares method rotating about the centroid, 27 objects are susceptible to smearing of landmarks in highly variable regions, and can require 1 time-consuming manual placement of landmarks. 2 3 In terms of the biological field of application, sperm shape analysis has proven useful in 4 three main interrelated areas: infertility, speciation, and toxicology. In infertility, while 5 abnormal sperm morphology is extremely common in infertile knockout lines, the role played 6 by specific types and extents of shape defect remains to be elucidated, as does the extent to 7 which teratozoospermia can be used as an indicator of other sperm defects (e.g. DNA 8 damage or defective motility [19]). Deregulation of reproductive processes is a major 9 contributor to speciation through the induction of hybrid male sterility [20]. In particular, 10 sperm shape abnormalities are a feature of house mouse hybrid sterility, with a range of 11 mapped quantitative trait loci known, particularly on the sex chromosomes but also on 12 autosomes [21][22][23][24]. 13 14 Sperm shape is used as an assessment of genotoxicity and/or reproductive toxicity of 15 compounds (e.g. [25,26]. These studies often carry out a manual classification of sperm into 16 various categories of morphological abnormality, based on previously described sperm 17 shapes. The manual element thus makes this application both time consuming, and prone to 18 operator bias. A further problem is that the classes of abnormality described are often 19 arbitrarily chosen, and vary between studies. Use of a scoring chart, based on the 20 morphological abnormalities typical for one experimental system, may therefore compromise 21 the ability to quantitate abnormalities in a different system. It would be far more useful to 22 have an automated and reproducible method that is able to discover categories of 23 morphological abnormality within a sperm population, without prior training. 24 25 To address these needs for unbiased measurement, analysis and categorisation of nuclear 26 morphologies, we have developed a new image analysis programme that generates 27 quantitative information on the underlying regions of the nucleus that differ within and 1 between samples, independent of nuclear size. 2 3 We have validated the software on different mouse lines, and can quickly analyse hundreds 4 of images. Here, we demonstrate the use of this software to compare a range of different 5 inbred, outbred and wild-derived lines (revealing the effects of inbreeding depression and 6 potentially hybrid dysgenesis), to unravel the morphological variation in a single sample 7 (revealing different classes of abnormality in an inbred line), and to trace genetic influences 8 on sperm morphology in a reciprocal F1 cross between CB57Bl6 and CBA lines (revealing 9 contrasting effects of the parental genomes on sperm size and shape). Committee (protocol 002-13) and were subject to local ethical review. Animals were sourced 5 as indicated in Table 1

Sperm collection and fixation 3
The vasa deferentia and caudae epididymes were dissected from each animal, and the 4 contents squeezed out into 1ml PBS (scaled up accordingly if multiple animals were pooled). 5 The sperm were transferred to a microfuge tube, and tissue clumps were allowed to settle. 6 Sperm were transferred to a new tube and pelleted at 500g for 5mins. The supernatant was 7 removed, and the sperm fixed dropwise with either 3:1 methanol-acetic acid or 2% 8 paraformaldehyde (PFA) in PBS. Sperm were again pelleted at 500g for 5mins, and washed 9 in fixative twice more. Samples were stored at -20°C (methanol-acetic acid) or 4°C (PFA). 10 11 Imaging 12 Samples were diluted in fixative as required to obtain an evenly-spread preparation, and 8μl 13 of sample dropped onto a slide and allowed to air dry. Slides were counterstained with 16μl 14 VectorShield with DAPI (Vector Labs) under a 22x50mm cover slip and imaged at 100x on 15 an Olympus BX-61 epifluorescence microscope equipped with a Hamamatsu Orca-ER 16 C4742-80 cooled CCD camera and appropriate filters. Images were captured using Smart-1 Capture 3 (Digital Scientific UK). To validate the reproducibility of the software, sample 2 images were also gathered on three other microscopes: (1) an Olympus BX61 with a 3 Hamamatsu C10600 orca r² camera, (2) an Olympus BX61 with a Hamamatsu Orca-03G 4 camera, and (3) a Nikon Microphot-SA epifluorescence microscope with a Photometrics 5 Metachrome II CH250 cooled CCD camera.

