Experimental rationale.

The rationale was to antagonize AR feedback pathways using selective antiandrogens that either permeate (flutamide) or do not permeate (bicalutamide) the blood-brain barrier (8,21). The degree to which the effect of flutamide on LH dynamics differed from that of bicalutamide and remained distinct from the known actions of estrogen (36) was taken as an index of AR-related CNS feedback control. In particular, feedback differences in drug actions (flutamide vs. bicalutamide) were quantified by specific deconvolution measurements of LH secretion and approximate entropy.

Subjects.

Each subject provided written informed consent approved by the Mayo Institutional Review Board (IRB). Volunteers were recruited from locally placed IRB-approved postings and received IRB-approved reimbursement for time spent in the study.

A total of 24 healthy men participated. The age range was 20–73 yr (median 51 yr), and body mass index (BMI) was 21–32 (median 26) kg/m². Ages (no. of subjects) by 15-yr intervals were <30 (4 men), 30–45 (6 men), 45–60 (6 men), and 60–75 (8 men). Mean dietary macronutrient distribution in the volunteers was 60% carbohydrate, 15% protein, and 25% fat. No subject had gained or lost >2 kg weight over the preceding month or during the study. None abused alcohol or acetaminophen. None had sexual dysfunction. Each denied concurrent or recent major life stressors (loss of employment, death in the family, divorce). Subjects described normal sleep habits. Medical inventory and physical examination (including testis size, libido, and potentia) were normal. There was no history of infertility, systemic disease, hormonal therapy, or psychoactive drug use. Fasting (0800) screening biochemical tests of endocrine, metabolic, hematological, hepatic, and renal function were normal.

Clinical protocol.

The study was a prospective, double-blind, and randomized crossover design. Each subject received oral placebo, flutamide (250 mg), and bicalutamide (50 mg) three times daily for 4 days, with intensive blood sampling on the 4th day. Placebo or drug was continued every 8 h during sampling. There was a minimum 1-mo drug washout (8). The Mayo IRB and the FDA approved the specific protocol for human use, contingent upon normal prestudy hepatic and renal function (transaminases and creatinine) and complete blood counts, the absence of allergy to either drug, normal prostate-specific antigen, and disallowance of concomitant exposure to alcohol or other potentially hepatotoxic drugs. A data safety and monitoring board was empaneled at Mayo to verify compliance.

An indwelling intravenous (iv) catheter was placed in a forearm vein at 0645 on the day of study, and blood samples (1.5 ml) were withdrawn every 10 min for 8 h beginning at 0800. The first 360 min of blood sampling served as a baseline (pre-GnRH injection). At 1400, GnRH (100 ng/kg) was given by bolus iv injection. Sampling ceased 120 min later.

Drug assays.

Serum concentrations of 2-hydroxyflutamide and bicalutamide were assayed on the day of blood sampling by liquid chromatography-tandem mass spectrometry, as described (8,34). Sensitivity was 0.1 μg/ml for both, and interassay variabilities were 10.3 and 10.8%. Drug assays were performed on three pooled samples from each volunteer comprising 0.1-ml aliquots of each 10-min blood sample (n = 49) across the three separate 8-h sampling intervals. This was done to verify subject compliance with drug ingestion.

Hormone assays.

