Testosterone and dihydrotestosterone reduce platelet activation and reactivity in older men and women

Design of the data analyses

Due to a relatively great abundance of the outcomes of the performed statistical analyses, they are presented as a hierarchy, i.e. starting from univariate and bivariate analyses (inference tests and simple correlations), primarily used to get a general idea of possible relationships, to multivariate analyses, which are employed to further explore the revealed links and to adjust for the presence of other covariates and confounders.

Comparisons between men and women

Table 1 lists selected aspects of blood morphology, as well as biochemical variables, determined in serum or plasma from men and women. It shows that numerous variables differed significantly between the sexes. The use of the bootstrap–boosted approach the analysis of covariance (ANCOVA) (performed on the Box-Cox-transformed data) revealed that men demonstrated significantly higher values for variables describing blood platelets than women. Total cholesterol and its lipoprotein fractions were reduced, while uric acid and Hcy increased in the serum or plasma taken from men. All the hallmarks of blood platelet reactivity in response to agonists, measured with the use of whole blood impedance aggregometry, were elevated in women. The plasma concentrations of androgens (T and DHT) remained significantly higher in plasma samples from men (Table 1).

Comparisons between individuals with lower and higher blood platelet reactivity

The dichotomised values of the van der Waerden normal scores of A_max cumulated through the agonists used (AA, collagen, ADP), referred to as ‘cumulative blood platelet reactivity’, were used to discriminate individuals with lower or higher blood platelet reactivity. The characteristics of blood morphology and biochemistry, and those of the functional state of blood platelets in subjects with lower or higher blood platelet reactivity, are given in Tables 2 and 3, respectively. The individuals with higher ‘cumulative blood platelet reactivity’ were characterized by very significantly elevated PLT and PCT, as well as increased MPV and P-LCR. Blood haematocrit was slightly, but significantly reduced, leukocyte count and neutrophil count remained higher (P< 0.05), whereas serum concentration of uric acid, T and DHT were reduced in this group. As expected, the blood platelet response to agonists was significantly increased in the group with higher ‘cumulative blood platelet reactivity’. Otherwise, the elevations in two surface membrane markers, P-selectin and the activated GPIIb/IIIa, appeared non-significant, although their cumulative measures (‘cumulative platelet activation’, as well as the agonist-induced ‘cumulative P-selectin expression’ and ‘cumulative GPIIb/IIIa expression’) demonstrated significantly increased values in the individuals with higher platelet reactivity (Tables 2 and 3).

Bivariate association analyses: androgens and blood morphology

The nonparametric correlation analysis was performed among the whole group of patients, including both sexes. A simple bivariate association analysis was performed, either not taking into account the possible confounders and accompanying variables, or with an adjustment for sex (partial Spearman’s correlation). While simple association analysis without adjustment demonstrated that plasma testosteronaemia was very significantly and positively associated with the variables of red blood cell morphology, including red blood cell count (R_{S LOO}= 0.336; P<< 0.0001), haemoglobin concentration (R_{S LOO} = 0.524; P<< 0.0001) and blood haematocrit (R_{S LOO} = 0.460; P<< 0.0001), the adjustment for sex revealed that partial Spearman’s correlation coefficients became considerably reduced, and were much below the level of statistical significance: RBC (R_{S LOO}= 0.020; NS), HGB (R_{S LOO} = 0.055; NS) and HCT (R_{S LOO} = 0.095; NS). Similarly, in a simple correlation analysis without adjustment, plasma levels of DHT were also significantly and positively associated with the variables related to erythrocyte morphology, including red blood cell count (R_{S LOO} = 0.320; P<< 0.0001), haemoglobin concentration (R_{S LOO} = 0.494; P<< 0.0001) and haematocrit (R_{S LOO} = 0.426; P<< 0.0001), and again, these significant associations disappeared upon adjusting for sex (R_{S LOO_RBC} = 0.005, R_{S LOO_HGB} = 0.016 and R_{S LOO_HCT} = 0.044, respectively; all NS).

On the contrary, plasma concentrations of T and DHT appeared very significantly negatively associated with the morphology of blood platelets for non-adjusted analysis. These significant relationships disappeared for all variables, but not for MPV and P-LCR, upon adjusting for sex (PLT: RS,LOO_T PLT= -0.215, PT PLT< 0.01 and RS,LOO_T PLT_sex-adjusted= -0.011, NS; RS,LOO_DHT PLT= -0.217, PDHT PLT< 0.01 and RS,LOO_DHT PLT_sex-adjusted= -0.007, NS; MPV: (RS,LOO_T MPV= -0.212, PT MPV< 0.01 and RS LOO_T MPV_sex-adjusted= -0.119, PT PLT= 0.070; RS,LOO_DHT MPV= -0.227, PDHT MPV< 0.005 and RS,LOO_DHT MPV_sex-adjusted= -0.147; PDHT MPV= 0.033; PCT: RS,LOO_T PCT= -0.305, PT PCT< 0.0001 and RS,LOO_T PCT_sex-adjusted= -0.002, NS; RS,LOO_DHT PCT= -0.311, PDHT PCT< 0.0001 and RS,LOO_DHT PCT_sex-adjusted= -0.028; NS; PDW: RS,LOO_T PDW= -0.189, PT< 0.02 and RS,LOO_T PDW_sex-adjusted= -0.085; NS; RS,LOO_DHT PDW= -0.199; PDHT PDW< 0.02 and RS,LOO_DHT PDW_sex-adjusted= -0.104, NS; P-LCR: RS,LOO_T P-LCR= -0.230; PT P-LCR< 0.005 and RS,LOO_DHT P-LCR_sex-adjusted= -0.141; PT P-LCR= 0.040; RS,LOO_DHT P-LCR= -0.246; PDHT P-LCR< 0.002 and RS,LOO_DHT_sex-adjusted= -0.170; PDHT P-LCR< 0.02).

No significant associations were found between the plasma concentrations of androgens and white blood cell morphology, or between the parameters of platelet or erythrocyte morphology and plasma concentration of E₂ (data not shown).

