Developing a tool to optimize research on antioxidants for rooster semen cryopreservation

Abstract

This study aimed to develop a tool for predicting the potential impact of research studies involving the effect of antioxidants in rooster semen freezing diluent, depending on the variables that have been studied. To achieve this, a comprehensive meta-analysis of fifty-eight research documents was performed. Sixty-two traits were sorted into four major categories: study demographics, study design-related parameters, rooster semen quality-related parameters, and fertility level indicators. The quartile determination of each research document was collected from the Journal Citation Reports database. After twenty-five testing rounds, all variables that showed multicollinearity problems were discarded from further analyses (VIF < 5). HOST, pellet volume, mass motility, light hours, and sperm concentration were the most influential traits for the classification of papers in different quartiles (Wilks’ lambda: 0.797, 0.891, 0.895, 0.896, and 0.904, respectively). The research was validated as reported in the cross-validation analysis, with 93.60% of papers being correctly classified within their group. The present research assists researchers not only in the decision-making process for journals in which to publish the outcomes of their studies, but also to seek for the inclusion of parameters which attract a wider interest for the matter from scientific readers. This leads to the optimisation of resources in studies evaluating the effect of antioxidants in poultry reproduction by identifying the most scientifically relevant variables and those which in trun will lead toa greater impact on research publications.

HIGHLIGHTS

• Using the discriminant canonical analysis, the scientific community can know which variables to take into account to achieve a greater impact in the cryopreservation studies of rooster sperm.

• The study of membrane functionality (HOST) could be a surrogate for in vivo fertility assessment as it predicts the survival rate of spermatozoa in the female reproductive tract.

• Pellet volume, mass motility, light hours for sperm-donor animals, and semen concentration have a higher discriminatory capacity for studies in different quartiles of the JCR database.

Keywords:

• Antioxidant effect
• research documents
• preservation of genetic material
• spermatozoa
• rooster

Introduction

Artificial insemination (AI) is a widely and successfully applied technique in the poultry industry since the first recorded use of hen intravaginal insemination in 1936 (Quinn and Burrows 1936). The use of cryopreserved semen in AI offers certain advantages over fresh semen, such as long-distance transportability and applicability across various farms and females (Siudzińska and Łukaszewicz 2008). Additionally, AI helps reduce disease transmission issues, serves as a crucial tool during periods of male sexual inactivity, and enables the dissemination of genetics from high-value individuals that are no longer alive (Gee et al. 2004).

However, when it comes to avian species such as chicken (Nakamura 2017) turkey (Woelders 2021), among others and, the fertility rate after AI with frozen semen (ranging from 0.8 to 2.1%) is significantly lower compared to other domestic species (from 30 to 80%) (Asano and Tajima 2017). This is due to the high content of polyunsaturated fatty acids (PUFAs), specifically docosatetraenoic (22:4n-6) and arachidonic (20:4n6) acids, present in avian spermatozoa (Surai et al. 2001; Long 2006; Longobardi et al. 2017). These fatty acids are prone to increased lipid peroxidation in the presence of reactive oxygen species (ROS), resulting in oxidative stress that adversely affects sperm functionality and ultimately impacts fertility (Aziz et al. 2004).

To address this issue, advancements in the sperm cryopreservation process are necessary. Many studies have focused on testing the supplementation of cryopreservation extenders with antioxidants to mitigate the damage caused by ROS production (Partyka and Niżański 2021). However, the inclusion of specific parameters in these studies can influence their impact on the scientific community. While certain traits related to rooster sperm evaluation, such as the number and age of sperm donor males, cryopreservation diluent composition, and sperm motility evaluation parameters, are typically included, other variables such as fertility and hatchability rates require more technical effort and infrastructure (Mehdipour, Kia, Najafi, et al. 2020; Mehdipour, Kia, Martínez-Pastor 2020; Thananurak et al. 2020; Valderrama et al. 2020).

Moreover, the cost and expertise required for evaluating seminal quality parameters, such as those assessed by flow cytometry, pose additional limitations (González Ariza et al. 2022; Brillard 2003; Aziz et al. 2005). Germplasm banks play a crucial role in preserving native poultry breeds, but studies focused on sperm conservation of local populations often face limitations in terms of infrastructure and animal availability due to budget constraints (Wolniak et al. 2017; González Ariza, Arando Arbulu, Navas González, Nogales Baena, et al. 2021; González Ariza, Navas González, Arando Arbulu, et al. 2022; Santiago-Moreno and Blesbois 2022).

To assess research impact, investigators utilise the Journal Impact Factor (JIF), a quantitative measurement tool widely used since its creation in 1955 (Garfield 2006). Grouping journals by quartiles simplifies the classification process. The quartiles range from the first quartile (Q1), representing journals with a JIF value in the top 25% for a given category, to the fourth quartile (Q4), encompassing journals with a JIF value in the lowest 25% for the same category (Liu et al. 2016). Although not about the improvement of rooster sperm cryopreservation, understanding the publication trends across different quartiles can provide valuable insights into research impact (Miranda and Garcia-Carpintero 2019; Valderrama et al. 2019).

