Data analyses, statistical methods

Experiment 1. Repeatability and method agreement analyses

The Bland-Altman statistical method (Bland and Altman 1986, Nagy et al. 2003a,b) was used to assess agreement between the CSB and TB stains. To investigate the repeatability and agreement of CSB/Giemsa and TB/Giemsa staining, I used altogether 30 semen samples from 10 stallions: four raw ejaculates, 5 fresh-diluted with NFDSM-Glucose extender, 5 fresh-diluted with EY-SM extender, 16 semen samples were centrifuged and diluted with an egg yolk + glycerol extender (7 after dilution, 9 after frozen/thawed were evaluated). A smear was made using each viability stain and two measurements were obtained for each method. Three hundred cells were counted on each slide and classified into five categories: intact head, intact tail and acrosome membrane (Intact); intact head, tail, damaged acrosome (IHITDA);

intact head, damaged tail (IHDT); damaged head, intact tail (DHIT); and damaged head, damaged tail, damaged acrosome (DHDTDA) (Fig. 5).

The analysis was calculated with the “intact” and the categories of the IHDT and DHDTDA merged, counting all damaged tails (DT). These two groups of cells clearly demonstrated the stained and unstained spermatozoa and were appropriate for comparing repeated measurements and the two different staining procedures.

Statistical analyses of repeatability and agreement were accomplished using Microsoft Excel 2000 software (Microsoft, Redmond, WA). To assess the repeatability of a method, we calculated the differences between pairs of 30 repeated measurements and the mean of these differences (d). As a measure of repeatability, the British Standards Institution (BSI) repeatability coefficient (British Standards Institution 1975; Petrie and Watson 1999) was calculated as twice the standard deviation (2 SD) of the differences. The differences between the repeated measurements were plotted against their average (Figs.15-16).

The mean of the differences (d) and the BSI repeatability coefficient (± 2 SD) are presented on the diagrams. We expected 95% of differences to be < 2 SD. Method agreement analysis between the two staining protocols was carried out using the same statistical method. The mean of the differences between the paired measurements on the same samples (d) was calculated to estimate the average bias of one method compared to the other. The SD of the differences had to be corrected in this case (Bland and Altman 1986).

The corrected SD of differences (corrSD) was where S₁ and S₂ were the standard deviation of differences between repeated measurements for each method separately, and SD was the standard deviation of the differences between the paired means for each method, which was approximately . We used this simplified formula for calculations. The 95% limits of agreement were calculated as d

± 2 corrSD (Bland and Altman 1986). The differences between the paired measurements were plotted against their average (Figs. 17-18). The mean of the differences (d) and the limits of agreement (± 2 SD) are presented on the diagrams.

Experiment 1. Densitometry (image analyses)

Twenty semen samples: five raw semen, five diluted with NFDSM extender, and 10 frozen using modified INRA82 extender (containing egg yolk and glycerol), from 15 different stallions were smeared and stained in parallel with TB and CSB. After viability staining, the entire procedure was the same according to the method described above. A Leica DM RA2 Microscope with a 100 x immersion objective, Leica Digital camera DC 500 and Leica IM50 1.20 image processing and archiving software were used to archive the photos. The settings were: 8 bit/channel color depth, 2600 x 2060 pixels image size, 16 shots, JPEG bitmap format. The digital images were stored for further analysis.Corel Draw 8.0; Corel Photo-Paint 8 software (Corel Corporation Ottawa, Ontario, Canada) was used for densitometry measurements. The parameters of the images were 440.3 x 348.8 mm (2600 x 2060 pixels), 24-bit RGB color, 150 dpi, JPEG bitmap format. Three digital images from each slide containing intact and “dead” cells stained with CSB or TB were saved to compare the color intensity of the stained vs. unstained tails and heads. The Magic Wand Mask tool of the Corel Photo-Paint 8 software was used to select the tail or head areas. The zoom level was 200%. For analysis of the individualdigital images, RGB histograms were drawn for each different area of spermatozoa (Figs. 11 and 12).

Figure 11. A spermatozoon with damaged head and intact tail membrane.

CSB/Giemsa staining. The unstained tail is outlined with dotted lines.

The histogram shows the RGB values of the selected area. Bar = 2 µm

Figure 12. A Spermatozoon with intact head and damaged tail membrane.

CSB/Giemsa staining. The stained tail is indicated with dotted lines.

The histogram shows the RGB values of the selected area. Bar = 2 µm

The histogram plotted the brightness value of every pixel in the selected area of the image. Values ranged from zero to 255 (from darkest to brightest), and the histogram indicated how many pixels were at each brightness level. Means of the composite RGB values of the selected area on each of the intact or damaged tails (midpiece and principal piece, at least 4000 pixels per selected area measurement) and the stained and unstained heads (without acrosome, at least 1000 pixels/measurement) were registered from each picture. Differences between means of RGB values of live vs.

dead tails and separately live vs. dead heads from each photo were used for comparing the two stains. Altogether, 120 photos were measured and 480 histograms of the total RGB value were made from the different areas. After evaluating the data for normality, the paired two-tailed T-test (SPSS 11.0. statistical analysis program, SPSS Inc. Chicago, IL) was used to compare the RGB differences between the stained and unstained tails or heads for CSB and TB staining.

In Experiment 2 paired T test statistical analysis was performed to compare the mean values of the percentages of different sperm categories in fresh, centrifuged and frozen samples using „R” software.

In Experiment 3 data (recovery rates and percentages of different cell types in the selected sperm after the 7 treatments) were arcsin transformed to achieve normality on the data and evaluated by GLM analysis of variance of SAS (SAS Inst. Inc., Cary, NC, USA). Differences among means were tested using Tukey's honestly significant difference (h. s.d.) procedure. In all cases, significance was set at p < 0.05 level. Data are presented as Least squares means and standard errors (LS means ± SE).

In Experiment 4 the evaluation of 10 subfertile stallion samples and the discussion with incorporation of previous data and observations from the stallions were interpreted in case reports. Mean values calculated from the results of the viability and morphology evaluations of stallions with good fertility (Fertile stallions), the average values, minimum requirements and the acceptable limits of the different sperm morphologic categories in fertile stallions according to the literature and to the guidelines of Hungarian standard for breeding stallion semen were considered as bases of comparison.