LH concentrations in the 10-min samples were measured in duplicate on the DxI automated two-site immunoenzymatic assay system (Beckman Instruments, Chaska, MN). Intraassay coefficients of variation (CVs) were 4.3 and 4.0% at 1.2 and 38.5 mIU/ml, respectively. Interassay CVs were 9.3, 6.0, and 4.2% at 1.4, 15.6, and 48.8 mIU/ml, respectively. Procedural sensitivity was 0.2 IU/l, and the upper analytic limit was 250 IU/l, using the World Health Organization Second International Reference Preparation 80/552 as standard. T was measured in 10-min samples on the same robotics system. Intrassay CVs were 6.5% at 69 ng/dl and 3.3% at 862 ng/dl. To convert ng/dl T levels to nmol/l, we multiplied by 0.0347. Interassay CVs were 7.4% at 1,116 ng/dl, 8.6% at 407 ng/dl, and 4.0% at 761 ng/dl. The analytic range was 54–1,650 ng/dl. The coefficient of determination wasr² = 0.98 between robotics and liquid chromatography-tandem mass spectrometry (LC-MS/MS; ThermoFisher Scientific, Franklin, MA, and Applied Biosystems-MDS Sciex, Foster City, CA), as described (31). Confirmatory LC-MS/MS was used to measure T in 0800 samples. Intraassay CVs were 3.3, 2.8, 2.2, and 2.0% at 16, 64, 184, and 927 ng/dl, respectively. Interassay CVs were 5.1, 3.8, 3.7, and 2.8% at 17, 65, 177, and 919 ng/dl, respectively. The analytic range is 7–2,000 ng/dl. Estradiol (E₂) concentrations were measured in the same mass spectrometry run: sensitivity 3.5 pg/ml, interassay CV 4.5% (24). To convert pg/ml E₂ to pmol/l, we multiplied by 3.69.

Sex hormone-binding globulin (SHBG) and albumin were measured in morning serum by Immulite 2000 [Diagnostic Products (Siemens), Los Angeles, CA] and Roche/Hitachi 912 (Roche Diagnostics, Basal, Switzerland), respectively. Inter- and intra-assay CVs were <6.0% at SHBG concentrations ranging from 5.4 to 96 nmol/l. Inter- and intra-assay CVs were <2.0% at albumin concentrations from 2.5 and 4.6 g/dl.

Calculation of free and bio T concentrations.

Free and bio T concentrations were calculated in each 10-min serum sample from measured total T in each sample and pooled (session-specific) albumin and SHBG concentrations, as described (refer to supplemental data in Ref.33).

Deconvolution analysis.

The 8-h LH concentration time series were analyzed using a recently developed automated deconvolution method described fully statistically (5). The pulse detection algorithm was empirically validated for LH studies using hypothalamo-pituitary sampling and simulated pulsatile time series (16). Sensitivity and specificity are both ∼93%. For preliminary pulse detection, the Matlab-based algorithm first detrends the data and normalizes concentrations to the unit interval (0, 1). Second, the program creates multiple successively decremental potential pulse time sets, each containing one fewer burst by a smoothing process (a nonlinear adaptation of the heat diffusion equation). Third, a maximum-likelihood expectation estimation method calculates all secretion and elimination parameters simultaneously conditionally on each of the candidate pulse time sets, using the original nondetrended LH concentration data. The rapid half-life of LH was represented as 18 min, constituting 63% of the decay amplitude, and the slow half-life 90 min (38). For T, corresponding values were 3.5 and 28 min (63% slow decay), based upon T infusions in eight men (41). Statistical model selection was performed using the Akaike information criterion. The parameters (and units) are basal (nonpulsatile) and pulsatile secretion rates (concentration units/session), mass secreted per burst (concentration units), and waveform mode (time delay to maximal secretion after objectively estimated burst onset, min). Total LH secretion was the sum of basal and pulsatile release. Percentage pulsatile secretion was defined as the percentage ratio of pulsatile-to-total LH secretion.

Approximate entropy measurement of regularity.

Approximate entropy (ApEn; 1, 20%) was used as a scale- and model-independent regularity statistic to quantify the orderliness (regularity) of hormone release (25). Higher ApEn denotes greater disorderliness (irregularity) of the secretion process. Mathematical models and clinical experiments establish that greater irregularity signifies decreased feedback control with high sensitivity and specificity (both >90%) (45). Cross-ApEn was used to test pairwise synchrony between LH and T (feedforward) as well as, conversely, between T and LH (feedback) (17,26).

Statistical comparisons.