The intricate study of sperm physiology and its multifaceted factors plays a pivotal role in shaping the landscape of reproductive science. Understanding the complexities inherent in sperm function is crucial for the development of meta-analysis studies aimed at unravelling the intricate web of variables influencing cryopreservation outcomes. Meta-analyses provide a comprehensive overview by synthesising data from multiple studies, allowing researchers to discern patterns and trends across diverse experiments. By delving into the intricacies of sperm physiology, researchers can identify key determinants that significantly impact the success of cryopreservation studies. These meta-analyses not only consolidate existing knowledge but also guide future research endeavours, enhancing the overall quality and success of publications in the field. Consequently, a nuanced comprehension of sperm physiology-related factors becomes indispensable in steering the course of scientific investigations focused on optimising cryopreservation techniques and bolstering advancements in reproductive medicine.

Given these considerations, it is crucial to aid researchers in selecting parameters from their studies and having more impact on the event of eventual publication. That is to determine how the inclusion of specific factors in studies involving the effects of antioxidants on cryopreserved rooster sperm can influence the research impact of papers dealing with the matter. This study aims to identify the most decisive parameters involved in the tailoring process of rooster semen cryopreservation studies to achieve a higher impact in the scientific community, using the JIF quartile as a reference. The results will aid researchers in selecting parameters to prepare their scientific studies involving the use of extenders in endangered breeds, and eventually achieve higher impacts on the publication of studies, but not about improvement the rooster sperm cryopreservation.

Materials and methods

Research strategy

Data collection was carried out in April 2022, following the premises described in previous research works (McLean and Gonzalez 2018; Iglesias Pastrana et al. 2020). Google Scholar (https://scholar.google.com) and Scopus (https://www.scopus.com) were the databases used for the search. However, the repositories PubMed (https://pubmed.ncbi.nlm.nih.gov) and Web of Science (www.webofscience.com/wos/woscc/basic-search) were not used since they do not permit data extraction for further assessment and public filters can be implemented are exhaustive enough to perform the proper analysis required for this retrospective observational longitudinal study over the period of time from October 2010 to April 2022, both inclusive. This period was chosen as October 2010 were the month and year when the oldest document in the topic was published and April 2022 when the research sampling process was stopped.

The keywords used were: antioxidant, rooster, semen, cryopreservation, and potential words comprised in their etymological fields. All the aforementioned keywords were searched for on all the databases. No limit in terms of language or date were set. The quartile determination of each research document was collected from the Journal Citation Reports database (JCR Clarivate; https://jcr.clarivate.com) and Scientific Journal Rankings (Scimago; https://www.scimagojr.com/journalrank.php). Papers were selected upon their relevance to the use of antioxidants in extenders for cryopreservation of rooster semen and further quality filters were not applied, given factor such as those related to indexing were registered as variables within the database.

The selected documents were incorporated into a database that included individual records for each article. Each record comprised details of the major categories of study demographics, study design-related parameters, rooster semen quality-related parameters, and fertility level indicators to determine the clustering patterns that the aforementioned factors define across JIF quartiles as clustering criteria. Traits evaluated in the documents were sorted into categories, as described in Supplementary Table S1. Major and minor categories, parameters, and definitions of each parameter registered per research document considered an used in the present study is reported in Supplementary Table S2. The JIF for each journal and year was recorded by referencing the Journal Citation Reports on the Web of Science. For articles published in non-indexed journals during their respective publication years, a value of zero was assigned in the database. As a result of the review process, 58 publications were selected to be introduced in the statistical analysis.

Data analysis

Testing differences in article number across quartiles

Shapiro–Wilk Francia’s W test routine of the Test and distribution graphics package of the Stata version 15.0 software process was used to test the normality. Levene’s test to test variance homogeny of the SPSS Statistics for Windows statistical program, version 24.0 was used to assess homoscedasticity. Following the analysis, it was observed that common parametric assumptions were violated (p > .05), suggesting that the data should follow a non-parametric approach. One-Sample Chi-Square also called the goodness-of-fit chi-square test was performed using the One sample routine of the Non-Parametric Package of the statistical software Statistical Package for the Social Sciences (SPSS Statistics) for Windows (Version 26.0, 2017, IBM Corp., Armonk, NY, USA) to determine the existence of differences in the number of papers dealing with antioxidants applied in reproduction in poultry across JIF quartiles.

Discriminant canonical analysis

The discriminant canonical analysis described in the present study was performed to develop a tool that evaluates linear combinations of the major categories of study demographics, study design-related parameters, rooster semen quality-related parameters, and fertility level indicators to determine the clustering patterns that the aforementioned factors define across JIF quartiles as clustering criteria. Traits evaluated in the documents were sorted into categories, as described in Supplementary Table S1. The explanatory variables considered for the discriminant canonical analyses were the following: rooster age, male breed, number of males, strain type, total females, female age, female per treatment, female breed, light hours, collection frequency, total ejaculates, replicates number, refrigeration duration, equilibration temperature, equilibration duration, antioxidant concentration, cryoprotectant, cryoprotectant concentration, osmotic pressure, antioxidant, pH extender, rack height, sperm concentration, sperm concentration per straw, pellet volume, thawing temperature, thawing duration, normal morphology, HOST (Hypo-osmotic swelling test), mass motility, subjective total motility, subjective progressive motility, TM (Total motility), PM (Progressive motility), VAP (Average path velocity), VCL (Curvilinear velocity), VSL (Straight line velocity), LIN (Linearity), BCF (Beat cross-frequency), STR (Straightness), ALH (Amplitude of lateral head displacement), WOB (Wobble), viability, acrosome integrity, DNA fragmentation, LPO (Lipid peroxidation), ROS (Reactive oxygen species), HMMP (High mitochondrial membrane potential), live, dead, apoptosis, SOD (Superoxide dismutase activity), GSH (Glutathione peroxidase activity), CAT (Catalase activity), ATP (Adenosine Triphosphate), TAC (Total antioxidant capacity), AI volume per female, hatchability, AI hour, AI procedure, fertility, AI concentration sperm per female. All the explanatory variables are described in Supplementary Table S2.