In addition, the variables describing blood platelet morphology demonstrated significant positive associations with selected markers of platelet reactivity. The most significantly associated were: platelet count (R_{S LOO_Amax_arachidonate}=0.284/R_{S LOO_Amax_arachidonate,sex-adjusted}=0.231, P=0.0002/0.002; R_{S LOO_Amax_collagen}=0.279/R_{S LOO_Amax_collagen,sex-adjusted}=0.231, P=0.0002/0.002; R_{S LOO_Amax_ADP}=0.402/R_{S LOO_Amax_ADP,sex-adjusted}=0.331, P<<0.0001/<<0.0001; R_{S LOO_cumulative plt reactivity}=0.354/R_{S LOO_cumulative plt reactivity,sex-adjusted}=0.294, P<<0.0001/0.0001), MPV (R_{S LOO_Amax_arachidonate}=0.142, P=0.036; R_{S LOO_Amax_ADP} =0.216/R_{S LOO_Amax_ADP,sex-adjusted}=0.216/0.159, P=0.003/0.024; R_{S LOO_cumulative plt reactivity}=0.178/ R_{S LOO_cumulative plt reactivity,sex-adjusted}=0.178/0.123, P=0.013/0.061; R_{S LOO_GPIIbIIIa_arachidonate}=0.190, P=0.011; R_{S LOO_GPIIbIIIa_collagen}=0.176, P=0.016) and the marker of large circulating platelets, P-LCR (R_{S LOO_Amax_arachidonate}=0.154, P=0.029; R_{S LOO_Amax_ADP}=0.233/R_{S LOO_Amax_ADP,sex-adjusted}=0.171, P=0.002/0.016; R_{S LOO_cumulative plt reactivity}=0.192/R_{S LOO_cumulative plt reactivity,sex-adjusted}=0.136, P=0.008/0.046; R_{S LOO_GPIIbIIIa_arachidonate}=0.201, P=0.007; R_{S LOO_GPIIbIIIa_collagen} =0.178, P = 0.016).

Associations of testosterone and dihydrotestosterone levels with selected plasma markers of atherogenesis – bivariate analyses

For the whole group, non-adjusted for confounding variables, plasma levels of T and DHT were significantly associated with blood plasma concentrations of several solutes, some of which being commonly acknowledged as pro- or anti-atherogenic factors. Negative associations were found between T or DHT and the concentrations of cholesterol subfractions, but only when the estimation was performed for both sexes analysed together (Table 4).

The plasma concentrations of both androgens, T and DHT, when both sexes were analysed together, were significantly positively associated with fasting glycaemia and homocysteinaemia, and a particularly significant positive correlation was revealed between uric acid and both androgens. Upon adjusting for sex, the significant relationships were maintained between both T and DHT level and fasting glycaemia and uric acid levels (Table 4).

In the excluded subgroup of women, a significant association between T and the concentration of uric acid was observed (R_{S LOO}= 0.304; P< 0.005 for women and R_{S LOO}= 0.411; P<< 0.0001 for the whole group), while DHT was significantly associated with the plasma concentrations of uric acid (R_{S LOO}= 0.330; P< 0.003 for women and R_{S LOO}= 0.403; P<< 0.0001 for the whole group) and Hcy (R_{S LOO}= 0.196; P< 0.02 for the whole group). These associations were enhanced by the use of the bootstrap procedure with resampling adjusted for the sample size of the overall group (n=155),, indicating significant correlations of T with uric acid, glucose and LDL-cholesterol and of DHT with uric acid and glucose (Table 4, the data with the superscript ^f).

For men, significant associations were found between plasma concentrations of T and fasting glycaemia (R_{S LOO} = 0.240; P< 0.05 and R_{S LOO}= 0.223; P< 0.003 for the whole group) and Hcy (R_{S LOO}= 0.234; P< 0.002 and R_{S LOO}= 0.234; P< 0.03 for the whole group), and also between the plasma concentration of DHT and fasting glycaemia (R_{S LOO}= 0.229; P< 0.05 and R_{S LOO}= 0.225; P< 0.002 for the whole group). Again, when a bootstrap procedure with resampling adjusted for the sample size of the overall group (n=155) was added, significant correlations were found between T and glucose or Hcy and also between DHT and glucose or Hcy (Table 4, the data with the superscript ^m).

E₂ was not significantly associated with any of the above mentioned plasma markers of atherosclerosis, either in a whole group of the examined individuals or upon adjustment for sex (not shown).

Associations of testosterone and dihydrotestosterone with the markers of platelet activation and reactivity

In the group of male and female subjects (analysed together) plasma concentrations of T and DHT remained significantly negatively associated with platelet aggregation in response to arachidonate, collagen or ADP (Table 5). Otherwise, plasma concentrations of E₂ demonstrated no significant association with the reactivity of blood platelets agonized with arachidonate, collagen or ADP (not shown).

The expression of the active form of GPII/IIIa glycoprotein (receptor of fibrinogen) on the surface membranes of non-activated (resting) circulating blood platelets and in vitro stimulated platelets were significantly negatively associated with plasma concentrations of T or DHT (Table 5), but not E₂ (not shown). Upon adjustment for sex, significant associations were still revealed between T or DHT level and blood platelet reactivity in response to AA, COL or ADP. Otherwise, for the extent of the activation of circulating platelets, the correlations were insignificant (T) or at the border of significance (DHT) (Table 5). The associations in separate subgroups of male and female subjects were further investigated using a resampling procedure with replacement adjusted for the overall sample size (n=155). In the men, the most significant associations were revealed between T and the activation of circulating platelets (P-selectin and the activated α_IIbβ₃ expression) or ADP-dependent aggregation, while circulating platelet activation (both surface antigens) and platelet aggregation dependent on AA and ADP correlated most strongly with DHT. In turn, in the women, associations were found between platelet activation (both antigens) and aggregation dependent on AA and COL (for T), or between platelet activation (both antigens) and aggregation triggered by AA, COL or ADP (for DHT) (Table5, the data with the superscripts ^m or ^f).

The association between testosterone and dihydrotestosterone and the markers of platelet activation and reactivity, adjusted for plasma markers of atherogenesis and blood morphology variables – multivariate analyses

Three different multivariate approaches were employed to better characterize the associations between plasma androgen concentrations and platelet function upon adjustment for confounders: multivariate regression, logistic regression and linear discriminant analysis.

