Pilot evaluation of sperm mobility for boar fertility classification using machine learning

Abstract

Predicting boar fertility remains a critical challenge for the swine industry, as standard semen quality measures such as motility and morphology do not accurately predict conception outcomes. This lack of accurate tools allows subfertile boars to enter breeding programs, reducing reproductive efficiency and causing economic loss. Therefore, there is a critical need for practical, readily deployable assays that complement current semen quality assessments and improve fertility prediction. The sperm mobility assay, originally developed for poultry fertility selection, was validated in boars by Vizcarra and Ford (2006), but its relationship to fertility outcomes has not been evaluated. This pilot study evaluated the predictive potential of the mobility assay to classify boars by conception rate (CR) group. Breeding doses from 21 commercial Duroc boars with known CR (>80%, HCR n = 11; <75%, LCR n = 10) were shipped overnight on day of collection, stored at 17 °C, and analyzed for mobility and CASA parameters across three weekly replicates. Across all boars, mobility was associated with CR (r = 0.47; P = 0.03) and moderately predictive of fertility group using binomial logistic regression (AUC = 0.77; P < 0.01), correctly ranking boars by CR group 77% of the time. Including stud as an interaction improved ranking to 83% (AUC = 0.83; P < 0.01), though model improvement was not significant. A random forest algorithm using CASA parameters and mobility reduced the model’s classification error compared to CASA alone following feature selection (28.6% vs 22.4%), but an AUC of 1.00 indicates overfitting and the need for more boars to confirm this finding. Mobility was also found to be reflective of kinematics (P < 0.05) including Amplitude of Lateral Head Displacement (ALH; r = −0.57) and Linearity (LIN; r = 0.53). Sperm Wobble (WOB) was the only significant parameter for both CR (r = 0.45) and mobility (r = 0.46). Because sperm kinematics reflect underlying flagellar function and membrane physiology, mobility may also have utility as a broader indicator of semen quality, though further validation is needed. While mobility values were associated with CR and improved classification when combined with CASA parameters, these models are preliminary and reflect proof-of-concept rather than finalized predictive tools. Validation in larger and more genetically diverse populations is essential before routine application in commercial settings.

This study shows the early potential of the sperm mobility assay to identify subfertile boars in commercial studs.

Introduction

The swine breeding herd has been challenged with the presence of a subfertile boar population over the last 50 years (Clark et al. 1989; Robinson and Buhr 2005; Minton et al. 2013). This issue stems in part from emphasis on progeny growth and carcass characteristics which are inversely correlated with typical semen quality estimates like sperm motility and concentration (Oh et al. 2005; Wolf 2009). Moreover, estimates for motility have a limited capacity to predict fertility outcomes. In a series of studies conducted by Broekhuijse et al., boar- and semen-related parameters were evaluated for their contribution to variation in fertility. Sperm cell motility assessed subjectively at collection accounted for only 4% of male-related fertility variation, while genetic line explained the largest proportion (Broekhuijse et al. 2012b). Computer-assisted sperm analysis (CASA), an objective method for quantifying motility that digitizes sperm movement, modestly improved this estimate with motility explaining 9–10% of variation in field fertility outcomes (Broekhuijse et al. 2012a). As a result, selection decisions based on traits explaining only 9-10% of fertility variation have contributed to the inclusion of subfertile boars in the breeding herd. Thus, improved assessments capable of capturing fertility variation more accurately than current tools would be highly valuable to the swine industry.

Recent investigations have highlighted the need to extend beyond conventional semen quality parameters. For example, 18% of all lipids characterized in the ejaculate lipidome were strongly associated with CASA-measured motility at collection (r > 0.70) (Mills et al. 2024). Similarly, flow cytometry assays evaluating capacitation as a predictor of boar fertility (chlortetracycline (CTC) staining, zinc signatures) have shown stronger associations with fertility outcomes than motility alone (Kwon et al. 2015a, 2015b, 2017; Kerns et al. 2018).

Within breeding programs, genetic improvement can be maximized through accurate identification of superior individuals that are selected as parents of the next generation, thereby achieving breeding goals (Mohammadabadi and Asadollahpour Nanaei 2021). Recent advancements in omics technologies further expand these opportunities by enabling the characterization of metabolites, proteins, RNA, and DNA to better understand how molecular components influence sperm structure, function, and fertility potential (Bordbar et al. 2022; Bond et al. 2024; Mills et al. 2024; Mohammadabadi et al. 2024a, 2024c). For example, the seminal plasma proteome differs significantly between high- and low-fertility boars, with specific proteins linked to traits such as farrowing rate and litter size (Mills et al. 2020). Likewise, comparative metabolomics of sperm cells before and after capacitation has revealed distinct shifts in metabolite profiles that mirror key physiological transitions relevant to fertilization capacity (Weide et al. 2024). Together, these findings illustrate that proteins and metabolites can serve as molecular fingerprints of fertility potential and capacitation status. Further, functional genomics is a field of research that aims to characterize the function and interaction of all the major components (DNA, RNA, proteins, and metabolites, along with their modifications) that contribute to phenotype: the set of observable characteristics of a cell or individual (Mohammadabadi et al. 2024b). However, omics data generation from these technologies can be too large and complex to handle through visual analysis or statistical correlations. Artificial neural networks have been proposed and encouraged to alleviate limitations of traditional regression methods and can be used to handle nonlinear and complex data, even when the data is imprecise and noisy and has encouraged the use of machine intelligence or artificial intelligence (Ghotbaldini et al. 2019; Bordbar et al. 2022).

Yet, as highlighted in recent reviews, such assays and technology remain constrained by labor intensity, cost, large-scale validation, and the need for standardized cutoffs, which limits their adoption in commercial boar studs (Keller and Kerns 2023). Currently, the most reliable way to assess individual boar fertility is through single-sire matings, which are inefficient, labor-intensive, and allow subfertile boars to impact sow productivity in the interim. Thus, providing further justification for commercially viable, immediately deployable tools that can improve fertility prediction alongside traditional semen analysis.

The sperm mobility assay, originally developed in poultry, provides such an opportunity. The assay measures the capacity of a sperm population to migrate through a resistance medium (Accudenz^®) at avian body temperature, therefore quantifying the “effective insemination dose” of the male (Froman and McLean 1996; Donoghue et al. 1998; Froman and Feltmann 1998; Froman et al. 1999; King et al. 2000b; Bowling et al. 2003; Froman 2006; Long 2014). Briefly, freshly collected semen is diluted to standardized concentrations (roosters: 5 × 10⁸ cells/mL, toms: 1 × 10⁹ cells/mL). An aliquot of diluted semen is overlaid pre-warmed Accudenz^® in a cuvette and incubated at body temperature for 5 min. Readings are then taken on a spectrophotometer and absorbance values indicate fertility status of the animal where higher readings indicate higher fertility. The assay revealed that mobility, not motility measured on CASA alone, was the functional trait most closely linked to fertility outcomes (King et al. 2000a; Froman et al. 2003). Supporting these findings, lipidomic profiling capable of distinguishing between fertility phenotype (Borges et al. 2018; Mills et al. 2021) identified early-production biomarkers of sperm mobility in broiler breeders, with several achieving >90% predictive accuracy for discriminating high- and low-mobility phenotypes (Bond et al. 2024).

The assay’s utility has also been explored in boars and stallions (Vizcarra and Ford 2006). In boars, protocol parameters were developed using different concentrations of viable cells/mL identified with fluorometric detection. Higher absorbance readings were indicative of a higher concentration of viable sperm cells/mL, with 5 × 10⁷ viable cells/mL being the optimal concentration for the boar assay. However, fertility validation was not performed due to limited field data acquisition. Given its demonstrated capacity to classify poultry fertility status, this assay may provide a valuable complement to semen evaluation tools such as CASA until high-throughput molecular screening becomes commercially feasible.