LH and T concentrations were averaged over either 6 (before GnRH) or 2 h (after GnRH). Statistical power was estimated as ≥90% to detect a 1.5 IU/l increment due to flutamide in mean 6-h LH concentrations by a two-sided pairedt-test if 25 men completed the study and ≥95% power to detect a 0.18 ApEn contrast atP < 0.05. The response to GnRH was a secondary endpoint, which was not powered. Deconvolution and ApEn measurements were compared parametrically after log transformation and nonparametrically without transformation (Kruskal-Wallis test). All primary outcomes were significant by both methods.

One-way analysis of covariance (ANCOVA) was used to test the impact of the two antiandrogens on log-transformed LH secretory parameters. The covariate was the subject's response on the placebo day. The covariate adjusts statistically for autocorrelation expected in a repeated-measurements design. Tukey's honestly significantly different test was applied post hoc for protected comparisons among means (47). Secretory parameters are given in the supplemental tables and text as the geometric mean (95% confidence intervals) and in the figures as box-and-whisker plots. Untransformed ApEn and hormone concentrations are presented as the arithmetic means ± SE. Experiment-wiseP < 0.05 was construed as statistically significant.

Age and BMI-dependent effects on LH secretion were analyzed via linear regression (version 9.0; Systat, Richmond, CA).

Baseline data.

Baseline (pretreatment) physical and endocrine data in the 24 men studied here are summarized in Supplemental Table S1 (Supplemental Material for this article can be found online at theAJP-Endocrinology and Metabolism web site). All subjects completed three 8-h sampling visits (comprising a total of 3,528 LH and 3,528 T samples) without adverse events. Compliance was verified by direct assay of serum drug levels on the day of study. Regression of baseline hormone concentrations linearly on age or BMI yielded expected relationships, showing typicality of the cohort (Supplemental Table S1).

Mean concentrations.

Six-hour pre-GnRH (baseline) and 2-h post-GnRH (stimulated) LH concentration vs. time profiles were obtained by 10-min sampling during each intervention (Fig. 1). Prior to GnRH injection, LH and T concentration curves appeared to separate in the descending rank order of flutamide > bicalutamide > placebo. ANCOVA confirmed this inference (flutamideP < 0.001 vs. placebo, bicalutamideP < 0.01 vs. placebo, overallP < 0.001, covariateP < 0.001;Fig. 2). By post hoc analysis, LH and T responses to flutamide exceeded those responses to bicalutamide byP = 0.002 andP = 0.017, respectively (Fig. 2,left). Changes in baseline total T concentrations were corroborated by mass spectrometry, viz. 454 ± 32 (placebo), 644 ± 34 (flutamide,P < 0.001 vs. placebo) and 563 ± 37 ng/dl (bicalutamide,P < 0.005 vs. placebo) (P < 0.015 for antiandrogen comparison). Concentrations of E₂ also rose during exposure to antiandrogens (P < 0.001 for both vs. placebo,P < 0.015 for drug comparisons; Supplemental Table S2). Estrone levels did not vary among the three sessions.

GnRH stimulation.

A submaximally stimulatory dose of GnRH (100 ng/kg) elicited similar absolute peak LH concentrations in all three treatment conditions (P = 0.31). Each response was significantly greater than the 6-h baseline mean (P < 0.001;Fig. 2,right). GnRH injection per se did not alter 2-h mean or peak T levels, which remained higher in the presence of flutamide than bicalutamide (P = 0.013) and bothP < 0.001 over placebo.

Deconvolution analysis of LH.

Pulsatile and basal (nonpulsatile) LH secretion rates were compared by ANCOVA. Flutamide and bicalutamide each stimulated basal LH secretion (P < 0.001 vs. placebo), with a nonsignificant trend toward a larger effect by flutamide (P = 0.065;Fig. 3,left). Only flutamide augmented pulsatile LH secretion and concomitantly inhibited GnRH-evoked LH release. Flutamide elevated only basal T secretion, and the magnitude of this effect exceeded that of bicalutamide (P = 0.032). Both antiandrogens increased LH pulse frequency (P < 0.001 vs. placebo), but flutamide induced a twofold larger incremental change (P = 0.013; Supplemental Table S3). Only flutamide stimulated 6-h pulsatile LH secretion (P = 0.003 vs. placebo;Fig. 3,middle). Neither antiandrogen significantly affected the size of LH secretory bursts, although there was a trend toward a decrease (P = 0.082;Fig. 4). Flutamide compared with bicalutamide abbreviated LH secretory bursts (decreased the mode,P = 0.035 overall,P = 0.032 for flutamide vs. bicalutamide; Supplemental Table S3). Although both flutamide and bicalutamide stimulated total (basal plus pulsatile) LH secretion (P < 0.001), flutamide had the greater effect (P = 0.002). Incremental (stimulus minus baseline) GnRH-induced pulsatile LH secretion was less after flutamide than placebo (overallP = 0.015 with a covariate effect ofP < 0.001;Fig. 3,right). This was not true for bicalutamide.