7
Nucleus detection and morphological analysis 8 Image analysis was performed using a custom program designed as a plugin for the freely 9 available image analysis program ImageJ [27]. The plugin, Nuclear Morphology Analysis. 10 The core software was developed using Java 8, with the user interface written using Swing. 11 The software is available at http://bitbucket.org/bmskinner/nuclear_morphology/wiki/Home/ 12 together with full installation instructions, an online wiki user manual, and example images 13 for testing. The analyses described here were conducted using software version 1.13.6. The profile for the entire perimeter is shown in (C). 4

Statistical analysis and clustering 5
Following segmentation, standard nuclear parameters were measured: area, perimeter and 6 aspect ratio, the width of the nuclear body versus the length of the hook as described in 7 other papers (e.g. [9], and the lengths of each perimeter segment. Data was exported for 8 further processing in R. Differences between datasets were tested using a pairwise Wilcoxon 9 rank sum test, with Bonferroni multiple testing correction. In order to quantify the variability of 10 the nuclear shapes, we developed a new per-nucleus measure defined as the root-mean-11 square difference between the per-nucleus angle profile and the median angle profile for the 12 dataset, averaged across the length of the angle profile. The coefficient of variability 13 (standard deviation / mean) was also calculated for each of the other measured parameters. 14 15 The 'average shape' of the nuclei was calculated by averaging the x and y coordinates at 16 consistent semilandmarks taken as fractions of the perimeter across all nuclei, vertically 17 aligned and with their centres of mass at (0,0). This yielded a 'consensus nucleus' 18 visualising the overall shape of the population. Clustering was implemented via the WEKA 19 data mining software library [29]. 20

1
Detection and quantification of sperm shape in C57Bl6 and CBA mice 2 The difference between CBA and C57Bl6 sperm is distinguishable to the trained eye, and 3 makes a useful demonstration of the software's features. The angle profiles generated are 4 distinct for each genotype (Figure 2A). CBA sperm have a larger cross-sectional area, are 5 longer, and also have slightly shorter hooks than C57Bl6 sperm ( Figure 2B/C). These 6 differences are reflected in the profiles; the long narrow tail in the CBAs appears as a 7 smooth curve at x=50 in the profile, while the shorter, wider C57Bl6s show a distinct dip 8 corresponding to the sharper curve of the dorsal angle before the acrosome. The shorter 9 hook of the CBAs is also seen as a narrow peak at x=10; the longer hook of the C57Bl6s 10 has a correspondingly wider peak. Automated segmentation of the nuclear profile allows 11 quantification and significance testing of the inter-line differences in each separate region of  Table 7) 17 but slightly smaller -as expected given that their measurements are for the entire sperm 18 head rather than just the nucleus. The body widths are within 0.3μm, and our bounding 19 heights are approximately 1.2μm smaller, consistent with our measurements lacking the 20 acrosomal cap (~0.15μm [30]), and the proteinaceous part of the sperm hook. We measured 21 the CBAs to be 12% longer than the C57Bl6s, again close to the previously published 22 13.5%. showing the median and interquartile range of the nuclear shape profiles. B) Consensus 3 nuclei from each population, and the overlap showing the regions differing. C) Size and 4 shape measurements between the lines. The prominent dorsal angle in C57Bl6 nuclei is 5 marked with an asterisk. 6 Comparison of sperm morphology and variability across lines demonstrates the effects of 1 inbreeding depression and hybrid dysgenesis 2 With the software tested on CBA and C57Bl6, we wanted to investigate the extent to which 3 sperm shape variability within and between lines is affected by two factors: inbreeding 4 depression and the complex inter-subspecific mosaic origin of classical laboratory strains. 5 We selected a panel of inbred laboratory lines and compared them to (a) outbred laboratory 6 lines, and (b) wild-derived inbred lines (Table 1). Biological replicate samples from the inbred 7 lines represent either single animals (lab lines) or a pool of two animals (wild-derived inbred 8 lines). For the outbred lines, several individuals were pooled to ensure we were capturing 9 the diversity of the population as a whole.  shape and the relief of inbreeding depression by heterosis. 22 The differences we saw between inbred and outbred laboratory lines made us curious as to 23 the impact of line background and genetic interactions thereof. We investigated one specific 24 reciprocal F1 cross, between C57Bl6 and CBAs. The use of F1 animals is important here as 25 it relieves the effects of inbreeding depression caused by fixation of deleterious recessive 26 variants in each of the parental lines, but still yields a uniform population of genetically 27 identical males from each cross. B6CBA mice are the F1 offspring of a female B6 with a 1 male CBA and CBAB6 mice are the reciprocal cross. Sperm morphology for both F1 lines 2 matches the CBA parental line closely, indicating a dominant effect of the CBA genotype 3 ( Figure 4A). In terms of sperm cross-sectional area, both types of F1 sperm are much more 4 similar to the CBA parent, while being fractionally larger than either parental line ( Figure 4B). 