Therefore, we aimed to build upon the findings of Vizcarra and Ford and determine the assay’s preliminary predictive potential for fertility status. However, assay parameters were established nearly 20 years ago, and both the genetic progress of commercial boars and industry practices have since evolved. After passing established semen quality thresholds, ejaculates are diluted with extender to final breeding dose concentration which can be lower than 5 × 10⁷ cells/mL. Ensuring the assay works with final breeding dose concentration would aid in its integration since it would remove an extra dilution step for laboratory technicians processing ejaculates. Additionally, acquiring breeding doses from boars with known fertility would require transport and storage at 17 °C. Stored doses would naturally exhibit a reduced number of viable sperm cells compared to freshly collected samples or the 5 × 10⁷ viable sperm cells used by Vizcarra and Ford to establish assay parameters (Lange-Consiglio et al. 2013; Henning et al. 2022). Revisiting parameters is therefore warranted to ensure relevance to modern boar physiology and production practices. We hypothesized that achieving comparable mobility readings and adequate separation of fertility phenotypes would require adjustments to the percentage or volume of Accudenz^® solution. Therefore, the objectives of this study were to: (1) adapt the sperm mobility assay for use with final breeding dose concentrations, and (2) evaluate its ability to classify boars into high- and low-conception rate categories in a pilot cohort using statistical modeling.

Materials and methods

Mobility assay reagent preparation and procedure

Reagents were prepared and mobility was determined in accordance with protocols described by Vizcarra and Ford (2006) and Froman and McLean (1996). Briefly, a 30% (w/v) stock solution (pH 7.4) of Accudenz^® (Accurate Chemical & Scientific Corp., Westbury, NY, USA) was prepared with 3 mM KCl (Sigma-Aldrich Co., St Louis, MO, USA) and 5 mM TES (N-[Tris(hydroxymethyl)methyl]-2-aminoethanesulfonic acid; Sigma-Aldrich Co.). In addition, a mobility buffer (pH 7.4) containing 111 mM NaCl, 25 mM glucose, and 4 mM CaCl₂ (Sigma-Aldrich Co.) in 50 mM TES was diluted in distilled water to an osmolality of 290 mM/kg. Finally, a 3% and 6% (w/v) working solutions were prepared by diluting 30% stock solution with mobility buffer at 1:10 and 1:5 ratios, respectively.

For experiment 1, four combinations of Accudenz^® concentration and volume were tested: 1) 3 mL of 6% (per Vizcarra and Ford 2006) 2) 1.5 mL of 6%, 3) 3 mL 3%, and 4) 1.5 mL 3%. Each solution was aliquoted into polystyrene macro cuvettes (Biosigma S.p.A., Cona (VE), Italy) for sperm mobility evaluation. In experiment 2, the same macro cuvettes were pre-filled with 1.5 mL 3% Accudenz^® prior to mobility readings. All pre-filled cuvettes were pre-warmed to 37 °C using a warming block (Animal Reproduction Systems, Chino, CA, USA) and slide warmer (VWR Scientific, Seacaucus, NJ, USA), consistent with the optimal temperature for boar sperm mobility assessment (Vizcarra and Ford 2006).

Animals

Experiment 1.

All samples used for experiments 1 and 2 were sourced during routine collections from boars enrolled in a commercial breeding program. Extended doses from sixteen Duroc boars representing four different genetic lines were sourced from a single commercial boar stud. Per standard industry protocol, boars were manually collected and ejaculates were delivered by the technician to the stud laboratory. Upon arrival, ejaculates were analyzed for semen quality estimates using Computer-Assisted Sperm Analysis (CASA). Ejaculates meeting pre-established minimum motility and morphology quality thresholds following collection (>75%; Flowers 1997; Foxcroft et al. 2008; Schulze et al. 2014; Knox 2016; Waberski et al. 2019) were extended to 3.5 billion total cells per breeding dose translating to a final volume of 70 mLs with 5 × 10⁷ cells/mL using NUTRIXCell+ (IMV Technologies, Osseo, MN USA), per standard industry protocol.

Samples were transported on the day of collection to a local sow farm and stored at 17 °C before being transferred in an insulated cooler containing 17 °C cooler packs to the Beltsville Agricultural Research Center (BARC). Upon arrival, doses were stored in a 17 °C refrigerator until use. A comparison of sample handling between the current study and Vizcarra and Ford’s original methods is provided in Table 1.

Table 1.

Summary of sample handling and processing differences between the present study and Vizcarra and Ford (2006).

	Vizcarra and Ford (2006  ), Experiment 3	Experiment 1	Experiment 2
Boars	USMARCa: four-breed composite of maternal Landrace, Yorkshire, Duroc, and terminal lean line landrace (n = 24)	Commercial Duroc boars representing four genetic sources housed in a single stud (n = 16)	Commercial Duroc boars from the same genetic source from two studs (Stud A & B; n = 21)
Sperm viability requirement	Fluorometric determination of viable sperm cells	Assumed viability based on standard motility and morphology thresholds (>75%) determined via CASA on day of collection	Assumed viability based on standard motility and morphology thresholds (>75%) determined via CASA on day of collection
Dilution medium	Androhep®, long-term BSA extender	NUTRIXcell+, long-term non-BSA Extender	AndroStar® Plus (Stud A) and Preserv® Xtreme (Stud B), long-term non BSA extenders
Assay ejaculate concentration	Extended to 1 × 108  viable cells/mL and re-diluted to 5 × 107  viable cells/mL after 24 h storage	Extended to 5 × 107 cells/mL for breeding dose concentration (3.5 billion cells per dose)	Extended to 3.3 × 107 cells/mL for breeding dose concentration (2 billion cells per dose)
Accudenz® concentration and volume	3 mL 6% Accudenz®	1.5 mL 3% Accudenz®	1.5 mL 3% Accudenz®
Wavelength	550 nm	530 nm	530 nm
Storage conditions prior to mobility assay	24 h at 15.9 ± 0.1 °C	24 h post-collection, transported and stored at 17 °C	24 h post-collection, transported and stored at 17 °C

^aU.S. Meat Animal Research Center, USDA-ARS.

Experiment 2.

Breeding doses from twenty-one commercial Duroc boars from the same genetic line and ≥50 single-sire matings were sourced from two commercial boar studs to evaluate the relationship between sperm mobility and fertility outcomes (Table 2). Prior to receiving samples, boars were allocated to high conception rate (HCR; >80% CR, n = 11) or low conception rate (LCR; <75% CR, n = 10) groups based on breeding program selection practices.

Table 2.

High and low phenotype group mean conception rate and age at collection and P-values—Experiment 2.

Boar parameters	HCR  a	LCR  b	P-value
Total  c	n = 11	n = 10
Age at collection (months)	23.5 ± 1.55	20.9 ± 1.52	0.25
Conception rate (%)	91.0 ± 1.21	67.3 ± 2.90	<0.05
Stud A  d	n = 8	n = 3
Age at collection (months)	25.0 ± 1.83	26.7 ± 2.17	0.57
Conception rate (%)	92.3 ± 0.97	72 ± 1.53	<0.01
Stud B	n = 3	n = 7
Age at collection (months)	19.47 ± 1.29	18.39 ± 0.89	0.52
Conception rate (%)	87.7 ± 3.18	65.3 ± 3.93	<0.01

^aHigh conception rate (>80%).

^bLow conception rate (<75%).

^cAll boars included in Experiment 2 have mobility readings on three weekly collections to account for variation. Values reported within the table represent the mean ± standard error.

^dBoars at stud A were older on average by 7 months (P < 0.01) and 15% higher average conception rate (P < 0.01) than Stud B.