Regularised forward stepwise multinomial logistic regression algorithms were used to perform the canonical variable selection. Priors were regularised according to the group sizes calculated using the prior probability of SPSS Statistics software for Windows (Version 26.0, 2017, IBM Corp., Armonk, NY, USA), instead of considering them the same, to avoid groups with different sample sizes affecting the quality of the classification (Tai and Pan 2007; González Ariza, Arando Arbulu, León Jurado, et al. 2021).

A minimum sample size of at least 20 observations for every 4 or 5 predictors, and the maximum number of independent explanatory variables of n-2 (where n is the observational sample size), should be inserted in the analysis so that possible distortion effects are palliated (González Ariza, Arando Arbulu, Navas González, Delgado Bermejo, et al. 2021). Initially, the observational sample used in this study did not overcome the minimum reported in the literature for statistical analysis outcomes to be robust (n = 344 observations against 62 independent explanatory variables were considered). For this reason, to ensure the independence of the explanatory variables and to discard redundant explanatory variables in order to reach appropriate ratios in agreement with the number of observational sample units, a multicollinearity analysis was run before the discriminant analysis. The variables chosen by the forward or backward stepwise selection methods were the same. In this context, a progressive forward selection method was run, given it is more time efficient than the backward selection method (González Ariza, Arando Arbulu, León Jurado, et al. 2021).

The Discriminant routine of the Classify package of SPSS Statistics software for Windows (Version 26.0, 2017, IBM Corp., Armonk, NY, USA) and the Discriminant Canonical Analysis routine of the Analysing data package of XLSTAT software (Addinsoft Pearson Edition 2014, Addinsoft, Paris, France) were used to perform the discriminant canonical analysis and multicollinearity preliminary testing.

Multicollinearity preliminary testing

A subroutine of the Discriminant Canonical Analysis routine of the Analysing data package of XLSTAT software (Addinsoft Pearson Edition 2014, Addinsoft, Paris, France) was used to compute the variance inflation factor (VIF), using the following formula: $$\text{VIF}=\text{1}/1-R^2$$

 Where R² is the coefficient of determination of the regression equation.

The discriminant canonical analysis must be run after the multicollinearity assumption has been tested. Thus, redundancies in the variables do not overinflate the explanatory potential of the variance.

To detect multicollinearity, VIF has been used as the most common indicator. For this, a maximum VIF value of 5 has been used in the present study following the premises used in previous studies (Rogerson 2001; González Ariza, Arando Arbulu, León Jurado, et al. 2021; González Ariza, Arando Arbulu, Navas González, Delgado Bermejo, et al. 2021). The tolerance is the proportion of variability in a certain independent variable that is not accounted for by other variables in a discriminant function and is equal to 1 – R² (Navas et al. 2021; González Ariza, Arando Arbulu, et al. 2022). Generally, multicollinearity problems might exist when a tolerance below 0.20 is obtained (Nanda et al. 2018).

Discriminant canonical analysis preliminary assumptions: efficiency and reliability

Wilks’ Lambda test (Rao’s approximation) was used to evaluate the variables that significantly contribute to the discriminant function. As Wilks’s lambda approaches 0, the variable’s contribution to the discriminant function increases. The functions can be used to explain group adscription if the significance value is ≤ 0.05 (Anuthama et al. 2011).

The assumption of equal covariance matrices was tested in the discriminant function analysis by Pillai’s trace criterion (Zhang et al. 2020). This is the only acceptable test to be used in cases of unequal sample sizes. It was computed as a subroutine of the Discriminant Canonical Analysis routine of the Analysing data package of XLSTAT software (Addinsoft Pearson Edition 2014, Addinsoft, Paris, France). A significance value of p ≤ .05 indicates that the set of predictors considered in the discriminant model is statistically significant; hence, the application of the discriminant canonical analysis is feasible.

Discriminant potential based on quartile differences in the mean: independent factors Wilk’s Lambda

As in multiple regression, the residuals (observed minus predicted values) should follow a normal distribution. Hence, the unidimensional test of equality of the means of the classes to test for difference in the means across quartiles once redundant variables have been removed. In this manner, those variables whose discriminant potential relies on the existing differences across quartiles were revealed. A better-discriminating power is indicated by greater values of F and consequently, lower values of Wilks’ Lambda. It was computed as a subroutine of the Discriminant Canonical Analysis routine of the Analysing data package of XLSTAT software (Addinsoft Pearson Edition 2014, Addinsoft, Paris, France).

Discriminant potential based on correct classification ability: standardised canonical coefficient, loading interpretation, and spatial representation

Afterward, once those variables whose discriminant potential relied on mean differences across quartiles, discriminant function analysis was used to identify those whose discriminant potential may itself rely on their capacity to determine greater percentages of allocation of an observation within its group (defined by the quartile in which the journal where the article comprising each sample unit observation was ranked). To perform discriminant analysis, the predictor variables need to be Standardised to ensure that the variables are on the same scale and that their relative importance is not affected by their original units of measurement. Values of ≥ |0.40| for the discriminant loading of the standardised coefficients of a certain variable can be considered to be substantially discriminating variables. The inclusion of non-discriminant (redundant) variables in the function was prevented using a stepwise procedure technique. Large absolute values in the loadings for standardised coefficients for each variable lead to greater discriminating ability and correct classification percentage.

