Impact of menopausal hormone formulations on pituitary-ovarian regulatory feedback

Abstract

Changes in pituitary-ovarian hormones across the menopausal transition have multiple physiological consequences. However, little is known about how the major types of postmenopausal hormone therapy (HT) affect pituitary-ovarian hormonal relationships. This study evaluated these relationships in recently menopausal women (52.45 ± 2.49 yr of age) in the Kronos Early Estrogen Prevention Study (KEEPS) who were compliant to randomized, double-blinded treatment with oral conjugated equine estrogen (o-CEE; n = 109), transdermal 17β-estradiol (t-E2; n = 107), or placebo (n = 146). Androstenedione, testosterone, 17β-estradiol, estrone, follicle-stimulating hormone (FSH), and luteinizing hormone (LH) were measured in serum before (baseline) and 48 mo after randomization to treatment. Descriptive summaries of hormone levels were performed, and multiple regression analyses were used to examine the effects of o-CEE, t-E2, and placebo on these hormone levels at 48 mo, adjusting for baseline levels. A network analysis examined the covariance of changes in hormone levels over the 48 mo within treatment groups. As expected, at 48 mo of treatment, hormone levels differed between women in the two active treatment groups compared with placebo, and network analysis indicated stronger relationships among hormone levels in the t-E2 and o-CEE groups compared with placebo. Associations among testosterone, 17β-estradiol, FSH, and LH differed between the o-CEE group compared with t-E2 and placebo groups. Thus, two common HT regimens differentially alter pituitary-ovarian hormone levels, altering feedback cycles and interhormonal associations in recently menopausal women. These interactions provide the basis for future studies investigating the impact of hormonal modulation of aging, including cognitive decline in women.

INTRODUCTION

The relationship among pituitary-ovarian hormones represents a classical negative feedback control of hormone release, i.e., the gonadotropins follicle-stimulating hormone (FSH) and luteinizing hormone (LH) released from the pituitary stimulates synthesis and secretion of 17β-estradiol (E2) by the ovaries, which in turn inhibits further release of LH and FSH. With advancing age the feedback system gradually changes, and with menopause ovarian follicle depletion is accompanied by dramatically decreased production of E2 and progesterone, subsequently increasing secretion of FSH and LH. In contrast, levels of androgens, including androstenedione and testosterone, remain at their relatively low premenopausal levels being released from the follicle-depleted ovary (2, 19, 24, 29, 31, 41) and the adrenal glands (12). Despite some aromatization of testosterone to E2, total gonadal hormone levels remain low and do not impact the increased secretion of FSH and LH.

Several studies have evaluated the impact of exogenous menopausal hormone therapy (HT) to relieve specific outcomes of menopause such as vasomotor hot flashes and night sweats, changes in cognition, depressed mood, and osteoporosis (4, 9, 20, 21, 23, 36, 39), but few have evaluated the effects of HT on hormones of the hypothalamic-pituitary axis (34). Nevertheless, changes in levels of endogenous pituitary-ovarian hormones resulting from exogenous HT may have important health-related effects. For example, lower relative levels of FSH are associated with an increased risk of cardiovascular disease, diabetes, and prediabetes in postmenopausal women, and higher levels of pituitary LH secretion may be associated with reduced intracerebral LH production, leading to adverse changes in cognitive function (6, 44, 45). Furthermore, HT may impact production of androgens (testosterone and androstenedione) from the postmenopausal ovary, which could be associated with physiological (2, 18) and neurobehavioral (8, 32) consequences. Thus, it is important to better understand the relationships among different formulations of HT and endogenous pituitary hormone levels to explore possible long-term effects of HT in postmenopausal women.

To address this question, this study analyzed serum collected from participants in the Kronos Early Estrogen Prevention Study (KEEPS) to investigate the associations among pituitary gonadotropins and ovarian hormones during up to 4 yr of treatment with either oral conjugated equine estrogen (o-CEE), transdermal 17β-estradiol (t-E2), or placebo in recently menopausal women.