5 Males from both directions of the F1 cross showed less variability in their sperm shape 6 compared to either parent line, suggestive of a degree of heterosis in the F1s. 7 8 The reciprocal cross data allows us to look for parent-of-origin effects on sperm shape. We 9 found two such differences, in sperm cross-sectional area and in bounding width. CBAB6s there is no significant difference between C57Bl6 and CBAB6 (p=0.18) or between CBA and 18 B6CBA (p=0.095). This suggests that this aspect of sperm shape may be influenced either 19 by sex chromosome or mitochondrial background or by autosomal imprinted loci. 20 21 1 Figure 4: Subtle differences can be seen between a CBAB6 (CBA mother) and a B6CBA 2 (C57Bl6 mother). Both are intermediate to the parental shapes, but CBAB6 sperm are wider, 3 and their shape is closer to that of the C57Bl6. A) Consensus nuclei B) Size measurements; 4 C) Overlay of consensus nuclei; D) comparison of angle profiles; the tail attachment region is 5 expanded in the inset. 6 Hierarchical clustering can separate samples based on shape differences 1 Next, we turned our attention to the analysis of morphological variation within a given 2 population. In particular, we considered that cluster analysis of the sperm from a single 3 sample would give an unbiased breakdown of the different morphological sub-populations 4 contained therein. We used a hierarchical clusterer, as implemented by the WEKA data 5 mining tool [29] to separate sperm based on their shape profiles. 6 7 We tested the clustering algorithm by pooling images from C57Bl6 and CBA and analysing 8 them as a single sample. Since C57Bl6 and CBA sperm are slightly different sizes, the 9 simplest partitioning of the mixed set is a binary cut-off at a given threshold for nuclear area. 10 Passing the nuclear areas to a hierarchical clusterer and selecting the two most distinct 11 clusters using the Ward clustering method was 83-85% accurate at separating the individual 12 sperm by line. To determine whether shape-based hierarchical clustering could improve 13 upon this, we sampled values from the angle profile for each nucleus at regular intervals 14 (corresponding to the original window proportion) and provided these as inputs to the 15 clustering algorithm. This clustering was markedly more accurate than a simple size-based 16 cut-off, and separated the two genotypes with 91-95% success (Supplementary table 6). In 17 head-to-head tests using a representative subset of 50 nuclei from each genotype, the 18 clusterer performed at least as well (96%) as experienced assessors (97% accuracy), and 19 substantially better than novice assessors (75% accuracy) (Supplementary figure 12). 20

21
Hierarchical clustering can detect morphological subgroups within a sample 22 Having demonstrated that cluster analysis can recover different shapes from a mixed 23 population of known composition, we looked at its use for novel shape discovery within a 24 single highly variable population. Since the BALB/c line showed the highest variability in our 25 line survey, we chose this as our test sample. A cluster analysis based on angle profile alone 26 found four major groups of sperm shape, from mostly normal through to severe hyper-1 condensation of the sperm ( Figure 5). The final class is still highly variable compared to the 2 other classes; clustering these nuclei further reveals a separation of two separate types of 3 hypercondensation (Supplementary figure 16) as previously described [31]. While the most 4 normal sperm had near-normal placement of the dorsal angle, and a normal tail attachment 5 site, the most heavily distorted sperm showed frequent presence of additional sharp angles 6 in the sperm outline, effacement of the tail attachment site due to compression of the rear of 7 the sperm head, and an ever more prominent and misplaced dorsal angle that may reflect 8 altered microtubule dynamics during nuclear shaping (see Discussion). 9 1 Figure 5: The overall population of BALB/c sperm appears distorted compared to other lines 2 (grey), but clustering reveals separate classes of morphology, from mostly normal (green) to 3 highly condensed (yellow). We present here a morphological analysis tool designed to study nuclear morphology, with 3 the ability to automatically identify key landmarks in the nuclear outline and quantitatively 4 measure a range of nuclear and sub-nuclear parameters. Here, we demonstrate the use of 5 this software to analyse the highly asymmetrical shape of the mouse sperm nucleus; 6 however it is a generally applicable tool suitable for analysis of all sizes and shapes of 7 nuclei. A companion paper ( [32], submitted for publication) demonstrates its use in 8 comparing sperm from boars judged to be suitable/unsuitable for use in artificial 9

insemination. 10