Discriminant function cross-validation

Upon analysing the overall confusion matrix for cross-validation results, the classification model demonstrated an accuracy of 82.6%, indicating that it made correct predictions for the majority of instances. The sensitivity, representing the model’s ability to detect positive instances, yielded an overall value of 82.6%. The specificity, reflecting the model’s capacity to identify negative instances accurately, reached 96.1%. These general values offer a comprehensive perspective on the model’s overall performance, suggesting a balanced ability to make correct predictions across both positive and negative instances.

The hit ratio parameter can be calculated to determine the probability that an observation of an unknown background is classified correctly in a particular group (González Ariza, Arando Arbulu, Navas González, Delgado Bermejo, et al. 2021). The leave-one-out cross-validation option is used by authors to consider if the different discriminant functions can be validated (Toalombo Vargas et al. 2019). Classification accuracy is achieved by the discriminant canonical analysis when the classification rate value is at least 25% higher than that obtained by chance.

These results obtained must be supported by Press’ Q statistic, which is a parameter that is able to compare the discriminating power of the cross-validated function by using the formula: $$\text{Press}’\text{Q}={\left[\text{n}–\left(n’K\right)\right]}^2/\left[\text{n}\left(\text{K}–\text{1}\right)\right]$$ where n is the number of observations in the sample; n’ is the number of observations correctly classified and K is the number of groups.

The value of Press’ Q statistic should be compared to the critical value of 6.63 for χ² with a degree of freedom in a significance level of 0.01. When Press’ Q exceeds the critical value of χ² = 6.63, the cross-validated classification can be considered significantly better than chance.

Results

Testing differences in article number across quartiles

Chi-Square tests for independent samples revealed significant differences in article count across quartiles (χ² = 193.724. df = 4. p < .001). The frequency distribution of articles is reported in Figure 1.

Figure 1. Frequency distribution of articles across quartiles.

Discriminant canonical analysis

Multicollinearity preliminary testing

Twenty-six rounds were needed until the set of explanatory variables considered in this study reached VIF-acceptable levels (VIF ≤ 5). After multicollinearity testing, 25 explanatory variables were deemed redundant and discarded from further analyses (Table 1). A total of 37 explanatory variables remain for the discriminant analyses (Table 2). Therefore, the present study used a relationship between observations and independent explanatory variables which was around 1.5 times higher than those described in the literature, hence a sufficient efficiency of the discriminant approaches could be presumed (Poulsen and French 2008).

Table 1. Variables discarded at preliminary multicollinearity analysis using variance inflation factor (VIF) of explanatory variables.

Explanatory variable	Multicollinearity testing round when it was discarded	VIF at discard
VCL (Curvilinear velocity)	1	405.342
VSL (Straight line velocity)	2	405.342
Antioxidant	3	369.527
AI procedure	4	216.842
Fertility	5	224.229
Dead	6	153.069
VAP (Average Path Velocity)	7	82.723
LIN (Linearity)	8	60.004
PM (Progressive Motility)	9	57.074
SOD (Superoxido Dismutase)	10	54.960
GSH (Glutathione)	11	54.960
Sperm concentration per straw	12	37.928
Equilibration temperature	13	36.176
BCF (Beat-cross frequency)	14	23.756
AI concentration sperm per female	15	19.282
STR (Straightness)	16	18.430
Apoptosis	17	13.705
TM (Total Motility)	18	10.603
Total females	19	8.802
Female per treatment	20	8.802
ALH (Amplitude of lateral head displacement)	21	7.173
Female breed	22	6.977
LPO (Lateral projection of the head)	23	6.204
Extender pH	24	5.777
Acrosome integrity	25	5.350

AI: Artificial Insemination.

Table 2. Variables that remained at preliminary multicollinearity analysis using variance inflation factor (VIF) of explanatory variables.

Statistic	Tolerance (1-R^2)	VIF (1/Tolerance)
Normal morphology	0.218	4.592
Subjective progressive motility	0.220	4.548
Mass motility	0.228	4.389
Sperm concentration	0.230	4.353
Refrigeration time	0.232	4.309
Thawing temperature	0.233	4.298
Subjective total motility	0.233	4.294
Equilibration time	0.235	4.259
Antioxidant concentration	0.241	4.150
Viability	0.241	4.146
Collection frequency	0.241	4.143
Thawing time	0.246	4.066
HOST	0.246	4.061
Pellet volume	0.254	3.933
AI volume per female	0.264	3.785
Rooster age	0.264	3.782
ATP	0.265	3.777
CAT	0.270	3.708
ROS	0.270	3.702
HMMP	0.277	3.614
Cryoprotectant	0.277	3.612
Male Breed	0.282	3.551
TAC	0.286	3.501
Female age	0.312	3.205
Hatchability	0.313	3.195
Live	0.342	2.920
Rack Height	0.347	2.878
Ligth hours	0.358	2.793
Cryoprotectant concentration	0.390	2.561
Osmotic pressure	0.391	2.556
DNA fragmentation	0.392	2.549
AI hour	0.402	2.488
Total ejaculates	0.467	2.143
WOB	0.487	2.054
Number of males	0.489	2.045
Strain type	0.000	0.000
Replicates number	0.000	0.000

HOST: Hypo-Osmotic Swelling Test; AI: Artificial Insemination; ATP: Adenosine triphosphate; CAT: Catalase; ROS: Reactive Oxygen Species; HMMP: High Mitochondrial Membrane Potential; TAC: Total Antioxidant Capacity; WOB: Waveform of the Beat.