Briefly, boars were collected weekly using a semi-automatic collection system, per each stud’s standard protocol. Following collection, the ejaculate was closed by using a band and sent to the laboratory via pneumatic tube or pass-through window. After passing pre-established semen quality motility and morphology parameters (>75%; Flowers 1997; Foxcroft et al. 2008; Schulze et al. 2014; Knox 2016; Waberski et al. 2019) using CASA, ejaculates were extended in long-term, non-BSA extenders to 2 billion total cells per breeding dose at 60 mL (3.3 × 10⁷ cells/mL). Doses from Stud A were extended in AndroStar^® Plus (Minitube, Verona, WI, USA) and Stud B in Preserv^® Xtreme (GenePro, Madison, WI, USA). Extended semen was stored at 17 °C and shipped overnight to BARC. Each boar was collected weekly for 3 consecutive weeks to capture within-boar variation in sperm mobility.

Experiment 1: Adapting parameters of the sperm mobility assay for breeding dose concentration

To reduce sample handling variability, pre-warmed cuvettes were arranged in warming blocks, allowing for all four assay treatments to be run simultaneously within each boar (Fig. 1). This strategy minimized the potential confounding effects of fluctuating storage temperatures, sample remixing and room-temperature exposure. Preliminary analysis determined that absorbance readings were not influenced by Accudenz^® concentration or volume, thus the spectrophotometer (530 nm; Micro-Reader I, IMV Technologies, Osseo, MN, USA) was calibrated using a blank containing 3 mL of 6% Accudenz^® prior to each assay run.

Fig. 1.

Layout of the warming blocks used in Experiment 1. Boars were evaluated in sperm mobility assay groups, were pre-determined based on the number of samples that could be measured accurately within 5-min intervals. Time between removal from 17 °C storage and assay loading was standardized to ensure all samples reached the same temperature prior to cuvette loading. Each boar was assigned a column, and a 300 µL aliquot of semen was overlaid onto each of the four pre-warmed Accudenz^® treatments (Trt): Trt 1 (3 mL of 6% Accudenz^®), Trt 2 (1.5 mL of 6% Accudenz^®), Trt 3 (3 mL of 3% Accudenz^®), and Trt 4 (1.5 mL of 3% Accudenz^®).

Approximately twenty-four hours post-transport, doses were removed from storage and gently inverted 5 times to resuspend settled cells. Boar doses were removed from storage in small groups to minimize variation in time at room temperature between the first and last boar run of the day. A maximum of two boars per technician was used to control the interval between sample loading and absorbance measurement. A 300 µL aliquot of semen from doses (5 × 10⁷ cells/mL) was carefully layered over the pre-warmed Accudenz^® solution in cuvettes in treatment order (1-4). Absorbance was measured at 530 nm after 1 min, followed by incubation with subsequent readings at 5, 10, 15, 20, and 40 min. Data were analyzed using a linear mixed model (REML, Satterthwaite’s method) in R. Fixed effects included treatment, time, and their interaction, with boar as a random effect. Post hoc comparisons at each reading time were performed using Tukey’s test. Significance was defined as P < 0.05 and values of 0.05 ≤ P ≤ 0.10 were considered trends.

Experiment 2: Comparison of absorbance values of boars with known fertility at different timepoints

To account for variation, three weekly ejaculates from 11 HCR and 10 LCR boars were evaluated using the 1.5 mL 3% Accudenz^® assay (Treatment 4 from Experiment 1). Although conception rate group was assigned in advance, all sample processing and analysis were conducted blinded to fertility status. Upon arrival, boar stud ID were randomized using a random generator to determine both assay order and cuvette position within the warming block (Fig. 2). Arbitrary ID numbers were assigned, starting with “1”, to blind technicans to boar identity throughout the experiment.

Fig. 2.

Layout of the warming block used in Experiment 2. Each cuvette slot was assigned to a boar based on an arbitrary identification number generated using a random number generator. This randomization was used to determined both the order of loading and the position of each sample within the warming block to ensure blinding throughout the assay.

Upon arrival, doses were removed from coolers and gently inverted to resuspend cells. A 300 µL aliquot (3.3 × 10⁷ cells/mL) was layered over a 37 °C cuvette containing 1.5 mL of 3% Accudenz^®. Absorbance was measured at 1, 5, 10, 15, 20, 25, and 40 min. The additional 25-min timepoint was added to assess absorbance beyond 20 min, based on variation observed in Experiment 1 and to test whether switching studs altered the optimal assay timepoint. On the day of mobility readings (24 h post-storage), doses were analyzed at BARC using Computer-Assisted Sperm Analysis (CASA; IVOS-II Hamilton-Thorne) for semen quality parameters including Amplitude of Lateral Head Displacement (ALH), curvilinear velocity (VCL), path velocity (VAP), straight line velocity (VSL), beat cross frequency (BCF), linearity (LIN), morphology, motility, and progressive motility.

Mobility readings were analyzed using a linear mixed model (REML; Satterthwaite’s method) implemented in R (v 4.5.0; lmer function, lme4 package), with fixed effects of time and conception rate group, and stud as an interactive effect. Conception rate group was coded as high conception rate (HCR; >80% CR) or low conception rate (LCR; <75% CR) based on breeding program selection practices (Table 2). Boar was included as a random effect to account for repeated measures. Tukey’s post-hoc tests using the emmeans package were used to compare absorbance values between groups at each timepoint. Correlation analysis between mobility readings, CASA parameters, and conception rate was first performed using each replicate as an independent data point and second using the average of three replicates per boar to allow assessment of both within-sample trends and between-boar trends across parameters. Since conception rate was calculated at the boar level, the averaged data were used for final interpretation to reflect the appropriate unit of biological inference and avoid inflating statistical significance due to pseudoreplication. Pearson correlation coefficients were calculated using rcorr() function from the Hmisc package in R. The resulting correlation matrix includes both correlation coefficients (r) and associated P-values, which indicate the statistical significance of each correlation.

Binomial logistic regression was used to evaluate whether mobility readings were predictive of low conception rate (LCR) status. Models were fitted in R using the glm function with a binomial link, where LCR animals were coded as the positive class. Unless otherwise specified (eg the balanced Stud A model evaluated at 10 min), absorbance values taken at 5 min were used for model fitting, based on preliminary analyses identifying this timepoint as having the greatest separation between conception rate group absorbance values. Given the goal was to assess the assay’s predictive potential rather than build a finalized predictive tool, model coefficients are considered preliminary and are not reported due to the small pilot cohort. To account for the potential of unstable estimates and inflated performance metrics due to overfitting, we evaluated predictive potential several ways. Two overall models were evaluated: one model with mobility as a sole predictor and one including stud as an interaction between mobility and stud to assess whether the predictive relationship differed by stud (HCR, n = 11; LCR, n = 10). Further, we created two within-stud models using a balanced subset of animals to evaluate the influence of unbalanced conception rate group sizes across studs and assess model robustness in a controlled structure. The top and bottom 3 individuals within their respective conception rate groups (mirrored standard deviation away from CR mean within stud) were compared. Predicted probabilities of LCR were generated using the predict () function (type = “response”), and used to visualize classification performance and construct fitted probability curves. Model outputs were further assessed using receiver operating characteristic (ROC) curve analysis and area under the curve (AUC) to evaluate the assay’s early predictive utility in identifying LCR boars (pROC package).