A multivariate regression analysis was used to determine the impact of confounding/co-explanatory variables on the modulation of the association(s) between the variables describing blood platelet functioning and the plasma concentrations of androgens. The following were used as dependent variables: the van der Waerden normal scores of A_max cumulated through the agonists used (AA, collagen, ADP), referred to as ‘cumulative blood platelet reactivity’, the van der Waerden normal scores of the surface membrane antigens (P-selectin and the active form of GPIIb/IIIa) in circulating platelets, referred to as ‘cumulative platelet activation’, or the van der Waerden normal scores of the surface platelet membrane expressions of P-selectin/the active form of GPIIb/IIIa, cumulated through the used agonists (AA, collagen), referred to as ‘cumulative P-selectin/cumulative GPIIb/IIIa expression’. The set of independent variables included age and gender, T or DHT, platelet and leukocyte counts, haemoglobin, total cholesterol, glucose, uric acid and Hcy.

The multivariate regression parameters for testosterone and the co-explanatory variables for the model of ‘cumulative blood platelet reactivity’ as a dependent variable are given in Table 6. The variables significantly contributing to the ‘cumulative blood platelet reactivity’ in the model with T were platelet count, uric acid, glucose, homocysteine and testosterone; the T level contributed mostly to explaining of the variability of the dependent variable (‘cumulative blood platelet reactivity’), as reasoned from the highest absolute value of the standardized beta coefficients. However, due to the considerable variability of T, the statistical significances of the partial correlation coefficient and semipartial correlation coefficient were rather low (P< 0.02 for each), which implies that other variables in the model have large compounding contributions to the association between T and ‘cumulative blood platelet reactivity’ (the univariate correlation coefficient between T and ‘cumulative blood platelet reactivity’ was r_P= -0.371, P<< 0.0001). When the multiple regression analysis was performed separately for men and for women, the relationships revealed for the overall group were largely maintained. In men, the absolute values of the standardized beta coefficients indicated that uric acid, glucose, platelet count and Hcy contributed to the highest extent to the variability of the ‘cumulative blood platelet reactivity’ in the model with T. Amongst these contributors, uric acid and glucose were characterized by the highest significance of partial and semipartial correlation coefficients, indicating that other variables in the model contributed much less to the association between ‘cumulative blood platelet reactivity’ and uric acid or glucose. In the same model for women, platelet count, T and uric acid level contributed to the highest extent, while platelet count, uric acid, Hcy and T demonstrated the highest partial and semipartial correlation with ‘cumulative blood platelet reactivity’ (Table 6, the data with the superscripts ^m and ^f). When the ‘cumulative platelet activation’ was used as a dependent variable, only haemoglobin appeared significant (resp., coeff. β: 0.251 + 0.107, P< 0.02), while both leukocyte count and T remained beyond statistical significance (coeff. β: -0.177 + 0.094, P= 0.061 and coeff. β: -0.716 + 0.403, P= 0.78). In a separate subgroup of men, the significant contributor was glucose (coeff. β: -0.311 + 0.085, P= 0.0004), and in a subgroup of women, it was Hcy (coeff. β: -0.288 + 0.083, P= 0.0007). For the van der Waerden normal scores of the ‘cumulative GPIIb/IIIa expression’ as a dependent variable, only Hcy remained statistically significant (coeff. β: -0.180 + 0.090, P< 0.05), while T was beyond significance (resp. coeff. β: -0.680 + 0.403, P= 0.094) in the overall group of patients. In men, Hcy and glucose (coeff. β_Hcy: -0.330 + 0.088, P< 0.0002 and coeff. β_glucose: -0.244 + 0.087, P< 0.005) were significant contributors, while in women, only Hb (coeff. β_Hb: 0.176 + 0.085, P= 0.041) contributed at the borderline significance to the variability of ‘cumulative GPIIb/IIIa expression’.

Multivariate regression parameters for dihydrotestosterone and co-explanatory variables for the model of ‘cumulative blood platelet reactivity’ as a dependent variable are given in Table 7. In the model with DHT, the significant variables in the model were platelet count, uric acid, glucose, Hcy and DHT. DHT demonstrated the highest absolute value of the standardized beta coefficient, which implies that it contributed to explaining the variability of the dependent variable (‘cumulative blood platelet reactivity) to the greatest extent. Both the partial correlation coefficient and semipartial correlation coefficient were moderate (P< 0.02), which suggests that the contribution of DHT was not dominating over other independent variables in the model (the univariate correlation coefficient between DHT and ‘cumulative blood platelet reactivity’ r_P= -0.393, P<< 0.0001). In the model with DHT, the multiple regression analysis performed separately for sexes revealed that in men, uric acid, glucose, platelet count, Hcy and DHT were the most important contributors to the variability of the ‘cumulative blood platelet reactivity’; in addition, uric acid, glucose, platelet count and Hcy demonstrated the most significant partial and semipartial correlations, indicating that they had the greatest individual impact in the model. In a separate group of women, it was found that platelet count, DHT and uric acid appeared the most important contributors to the ‘cumulative blood platelet reactivity’ variability, while platelet count and DHT demonstrated the highest individual impact in the model (Table 7, the data with the superscript ^m and ^f).

For the ‘cumulative platelet activation’, used as a dependent variable, only haemoglobin and leukocyte count appeared to be significant independent variables (resp., coeff. β: 0.237 + 0.107, P< 0.03 and coeff. β: -0.189 + 0.095, P< 0.05), while DHT remained beyond significance (coeff. β: -0.453 + 0.273, P= 0.099). In a separate subgroup of men, only glucose contributed significantly to the overall variability of the model (coeff. β: -0.324 + 0.085, P= 0.0002), while in the subgroup of women, it was Hcy and DHT (coeff. β_Hcy: -0.283 + 0.082, P< 0.001 and coeff. β_DHT: -0.172 + 0.085, P< 0.05).

When the van der Waerden normal scores of the ‘cumulative GPIIb/IIIa expression’ were employed as a dependent variable, Hcy remained the only significant independent variable (coeff. β: -0.183 + 0.090, P< 0.05), while DHT was beyond significance (coeff. β: -0.516 + 0.272, P= 0.059). In men, Hcy and glucose contributed significantly to the variability of ‘cumulative GPIIb/IIIa expression’ (coeff. β_Hcy: -0.327 + 0.087, P= 0.0002 and coeff. β_glucose: -0.237 + 0.086, P= 0.006), whereas haemoglobin was the only significant independent variable in women (coeff. β: 0.170 + 0.082, P= 0.042). To sum up this part, both T and DHT contribute significantly to blood platelet reactivity, both in the overall group of patients and in separate subgroups of men and women. However, it is important to note that the extent of such a contribution is strongly affected by confounding factors. The concentrations of both androgens are not, however, significant predictors of platelet activation or expression of P-selectin or GPIIb/IIIa on blood platelets.