Discriminant canonical analysis efficiency and model reliability

Wilks’ lambda test determined that the functions can be used to explain group adscription (Wilk’s Lambda = 0.117; F (obs)=8.767; F (critical)=1.256; df1 = 105; df2 = 881.309; p < .0001). Significant Pillai’s trace criterion (Pillai’s trace criterion = 1.470; F (obs)=8.121; F (critical)=1.256; df1 = 105; df2 = 888; p < .0001) determined the validity of the discriminant canonical analysis.

Significant discriminant abilities were reported for the three functions revealed after the discriminant analysis as reported in Table 3. The discriminatory power of the F1 and F2 functions was high (eigenvalue of 1.709 and 1.257 respectively; Figure 2) with 50.893% and 37.425% of the variance significantly explained, respectively, which accounted for a total of 88.317% of the total explained variability by F1 and F2 (Figure 2).

Figure 2. Eigenvalue and cumulative variability explanatory potential of independent explanatory variables.

Table 3. Wilks’ lambda test of discriminant functions.

Test of Function(s)	Wilks’ lambda	Chi-square	Df	Sig.	Eigenvalue	Discrimination (%)	Cumulative (%)
1 through 3	0.163	576.269	69	<0.001	1.709	50.893	50.893
2 through 3	0.384	303.958	44	<0.001	1.257	37.425	88.318
3	0.801	70.574	21	<0.001	0.392	11.682	100.000

Independent factor discriminant potential evaluation

The different variables studied in this research were ranked according to their discriminating ability. The unidimensional test for equality of the means of the classes to test for differences in the means across quartiles once redundant variables have been removed as shown in Table 4.

Table 4. Results for the unidimensional test of equality of the means of the classes to test for difference in the means across sample groups once redundant variables have been removed (higher F and lower wilks’ lambda values are indicative of a greater discriminatory ability).

Variable	Rank	Wilks’ Lambda	F	DF1	DF2	p-value
HOST	1	0.7970	27.7890	3	328	<.0001
Pellet volume	2	0.8910	13.4340	3	328	<.0001
Mass motility	3	0.8950	12.8760	3	328	<.0001
Ligth hours	4	0.8960	12.6950	3	328	<.0001
Sperm concentration	5	0.9040	11.5590	3	328	<.0001
Subjective progressive motility	6	0.9070	11.1480	3	328	<.0001
Antioxidant concentration	7	0.9130	10.3910	3	328	<.0001
Equilibration time	8	0.9180	9.7310	3	328	<.0001
Osmotic pressure	9	0.9250	8.8810	3	328	<.0001
Refrigeration duration	10	0.9400	6.9990	3	328	<.0001
DNA fragmentation	11	0.9410	6.8210	3	328	<.0001
Rooster age	12	0.9470	6.0870	3	328	<.0001
HMMP	13	0.9490	5.8520	3	328	.0010
Collection frequency	14	0.9500	5.7220	3	328	.0010
Viability	15	0.9550	5.1040	3	328	.0020
CAT	16	0.9580	4.7770	3	328	.0030
Live	17	0.9620	4.2930	3	328	.0050
Cryoprotectant concentration	18	0.9630	4.2070	3	328	.0060
Rack height	19	0.9700	3.3420	3	328	.0200
Total ejaculates	20	0.9750	2.8370	3	328	.0380
ATP	21	0.9760	2.7340	3	328	.0440
Female age	22	0.9780	2.4370	3	328	.0650
TAC	23	0.9780	2.4370	3	328	.0650
AI hour	24	0.9800	2.2440	3	328	.0830
Number of males	25	0.9800	2.1950	3	328	.0880
Cryoprotectant	26	0.9830	1.8440	3	328	.1390
Male breed	27	0.9850	1.6770	3	328	.1720
Subjective total motility	28	0.9860	1.5530	3	328	.2010
ROS	29	0.9860	1.5370	3	328	.2050
Normal morphology	30	0.9890	1.1840	3	328	.3160
AI volume per female	31	0.9920	0.8840	3	328	.4500
Hatchability	32	0.9930	0.7790	3	328	.5060
Thawing duration	33	0.9940	0.6260	3	328	.5990
WOB	34	0.9960	0.4030	3	328	.7510
Thawing temperature	35	0.9960	0.3910	3	328	.7600
Strain type	36	–	–	3	328	Inconclusive results
Replicates number	37	–	–	3	328	Inconclusive results

HOST: Hypo-Osmotic Swelling Test; AI: Artificial Insemination; ATP: Adenosine triphosphate; CAT: Catalase; ROS: Reactive Oxygen Species; HMMP: High Mitochondrial Membrane Potential; TAC: Total Antioxidant Capacity; WOB: Waveform of the Beat.]