Comparison of this method with other nuclear shape analysis methods 11 The key advantage offered by the software presented here is automation of the steps 12 involved in object detection, shape decomposition and comparison. At the object detection 13 stage, we use an edge detection algorithm that is markedly more effective than the fixed-14 threshold detection used in other packages, particularly in the presence of inhomogeneous 15 staining of the bright chromocenter and dim apical hook. At the shape decomposition step, 16 we introduce a modification of the Zahn-Roskies transform [28] that sensitively detects the 17 various angular landmarks around the sperm periphery without the need for manual 18 intervention. Together, these innovations massively increase the number of nuclei that can 19 be quantified and compared to each other, with a total of 8,749 nuclei being measured 20 during this study, and over 22,000 nuclei in the companion paper analysing boar sperm [32]. 21 This for the first time permits the use of sample sizes that accurately capture not only fixed 22 size and shape differences between samples, but also the detection and classification of 23 intra-sample variability. Our method is robust to differences between camera and 24 microscope setups and fixation techniques, making it amenable to analysis of large numbers 25 of images, and potentially to automated image capturing from whole slide scanners. 26 1 While there are other features of sperm morphology that we do not yet address in this 2 package, the modular design of our software allows additional analysis pipelines to be added 3 at a subsequent date, and for features from different fluorescence channels to be associated 4 with specific nuclei and analysed in relation to them. We anticipate that other sperm 5 morphological features such as the extent and thickness of the acrosome, the proteinaceous 6 tip of the hook, the presence of cytoplasmic droplets, and the length and morphology of the 7 tail will be amenable to our approach by combining nuclear staining for orientation with 8 phase contrast imaging, tubulin immunostaining, MitoTracker, SpermBlue or other stains. 9 Since we are imaging fixed cells, the nuclei also remain available for interrogation by 10 chromosome painting or other molecular cytogenetic approaches, e.g. to detect aneuploid 11 cells and correlate their chromosomal status with their nuclear morphology. 12

Comparison of sperm shape within and between lines 13
Our observations support previous studies (e.g. [4,9]), add further information on the precise 14 regions of the sperm head that that differ between lines, and demonstrate the variability of 15 sperm morphology within each given line. In particular, we examined the presence and 16 placement of the dorsal angle of the sperm. This feature is created by pressure from the 17 manchette: a cone-shaped array of microtubules that forms behind the nucleus and slides 18 backwards during spermiogenesis, shaping the rear of the sperm head in the process.

Comparison of sperm variability within and between lines 2
The greatest variability we saw was in the BALB/c animals. This line is known to have poor 3 sperm morphology and high levels of sperm aneuploidy. Kishikawa et al [31] observed 4 different classes of abnormality, which we were able to recapitulate. In their analysis, the 5 authors found chromosomal abnormalities in 35% of highly abnormal sperm, but also in 15% 6 of sperm that were morphologically 'normal' by their criteria. Given that our new analysis 7 detects additional classes of more subtle shape difference that were not discriminated in the 8 earlier analysis, we hypothesise that these new abnormal classes may also be enriched for 9 chromosomal defects compared to the most normal sperm. Further differences await varying extents in each line [37]. 6 7 X/Y mismatch is a strong potential contributor to regulatory disruption, since most laboratory 8 lines carry a musculus Y on a predominantly domesticus background [35]. The copy number 9 of the ampliconic genes on the X and Y chromosomes varies markedly between musculus 10 and domesticus subspecies, and the relative copy number of these genes is known to be 11 important for normal sperm morphology [38][39][40]. However, while most of the laboratory lines 12 we examined do indeed have mismatched X/Y chromosomes [35,41,42], the FVB X and Y 13 are both of domesticus origin, indicating that the alterations in sperm shape in this line are 14 not due to X/Y mismatch. 15 16 An alternative but not mutually exclusive explanation for the difference between classical 17 laboratory inbred lines and wild-derived inbred lines is that the classical lines have been 18 selected over multiple generations for their ability to breed well in captivity -indeed FVB is 19 particularly known for its fecundity [43]. It may seem paradoxical that selection for high 20 fecundity could adversely affect male fertility parameters: however, under laboratory 21 conditions of non-competitive mating, co-housing a single male with one or more females, it 22 is likely that reproductive output is driven largely by maternal factors. Thus, even though 23 laboratory lines are fertile under lab breeding conditions, their sperm may be uncompetitive 24 in mixed mating experiments compared to a pure species background. The morphology of 25 the FVB zygote pronucleus is independent of the paternal genetic background, and the 26 efficiency of FVB sperm for IVF appears unexceptional [44]. Sperm morphology and 27 fertilisation success in laboratory mice has been shown to evolve rapidly in response to 28 competitive mating experiments, indicating that the baseline competitive ability of laboratory 1 line sperm is sub-optimal [45,46]. Intriguingly, it has even been shown that in lines 2 experimentally selected for high fecundity, male fertility and sperm morphology/motility 3 parameters are compromised, suggestive of a trade-off between the male and female factors 4 necessary for high fecundity in a laboratory environment [47]. 