A logistic regression analysis was performed to determine how selected analysed (confounding/co-explanatory) variables contribute to lower or higher blood platelet reactivity cumulated through the used agonists (AA, collagen and ADP in the case of aggregometry, AA and collagen in the case of flow cytometry). The dependent variables were the dichotomised values of the variables referred to as ‘cumulative blood platelet reactivity’, the dichotomised values of the variables referred to as ‘cumulative platelet activation’ or ‘cumulative P-selectin/GPIIb/IIIa expression’.

In the whole group, both T and DHT, adjusted for sex and age, appeared as significant predictors of the ‘cumulative blood platelet reactivity’ (OR_T = 1.002*10^-10 [95%CI: 4.956*10^-19 – 0.202*10^-1, P< 0.02] and OR_DHT = 6.842*10^-12 [95%CI: 1.244*10^-19 – 0.376*10^-3, P< 0.005]). Sex- and age-adjusted T also remained a significant predictor when standardized individually for blood haemoglobin or leukocyte count (for both P< 0.02), platelet count (P< 0.005), mean platelet volume (P< 0.025) or plateletcrit (P< 0.01), total or HDL-cholesterol (for both P< 0.02), glucose (P< 0.01) or Hcy level (P< 0.02), but not upon standardization for uric acid (P= 0.061). Upon the overall multiple post hoc standardization for platelet and leukocyte counts, haemoglobin, cholesterol, Hcy, glucose and uric acid the resultant multivariate OR_T,multivar was 3.618*10^-11 [95%CI: 1.145*10^-20 – 0.114, P< 0.03] (P_{Hosmer-Lemeshow}= 0.538). When adjusted for gender and age, DHT remained a significant predictor of the dichotomised ‘cumulative blood platelet reactivity’ upon its post hoc individual standardization for blood platelet count or plateletcrit (for both P< 0.0005), haemoglobin (P< 0.04), total or HDL-cholesterol (for both P< 0.01), glucose (P< 0.01), uric acid (P< 0.02) or Hcy (P< 0.005), but not upon standardization for leukocyte count (P= 0.051). Upon post hoc multiple standardization for platelet and leukocyte counts, haemoglobin, cholesterol, homocysteine and uric acid, the overall age- and sex-adjusted multivariate OR was OR_DHT,multivar = 1.318*10^-13 [95%CI: 2.270*10^-23 – 0.627*10^-3, P< 0.01] (P_{Hosmer-Lemeshow}= 0.744). Thus, concentrations of T and DHT appear to be significant predictors of lower platelet aggregability, and remained so upon adjustment of the models, not only for sex and age, but also for several other studied variables.

Otherwise, only DHT adjusted for age appeared to be the significant predictor of the ‘cumulative blood platelet reactivity’ in separated subgroups of men (OR_DHT = 0.585*10^-10 [95%CI: 0.536*10^-20 – 0.638, P< 0.05]) and women (OR_DHT = 1.091*10^-12 [95%CI: 1.106*10^-19 – 1.076*10^-5, P< 0.001]): no significant relationships were observed for T adjusted only for age, in neither men nor women. In men, age-adjusted DHT also maintained a significant impact when standardized individually for blood haemoglobin (P= 0.05), mean platelet volume and total, HDL- or LDL-cholesterol (for all P< 0.05), platelet count (P< 0.004), plateletcrit (P< 0.01), uric acid or Hcy (for both P< 0.04), but not upon standardization for leukocyte count (P= 0.094). In turn, in women, DHT maintained a significant impact also upon individual standardization for plateletcrit, leukocyte count, HDL-cholesterol or Hcy (for all P< 0.001), blood haemoglobin, mean platelet volume, total and LDL-cholesterol or glucose (for all P< 0.002), platelet count (P< 0.0005) and uric acid (P< 0.003). When the model with age-adjusted ‘cumulative blood platelet reactivity’, used as a dichotomous dependent variable, was subjected to post hoc multiple-standardization for a set of independent predictors including T or DHT, platelet and leukocyte counts, haemoglobin, cholesterol, homocysteine and uric acid level, the multivariate OR appeared particularly significant for DHT in both sexes (OR_{DHT-men,multivar} = 1.002*10^-22 [95%CI: 9.423*10^-37 – 1.065*10^-8, P= 0.002]; P_{Hosmer-Lemeshow}= 0.123 and OR_{DHT-women,multivar} = 4.472*10^-17 [95%CI: 1.864*10^-27 – 1.072*10^-6, P= 0.01]; P_{Hosmer-Lemeshow}= 0.731), and less significant for T (OR_{T-men,multivar} = 2.374*10^-15 [95%CI: 1.926*10^-27 – 2.926*10^-3, P= 0.018]; P_{Hosmer-Lemeshow}= 0.195 OR_{T-women,multivar} = 3.809*10^-17 [95%CI: 3.789*10^-27 – 3.830*10^-7, P= 0.001]; P_{Hosmer-Lemeshow}= 0.063).

For the ‘cumulative platelet activation’, the model with DHT as an explanatory variable, adjusted for age and gender and subjected to post hoc normalisation for additional confounders, appeared at the borderline of significance for individually-included confounders (P> 0.08). The multivariate model with sex, age, platelet and leukocyte counts, haemoglobin, total cholesterol, glucose, uric acid and Hcy also remained beyond statistical significance (P= 0.078) (P_{Hosmer-Lemeshow}= 0.619). Otherwise, when T was chosen as an explanatory variable in the model, adjusted for age and gender and subjected to post hoc normalisation for additional confounders, the significance was even poorer for the models with individual confounders and for the multivariate model (P= 0.089) (P_{Hosmer-Lemeshow}= 0.404).