METHODS

The Kronos Early Estrogen Prevention Study (KEEPS; NCT00154180) was a randomized, placebo-controlled, double-blinded multicenter clinical trial to evaluate the cardiovascular and cognitive effects of o-CEE, transdermal 17β-estradiol, and placebo in healthy women who experienced natural menopause between 42 and 58 yr of age. Enrollment occurred at nine US sites between August 2005 and July 2008, with final visits completed in 2012. To be eligible for KEEPS, women could not have taken estrogen- or progestin-containing medications within 6 mo of randomization and had to have had their last menstrual period within 6–36 mo, also having a plasma follicle-stimulating hormone (FSH) level of ≥35 ng/mL and plasma E2 level of <40 pg/mL. The women were generally healthy, as determined by exclusion criteria published previously (21).

Institutional Review Board numbers for KEEPS institutions.

The central KEEPS and Phoenix KEEPS [Institutional Review Board (IRB) protocol by the Western IRB] numbers were as follows: study no. 1058663 and Western IRB pro no. 20040792KEEPS (main study and cognitive substudy), no. 10-02980 and MDBHAS no. 11-05383; Brigham and Women’s Hospital (Partners): no. 2004-P-002144 BWH; Mayo Clinic: no. 2241-04; Columbia: IRB no. AAAA-8062; Yale: 0409027022; University of Utah: 13257; Einstein/Montefiore: 04–08–213; University of Wisconsin: H-2005–0059; University of California San Francisco: KEEPS (main study and cognitive substudy) no. 10-02980; University of Washington IRB no. 26702; Phoenix Veterans Affairs Health Care System IRB no. 01048.

Participants were randomized to receive ≤4 yr of double-blinded treatment with 0.45 mg/day o-CEE daily, 50 µg/day t-E2 twice weekly with both oral 200 mg/day micronized progesterone for the first 12 days of each month to protect the uterus, or placebo pills and patches. Institutional review boards (IRBs) at the participating sites approved the study procedures and all participants provided written, informed consent.

Of the 727 women participating in KEEPS, serum was evaluated from 362 women who had samples available and who were deemed compliant to treatment (by counts of pills and patches) from all enrollment centers. Androstenedione (ng/dL), testosterone (ng/dL), E2 (pg/mL), estrone (pg/mL), FSH (IU/L), and LH (IU/L) were measured in serum samples collected at baseline (before randomization) and 48 mo after randomization. The protocol for the blood drawing included that it was in the week before the addition of progesterone, so the progesterone effect contaminated the results least. Hormone levels were measured at the clinical core laboratory at the Mayo Clinic, Rochester, MN.

Seventeen β-estradiol and estrone were extracted with methylene chloride, and after derivatization with dansyl chloride, high-pressure liquid chromatography (HPLC) was used before introduction of the derivatized sample extract into the tandem mass spectrometry (LC-MS/MS) (Agilent Technologies, Santa Clara, CA). Testosterone was measured by liquid chromatography-tandem mass spectrometry (LC-MS/MS; Agilent Technologies). FSH and LH were measured by respective, specific, two-site immunoenzymatic assays performed on a DxI 800 automated immunoassay system (Beckman Instruments, Chaska, MN). The intra-assay and interassay coefficients of variability (CVs) were within acceptable ranges for high-sensitivity testosterone, E2, estrone, FSH, and LH (Table 1).

Table 1.

Hormone intra- and interassay CVs

Hormone	Intra-Assay CV						Interassay CV
17β-Estradiol, pg/mL	0.23	0.50	0.74	35	151	405	0.29	0.50	0.77	32	140	382
17β-Estradiol variation, %	11.8	7.3	6.0	1.6	1.5	1.4	10.8	8.5	6.9	5.1	4.6	4.8
Estrone, pg/mL	0.30	0.50	0.84	32	142	389	0.25	0.51	0.85	30	131	355
Estrone variation, %	17.8	7.5	6.1	2.5	1.7	1.2	12.0	9.5	7.9	7.4	7.1	6.6
Follicle-stimulating hormone, mIU/mL	8.6	47.1					6.5	16.7	58.0
Follicle-stimulating hormone variation, %	32.	2.8					3.6	3.2	4.7
Luteinizing hormone, mIU/mL	1.2	38.5					1.4	15.6	48.8
Luteinizing hormone variation, %	4.3	4.0					9.3	6.0	6.0
Testosterone, high sensitivity, ng/dL	0.65	4.3	48	118	832		0.69	4.3	45	117	841
Testosterone, high sensitivity, variation, %	7.4	6.1	9.0	2.3	0.9		8.9	6.9	4.0	3.6	3.5

CV, coefficient of variability.