A Random Forest (RF) classifier was implemented to evaluate one model with CASA measurements and one with CASA + mobility using the randomForest package in R (version 4.5.0) to assess whether assay integration could improve CASA’s predictive ability for conception rate group. Average mobility and CASA measurements across both studs, no stud interaction, (HCR, n = 11; LCR, n = 10) were used to train the RF algorithm. The algorithm was trained 10 times, each with 1000 trees and different random seeds, to account for stochastic variability in bootstrap sampling and feature selection. Out-of-bag (OOB) error rates were recorded for each run to estimate model performance, and Mean Decrease Accuracy (MDA) values were calculated to assess variable importance within the model. The average OOB error and average MDA across 10 iterations were reported to evaluate consistency and identify key predictive features comparing CASA to mobility. Significance was defined as P < 0.05 and values of 0.05 ≤ P ≤ 0.10 were considered trends. AUC values were interpreted using standardized performance benchmarks.

Results

Experiment 1

The overall trend of absorbance values related to 3 mL of Accudenz^®, regardless of percentage was a slow increase and peak absorbance at 15 min and a slow decrease in absorbance value to 40 min (Fig. 3). Samples loaded onto cuvettes containing 3 mL of Accudenz^® solution also had very low absorbance values throughout the assay and were never able to exceed an absorbance reading of 0.050 For samples loaded onto 1.5 mL of Accudenz^®, peak absorbance was achieved at 10 min and overall absorbance values were higher in 3 mL Accudenz^® treatments. Moreover, 1.5 mL of 3% Accudenz^® yielded the highest absorbance values that reflected more closely to the values determined by Vizcarra and Ford using 5 × 10⁷ viable cells/mL from fresh samples at the same timepoints.

Fig. 3.

Mean absorbance values (530 nm; empty symbols) over time for boar sperm migrating across four Accudenz^® treatments in Experiment 1, compared to data adapted from Vizcarra and Ford (2006; solid symbol; 5 × 10⁷ viable cells/mL; 550 nm). Each point represents the mean ± standard error of 16 ejaculates measured at multiple timepoints in Experiment 1. All treatments were run simultaneously within each boar. Accudenz^® treatments (Trt) were as follows: Trt 1 (3 mL of 6% Accudenz^®), Trt 2 (1.5 mL of 6% Accudenz^®), Trt 3 (3 mL of 3% Accudenz^®), and Trt 4 (1.5 mL of 3% Accudenz^®). Absorbance values for the 1.5 mL, 3% Accudenz^® treatment (marked by “X”) closely matched those reported by Vizcarra and Ford (2006) at 5 and 10 min (solid triangles).

Experiment 2

Associations among conception rate, mobility readings, and CASA parameters.

Differences in mobility values between HCR and LCR were found to be associated with both morphology and kinematic parameters. Both replicate-level and average-based correlation analysis demonstrated significant associations between 5-min mobility readings and CR. However, the analysis using averaged values per boar produced slightly stronger correlation coefficients (r = 0.47; P = 0.03) compared to replicate-level (r = 0.38; P < 0.01) and was selected for reporting to reduce inflated sample size from non-independence of replicates. When comparing boars across all studs, 5-min mobility was positively related to CR (Fig. 4), with the visual highlighting the >80% HCR and <75% LCR thresholds (orange dashed line) and stud-specific point shapes.

Fig. 4.

Relationship between 5-min sperm mobility readings and conception rate (CR) across all boars from Experiment 2. Each data point represents the average of three weekly mobility measurements for an individual boar, taken at 5 min of incubation. CR is presented as a continuous variable on the x-axis (%). An orange dashed line denotes approximate thresholds for high conception rate (HCR; >80%) and low conception rate (LCR; <75%). Point shapes indicate stud of origin, allowing visualization of potential stud-level clustering. While modest, a positive relationship is evident between mobility and CR, supporting the selection of 5-min absorbance as a key mobility measure for subsequent analyses in the future.

Traditional CASA parameters commonly used to evaluate fresh semen, such as normal morphology, motility, and progressive motility, were not significantly associated with CR or 5-min mobility readings (P < 0.05; Table 3). Although the directionality of most CASA traits was generally conserved between CR and mobility, the specific parameters significantly associated with each outcome differed. The only parameter significant for both outcomes was average wobble (WOB mean).

Table 3.

Pearson correlation coefficients between mobility values, CASA parameters, and conception rate—Experiment 2.

Across Studs (HCR n = 11; LCR n = 10)a	Conception rate (%)	5-min mobility
Conception rate	–	0.47*
Motility	0.26	−0.27
Progressive motility	0.16	−0.03
Morphology	−0.01	0.01
Bent tail	0.10	0.47*
ALH meanb	−0.25	−0.57*
ALH median	−0.22	−0.59*
BCF mean	0.05	0.45*
DCL median	−0.29	−0.44*
LIN mean	0.14	0.53*
LIN median	0.11	0.43*
LIN SD	0.40	0.59*
STR SD	0.52*	0.28
VAP median	−0.19	−0.45*
VAP SD	0.43*	0.02
VCL mean	−0.19	−0.45*
VCL median	−0.28	−0.53*
VCL SD	0.49*	0.10
WOB mean	0.45*	0.46*
WOB median	0.48*	0.41
WOB SD	0.16	0.47*

Stud A (HCR n = 8; LCR n = 3)c	Conception rate (%)	5-min mobility
Conception rate	−	−0.09
20-min mobility	−0.49	0.70*
Motility	0.30	−0.18
Progressive motility	0.27	−0.05
Morphology	−0.23	−0.36
Mean area, motile	0.66*	−0.02
Median area, motile	0.67*	−0.07
Mean area, total	0.62*	−0.04
Median area, total	0.68*	−0.05
Concentration mil/mL, motile	−0.02	0.61*
Concentration mil/mL, total	−0.05	0.66*

Stud B (HCR n = 3; LCR n = 7)	Conception rate (%)	5-min mobility
Conception rate	−	0.26
Motility	0.08	−0.78*
Progressive motility	−0.22	−0.52
Morphology	−0.24	−0.36
Bent tail	0.11	0.78*
DMR	0.68*	0.11
Proximal droplet concentration mil/mL	0.65*	0.29
ALH median	0.16	−0.64*
ALH SD	0.65*	−0.18
BCF mean	−0.64*	0.42
BCF median	−0.67*	0.45
DAP mean	−0.03	−0.66*
DCL mean	−0.03	−0.64*
Elongation SD, motile	0.18	0.75*
STR SD	0.69*	−0.34

^aPearson correlation coefficients (r) were calculated to evaluate relationships between mobility assay readings (5-min values unless otherwise noted), conception rate (%), and computer assisted sperm analysis (CASA)-derived parameters within and across studs. Positive r values indicate that as one variable increases, the other tends to increase, while negative r values indicate an inverse relationship. Asterisks denote statistically significant correlations at P < 0.05. Average mobility and CASA values per boar (calculated across repeated measures) were used for correlation analysis. Mobility and CASA measurements were taken ∼24 h post-collection following transport and storage in 17 °C.

^bALH—Amplitude of Lateral Head Displacement; BCF—Beat Cross Frequency; DAP—Distance Along the Path; DCL—Distance Curvilinear; LIN—Linearity; STR—Straightness; VAP—Average Path Velocity; VCL—Curvilinear Velocity; WOB—Wobble; SD—Standard deviation; DMR—Distal Midpiece Reflex.

^cLCR boars in Stud A exhibited higher average mobility values than HCR boars, whereas the opposite trend was observed in Stud B.

Additional traits correlated with CR only included median WOB and the standard deviations of straightness (STR SD), velocity of average path (VAP SD), and curvilinear velocity (VCL SD), supporting the idea that intra-ejaculate variability in sperm motion may be linked to fertility potential. In contrast, parameters associated only with 5 min mobility included bent tail morphology, mean beat cross frequency (BCF), mean and median amplitude of lateral head displacement (ALH), mean and median linearity (LIN), mean and median curvilinear velocity (VCL), median distance along curvilinear path (DCL), median VAP, and the standard deviations of motile linearity (LIN SD) and wobble (WOB SD). These traits appear to contribute to short-term mobility classification but do not necessarily reflect fertilization success.