For the ‘cumulative GPIIb/IIIa expression’ the models with testosterone or DHT as the explanatory variables, adjusted for age and gender and subjected to post hoc normalisation for additional confounders, were significant neither for individually-included confounders nor for the multivariate model comprising sex, age, platelet and leukocyte counts, haemoglobin, total cholesterol, glucose, uric acid and Hcy (P> 0.1) (P_{Hosmer-Lemeshow}= 0.163 and P_{Hosmer-Lemeshow}= 0.202 for T and DHT, resp.).

Finally, a linear discriminant analysis (LDA) was performed to determine which analysed confounding/co-explanatory variables contribute the most to the discrimination between lower and higher blood platelet reactivity (dichotomised values of ‘cumulative blood platelet reactivity’, ‘cumulative platelet activation’ and ‘cumulative GPIIb/IIIa expression’ used as grouping variables) depending on sex. In general, based on the values of partial Wilk’s lambda estimated for all patients together (^mλ_{partial Wilks} and ^fλ_{partial Wilks} denote the respective values of partial Wilk’s lambda for separate subgroups of men and women adjusted for the sample size of n= 155), the most discriminative variables for dichotomised A_max cumulated through agonists (‘cumulative blood platelet reactivity’), in the standard LDA model appear to be platelet count (λ_{partial Wilks, plt} =0.912, P< 0.0003; ^mλ_{partial Wilks, plt} =0.896, P< 0.0002; ^fλ_{partial Wilks, plt} =0.869, P<< 0.0001), uric acid (λ_{partial Wilks, uric acid} = 0.924, P<0.001; ^mλ_{partial Wilks, uric acid} =0.925, P< 0.002; ^fλ_{partial Wilks, uric acid} =0.936, P< 0.002), DHT (λ_{partial Wilks, DHT} = 0.884, P<< 0.0001; ^mλ_{partial Wilks, DHT} =0.899, P< 0.0002; ^fλ_{partial Wilks, DHT} =0.908, P< 0.0002), T (λ_{partial Wilks, T} = 0.907, P< 0.002; ^mλ_{partial Wilks, T} =0.946, P< 0.01; ^fλ_{partial Wilks, T} =0.945, P< 0.003) and glucose (λ_{partial Wilks, glucose} = 0.971, P< 0.03; ^mλ_{partial Wilks, glucose} =0.925, P< 0.002; ^fλ_{partial Wilks, glucose} =0.980, P= 0.076). However, the only significant variables identified in the model using the backward stepwise approach were platelet count (λ_{partial Wilks, plt} = 0.895, P<< 0.0001; ^mλ_{partial Wilks, plt} =0.871, P<< 0.0001; ^fλ_{partial Wilks, plt} =0.842, P<< 0.0001), DHT (λ_{partial Wilks, DHT} =0.853, P<< 0.0001; ^mλ_{partial Wilks, DHT} =0.919, P< 0.001; ^fλ_{partial Wilks, DHT} =0.903, P< 0.0001) and T (λ_{partial Wilks, T} =0.877, P<< 0.0001; ^mλ_{partial Wilks, T} =0.962, P< 0.02; ^fλ_{partial Wilks, T} =0.907, P< 0.0001). For men DHT (λ_{partial Wilks}= 0.893, P<< 0.0001), platelet count (λ_{partial Wilks}= 0.911, P< 0.0005) and uric acid (λ_{partial Wilks}= 0.961, P< 0.02) appeared the most discriminative variables for the same grouping variable (dichotomised ‘cumulative blood platelet reactivity’), whereas also DHT (λ_{partial Wilks} =0.859, P<< 0.0001) and platelet count (λ_{partial Wilks}= 0.897, P<< 0.0001) discriminated the best in the backward stepwise approach. In women, platelet count and uric acid most clearly discriminated (with λ_{partial Wilks}= 0.911, P< 0.015 and λ_{partial Wilks}= 0.946, P< 0.05) for the same grouping variable (dichotomised ‘cumulative blood platelet reactivity’).

For the ‘cumulative platelet activation’ as the grouping variable, upon adjustment for sex and age, T and DHT remained the best discriminators (with λ_{partial Wilks}= 0.970, P< 0.04 and λ_{partial Wilks}= 0.965, P< 0.03 for separate models; stepwise forward analyses). When the expression of P-selectin (cumulated through agonists) was used as a grouping variable, testosterone and DHT), adjusted for sex and age, also appeared the most significant discriminators (with λ_{partial Wilks}= 0.960, P< 0.02 and λ_{partial Wilks}= 0.958, P< 0.02, respectively; stepwise forward approach).

Overall, both tested androgens, DHT and T appeared to significantly discriminate platelet ‘lower’ and ‘higher’ reactivity in the overall group including both sexes, however, the discriminative potential of T was markedly less expressed, particularly in sex subgroups. The androgens also remained significant predictors of platelet surface membrane GPIIb/IIIa expression. In the whole group of patients (incl. both males and females), as well as in the separated sex subgroups, platelet count, DHT/T and uric acid appeared the leading modulators of platelet reactivity.

In vitro effects of sex steroids on whole blood platelet aggregation

All the tested hormones demonstrated in vitro inhibitory effects on agonist-stimulated platelet aggregation, even at very low steroid concentrations. In general, sex was not found to have any significant effect when analysing the inhibitions of platelet aggregation by T, DHT and E₂ in separate groups of men and women.

Briefly, T appeared to be a potent inhibitor of 1 µg/ml collagen-dependent aggregation of blood platelets in both men and women. After the addition of T at final concentrations of 0.1, 0.5, 1, 2.5, 5 or 10 ng/ml, respective significant inhibition of platelet aggregation by 30, 45, 44, 52, 38 and 42% was observed in men (Fig. 1B), and by 50, 36, 41, 56, 54 and 31% in women (Fig. 1D), compared to control vehicle-treated samples.