Androstenedione was assayed by adding a known amount of deuterated d7-androstenedione to a 0.1-mL plasma sample as an internal standard, followed by extraction of androstenedione and the internal standard using a solid-phase cartridge, from which they were eluted with methanol. The extracts were then dried down under nitrogen reconstituted with 75 µL of 70/30 methanol/H₂O containing 1 µg/mL of estriol to prevent adsorption to equipment components and analyzed by liquid chromatography-tandem mass spectrometry using multiple-reaction monitoring (3, 11, 22, 37, 43, 46).

Data analysis.

Multiple regression analyses were conducted to examine the effects of o-CEE and t-E2 therapy on hormone levels in month 48, adjusting for baseline hormone levels. FSH and LH levels were normally distributed at both data collection time points, but androstenedione, testosterone, E2, and estrone measures required log transformation to normalize their distributions. The treatment variable was coded using two indicator variables: o-CEE (1 = yes, 0 = no) and t-E2 (1 = yes, 0 = no); the placebo group was the “reference” group represented by zeroes for all two indicator variables. To evaluate the influence of time since menopause on therapy response, time in months since menopause was included in the regression models. Linearity assumptions underlying the regression model were checked by using scatterplots and partial regression plots. Plots of the studentized residuals against the unstandardized predicted values were used to confirm homoscedasticity. Normality of residuals was inspected using Q-Q plots of the studentized residuals. An adjustment for multiple comparisons was taken into account using the Benjamini and Hochberg (5) method with a false discovery rate set at 0.05. All models were estimated using R 3.5.1 (35).

Exploring associations among gonadotropins and ovarian hormones: network analysis.

A network analysis technique (14, 16) was used to gain insight into the covariance of changes from baseline in hormone levels and describe the pattern of relationships within treatment groups. That is, the network approach allowed us to visualize the “interconnectivity” of the hormones or how these hormones are related among themselves within each HT group. It also permitted the identification of specific hormones that were more interconnected within group. The analysis produced a network of associations comprised of nodes representing hormone changes from baseline and edges representing the associations between the changes in hormones. The edges can directly be interpreted as non-zero partial correlation coefficients; thus, for example, the edge between hormone A and hormone B is their association after controlling for all other non-zero edges in the group. In the network plot, thicker (or saturated) edges represent stronger associations. Positive and negative associations were plotted in different colors to facilitate interpretation. The Fruchterman-Reingold algorithm (17) was used to plot nodes with more connections closer together. The partial correlations were estimated using a Gaussian Graphical Model and a least absolute shrinkage and selection operator (LASSO) (40) regularization technique that estimates all pairwise associations that are relevant (i.e., non-zero) to the underlying network structure (16) and the covariation in the data. The tuning parameter that was used with LASSO to control for the applied regularization was selected by minimizing the extended Bayesian information criterion (10, 15). The EBIC criterion was used to select the best-fitting regression function defining each node in the final network structure. Using 1,000 samples, we calculated nonparametric bootstrapped confidence intervals for the estimated edges (partial correlations). The structure of the multigroup networks was estimated using the software package igraph (22a). A residualized change model was used instead of a difference score model to estimate change from baseline using data from 362 participants with hormone data at both time points (baseline and month 48). That is, a linear regression model was applied using baseline hormone level as a predictor. This model has a higher power than difference scores (42, 48). The unstandardized residuals for each of the six hormones were used as outcomes to conduct the network analysis.

RESULTS

Of the 362 women who were compliant with treatment at 48 mo of the study, 109 were in the o-CEE group, 107 were in the t-E2 group, and 146 were in the placebo group. Demographics of the participating cohort in this analysis are presented in Table 2. The demographic and clinical characteristics of the sample included in the present study were not significantly different from the sample of participants enrolled in the KEEPS study who were noncompliant and/or missing hormone level data either at baseline or in month 48, Also, they did not differ in apolipoprotein E-ε4 (APOE-ε4) carrier status [χ² (1), P = 0.728], ethnic/racial background [χ²(7), P = 0.051)], or years since menopause [t(723) = 0.171, P = 0.864]. The participants included in this study were slightly younger (M = 52.48, SD = 2.48) compared with those in the primary KEEPS study (52.88, SD = 2.63) [t(725) = 2.390, P = 0.017].