When comparing mobility across all boars, average absorbance values were higher in HCR than LCR boars at 5 min (P < 0.01) and remained numerically higher throughout the 40 min incubation. The absorbance trajectory in HCR boars mirrored patterns observed in Experiment 1, peaking at 10 min and gradually declining through 40 min (Fig. 5A). LCR boars showed a delayed peak at 20 min and slower decline through the end of the assay. At 40 min, absorbance values for LCR were numerically higher than HCR, though not significant (P > 0.05). The percentage of bent tails was positively associated with 5-min mobility (r = 0.47; P = 0.03). For 20, 25, and 40 min, mobility was increasingly related to the presence of distal droplets (r = 0.47, 0.62, 0.71; P < 0.05).

Fig. 5.

Absorbance trajectories from sperm mobility assays over 40 min of incubation, with data averaged across repeated measurements from Experiment 2. (A) Overall mean mobility curves for HCR (solid line) and LCR (dashed line) boars across both studs. Each point represents the mean absorbance value for all boars within a CR group at the indicated timepoint, calculated as the average of three weekly measurements per boar. Error bars represent the standard error of the mean (SEM). HCR boars reached peak mobility earlier (10 min) than LCR boars (20 min), followed by a gradual decline in absorbance values for both groups. (B) Stud-specific mobility trajectories for HCR and LCR boars. Each point represents the mean absorbance within a CR group for that stud, averaged from three weekly measurements per boar. Error bars represent SEM. Stud A showed earlier group separation with an HCR peak at 10 min and LCR peak at 15 min, while Stud B exhibited a delayed but consistent difference between groups, with HCR peaking earlier than LCR.

Mobility profiles differed notably between studs (Fig. 5B). In Stud A, HCR boars exhibited a peak in absorbance at 10 min, followed by a gradual decline through 40 min, consistent with the overall pattern observed across all boars and in Experiment 1 (Fig. 5). However, group separation between HCR and LCR did not occur until 10 min, with the LCR group peaking later at 15 min and maintaining a numerically higher absorbance values throughout the assay. Elevated mobility at 10 min was associated with higher concentrations of proximal droplets (r = 0.66; P = 0.03) and slow cells (r = 0.63; P = 0.04). Further, the concentration of bent tails and distal droplets showed a tendency toward a positive association with mobility at 10 min (r = 0.57; P = 0.07). Mobility at 5 min for Stud A was not associated with CR. The mobility value with the strongest correlation to CR was 20 min, but was not significant (r = −0.49; P = 0.13). CASA parameters associated with CR in Stud A included mean and median area for motile and all cells (Table 3). Mobility at 5 min in Stud A was associated with the concentration of motile and total cells.

Stud B’s absorbances both peaked later in the assay compared to Stud A, but the trajectory of mobility by CR group was conserved where HCR peaked earlier than LCR at 20 and 40 min, respectively. Mobility values increased similarly in both groups during the early phase of the assay, with the most pronounced differences observed at 5 and 10 min, with HCR being higher than LCR (P < 0.05). Mobility values at 5 min were not significantly associated with CR (r = 0.26; P > 0.10). Lower mobility values in LCR boars at 5 min were associated with higher median ALH (r = −0.64; P = 0.05), mean distance along the average path (DAP; r = −0.66; P = 0.04), and mean distance along the curvilinear path (DCL; r = −0.64; P = 0.05). Traits that were negatively correlated to CR in Stud B included mean and median BCF. Higher mobility values were associated with a greater concentration of proximal droplets, DMR, and standard deviations of STR (Table 3).

Predictive performance of mobility readings and CASA parameters.

According to binomial logistic regression across all boars, mobility readings at 5 min were significantly associated with the LCR group when not adjusted for stud effect (P = 0.01), with a corresponding AUC of 0.77 (95% CI: 0.56-0.99; P < 0.01), indicating moderate discriminative ability. When stud was incorporated into the model as an interaction term with absorbance, the main effect for Stud B was significant (P = 0.05), suggesting differences in baseline LCR probability between studs, while the absorbance main effect (P = 0.77) and the interaction effect itself (P = 0.14) were not significant. ROC analysis for this interaction model produced a slightly higher AUC of 0.83 (95% CI: 0.65-1.00; P < 0.01), but the improvement in AUC over the simpler model was not significant (ΔAUC = 0.06, 95% CI: –0.22 to 0.11, P = 0.52). This lack of improvement was likely influenced by the unbalanced sample sizes between conception rate groups, with Stud A having only three LCR boars compared to seven in Stud B, and by the weaker separation in absorbance values at 5 min between HCR and LCR boars in Stud A. Model performance across all configurations is summarized in Table 4.

Table 4.

Summary of binomial logistic regression models used to predict low conception rate (LCR) based on mobility readings—Experiment 2.

Model type  a	Stud	Time (min)	AUC	CI Low	CI High	P (AUC > 0.5)	HCR n  b	LCR n  c
Overall—absorbance only	A + B	5	0.77	0.56	0.99	0.01	11	10
Overall—absorbance × stud	A + B	5	0.83	0.65	1.00	<0.01	11	10
Per-stud	A	5	0.63	0.24	1.00	0.52	8	3
	B	5	0.81	0.41	1.00	0.13	3	7
Balanced subset	A	5	0.89	0.58	1.00	0.01	3	3
	A	10	0.89	0.58	1.00	0.01	3	3
	B	5	0.78	0.29	1.00	0.26	3	3

^aBinomial logistic regression models are preliminary and serve as proof-of-concept for the assay and are not final predictive models. Multiple model configurations were evaluated to compare predictive performance under different data structures, including overall models (all boars combined), within-stud models, and balanced subsets (top and bottom three boars per stud by conception rate). For the balanced model, Stud A was evaluated at both 5 and 10 min because conception rate groups in this stud began to separate later in the assay, with clearer divergence emerging at 10 min; in contrast, Stud B showed its strongest separation at 5 min, so only that timepoint was analyzed. The aim was to assess the robustness of assay performance across varying sample sizes, population distributions, and model complexities. Model performance metrics include: AUC—area under the ROC (Receiver Operating Characteristic) curve, where 1.0 indicates perfect classification and 0.5 indicates random classification; CI Low/High—lower and upper bounds of the 95% confidence interval for AUC; P (AUC > 0.5) – probability that the model’s predictive ability (AUC) is greater than chance. Average mobility values per boar (calculated across repeated measures) were used to generate model outputs.

^b,cNumber of boars in high and low CR groups, respectively.

When binomial models were fit with each stud individually, predictive performance decreased in both cases, as expected due to reduced sample size. Stud A yielded an AUC of 0.63 (95% CI: 0.24–1.00, P = 0.52), suggesting a near-random classification, while Stud B retained moderate classification ability with an AUC of 0.81 (95% CI: 0.41–1.00, P = 0.13). The ΔAUC between studs was 0.18, although this difference was not significant given the wide confidence intervals. This result aligns with observed data patterns, where Stud B showed a stronger and more consistent separation between conception rate groups across timepoints, particularly at 5 min (Fig. 5B). Further exploration of model behavior across absorbance thresholds revealed that when all boars were modeled together (including the stud interaction), most LCR animals were assigned predicted probabilities > 0.5 when absorbance values were below 0.200 (Fig. 6A).

Fig. 6.

Binomial logistic regression models predicting low conception rate (LCR) probability from 5-min sperm mobility readings, based on averaged per-boar data from Experiment 2. (A) Model fitted across all boars without stud adjustment. Points represent individual boars (average of three weekly measurements), with predicted probabilities of being classified as LCR plotted against observed absorbance values at 5 min. Solid lines show model-predicted probabilities, shaded ribbons indicate 95% confidence intervals. (B) Model including the interaction between absorbance and stud. Predicted probabilities were calculated from per-boar averages. Shapes indicate stud of origin; solid and dashed lines represent model fits for each stud. Interaction modeling slightly improved the area under the ROC curve (AUC) but was not statistically significant.