Deconvolution analysis of T.

Flutamide elevated basal (nonpulsatile) T secretion significantly compared with both placebo (P < 0.001) and bicalutamide (P = 0.032) (Fig. 3,bottom). Six-hour pulsatile T secretion was not affected (P = 0.31). Both antiandrogens stimulated total (pulsatile plus basal) T secretion (P < 0.001) also with a greater effect of flutamide than bicalutamide (P = 0.019; Supplemental Table S4). Neither AR antagonist altered pulsatile T secretion after GnRH injection (P = 0.47;Fig. 4,bottom right). T secretory burst number, size (mass), and shape (mode) were not affected by antiandrogens.

ApEn.

Baseline LH ApEn increased the most during flutamide administration (P < 0.001 vs. placebo,P = 0.044 vs. bicalutamide, covariateP < 0.001), albeit significantly during bicalutamide exposure (P = 0.006 vs. placebo;Fig. 5,top left). Only flutamide increased LH ApEn after GnRH injection (P = 0.012 overall,P < 0.001 covariate effect,P = 0.009 for flutamidevs placebo,P = 0.27 for bicalutamide vs. flutamide). Antiandrogens did not affect T ApEn values before (P = 0.15) (Fig. 5,top right) or after (P = 0.11) GnRH injection.

Cross-ApEn, which is a measurement of joint-series irregularity (viz., paired-series asynchrony), revealed that LH → T feedforward asynchrony was higher (pairwise synchrony was lower) during flutamide than bicalutamide administration (P = 0.034 overall,P = 0.045 for flutamide > bicalutamide). This may reflect less effective LH pulsations in driving T in that T → LH feedback asynchrony did not change (P = 0.14;Fig. 5,bottom).

Linear regression analysis.

There was a consistently negative effect of age on the mode (time delay to maximal LH secretion within a burst) determined over 8 h under all three study conditions (P = 0.005 placebo,P < 0.001 flutamide,P = 0.004 bicalutamide,r² = 0.30 to 0.47;Fig. 6). The slope on age was more negative during flutamide than placebo or bicalutamide administration (P ≤ 0.025).

During flutamide administration, BMI was a positive, and age a negative, joint correlate of percentage pulsatile LH secretion (overallP = 0.0004,r² = 0.52, BMIP = 0.028, ageP = 0.0011; Supplemental Fig. S1). The negative effect of age, but not the positive effect of BMI, was evident during placebo (P = 0.012,r² = 0.25) and bicalutamide (P = 0.016,r² = 0.25) treatment. Conversely, during flutamide exposure, BMI was a negative determinant (P = 0.0061) and age a positive determinant (P = 0.0083) of basal LH secretion (overallP = 0.0009,r² = 0.59; Supplemental Fig. S2).

Serum concentrations of 2-hydroxyflutamide averaged 1.1 ± 0.09 μg/ml (median 0.93, range 0.45–2.3) and bicalutamide 7.9 ± 0.26 μg/ml (median 7.5, range 5.8–10.9). Neither correlated with age (P > 0.70). However, bicalutamide levels correlated negatively with BMI (r² = 0.18,P = 0.041). 2-Hydroxyflutamide levels behaved similarly but at a nonsignificant trend level (r² = 0.14,P = 0.073; Supplemental Fig. S3).