Table 2.

Participant demographics

Characteristic	Total Sample	o-CEE	t-E2	Placebo
Age [means (SD)]	52.45 (2.49)	52.51 (2.62)	52.53 (2.50)	52.35 (2.38)
Years since menopause at baseline [means (SD)]	1.43 (0.719)	1.48 (0.78)	1.49 (0.65)	1.35 (0.71)
Ethnicity [n (%)]
Asian/Indian	3 (0.9)	2 (1.9)		1 (0.7)
Black/African American	19 (5.5)	4 (3.8)	3 (3.1)	12 (8.5)
Caucasian/White	293 (85.4)	91 (87.5)	85 (86.7)	117 (83.0)
Chinese	4 (1.2)	1 (1.0)	2 (2.0)	1 (0.7)
Filipino	1 (0.3)	–	–	1 (0.7)
Hispanic	18 (5.2)	5 (4.8)	5 (5.1)	8 (5.7)
Japanese	1 (0.3)		1 (1.0)
Other	4 (1.2)	1 (1.0)	2 (2.0)	1 (0.7)
APOE-ε4 [n (%)]	86 (25.9)	32 (32.8)	29 (30.5)	25 (18.0)

APOE-ε4, apolipoprotein E-ε4; o-CEE, oral conjugated equine estrogen; t-E2, transdermal 17β-estradiol. Note that treatment groups did not differ in age (F = 0.22, P = 0.81), years since menopause at baseline (F = 1.50, P = 0.26), ethnic composition (χ² = 12.14, P = 0.60), or educational attainment (χ² = 14.70, P = 0.14). The proportion of APOE-ε4 carriers differed by treatment group (χ² = 7.92, P = 0.02).

At baseline (before initiation of study treatment), both mean and median hormone levels were similar among the assigned treatment groups (Table 3). After 48 mo of treatment, observed hormone levels (Table 3) varied with treatment compared with placebo. Parameter estimates for the separate multiple regression models evaluating the differential effects of o-CEE and t-E2 therapy on individual hormone levels in month 48 after adjusting for baseline hormone levels are shown in Table 4. Compared with the placebo group, in month 48 the t-E2 group had significantly lower levels of FSH (β = −24.703, P < 0.001) and LH (β = −4.557, P < 0.001) and higher levels of estrone (β = 0.583, P < 0.001) and 17β-estradiol (β = 1.511, P < 0.001). Androstendione levels in month 48 were marginally higher for the E2 group compared with the placebo group, but the difference was not statistically significant after adjustment for multiple comparisons (P = 0.044 > 0.039). The o-CEE group had significantly higher levels of testosterone (β = 0.158, P < 0.001), as well as estrone (β = 1.123, P < 0.001) and E2 (β = 0.707, P < 0.001), than the placebo group. Androstenedione levels in the o-CEE group were similar to those of the placebo group. The levels of FSH in month 48 were significantly lower for the o-CEE treatment group (β = 19.360, P < 0.001), but LH levels did not differ (β = 0.198, P = 0.879) compared with placebo. There was a marginal average effect of time since menopause on FSH in month 48 (Table 3). However, interactions between treatment groups and time since menopause for FSH were not significant, indicating that time since menopause did not have an independent effect on hormone levels in month 48.

Table 3.

Hormone levels at baseline and month 48 by time since menopause at baseline and treatment group