To evaluate the impact of class imbalance more directly, balanced subsets were created for each stud using the top and bottom three boars based on CR. In Stud A, creating a balanced model improved model performance from the unbalanced per-stud AUC of 0.63 to 0.89 at both 5 and 10 min (95% CI: 0.58–1.00, P = 0.01), reflecting the timepoints at which absorbance divergence between groups began. However, when the model was refit using 10-min mobility readings, AUC improved from 0.54 to 0.77, reflecting the timepoint at which absorbance divergence between groups became more pronounced (Fig. 7A). In contrast, balancing the samples in Stud B using the same selection parameters slightly reduced AUC from 0.81 to 0.78 at 5 min (95% CI: 0.29–1.00, P = 0.26) (Fig. 7B), suggesting the full dataset retained stronger discriminatory power and that LCR animals in Stud B may exhibit more consistent early mobility decline.

Fig. 7.

Binomial logistic regression models for balanced subsets of boars, selected to include the top and bottom three individuals by conception rate within each stud, using averaged per-boar measurements from Experiment 2. (A) Stud A balanced model performance at 5 and 10 min. Points represent per-boar predicted probabilities of being classified as LCR based on absorbance values averaged from three weekly measurements. Model performance improved substantially compared to the unbalanced dataset, with the strongest separation between CR groups observed at 10 min. (B) Stud B balanced model performance at 5 min. Points represent per-boar predicted probabilities using averaged absorbance values. Balancing resulted in a slight decrease in area under the ROC curve (AUC) compared to the unbalanced model, indicating that the full dataset retained stronger discriminatory power for Stud B.

For conceptual validation purposes only, we implemented a series of random forest (RF) classifiers to predict HCR and LCR based on CASA parameters, with a focus on evaluating how variable selection and inclusion of mobility readings affected classification accuracy (Table 5). The initial RF model, trained on filtered CASA data (excluding low-variance and highly correlated predictors), resulted in a mean out-of-bag (OOB) error rate of 45.71% (SD = 3.3%). Class-specific error rates were 45.5% for HCR and 50.0% for LCR, indicating weak predictive performance. Several predictors were identified with negative Mean Decrease Accuracy (MDA), suggesting they decreased model performance. These variables were removed, and the model was retrained. After removing uninformative predictors, the reduced model achieved a lower OOB error rate of 28.57% with zero variation across replicates (SD = 0%). Class error dropped to 18.2% for HCR and 40.0% for LCR. ROC analysis showed an AUC of 1.00 indicating perfect classification. However, this high AUC is likely due to overfitting given the relatively small sample size and lack of external validation. As such, the model was not used for deployment, but rather as a conceptual validation tool, providing insights into the impact of feature selection on classification error.

Table 5.

Summary of random forest models used to classify boars as high or low conception rate (HCR/LCR) based on CASA and mobility parameters—Experiment 2.

Model type  a	Variables	Mean OOB error, %	AUC	HCR error, %	LCR error, %
CASA onlyb	CASA parameters (n = 19)	45.7	1.00c	45.5	50.0
CASA only (final)	Low-MDA CASA parameters removed (n = 9)c	28.6	1.00	18.2	40.0
CASA + mobility	CASA + 5-min mobility (n = 20)	37.6	1.00	36.4	40.0
CASA + mobility (final)	Low-MDA CASA + 5-min mobility parameters removed (n = 9)	22.4	1.00	9.10	30.0

^aRandom forest models are preliminary and serve as proof-of-concept for the assay and are not final predictive models. Perfect AUC values (1.00) likely reflect model overfitting in small datasets.

^bMultiple model configurations were evaluated to compare predictive performance using different predictor sets, including raw CASA parameters, CASA parameters with low mean decrease accuracy (MDA) removed, and these same CASA configurations combined with 5-minute mobility readings. Model performance metrics include: Mean OOB Error (%) – average classification error rate estimated from out-of-bag samples during model training; AUC—area under the ROC (Receiver Operating Characteristic) curve, where 1.0 indicates perfect classification and 0.5 indicates random classification; HCR Error (%)—misclassification rate for high conception rate boars; LCR Error (%)—misclassification rate for low conception rate boars. Average mobility values and CASA parameters per boar (calculated across repeated measures) were used to generate these model outputs.

^cVariables with low-mean decrease accuracy (MDA) were excluded to improve model performance and reduce overfitting where possible.

To determine whether incorporating mobility data would enhance classification performance, a second modeling pipeline was run using the same filtering and feature selection approach, but including the mobility readings at 5 min. The initial RF model that included 5 min mobility yielded a mean OOB error of 37.6% (SD = 4.2%). After filtering out predictors with negative MDA and retraining the model, the final OOB error rate dropped to 22.38% (SD = 3.21%), which class-specific error rates of 9.1% for HCR and 30.0% for LCR. This represents an improvement over the CASA-only model, particularly in accurately classifying HCR boars.

As with the previous model, the ROC curve for predicted probabilities again produced an AUC of 1.00, confirming perfect discrimination within the training data. However, the absence of cross-validation or external testing reinforces that this high AUC is likely a result of overfitting, rather than a generalizable predictive performance. These findings should therefore be interpreted with caution and viewed primarily as proof-of-concept (Table 5).

Discussion

The presence of subfertile boars in commercial breeding herds continues to undermine reproductive efficiency and cause significant economic losses. Current semen quality estimates such as motility and morphology lack the predictive power needed to identify boars with reduced fertility traits (eg conception rate and litter size) prior to their use in breeding programs. Advances in high-throughput technologies have led to the identification of promising predictors of fertility, including capacitation status markers and omics-based biomarkers. However, integration of these technologies into commercial practice remains several years away and does not address the immediate need for practical tools. Developing fertility prediction assays that can complement current semen assessments and bridge the gap between today’s evaluations and tomorrow’s high-throughput screenings therefore remains a critical industry priority.

The sperm mobility assay, originally developed and validated in poultry (Froman and McLean 1996), is one such approach. In poultry, mobility values independently determine fertility status within minutes, with higher mobility values consistently linked to improved reproductive outcomes. The assay is based on a simple and objective measurement of sperm penetration through a density gradient medium (Accudenz^®), which is mechanistically distinct from motility demonstrating that it offers complementary information rather than simply refining motility measures. Vizcarra and Ford (2006) subsequently adapted the assay for use in boars, demonstrating feasibility but not establishing fertility correlations due to the absence of accessible fertility records at that time. The present study extends this work by evaluating sperm mobility in relation to known conception rate outcomes, providing the first assessment of its predictive value in swine.

From a practical standpoint, the assay requires minimal infrastructure such as a photometer capable of measuring absorbance, disposable cuvettes, a pipette, and a means to incubate samples (eg water bath, slide warmer with warming block, or incubator). The cuvettes containing Accudenz^® must be pre-warmed for at least 30 min prior to the assay, but this step can be completed in advance at the start of the workday. Once warmed, the assay can be run as animals are collected throughout the day. Turnaround time for the assay is rapid and results are generated after 5 min, enabling same-day decision making. The cost per test is minimal once reagents are prepared, as the assay consumes only small volumes of Accudenz^® and uses standard laboratory disposables. The primary expense lies in buffer preparation, which could be streamlined in the future through substitution with commercial buffers manufactured under controlled osmolality. Notably, Animal Reproduction Systems, Inc., markets a complete standalone mobility kit for poultry containing buffers, pre-filled cuvettes with Accudenz^®, a warming block, and a densimeter and its existence underscores the feasibility of commercializing the mobility assay for industry integration. The training requirement is also modest. A technician already familiar with routine semen handling and pipetting can be trained to perform the assay reliably with only brief instruction, as the procedure is based on straightforward steps (pipetting, incubation, and absorbance measurement) and produces objective, instrument-based readouts. Given its speed, simplicity, and relatively low training threshold, the mobility assay represents a semi-ready tool with strong potential to bridge the gap between current semen quality evaluations and the high-throughput fertility screens of the future.