The present analysis revealed that the following variables contributed significantly (p < .05) to the discriminant functions: HOST, pellet volume, mass motility, light hours, sperm concentration, subjective progressive motility, antioxidant concentration, equilibration duration, osmotic pressure, refrigeration duration, DNA fragmentation, rooster age, HMMP, collection frequency, viability, CAT, live, cryoprotectant concentration, rack height, total ejaculates, and ATP. On the contrary, female age, TAC, AI hour, number of males, cryoprotectant, male breed, subjective total motility, ROS, normal morphology, AI volume per female, hatchability, thawing duration, WOB, and thawing temperature did not contribute significantly (p > .05) to the discriminant functions. Finally, strain type and replicates number for which Tolerance values were 0, a sign of perfect multicollinearity occurring among the variables within the model. Perfect multicollinearity occurs when one of the independent variables can be expressed as a linear combination of the other independent variables. This makes the estimation of the regression coefficients impossible, as there is no unique solution. Therefore, when tolerance is equal to 0, the regression model is invalid and needs to be revised, hence, such variables did not show conclusive results (Figure 3).

Figure 3. Standardised canonical discriminant function coefficients.

Discriminant function cross-validation

Accuracy, sensitivity and specificity were calculated from the confusion matrix for cross-validation results. The computation of accuracy, sensitivity, and specificity is fundamental in assessing the overall performance of a classification model based on a confusion matrix. Accuracy, a measure of overall correctness, is calculated by dividing the sum of correct predictions (true positives and true negatives) by the total number of predictions. Sensitivity, also known as recall or the true positive rate, is determined by dividing the number of true positives by the sum of true positives and false negatives. It assesses the model’s ability to correctly identify positive instances. Specificity, referred to as the true negative rate, is calculated by dividing the number of true negatives by the sum of true negatives and false positives, measuring the model’s proficiency in correctly identifying negative instances. These metrics collectively provide insights into the model’s effectiveness in making correct predictions and discerning between positive and negative instances.

A Press’ Q value of 863.50 (n = 344; n’=322; K = 4) was computed. Thus, predictions can be considered to be better than chance at 95%. Table 5 shows the cross-validation of discriminant classification results.

Table 5. Cross-validation of classification results.

From/To	Q1	Q2	Q3	Q4	Total	% Correct
Q1	162	9	0	0	171	94.74
Q2	0	122	13	0	135	90.37
Q3	0	0	35	0	35	100.00
Q4	0	0	0	3	3	100.00
Total	162	131	48	3	344	93.60

Discussion

The sampling approach employed in this study introduced several potential limitations and biases which demanded consideration. While Google Scholar and Scopus were the chosen databases for the literature search, the exclusion of PubMed and Web of Science due to perceived data extraction restrictions may introduce selection bias by omitting relevant studies. However, Scopus, which was considered, is the source from which other databases such as Web of Science, collect their information. On the other hand, language bias is mitigated by the decision not to set language limits, allowing for a more inclusive representation of non-English literature. The use of a diverse set of keywords, including those within etymological fields, attempts to capture a comprehensive range of relevant literature and counteract potential bias. The determination of quartiles using Journal Citation Reports and Scientific Journal Rankings, which comprehensively provides a transparent evaluation of publication quality.

The improvement of semen cryopreservation techniques in roosters is a milestone for the poultry industry, allowing greater and more efficient use of cryopreserved semen in artificial insemination (AI) and providing various advantages and more efficient use of cryopreserved semen in AI with the advantages that this entails (Sun et al. 2022). In addition to its reproductive purposes, the ability to cryopreserve semen would enable the creation of reliable and effective germplasm banks for both endangered populations and highly valuable individuals in animal breeding (Sun et al. 2021). Therefore, studying the different factors involved in avian sperm cryopreservation techniques, particularly the inclusion of antioxidants in the diluent, plays a crucial role in better planning the methodologies used in research design which eventually leads to greater success in the publication outcomes in highly impacted research sources.

Chi-Square tests have revealed significant differences in the number of papers related to evaluating the effect of antioxidants added to the diluent for cryopreserving rooster sperm across Journal Impact Factor (JIF) quartiles. Historically, veterinary research studies have faced difficulties in publishing in high-impact journals due to the high variability within the field of Veterinary Science, which encompasses areas such as general medicine, biology, zootechnics, and animal management (Krauskopf et al. 2017). Production sciences in this research area often play a secondary role compared to medical studies (Choudhary et al. 2017).

This becomes more patent in those research areas in which the manner in which studies are designed mostly focus on the replication of aforedeveloped methodologies testing different extenders. Research is normally incomplete in terms of the factors that are eventually included, due to certain limitations of scientific skills to process data, personnel or budget limitations. However, contrary to the initial hypothesis and to the circumstances occurring in other production areas related research, most of the articles found in this study have been published in Q1 and Q2 journals. The use of antioxidants in the cryopreservation extender aims to improve the quality of post-thawed sperm, leading to comprehensive studies with advanced techniques being applied to control and enhance sperm quality (Dai et al. 2021). The use of more specialised techniques incurs higher economic costs, which motivates research groups to make greater efforts to gain visibility for their work.

Discriminant canonical analysis was used to detect the parameters that had the greatest influence on the publication of research documents in a given quartile. Prior to the analysis, a multicollinearity test was performed, which identified and eliminated several variables showing redundancy problems. Regarding sperm motility, most of the related variables were discarded, except for mass motility and subjective parameters, including WOB (wave, orbit, and beat). These traits are widely applied in the industrial sector as simple techniques that require only a microscope, and they have been reported to be directly related to ejaculate mass motility and fertilising capacity in the ovine species (David et al. 2015; Van de Hoek et al. 2022). These observations suggest a potential transferability to poultry species, leading to potential multicollinearity issues due to the close relationship between subjective and objective motility parameters. Previous studies on pigeon and canine species have also reported positive correlations between objective and subjective motility parameters, indicating the potential for standardising subjective motility estimation with proper training of laboratory personnel (Yeung et al. 1997; Rijsselaere et al. 2003; Vyt et al. 2004; Klimowicz et al. 2008).