	Baseline Hormone Levels						Hormone Levels in Month 48
Treatment	Andro, ng/dL	FSH, IU/L	LH, IU/L	Testo, ng/dL	E1, pg/mL	E2, pg/mL	Andro, ng/dL	FSH, IU/L	LH, IU/L	Testo, ng/dL	E1, pg/mL	E2, pg/mL
	6–10.8 mo Since menopause (n = 114)
o-CEE (n = 32)
Mean	75.61	77.78	36.94	20.74	42.94	25.07	66.00	62.91	34.79	21.50	103.32	15.72
SD	35.31	27.30	13.38	13.51	24.80	32.54	27.96	23.57	10.92	9.54	71.32	10.60
t-E2 (n = 25)
Mean	62.96	80.64	38.28	16.00	36.60	26.03	69.50	60.95	30.74	17.56	52.16	44.09
SD	27.25	41.22	16.63	6.64	17.83	34.07	40.07	27.04	9.63	9.03	22.75	29.50
Placebo (n = 56)
Mean	61.23	76.27	37.44	15.77	43.70	32.89	59.76	85.58	36.12	16.61	29.19	6.94
SD	26.94	32.39	16.82	6.03	34.67	54.23	22.84	31.89	14.33	11.08	12.83	3.48
	12–21.6 mo Since menopause (n = 134)
o-CEE (n = 42)
Mean	62.07	91.09	41.77	16.56	32.35	11.52	53.55	68.31	33.28	18.95	118.19	16.99
SD	25.29	32.58	14.48	7.42	14.87	14.76	20.92	33.47	14.38	8.66	118.16	10.98
t-E2 (n = 45)
Mean	69.18	99.29	44.09	18.67	34.53	19.62	67.20	70.92	32.84	18.52	53.91	48.04
SD	34.77	40.23	14.24	10.10	20.48	30.82	36.07	35.94	12.29	9.78	28.05	44.56
Placebo (n = 47)
Mean	57.21	83.51	39.23	18.04	31.82	12.89	55.32	81.92	32.74	22.05	28.79	7.58
SD	25.46	31.60	15.54	9.04	24.10	28.38	20.68	31.78	11.98	27.67	10.81	3.35
	22.8–36 mo Since menopause (n = 119)
o-CEE (n = 37)
Mean	52.51	105.99	43.88	16.17	25.92	6.79	54.00	74.41	38.49	21.34	93.47	13.72
SD	18.13	37.05	15.95	8.05	10.99	3.37	17.70	26.16	14.61	11.75	47.85	6.32
t-E2 (n = 37)
Mean	53.92	97.49	42.42	14.87	26.88	11.24	58.19	60.08	29.82	15.37	51.54	47.89
SD	22.02	40.88	17.07	6.92	22.00	31.90	21.28	20.92	10.67	6.88	27.63	40.50
Placebo (n = 45)
Mean	54.86	95.33	40.20	17.79	28.62	7.34	51.36	87.16	34.07	15.77	27.58	7.07
SD	24.15	35.02	17.05	16.18	10.93	3.79	25.10	35.55	13.54	7.34	11.13	4.45

Andro, androstenedione; E1, estrone; E2, estradiol; FSH, follicle-stimulating hormone; LH; luteinizing hormone; o-CEE, oral conjugated equine estrogen; SD, standard deviation; t-E2, transdermal estradiol; Testo, testosterone.

Table 4.

Model estimates by outcome measure after adding time since menopause (in mo)

					95% CI
Model (Outcome)	Estimate	SE	t	P Value†	Lower	Upper
Model 1 (LnAndro_48)
(Intercept)	1.709	0.176	9.701	<0.001	1.363	2.056
LnAndro_0	0.567	0.041	13.676	<0.001	0.486	0.649
o-Cee	−0.010	0.043	−0.223	0.823	−0.094	0.074
t-E2	0.087	0.043	2.023	0.044	0.002	0.171
Time since menopause	−0.001	0.002	−0.619	0.536	−0.005	0.003
Model 2 (FSH_48)
(Intercept)	43.524	3.993	10.899	<0.001	35.670	51.377
FSH_0	0.546	0.036	15.372	<0.001	0.476	0.616
o-Cee	−19.360	3.037	−6.375	<0.001	−25.332	−13.388
t-E2	−24.703	3.061	−8.071	<0.001	−30.722	−18.683
Time since menopause	−0.308	0.149	−2.062	0.040	−0.601	−0.014
Model 3 (LH_48)
(Intercept)	16.884	1.817	9.291	<0.001	13.310	20.457
LH_0	0.491	0.034	14.295	<0.001	0.424	0.559
o-Cee	0.198	1.305	0.152	0.879	−2.368	2.764
t-E2	−4.557	1.315	−3.466	<0.001	−7.143	−1.972
Time Since Menopause	−0.097	0.063	−1.535	0.126	−0.221	0.027
Model 4 (LnTesto_48)
(Intercept)	0.726	0.125	5.803	<0.001	0.480	0.972
Lntesto_0	0.733	0.042	17.602	<0.001	0.651	0.815
o-Cee	0.158	0.047	3.380	<0.001	0.066	0.251
t-E2	0.010	0.047	0.206	0.837	−0.083	0.102
Time since menopause	0.001	0.002	0.202	0.840	−0.004	0.005
Model 5 (LnE1_48)
(Intercept)	2.113	0.216	9.798	<0.001	1.689	2.537
LnE1_0	0.317	0.056	5.630	<0.001	0.207	0.428
o-Cee	1.123	0.070	16.148	<0.001	0.987	1.260
t-E2	0.583	0.070	8.325	<0.001	0.445	0.720
Time since menopause	0.005	0.003	1.411	0.159	−0.002	0.012
Model 6 (LnE2_48)
(Intercept)	1.609	0.154	10.425	<0.001	1.305	1.912
LnE2_0	0.086	0.044	1.944	0.053	−0.001	0.174
o-Cee	0.707	0.096	7.370	<0.001	0.518	0.895
t-E2	1.511	0.096	15.671	<0.001	1.321	1.700
Time since menopause	0.004	0.005	0.777	0.437	−0.006	0.013