Differences in modified assay versus Vizcarra and Ford (2006) assay performance

While the mobility assay offers a simple, low-input option that can be integrated into boar stud workflows, its performance may be influenced by stud-specific management practices and fertility baselines, making it important to evaluate outcomes across different production environments. In Experiment 1, absorbances in the present study overall were lower than observed by Vizcarra and Ford (2006). While they used a 550 nm spectrophotometer to take their mobility readings, ours were taken on a 530 nm spectrophotometer that is regularly used for poultry sperm mobility readings for experiments conducted at BARC. Though the difference between these wavelengths is minor, it may still contribute to the reduced absorbance in our assay.

Another key difference is Vizcarra and Ford used a sperm concentration of 5 × 10⁷ viable cells/mL confirmed after 24 h of storage using fluorometric reassessment—essentially mimicking what a freshly collected and extended sample would read on the mobility assay. Knowing that samples from boars with fertility data would have to be shipped overnight, we evaluated the assay’s predictive capabilities based on extended breeding dose concentration after transport and 24 h in 17 °C storage. Time in storage reduces parameters that would impact mobility like viability, morphology, and motility (Lange-Consiglio et al. 2013). Therefore, a small pilot study was conducted using 3 commercial Duroc boars sourced from the Stud in Experiment 1 (NUTRIXcell+) to evaluate the reduction in mobility readings taken on the day of collection (approximately 8 h at 17 °C between processing and transport to BARC) and readings 24 h post-storage. Overall, mobility readings taken on the day of collection were higher in both 1.5 mL Accudenz^® concentrations compared to mobility readings taken at 24 h post-storage, but not different for those using the 3 mL of Accudenz^®, indicating the assay at that volume would be unable to detect shifts in viability. For 1.5 mL of 3% and 6% Accudenz^®, mobility readings taken on the day of collection produced a maximum mobility reading of 0.312 at 15 min (0.210 at 10, 24 h) and 0.205 at 25 min (0.130 at 20, 24 h), respectively. Due to the magnitude of change in absorbance values using 1.5 mL of 3% Accudenz^®, we felt the assay would be more sensitive to reductions in viability and motility over time. The reduction in mobility readings taken 24 h following storage is consistent with other literature where sperm cell viability and motility are reduced after 24 h in storage (Frydrychová et al. 2010; Becherer et al. 2014), further supporting that the 1.5 mL of 3% Accudenz^® mobility assay is more likely to reflect changes in viability and motility. The reduction in mobility readings pose an interesting question regarding current semen quality assurance (QA) assessment timepoints. Breeding doses are not evaluated for semen quality parameters upon arrival to the sow farm, often remaining in 17 °C storage for 3-7 days before breeding. Therefore, it may be advantageous to compare mobility values taken on day of breeding to determine a benefit, if any, for sow farm testing. Further studies outlining the relationship between testing timepoint, concentration of viable sperm cells confirmed via fluorometric detection, and fertility outcomes should be evaluated.

In another interesting observation, mobility readings from samples extended to 5 × 10⁷ cells/mL measured on day of collection in our 3-boar pilot study were similar to those from the mobility readings of 1 × 10⁷ viable cells/mL in Vizcarra and Ford’s work when the assay was conducted over 3 and 1.5 mL of 6% Accudenz^®. It is also important to note that there has been almost 20 years’ worth of genetic progress, feeding, and management practices since the previous study was published and could all be reasons pointing towards the need to “reduce resistance”. As a result, we may have indirectly selected for reduced mobility and the implications of this should be explored considering the lowest mobility readings were reflective of the larger LCR cohort, potentially pointing to a reduced ability to navigate the female reproductive tract.

Another variable in the present study that impacts viability and semen quality following storage is the type of semen extender used between Vizcarra and Ford, Experiment 1, and Experiment 2 (Stud A vs Stud B). Vizcarra and Ford used a long-term extender with bovine serum albumin (BSA) (Androhep^®, Minitube, Verona, WI, USA), an ingredient known for preventing damage from cold storage and protecting against the dilution effect (Weitze 1991; Gadea 2003) which could have been another reason for higher mobility readings in general compared to the present study where we used long-term, non-BSA extenders in Experiment 1 (NUTRIXcell+), Stud A in Experiment 2 (Androstar^® Plus), and Stud B in Experiment 2 (Preserv^® Xtreme).

Studies that directly compared the percentage of viable sperm cells in ejaculates extended in several commercially available products were not different between extender types after 24 h of 17 °C storage (Frydrychová et al. 2010). The average reduction in viability from fresh to 24 h after storage was around 10% and consistent across all extenders regardless of brand or presence of BSA. A comparison of the capacitation responses of boar sperm extended in Beltsville Thawing Solution (BTS) vs. Androstar^® Plus revealed no differences when stored at 17 °C for 24 h in membrane integrity, motility, straight-line velocity, or response to capacitating conditions (Schmid et al. 2013). Therefore, it was assumed that mobility readings at 24 h after 17 °C storage would remain relatively consistent regardless of extender brand and anticipated a difference in mobility readings between Experiment 1 and Experiment 2 due to final breeding concentration of 3.5 billion vs 2 billion cells, respectively. Interestingly, samples extended in NUTRIXcell+ and Androstar^® Plus had similar mobility reading patterns over time and samples extended in Preserv^® Xtreme had lower mobility readings overall. These findings could indicate one of two things, NUTRIXcell+ and Androstar^® Plus maintain factors that influence mobility readings similarly regardless of total cell concentration or the concentration of viable sperm cells in Stud B was lower compared to the others. In a study evaluating extender storage performance over time, Xcell (IMV Technologies, Osseo, MN USA) and Androstar^® Plus performed similarly when evaluated for capacitation status, acrosomal and membrane integrity, and motility, over the course of 1 to 15 days in 17 °C storage (Becherer et al., 2014). However, their findings were consistent with Frydrychová et al. (2010) where major differences in semen quality amongst brands and extender-type were observed much later (∼11 days). Taken together, semen quality is relatively unaffected for the first 24 h of 17 °C storage regardless of extender type.

Observations taken on the CASA on the day of the mobility assay revealed that CASA measurements are not as strongly related to conception rate when compared to mobility readings at 5 min. This further suggests that the mobility assay likely encompasses viability which is the basis of the poultry and boar mobility assay (Froman and McLean 1996; Vizcarra and Ford 2006). While final breeding dose concentration is different between the studs in Experiment 1 and 2 due to collection differences, high conception rates should produce ejaculates with comparable proportions of viable sperm cells capable of fertilization, except for subfertile males. Therefore, we only anticipated lower mobility readings for LCR boars like we would see in poultry with low fertility (King et al. 2000b; Froman 2006; Bond et al. 2024). Semen extended in NUTRIXcell+ and AndroStar^® Plus produced similar absorbance patterns, despite differences in final cell concentrations, genetic line, and stud source. In contrast, HCR semen extended in Preserv^® Xtreme exhibited a delayed overall peak in absorbance, but maintained similar absorbance to HCR in Stud A and yielded the greatest separation of high and low conception rate (HCR and LCR) within stud. Although this separation indicates Preserv^® Xtreme would be an ideal extender for separating mobility for HCR and LCR boars, mobility reading differences are unlikely related to extender alone due to the mobility parallels between HCR in Stud A and Stud B.