In vivo tests, such as fertility and hatchability tests, provide comprehensive study designs in the field of animal reproduction, and fertility results are valuable information for breeders (Arando et al. 2020; Salgado Pardo et al. 2022). However, in the present study focusing on evaluating the effect of antioxidants, most female-related and AI parameters were discarded in the multicollinearity analysis. Fertility-related traits in poultry populations are primarily associated with semen and sperm qualities, including semen volume, sperm concentration, sperm viability, sperm motility, sperm forward progression, and sperm fertilising capacity (Rengaraj and Hong 2015). These semen quality-related traits can also encompass other external factors such as endocrine-disrupting chemicals, which can enter the organism through diet, respiration, and skin contact (Rengaraj et al. 2015).

Another variable that was discarded was apoptosis. Mitochondria play a key role in this process, as defects in the apoptotic machinery are associated with a decrease in mitochondrial membrane potential (Martin et al. 2004). Therefore, the variable HMMP, which has been retained in the analysis, may add further information to the performed analysis.

During the cryopreservation process, reactive oxygen species (ROS) are generated, leading to lipid peroxidation (LPO) of the sperm plasma membrane (Bréque et al. 2003). On the other hand, certain enzymes such as SOD, GSH, and CAT are part of the avian seminal plasma’s antioxidant system, and their function is to cope with the generated ROS (Partyka et al. 2012; Surai 2016). Hence, it makes sense that the LPO variable was discarded due to its high correlation with ROS. Similarly, SOD and GSH, as additional enzymes of the endogenous antioxidant system, would generally fulfill the same function as CAT, which remains in the analysis. CAT acts by detoxifying intracellular and extracellular H2O2 into water and oxygen, resulting in a decrease in oxidative stress (Partyka et al. 2012). The efficacy of CAT has been demonstrated in several species, while the effect of SOD is controversial (Amini et al. 2015).

Various factors can influence the quality of sperm following cryopreservation, including the type of sperm packaging (Najafi et al. 2022). In poultry species, 0.25 ml straws have generally been more commonly used than 0.5 ml straws due to the small volume of semen in each rooster, ensuring convenience for artificial insemination (AI) and semen storage (Zong et al. 2023). However, some studies have reported better quality results when using 0.5 ml straws, as they have a higher surface-to-volume ratio compared to 0.25 ml straws, allowing for more uniform freezing and thawing temperatures across the sample (Nöthling and Shuttleworth 2005; Mocé et al. 2010; Stuart et al. 2019; Najafi et al. 2022). In this study, the variable ‘sperm concentration per straw’ could have been discarded due to redundancy problems with the ‘sperm concentration per mL’ variable, as the latter is more explanatory and technically accurate.

The equilibration temperature during cryoprese vation is usually around 4-5 °C (Najafi et al. 2022; Sun et al. 2022). Different protocols can be used to add cryoprotectants before freezing, aiming to minimise interaction with sperm metabolism (Zong et al. 2023). Thus, the equilibration temperature variable was discarded due to high variance inflation factor (VIF) values, indicating multicollinearity issues. It is closely linked to the equilibration time when protocols are applied.

Although the pH of the extender can be modified by adding different concentrations of NaOH, voluntary pH modification of the extender has only been practiced in research studies, as it has been reported not to affect avian sperm quality (Saint Jalme et al. 2003). Therefore, the elimination of the extender pH variable after multicollinearity testing can be explained by the fact that the extender composition variable, which remained in the discriminant canonical analysis, provides more information than the pH value itself, as it can be calculated from the aforementioned extender composition variable.

Finally, the variable ‘acrosome integrity’ was eliminated after the multicollinearity analysis. The acrosome is a unique organelle found only in sperm cells and facilitates the penetration of the zona pellucida during the acrosome reaction, a process that involves fusion of the outer acrosomal and plasma membranes, resulting in the release of acrosomal enzymes (González et al. 2019; Kruit et al. 2022). While capacitated spermatozoa can carry out fertilisation processes, they have a limited lifespan before eventual death (Maxwell et al. 1999). Therefore, the redundancy between the closely related variables of ‘sperm viability’ and ‘acrosome integrity’ suggests that the information provided by sperm viability is sufficient.

After identifying variables with multicollinearity problems, they were eliminated from further analysis. The remaining variables were ranked based on their discriminant power. The variables with the highest discriminant power were HOST, pellet volume, mass motility, light hours, and sperm concentration. However, the discriminant canonical analysis did not provide conclusive results for the variables ‘strain type’ and ‘replicates number,’ indicating that a higher number of studies must be analysed to determine their discriminative power (González Ariza, Arando Arbulu, León Jurado, et al. 2021; González Ariza, Arando Arbulu, Navas González, Delgado Bermejo, et al. 2021).

The inclusion of the HOST variable in the study design when assessing the addition of antioxidants in the semen extender is crucial, as it is closely related to fertility, as reported in bovine species (Revell and Mrode 1994). HOST can be used as a substitute for in vivo fertility studies and measures the resistance of the sperm plasma membrane to damage during the cryopreservation process, reflecting its role in the survival and fertilising ability of sperm (Bucak et al. 2009).