CI, confidence interval; FSH, follicle stimulating hormone, LnAndro, natural logarithm of androstenedione; LH, luteinizing hormone; LnE1, natural logarithm of estrone; LnE2, natural logarithm of estradiol; o-CEE, oral conjugated equine estrogen; t-E2, transdermal estradiol; Testo, testosterone.

^†Unadjusted P values.

Exploratory network analysis results.

Associations among the changes in gonadotropins and steroid hormones after 48 mo of randomization differed by treatment assignment (Fig. 1), as tested by network analysis of the associations (partial correlations) among all of the hormones across HT groups. More hormone interconnectivity was observed in the t-E2 and the o-CEE groups compared with the placebo group. In the t-E2 group, changes in estrone levels showed more interconnection with other hormone levels, as demonstrated by more edges/lines in Fig. 1. In the placebo group, there was a stronger negative association between changes in estrone and FSH than in both the t-E2 and o-CEE groups. The association between estrone and testosterone was positive for all of the groups but was weaker in the o-CEE group.

Fig. 1.

Empirical network structure for pituitary and ovarian hormones. Blue circles represent pituitary hormones, and orange circles represent ovarian hormones. Thickness of the lines indicates the strength of the association between the variables. Thicker lines represent stronger associations. Line colors indicate the direction of the association; green lines are positive associations, and red lines are negative associations. o-CEE, oral conjugated equine estrogen; t-E2, transdermal 17β-estradiol.

In each of the three randomized groups, the associations among testosterone, androstenedione, and estrone were positive, and a slight positive association was seen between LH and androstenedione as well. Overall, the number of nonzero associations between the hormones was greater in the t-E2 group than in the placebo group, such that associations emerged in the t-E2 group that were not present in the placebo group. In the t-E2 group, the negative association between FSH and E2 was stronger, whereas the positive association between androstenedione and estrone was weaker, than that seen in the placebo group. There was a negative association in the t-E2 group between estrone and LH that was not noted in the placebo group. The association between estrone and androstenedione was weaker in the t-E2 group than in the placebo group. Additionally, in the t-E2 group, a negative association was found between testosterone and E2 that was not observed in the placebo group.

In the o-CEE group, the negative association between FSH and 17β-estradiol was weaker than the same association in the t-E2 group. Also, in the o-CEE group, a negative association was found between FSH and testosterone, whereas a positive association emerged between testosterone and LH; these effects were not seen in the other groups. Finally, a weak positive association was found between E2 and testosterone, whereas no association was found between estrone and androstenedione in the o-CEE group.

DISCUSSION

The current analysis provides the first graphical depiction and overview of the impact of two different types of HT on associations among circulating gonadotropic and reproductive hormones in recently menopausal women. The relationships among the changing hormone levels across two time points demonstrated expected negative feedback control between E2 and FSH. Time since menopause at enrollment did not appear to have an independent effect on hormone levels in month 48. Although the expected negative association between E2 and FSH was seen in both treated groups, there was a weaker association in the o-CEE group. o-CEE consists of numerous biologically active hormones, some of which are nonhuman (33). As seen in oral preparations, some of these components would undergo metabolism in the liver before reaching the peripheral circulation due to the gastric-hepatic circulation. The principle estrogen measured in the serum of the o-CEE group was estrone, which is less biologically active than E2. Thus, these estrogen metabolites may alter the relationship between E2 and FSH. It is not surprising that it is difficult to fully explicate the hormonal associations for the o-CEE group, given the interconversion between estrone and E2.