Figure 1. Testosterone affects platelet aggregation induced by arachidonate or collagen in blood taken from men and from women. Data is presented as medians (thick horizontal lines) and interquartile ranges (IQR) (boxes, from lower quartile [25%] to upper quartile [75%]). Raw data is presented as black solid triangles or grey solid circles (outliers, by two-sided Tukey’s test: 1.5*[IQR]) for whole blood platelets stimulated with either arachidonate (0.5 mmol/l) (A, C) or collagen (1 µg/ml) (B, D) in men (A, B) and women (C, D). For experimental details, see Materials and methods. The significance of differences was estimated for Box-Cox-transformed data by the bootstrap-boosted (10000 iterations) ANOVA for repeated measures and the paired Student’s t-test with Bonferroni’s correction for post hoc multiple comparisons: P < 0.0005, µ₀ ≠ µ_0.1 = µ_0.5 = µ₁ = µ_2.5 = µ₅ = µ₁₀ for arachidonate-activated platelets in men; P < 0.001, µ₀ ≠ µ_0.1 = µ_0.5 = µ₁ = µ_2.5 = µ₅ = µ₁₀ for collagen-activated platelets in men; P < 0.001, µ₀ ≠ µ_0.1 = µ_0.5 = µ₁ = µ_2.5 = µ₅ = µ₁₀ for arachidonate-activated platelets in women; P < 0.05, µ₀ ≠ µ_0.1 = µ_0.5 = µ₁ = µ_2.5 = µ₅ = µ₁₀ for collagen-activated platelets in women.

Likewise, T effectively attenuated the extent of aggregation of male and female platelets induced with 0.5 mmol/l arachidonic acid: it reduced arachidonate-dependent platelet aggregation by 27, 41, 36, 45, 35 and 49%, respectively, in men (Fig. 1A) and by 26, 23, 31, 29, 37 and 27%, respectively, in women (Fig. 1C). Characteristically, the degree of T-induced anti-aggregatory effect was very similar regardless of T concentration.

When blood obtained from men or women was incubated with DHT, significant anti-aggregatory effects were observed in the case of collagen-dependent platelet reactivity. DHT at concentrations of 0.01, 0.1 and 1 ng/ml inhibited aggregation to similar extents, i.e. by 32, 39 and 47%, respectively, in men (Fig. 2B) and by 48, 59 and 38%, respectively, in women (Fig. 2D).

Figure 2. Dihydrotestosterone affects platelet aggregation induced by arachidonate or collagen in blood taken from men and from women. Data is presented as medians (thick horizontal lines) and interquartile ranges (IQR) (boxes, from lower quartile [25%] to upper quartile [75%]). Raw data is presented as black solid triangles or grey solid circles (outliers, by two-sided Tukey’s test: 1.5*[IQR]) for whole blood platelets stimulated with either arachidonate (0.5 mmol/l) (A, C) or collagen (1 µg/ml) (B, D) in men (A, B) and women (C, D). For experimental details see Materials and methods. The significance of differences was estimated for Box-Cox-transformed data by the bootstrap-boosted (10000 iterations) ANOVA for repeated measures and the paired Student’s t test with Bonferroni’s correction for post hoc multiple comparisons: P < 0.001, µ₀ ≠ µ_0.01 = µ_0.1 = µ₁ for arachidonate-activated platelets in men; P < 0.01, µ₀ ≠ µ_0.01 = µ_0.1 = µ₁ for collagen-activated platelets in men; P < 0.01, µ₀ ≠ µ_0.01 = µ_0.1 = µ₁ for arachidonate-activated platelets in women; P < 0.01, µ₀ ≠ µ_0.01 = µ_0.1 for collagen-activated platelets in women.

Also, in the case of arachidonate-induced aggregation, DHT acted as an efficient inhibitor in blood samples taken from men (inhibition by 40, 35 and 29%, resp. for the increasing DHT concentrations) or women (inhibition by 39, 40 and 30%, resp. for the increasing DHT concentrations) (Fig. 2A, C).

E₂ inhibited platelet aggregation induced by either collagen or AA. When used at increasing concentrations (0.01, 0.1 and 0.3 ng/ml), E₂ significantly inhibited collagen-triggered platelet aggregation by 8, 38 and 19% in men (Fig. 3B) and by 23, 38 and 43% in women (Fig. 3D). Also, according to our in vitro aggregatory measurements, E₂ exhibited significant antiplatelet properties in the blood of both male and female subjects, when arachidonate was used to trigger platelet aggregation. E₂ used at final concentrations of 0.01, 0.1 or 0.3 ng/ml inhibited arachidonic acid-induced platelet aggregation by 29, 35 and 21% in men and by 28, 27 26% in women (Fig. 3A, C).

Figure 3. Oestradiol affects platelet aggregation induced by arachidonate or collagen in male and female blood. Data s presented as medians (thick horizontal lines) and interquartile ranges (IQR) (boxes, from lower quartile [25%] to upper quartile [75%]). Raw data is presented as black solid triangles or grey solid circles (outliers, by two-sided Tukey’s test: 1.5*[IQR]) for whole blood platelets stimulated with either arachidonate (0.5 mmol/l) (A, C) or collagen (1 µg/ml) (B, D) in men (A, B) and women (C, D). For experimental details, see Materials and methods. The significance of differences is estimated for Box-Cox-transformed data by the bootstrap-boosted (10000 iterations) ANOVA for repeated measures and the paired Student’s t-test with Bonferroni’s correction for post hoc multiple comparisons: P < 0.01, µ₀ ≠ µ_0.01 = µ_0.1 = µ_0.3 for arachidonate-activated platelets in men; P < 0.01, µ₀ ≠ µ_0.1 = µ_0.3 for collagen-activated platelets in men; P < 0.01, µ₀ ≠ µ_0.1 = µ_0.3 for arachidonate-activated platelets in women; P < 0.05, µ₀ ≠ µ_0.01 = µ_0.1 = µ_0.3 for collagen-activated platelets in women.

Chemicals

LC–MS grade, acetonitrile (MeCN), methanol (MeOH), tetrahydrofuran (THF), dimethyl sulfoxide, MS grade formic acid (HCOOH) and all the analytical standards of testosterone, dihydrotestosterone, α-estradiol, β-estradiol, methyltestosterone, as well as the hormone preparations used for in vitro measurements of platelet activation/reactivity were obtained from Sigma (St. Louis, MO, USA) and had a minimum purity specification of 99%. Nitric acid (HNO₃) was purchased from POCH (Gliwice, Poland). Dichloromethane (DCHM, HPLC grade) was provided by VWR International (Radnor, PA, USA).