Sperm mobility, kinematics, and fertility phenotype

Although the sperm mobility assay is interpreted as a proxy for fertility in poultry, our findings suggest it functions more precisely as a composite measure of motility kinematics weighted by viability. Vizcarra and Ford (2006) parameterized the boar sperm mobility assay against viable sperm cell concentration using ethidium bromide (EtBr) exclusion across three different concentrations: 1 × 10⁸, 5 × 10⁷, and 1 × 10⁷ viable cells/mL and adopting 5 × 10⁷ viable cells/mL with a 5-min read as their standard condition. Establishing the assay based on EtBr anchored its signal to the membrane-intact fraction of sperm cells rather than the full spectrum of functional states that would be present in breeding doses. While concentration of viable sperm cells is related to fertility it is not the entire story. In order to successfully fertilize the egg, a viable sperm cell must complete capacitation to detect and bind to the zona pellucida of the oocyte (Kerns et al. 2020; Keller and Kerns 2023). It has also been shown that the capacity of the sperm cell to complete capacitation has been associated with boar fertility traits (Kerns et al. 2018). In our dataset, 5-min mobility values tracked differences in CASA kinematic profiles, indicating that the assay preferentially reflects the performance of cells that both remain membrane-competent and sustain coordinated movement which could explain why there was only a modest relationship to CR.

We did not co-measure fluorometric markers in this study. Therefore, we cannot ascribe specific cellular mechanisms or processes like capacitation status, acrosomal status, or mitochondrial membrane potential to the observed mobility differences. Nevertheless, our kinematic patterns are concordant with prior reports in which CASA outputs covary with various fluorometric measures of membrane and acrosomal status. At 5-min, mobility correlated negatively with ALH, VAP, and VCL, and positively with LIN, WOB, and BCF. External evidence indicates that capacitation conditions shift kinematics toward higher ALH, WOB, LIN, BCF, and increased VCL and VAP relative to sperm in non-capacitating media (Serrano et al. 2020). In the same study, a higher proportion of PNA+/PNA- cells were compatible with acrosome reaction after 1 h of incubation in capacitation media. Another study combining CASA with FITC-PNA/PI/JC-1 staining have described kinematics related to destabilized sperm membranes, membrane potential, and dead cells during time in 17 °C storage (Lange-Consiglio et al. 2013). Thus, the increase in ALH and VCL and reduction in LIN and WOB in our LCR boars from Stud B could reflect a subpopulation of destabilized sperm that, while still motile, had reduced fertilizing capacity.

Another potential application for the mobility assay lies in further breeding dose QA after storage or transport. After semen collection, there are opportunities for the capacitation process or concentration of viable sperm to be influenced by factors external to individual boar fertility. Previous works have shown that 11% of boar fertility variation can be attributed to laboratory technician and 7% to the AI center using microscopic evaluation of semen quality, but variation due to AI center was removed when CASA was implemented (Broekhuijse et al. 2012b, 2012c). Although both studs in the present study utilized CASA to evaluate semen quality parameters which should remove AI center variation, we observed a strong effect of stud on mobility values with average fertility lower for Stud B than Stud A. With ejaculates meeting quality estimates during initial processing, low mobility reflective of kinematic patterns in Stud B could be the result of damage incurred during storage or transport. Purebred boars, specifically Durocs, are more sensitive to membrane remodeling and damage during 17 °C storage when compared to crossbred boars (Duroc x Pietrain) (Wysokińska and Szablicka 2021). Further, another factor to consider is bacteriospermia in extended doses which decreases dose longevity in storage and increases susceptibility to premature acrosome reaction and damage (Althouse et al. 2000; Althouse and Lu 2005; Kuster and Althouse 2016; Waberski et al. 2019; Tvrdá et al. 2022). Common opportunistic pathogens like Pseudomonas aeruginosa can have little to no effect of motility after 24-48 h in storage, but can quickly decrease viability depending on starting concentration (Sepúlveda et al. 2014; Ngo et al. 2023). Typical QA practices require a minimum of 24 h post-processing followed by microbial screening with agar plates where contamination is confirmed by bacterial growth (Althouse 2024). In the present study we did not culture breeding doses to determine if bacterial load was different between studs. However, if thresholds for pathogenic bacterial levels could be established, the assay would become a valuable tool for AI centers, enabling faster and more informed management decisions. In all contexts, the mobility assay may indirectly capture such physiological differences, serving as a functional proxy for viability or capacitation status regardless of cause. Our findings warrant future studies outlining the effects of time in storage, storage temperature, bacterial load, and differing transport conditions on mobility values as they relate to sperm function using flow cytometry-based assays to validate the physiological basis of the signal.

Predictive performance of the mobility assay and model limitations

Although these physiological associations provide a strong biological rationale, the predictive strength of the mobility assay ultimately depends on how consistently it can discriminate fertility groups, which we evaluated through model-based approaches. Our preliminary results support the use of 5-min absorbance as the most informative timepoint for distinguishing between HCR and LCR boars. In binomial logistic regression evaluating all boars, the model produced predicted probabilities ≥0.5 when mobility readings were below 0.200. However, it is important to note that the overall discriminatory power of the models was limited by sample size, particularly the distribution of LCR boars across studs. As such, model strength was primarily driven by the HCR mobility values across studs and the LCR cohort in Stud B. Within Stud A, three LCR boars displayed conception rates just below the 75% conception rate selection cutoff (74%, 73%, and 69%) and yielded mobility values similar to HCR boars, suggesting that the assay may not be sensitive enough to distinguish fertility differences when conception rate is relatively high (≥70%). Future work should therefore include boars with conception rates at the lower extreme to better evaluate sensitivity and discriminatory thresholds.

To assess whether mobility values could improve the predictive ability of CASA parameters, random forest classifiers were utilized using different predictor variables. Models using CASA parameters alone had relatively poor predictive performance, misclassifying nearly half of the samples. Even when low-importance parameters were removed, the classifier still misidentified one-third of the cohort. Interestingly, the random forest classifier never identified motility or morphology as a variable of importance out of all 55 CASA parameters measured when it was trying to categorize boars as high or low conception rate. However, 5-min mobility and specific kinematics consistently emerged as a top variable of importance and reflects our correlation coefficients. Although all models produced AUC values of 1.00 indicating a perfect model, this reflects overfitting due to the small dataset size rather than perfect classification. These results should therefore be interpreted as proof-of-concept, demonstrating the possibility that inclusion of mobility data could strengthen CASA-based models, but requiring validation in larger populations with more balanced fertility phenotypes.

Conclusion

This pilot evaluation provides the first evidence that the sperm mobility assay, adapted for extended boar semen doses, can partially distinguish between high and low fertility groups and reduce model classification error when integrated with CASA parameters. Importantly, the strength of the models was limited by small sample size, stud-specific effects, and potential overfitting in machine learning outputs. These results should therefore be interpreted as proof-of-concept rather than as a validated diagnostic test. Future studies must include larger, more diverse populations across breeds, studs, and extender types to establish thresholds, assess robustness, and determine the assay’s reliability under real-world commercial conditions. Only after such validation can sperm mobility be considered a practical addition to fertility prediction and semen quality assurance pipelines.

Funding

This research was funded by the in-house USDA-ARS CRIS project number 8042-31000-111-00D to Kayla Mills.

Author contributions

Kayla Mills (Conceptualization, Formal Analysis, Investigation, Methodology, Project Administration, Supervision, Visualization, Writing—original draft), Amanda Minton (Resources, Writing—review & editing), J.M. Magee (Resources, Writing—review & editing), and Julie A. Long (Methodology, Writing—review & editing)

Conflicts of interest

None declared.