Many of the associations seen in the t-E2 group would be expected based on established hormonal regulatory processes in premenopausal women (38). For example, the associations among testosterone, androstenedione, and estrone were, as expected, positive, since androstenedione converts to estrone via the aromatase enzyme, and it is also an intermediate in the conversion of dehydroepiandrosterone (DHEA) to testosterone. Also, as expected, FSH levels were lower in month 48 than at baseline in both treatment groups compared with the placebo group. Other studies have shown associations between FSH levels and various disease states such as cardiovascular disease, diabetes, and adiposity (28, 30, 44, 45, 47). Elevated FSH levels are not known to have direct effects on these adverse outcomes in the postmenopausal state, but instead, elevated FSH levels seem more likely to be a marker of other causal factors. The greater reduction in circulating LH with t-E2 than with o-CEE treatment may have implications for preservation of cognitive function by HT (6), but further research is needed. It is unclear whether pituitary hormone levels and their association with outcomes are related solely to levels of hormone and/or to alterations in hypothalamic-pituitary-adrenal axis responsiveness.

In experimentally induced ovarian follicle-depleted rats, higher levels of circulating androstenedione were associated with cognitive deficits (1, 8). In the current study, androstenedione levels differed by treatment group; however, after adjustment, no statistically significant change was seen in the treatment groups compared with placebo in month 48. How circulating levels of androstenedione associate with cognitive function or menopausal symptoms in postmenopausal women is an important question that remains to be answered.

The exploratory graphical network analyses across treatment groups reveal very strong connections (partial correlations) between estrone and E2, testosterone, and androstenedione as well as LH and FSH. (The bootstrapped confidence intervals are shown in Fig. 2.) The first two relationships are not surprising from a biochemical point of view, given the equilibrium interconversion of the two estrogens and the fact that androstenedione is a precursor directly converted to testosterone. In the third case, FSH and LH are secretory products of the same pituitary cells (gonadotrophs) and can be secreted in tandem. Particularly striking is the observation of moderate to strong relationships of E2 to testosterone in the t-E2 group and of estrone to androstenedione in the placebo group, especially given that these relationships are reciprocally present, albeit weaker, in the opposite groups. These relationships may be explained by androstenedione being an immediate precursor of estrone and testosterone being an immediate precursor of E2. The relative strengths of the relationships may reflect the greater abundance of estrone in the placebo group and, after treatment, the greater abundance of E2 in the t-E2 group. It is noteworthy that neither of these estrogen-to-androgen relationships was detected in the o-CEE group most likely influenced by the o-CEE, resulting in high circulating levels that would interfere with feedback mechanisms. Collectively, these outcomes suggest that treatment with o-CEE disrupts the pathways connecting these two important classes of sex steroids.

Fig. 2.

Bootstrapped confidence intervals of estimated edges (partial correlations) for the estimated network of 6 hormones by randomized group. Black line indicates the bootstrapped values, red line shows the sample values, and gray area shows the bootstrapped confidence intervals. The x-axis shows the partial correlations. E2, 17β-estradiol; o-CEE, oral conjugated equine estrogen.

Because the relationships of the hormone associations in the network analysis are different within the three treatment groups, it is likely that the hormone treatments (or lack thereof in the placebo) have influenced the connections. Together, the observed differences in the hormonal network connections may help explain differences in both the short- and long-term outcomes associated with different HT formulations. For example, brain ventricular volumes increased and cortical volumes decreased to a greater extent in women on o-CEE than on placebo (26). Furthermore, in the subset of women who were APOE-ε4 carriers, those randomized to transdermal E2 had less accumulation of β-amyloid (a putative marker of Alzheimer’s disease) than those on placebo and o-CEE (25, 27). As expected, women treated with hormones report improvement in vasomotor symptoms and in the o-CEE group, an additional improvement in mood (21). Systematic study of the associated changes with cognitive and mood data could reveal which hormones, or relationships among hormones, associate with cognitive and mood changes.

Limitations.