PBS was from Avantor Performance Materials Poland S.A. (Gliwice, Poland). Arachidonate, equine tendon collagen and ADP were from Chrono-Log Corp. (Havertown, PA, USA). Fluorolabelled monoclonal antibodies (moAbs): anti-CD61/PerCP, antiCD62/PE, PAC-1/FITC, isotype controls IgG/PE and IgM/FITC, as well as CellFix (1% formaldehyde in PBS) were from Becton Dickinson (San Diego, CA, USA). Ultrapure water was obtained from Milli-Q purification system (Millipore, Bedford, MA, USA). Nitrogen (NM32LA Nitrogen Generator, Peak Scientific Instruments, Billerica, MA, USA) was used as a drying gas.

Ethics statement

All steps of the experiments with the participation of human subjects were undertaken in accordance with the guidelines of the 1975 Helsinki Declaration for human research. The study was approved by the Committee on the Ethics of Research in Human Experimentation at the Medical University of Lodz. A written abstract of the experiment, including detailed information regarding the study objectives, study design, risks and benefits, was given to each of the volunteers in the course of recruitment. Informed written consent was obtained from each individual at the beginning of each experiment.

Design of the study

This study presents the results obtained in subgroups of volunteers participating in the project entitled “The occurrence of oxidative stress and selected risk factors for cardiovascular risk and functional efficiency of older people in the context of workload”, funded by the Central Institute For Labour Protection – National Research Institute (Warsaw, Poland) and supervised by the Clinic of Geriatrics at Medical University in Lodz (Poland). Participants were recruited through announcements given on local TV, radio and newspapers. Two basic inclusion criteria were: age within the range of 60 to 65 years and the willingness to participate. The recruitment criteria are described in more detail in an earlier report [50]. To ensure approximately equal participation of men and women, a stratified probability sampling algorithm was used. The research group included roughly 320 subjects (approx. equal sex proportions of 160 men and 160 women), aged from 60 to 65 years, who responded to the announcements (source population). For the purpose of the present study, only those who did not use antiplatelet drugs (acetylsalicylic acid, thienopyridines) within three weeks prior to blood withdrawal were selected (target population). Based on a preliminary pilot study, minimal correlation between the cumulative measure of platelet aggregation and plasma androgen concentration was assumed to be at least 0.30-0.32, the statistical test power, i.e. the likelihood of rejecting false null hypothesis, was assumed to be 90%, and the significance level was assumed to be at least 1% [51]. Of the enlisted target population, 82 women and 73 men were randomly selected (simple unrestricted randomization) to create the subgroup for studying sex steroid level.

Our study consisted of essentially two separate approaches: (1) the monitoring of the in vivo associations between platelet function and plasma solutes, including androgens and oestrogens, and (2) the monitoring of the in vitro effects of selected sex hormones on the agonist-mediated whole blood platelet aggregation. Blood samples from eight to ten randomly-selected individuals participating in the in vivo part of the study (approach 1) were used to perform the incubations with the increasing hormone concentrations under in vitro conditions (approach 2).

Blood sampling, isolation of blood plasma, measurements of blood morphology and plasma biochemistry

Blood was taken after overnight fasting and 15 minute rest directly before blood donation. Blood was collected by aspiration either to vacuum tubes (Sarstedt, Nümbrecht, Germany) supplemented with 0.105 mol/l buffered sodium citrate (citrate:blood ratio = 1:9, v/v) for the measurement of platelet activation and reactivity, or to tubes coated with EDTAK₃ in the case of samples taken for blood morphology analysis, or to tubes without any anticoagulant when the serum needed to be isolated for further biochemical determinations. In all cases, blood was collected from a peripheral vein cannulated with an 18-gauge needle.

Blood morphology parameters were measured with a 5-Diff Sysmex XS-100i haematological analyser (Sysmex, Kobe, Japan) and serum biochemical parameters were evaluated with a DIRUI CS 400 analyser (Dirui, Changchun, China). Homocysteine concentration was measured with an Immulite 2000 XPi analyser (Siemens, Erlargen, Germany).

In order to obtain blood serum, whole blood taken to the tubes without an anticoagulant was incubated for 30 minutes at 37°C and centrifuged (2000 x g/15 min/4°C). Supernatant (serum) was aspirated and used in further analyses.

Blood plasma was obtained from citrated blood (0.1 mol/l citrate, blood:citrate = 9:1 (v/v)), centrifuged immediately after withdrawal (1000 x g/15 min/4°C), and 1 ml aliquots of plasma were immediately frozen at -80°C and used within the following six months.

Measurements of whole blood platelet aggregation

Platelet aggregability was measured with a Multiplate Analyser impedance aggregometer (Dynabyte, Munich, Germany), in citrated blood following a 10-minute blood reposition at 37°C to avoid the interference of artefactual platelet activation caused by aspiration. “Tranquilized” 300 µl aliquots of whole blood sample were mixed with an equal volume of PBS for three minutes at 37°C. The mixed sample then was supplemented with one of the used agonists: 0.5 mmol/l arachidonate, 1 µg/ml collagen or 10 µmol/l ADP were added to trigger platelet aggregation. The recording of platelet clumping started immediately thereafter and was tracked for at least 15 minutes. The area under the aggregation curve (AUC) and maximal aggregation (A_max) were used to describe blood platelet aggregability.

Measurements of the surface membrane markers of platelet activation and reactivity with the use of flow cytometry

Flow cytometric measurements of the activation of circulating platelets (not agonized in vitro either arachidonate nor collagen) included the determinations of the expressions of two surface membrane complexes: the activated glycoprotein complex GPIIb/IIIa (receptor for fibrinogen) and P-selectin. Flow cytometric analyses were performed according to the protocol described in greater detail elsewhere [52].

Preparation of standard solutions and extraction mixture for hormone determinations with the use of the LC-ESI/MS method

Standard solutions of all the analyzed steroids at the concentrations of 10 µg/ml were prepared by diluting stock solutions containing 1 mg of steroid per 1 ml of methanol. The stock standard solutions of steroids were stored at −20°C until analysis, but no longer than one month. The extraction mixture was prepared by mixing 3 ml of THF with 1.2 ml of DCHM, to ensure the THF/ DCHM volumetric ratio of 5:2.

Conditions of LC-MS analysis with the use of the LC-ESI/MS method

LC–MS analyses were performed on an Agilent 1260 Infinity HPLC system (Agilent Technologies, Santa Clara, CA, USA) that included a binary pump, an autosampler and a UV variable wavelength detector coupled to an Agilent 6120 mass spectrometry MS (Single Quad, Agilent Technologies, Santa Clara, CA, USA). The mass spectrometer was equipped with the electrospray (ESI) ion source and single quadrupole analyzer, operating in a positive ion mode.