5 6 Relevance for speciation, fertility, and toxicology studies 7 Abnormal sperm head morphology has emerged as a common form of hybrid male sterility in 8 mice [21][22][23][24]48]. Some sterility factors broadly impair spermatogenesis, resulting in reduced 9 sperm counts, lower motility, and abnormal morphology. However, several studies have now 10 shown that hybrid sterility QTL in mice often correspond to specific reproductive phenotypes 11 [24]. The challenges of manually quantifying morphology in large mapping panels has 12 necessitated the use of crude categorical scores [21,23,48], hampering quantitative 13 precision and likely limiting the ability to draw causal links between hybrid incompatibilities 14 and specific aspects of sperm morphological development. 15 16 Our approach assists in two ways: firstly by enabling more rigorous quantitation of sperm 17 shape, and secondly by enabling the large sample sizes and systematic approach needed 18 for mapping studies. As a proof of principle, we have compared males from a reciprocal 19 cross between C57Bl6 and CBA mice, and identified a dominant effect of the CBA genotype 20 on sperm shape. Within this, however, there are subtle differences between the CBAB6 and 21 B6CBA animals, suggesting an effect of either chromosome constitution or imprinting on 22 sperm bounding width. This demonstrates the usefulness of this approach for understanding 23 subtle features of mouse sperm nuclear development, and the potential to use this software 24 for genetic mapping of the various determinants of mouse sperm head shape. 25 26 Fertility rate and IVF efficiency has been correlated with the genetic background of sperm 1 among inbred mouse lines [49]. Furthermore, many studies have shown that the genetic 2 background of a line can influence sperm morphology. For example, deletion of the long arm 3 of the Y chromosome results in a more severe phenotype on B10.BR background than on 4 CBA [50]. Mashiko et al [16] have suggested morphology of sperm is associated with 5 fertilising efficiency in at least two mouse lines (B6D2F1 and C57Bl6/N). Since particular 6 genetic mutations in mouse sperm shape are associated with characteristic nuclear shape 7 abnormalities (e.g. [19]), detailed examination of sperm from natural mutant and/or targeted 8 knowckout animals may point to pathways of interest for understanding spermiogenesis and 9 male fertility more generally. 10 11 In toxicological analysis, rodent sperm are conventionally manually classified into classes of 12 predefined morphological abnormality (e.g. [26,51]). The hierarchical clustering implemented 13 within the software is able to separate nuclei based on shape as accurately as an 14 experienced manual sperm scorer; however it is much faster and more consistent. This may 15 be of use in samples where the nature and degree of abnormalities is hard for humans to 16 reliably quantify. It is also important to understand and quantify normal morphological 17 variation between lines since different lines can have different responses to toxicological 18 agents [52]. While many studies of toxicology using rodent models are conducted on rats, 19 the extra information available in the mouse sperm head still makes them a useful model 20 system. The fact that specific genetic lesions cause specific shape changes means that the 21 sperm shape might in principle give information not just about the presence/absence of 22 toxicity but also its mode of action. This level of analysis would complement existing studies 23 of sperm function, which, in clinical settings or in automated CASA platforms (e.g. [53]), is 24 still lacking detailed morphological data [10]. We present a new software package for the rapid, high-throughput, replicable analysis and 2 comparison of nucleus shape in mouse sperm. By using a range of mouse lines, we have 3 demonstrated the ability of the software to discriminate subtle differences between lines, and 4 to reproducibly separate the nuclei into morphological groups. This has applications for 5 studies of speciation, fertility and understanding the impact of genotoxic compounds. The 6 analysis steps are generalisable and will work on many symmetric or asymmetric shapes of 7 nuclei including, but not limited to sperm from other species. 8 9 Acknowledgements 10 We thank the animal handling staff at the University of Kent, University of Cambridge, 11 University of Montana and Charles River Laboratories. We also thank the experimental test 12 subjects who volunteered to classify C57Bl6 and CBA sperm.