Women enrolled in KEEPS had undergone natural menopause and had their uteruses and ovaries at the time of study; thus, the relationships noted here may not apply to women who underwent oophorectomy or hysterectomy before or after the age of natural menopause. Also, reproductive history was not collected on this cohort, nor was a history of other conditions that might have affected ovarian function before menopause (e.g., polycystic ovarian syndrome, endometriosis). Additionally, other hormones such as progesterone, thyroid hormones, sex hormone-binding globulin, and corticosteroids, which might also affect the interpretability of the observed associations and menopausal state of the gonadotropic-pituitary axis, were not included in this analysis. Because sex hormone-binding globulin was not included in the analysis, serum hormone levels reflect the total hormone level and not the free or bioavailable state. Finally, our findings are based on observation of associations; thus, the results are hypothesis generating and may set the stage for future studies that could lead to enhanced understanding of hormonal interactions in women, with and without exogenous hormone treatment, as they age.

Perspectives and Significance

Levels of ovarian and pituitary hormones and their interrelationships differ among women who have recently undergone natural menopause during administration of two common regimens of menopausal HT. Further studies evaluating the relationships among these changes in circulating hormones with other postmenopausal changes, such as cognition, neuroimaging, and menopausal symptomology, may help elucidate the mechanisms of differential effects of various HT regimens on these clinical outcomes. Importantly, further definition of the interrelationships among hormonal and clinical changes with HT could identify new approaches to managing menopause-related symptoms and pathophysiological events.

GRANTS

This work was supported in part by a Mayo Clinic Mentored Research Award as well as Grant National Institute on Aging (NIA) RF1 AG057547 to K. Kantarci, C. E. Gleason, and V. M. Miller; NIA AG-028084, ADHS14-052688, and the National Institutes of Health (NIH) Alzheimer’s Disease Centers P30-AG-019610 to H. A. Bimonte-Nelson; P50-AG-033514 to C. E. Gleason; and the Mayo Clinic Foundation for Research and Education. KEEPS was funded by grants from the Aurora Foundation to the Kronos Longevity Research Institute, from National Heart, Lung, and Blood Institute Grant HL-90639 to V. M. Miller, Mayo Clinic CTSA 1 UL1 RR024150, the Mayo Foundation, Brigham and Women’s Hospital/Harvard Medical School CTSA, CTSA UL1 RR024139 and UCSF CTSA UL1 RR024131 from the National Center for Advancing Translational Sciences (NCATS), a component of the NIH and NIH Roadmap for Medical Research.

The content of this article is solely the responsibility of the authors and does not necessarily represent the official view of the NCATS or NIH. Study medications were supplied in part by Bayer Health Care and by Abbott Pharmaceuticals. Funding for FSH/LH assays was made available through support offered by the Department of Obstetrics, Gynecology, and Reproductive Sciences at Yale and the Benneck-Polan Family Foundation.

DISCLOSURES

Kejal Kantarci serves on the data safety monitoring board for Takeda Global Research and Development Center, Inc., and receives research support from Avid Radiopharmaceuticals and Eli Lilly and funding from NIH and the Alzheimer’s Drug Discovery Foundation. Other authors have nothing additional to disclose.

The Aurora Foundation, Bayer HealthCare, Abbott Pharmaceuticals, and Pfizer Pharmaceuticals had no input into the design or conduct of the study or the writing, review, or approval of this article.

AUTHOR CONTRIBUTIONS

J.M.K., H.A.B.-N., and V.M.M. conceived and designed research; N.M.D. performed experiments; N.M.D. analyzed data; J.M.K., N.M.D., H.A.B.-N., C.E.G., K.K., J.E.M., H.S.T., E.A.B., R.A.L., M.L.C., L.P., G.N.-P., F.N., S.M.H., and V.M.M. interpreted results of experiments; N.M.D. prepared figures; J.M.K. drafted manuscript; J.M.K., N.M.D., H.A.B.-N., C.E.G., K.K., J.E.M., H.S.T., E.A.B., R.A.L., M.L.C., L.P., G.N.-P., F.N., S.M.H., and V.M.M. edited and revised manuscript; J.M.K., N.M.D., H.A.B.-N., C.E.G., K.K., J.E.M., H.S.T., E.A.B., R.A.L., M.L.C., L.P., G.N.-P., F.N., S.M.H., and V.M.M. approved final version of manuscript.