Chromatographic separation of the analytes was carried out with the use of a Poroshell 120 EC-C18 column (Agilent Technologies, St. Clara, CA, USA) (3.0 x 100 mm, particle size 2.7 μm), at temperature of 25°C. The chromatographic results were collected using ChemStation software (Agilent Technologies, Santa Clara, CA, USA).

The mobile phase was composed of 0.1% formic acid in a Millipore grade water (eluent A) and 0.1% formic acid in an LC–MS grade acetonitrile (eluent B). The gradient started linearly from 33.1 to 55.1% B for 12 minutes, followed by an increase to 100% B for two minutes, remained at a constant level of 100% B for two minutes, before decreasing to 33.1% for 6 s. Then, the column was re-equilibrated for 3.9 minutes at initial chromatographic conditions. The total analysis run time was 25 minutes. The injection volume was 5 µl and the flow rate of the mobile phase was set at 0.5 ml/minute. The column was thermostated at 25°C, whereas the temperature at the autosampler storage box was 4°C.

The optimized MS parameters were as follows: ion source temperature 350 °C, drying gas (N₂) flow rate of gas 12 l/min, nebulizer gas (N2) pressure 60 psi and capillary voltage 6 kV. The fragmentor voltage was set individually for each analyzed compound within the range of 100 to 175 V. The selected ion monitoring (SIM) values chosen for quantitative assay development were (m/z): 363 for cortisol, 289 for testosterone, 255 for β-estradiol, 291 for dihydrotestosterone, 361 for aldosterone, 373 and 303, for two internal standards, betamethasone and methyltestosterone, respectively.

Sample preparation with the use of dispersive liquid-liquid microextraction (DLLME) method for LC-ESI/MS hormone determinations in blood plasma samples

Human blood plasma samples were stored at -80°C until analysis. Before the analysis, samples were thawed at room temperature and deproteinized by mixing equal volumes (200 µl) of human plasma and 0.1 M HNO3, then filled with water to 1 ml, vortexed and centrifuged at 10,000 rpm for seven minutes. The 980 µl aliquots of supernatants were transferred into 1.5 ml Eppendorf tubes and spiked with internal standard solutions in methanol, 10 µl of methyltestosterone and 10 µl of betamethasone, both at concentration of 10 µg/ml was added.

To extract the steroids, 700 µl of the extraction mixture (containing 500 µl of the extraction solvent (THF) and 200 µl of the disperser solvent (DCHM) was rapidly injected into 1 ml of plasma sample. The mixture was shaken vigorously for about 10 s and then frozen for 10 minutes. Next, the emulsion was centrifuged for 10 minutes at 8000 rpm to enable phase separation. A droplet of DCHM (ca. 150 ± 10 µl in volume), containing the analyzed steroids, was collected and evaporated to dryness at 45^oC using the CentriVap vacuum rotary evaporator (Labconco, Kansas City, Missouri, USA). The residue was dissolved in 100 µl of methanol, moved to a vial insert and submitted for LC-MS analysis.

Statistical analysis

The data was expressed as mean + SD or median and IQR (interquartile range: lower [25%] to upper quartile [75%]), depending on data distribution (the Shapiro-Wilk test). Heteroscedasticity was verified with the Brown-Forsythe test. As the data distribution and residual distribution were not normal in the case of numerous variables, the data was subjected to Box-Cox transformation for further analyses. The analysis of outliers was performed with the use Grubb’s and Tukey’s tests; the k nearest neighbours analysis was employed as the imputation method when the fraction of missing data did not exceed 3% per a given variable. The principal components analysis and quantile-quantile plot were used to visualize possible batch effects. The block ANOVA was used to correct for possible batch effects in single variables. The Student’s t-test for independent samples, Welch test or Mann-Whitney U-test were used as the inference tests, according to assumptions of data normality and homoscedasticity.

The comparisons with the adjustments for dummy or confounding variables were performed using an analysis of covariance (ANCOVA). The nonparametric Spearman partial correlations method was used to adjust for the effect of gender in simple rank associations. Simple nonparametric correlation estimates were validated using the leave-one out (LOO, jackknife or d-jackknife) techniques.

Multiple logistic regression models were used to reason on the outcome prediction, and the goodness of fit for the models was evaluated using the Hosmer-Lemeshow test [53]. Multivariate regression models were used with continuous dependent variables as the outcomes. The multivariate regression models were validated using bootstrap-boosted procedures (1000-10000 iterations). Multiple group comparisons were performed with various models of the analysis of variance, followed by multiple comparisons post hoc tests (Tukey’s honest significant difference test or Dunnett’s test, depending on the tested contrast). For multivariate correlations both the partial correlation coefficient (which reflects the association between a given independent variable and a dependent variable, having eliminated the contribution of other associations of this independent variable with other independent variables in the model) and semipartial correlations (which, in turns, reflects the individual contribution of a given independent variable associated with the overall variability of a dependent variable) were evaluated.

Some bivariate and multivariate analyses, performed separately for men and for women, used a resampling approach with replacement adjusted to sample size. This was intended to overcome any bias related to sample size and to standardize the sample sizes of separate sexes to the overall sample size of the studied group.

The variables describing platelet aggregability in response to agonists and platelet surface membrane glycoprotein expressions were subjected to normalization according to the van der Waerden method. Cumulative normal scores through the used agonists were determined for the markers of whole blood aggregometry (A_max, AUC, A_max*AUC/1000) and two flow cytometry markers (P-selectin and the activated GPIIb/IIIa complex) (see the description in the ‘Results’ section for details); they were then dichotomised according to the median in such a way that the values below or equal to the median (≤Me, rank 0) corresponded to the “lower overall blood platelet reactivity”, while the values above the median (>Me, rank 1) corresponded to the “higher overall blood platelet reactivity”. Such categorical measures of platelet reactivity were accepted as the grouping variables or dichotomous dependent variables in some multivariate analyses (logistic regression, discriminant analysis). The following software was used for statistical analyses: Statistica v. 13.1 (Statsoft), StatsDirect v.3.0.182, Resampling Stats Add-in for Excel v.4, GraphPad Prism v.5.