Pellet volume was determined to be the second variable with the highest discriminant power. This method of semen packaging has been less widely used in poultry semen compared to straw packaging. However, it may attract attention from the scientific community as it offers advantages such as ease of application and not requiring specialised equipment. The volume used for pellet formation directly affects post-thaw sperm quality, particularly sperm motility in roosters.

Mass motility is a controversial parameter among investigators (Gillan et al. 2008; Vincent et al. 2014; David et al. 2015). It is an easily observable parameter with basic and inexpensive equipment requirements. However, it is a subjective measure and can vary depending on the technician performing the assessment. Objective systems like the CASA system assess individual spermatozoa’s movement, while mass motility assesses the overall movement of the ejaculate in 3D (Lin et al. 2019).

The variable ‘light hours’ showed good di’criminating power in the present analysis. Photoperiod is a crucial environmental factor in regulating the reproductive season of hens and roosters (Santiago‐Moreno et al. 2012). Poultry semen quality improves with increasing daylight hours, and sperm freezability varies with the season, with the best post-thaw ejaculate quality in terms of motility observed in spring (Santiago-Moreno et al. 2009; Santiago‐Moreno et al. 2012). Therefore, controlling light hours per day in studies is of great interest and may affect the impact of published papers.

Sperm concentration is a determining trait that should always be considered in study methodologies. It has a direct effect on post-thawing sperm quality. Previous studies have reported better results for motility, viability, and membrane integrity parameters when rooster semen is cryopreserved at a concentration of 400 × 10⁶ sperm/mL compared to other concentrations (Moghbeli et al. 2016). High sperm concentrations generally have a negative effect on post-thaw semen quality in various species (Peña and Linde-Forsberg 2000; Nascimento et al. 2008; Alvarez et al. 2012). Sperm concentration also influences the production of ROS during cryopreservation, which provides valuable information on post-thawing sperm quality in antioxidant evaluation studies (Alvarez et al. 2012).

Finally, the present discriminant tool enables efficient prediction of the potential quartile in which a study can be published based on the variables considered. With 93.60% of observations correctly classified within their respective groups, the scientific community can use the afore-discussed model to optimise resources in studies evaluating the use of antioxidants in rooster sperm freezing extenders by identifying the variables that will have the greatest impact on research publications.

This way, the study offers valuable insights for researchers engaged in avian sperm cryopreservation. Firstly, it emphasises the importance of comprehensive literature searches by advocating for the use of multiple databases to mitigate bias. The study also encourages inclusivity by recommending the absence of language limits in order to capture a diverse range of relevant literature. Addressing biases related to high-impact journals is another crucial aspect highlighted, urging researchers to challenge preconceptions and carefully consider factors influencing journal selection. Multicollinearity issues in the variable analysis are addressed, with a specific focus on prioritising key variables such as HOST, pellet volume, mass motility, light hours, and sperm concentration for study designs. Additionally, the study underscores the significance of parameters like photoperiod and sperm concentration in shaping study designs, acknowledging the practical implications of environmental factors. Lastly, the application of antioxidants in cryopreservation techniques is emphasised, with a call to consider economic costs and motivations for visibility in study design. Overall, the recommendations encompass a holistic approach, urging researchers to enhance inclusivity, challenge biases, address statistical considerations, and prioritise key parameters for impactful and resource-efficient studies in avian sperm cryopreservation.

Conclusions

The present discriminant method has been validated as an effective tool that allows us to understand the variables that the scientific community prioritises when tailoring rooster semen cryopreservation studies, predicting the future impact of research based on these variables. The extensive research on antioxidants in cryopreservation diluents has increased the significance of studies in the Veterinary Science field. Standardising procedures for evaluating semen motility among laboratory personnel would enable the adoption of a quick, inexpensive, and reliable technique in poultry reproduction laboratories. Consulting the multicollinearity analysis can optimise and streamline research methodologies, enhancing the effectiveness of research projects. The analysis suggests that the HOST variable can substitute for in vivo fertility studies by predicting the survival rate of spermatozoa in the female reproductive tract. Additionally, researchers should consider factors like pellet volume, mass motility, light hours for sperm-donor animals, and semen concentration, as these variables have a greater discriminatory capacity for studies in different quartiles of the JCR database.

Ethicial Approval

The present study is out of the scope of evaluation of the Ethics review board of the University of Córdoba, as it does not fall under the legislation for the protection of animals used for scientific purposes. In these regards, the national Royal Decree Law 113/2013, of February 1, which establishes the basic rules applicable to the protection of animals used in experimentation and other scientific purposes, including teaching (Directive 2010/63/EU of the European Parliament and of the Council of 22 September 2010 on the protection of animals used for scientific purposes) states that it shall apply until the animals referred to in the first section have been sacrificed, relocated or reintegrated into a suitable habitat or zootechnical system. Hence, contextually, the reasons for this exemption are that this study involves practices that are unlikely to cause pain, suffering, distress or lasting harm equivalent to or greater than that caused by the introduction of a needle in accordance with good veterinary practice, that practices were carried out for recognized zootechnical purposes and that animals were not sacrificed because the data was collected during the application of regular zootechnical procedures at the farms where the animals are housed.

Acknowledgements

This work would not have been possible if it had not been for the assistance of the Provincial Agricultural Centre of the Cordoba Provincial Council and the PAIDI AGR 218 research group.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

Additional information

Funding

The present research was carried out during the covering period of a Ramón y Cajal Post-Doctoral Contract with the reference MCIN/AEI/10.13039/501100011033 and the European Union ‘NextGenerationEU’/PRTR.