Long-term changes in plasma proteomic profiles in pre- and postmenopausal Black and White women: The Atherosclerosis Risk in Communities (ARIC) study

Abstract

OBJECTIVE:

The activity, localization, and turnover of proteins within cells and plasma may contribute to physiologic changes during menopause and may influence disease occurrence. We examined cross-sectional differences and long-term changes in plasma proteins between premenopausal and naturally postmenopausal women.

METHODS:

We used data from 4,508 (19% Black) women enrolled in the Atherosclerosis Risk in Communities (ARIC) study. SOMAscan multiplexed aptamer technology was used to measure 4,697 plasma proteins. Linear regression models were used to compare differences in proteins at baseline (1993–95) and 18-year change in proteins from baseline to 2011–13.

RESULTS:

At baseline, 472 women reported being premenopausal and 4,036 women reported being postmenopausal, with average ages of 52.3 and 61.4 years, respectively. A greater proportion of postmenopausal women had diabetes (15 vs. 9%), used hypertension (38 vs. 27%) and lipid-lowering medications (10 vs. 3%) and had elevated total cholesterol and waist girth. In multivariable adjusted models, 38 proteins differed significantly between premenopausal and postmenopausal women at baseline, with 29 of the proteins also showing significantly different changes between groups over the 18-year follow-up as the premenopausal women also reached menopause. These proteins were associated with various molecular/cellular functions (cellular development, growth, proliferation and maintenance), physiological system development (skeletal and muscular system development, and cardiovascular system development and function), and diseases/disorders (hematological and metabolic diseases and developmental disorders).

CONCLUSIONS:

We observed significantly different changes between pre- and postmenopausal women in several plasma proteins that reflect many biological processes. These processes may help to understand disease development during the postmenopausal period.

INTRODUCTION

Menopause marks a milestone in the lives of women as it signifies the end of reproductive capacity ¹. The loss of ovarian function leading to diminished exposure to endogenous ovarian estrogen has also been reported to be associated with the occurrence of several chronic conditions such as osteoporosis, breast and endometrial cancers, and cardiovascular disease ^2–4. However, recent evidence suggests that health-related factors such as obesity, hypertension, and smoking occurring long before menopause may also influence physiologic changes during the menopausal transition and afterwards, and may contribute to the role of menopause in chronic diseases, especially cardiovascular diseases ^5,6.

Several population-based studies have reported adverse changes in glucose metabolism, fat composition and deposition, blood pressure, lipids, inflammatory and coagulation factors during the menopausal transition ^2,7–9. These physiologic changes are modulated by several regulatory processes. These include genetics, the activity, localization and turnover of proteins within cells ¹⁰, and adverse aging-related physiological changes associated with plasma levels of key proteins such as C-reactive protein, adiponectin, leptin, retinol-binding protein 4, fetuin-A, and sex hormone-binding globulin ¹¹. The profiling of human plasma proteome is an area of great interest, especially as many disease processes can cause changes in the concentrations of plasma proteins ^12–14. Therefore, the determination of plasma protein concentrations and changes in plasma protein concentrations over time as women transition from premenopausal to postmenopausal status may offer important diagnostic information to better understand the mechanisms underlying the etiology of chronic diseases that occur during the menopausal transition and in the postmenopausal years, as well as offer potential therapeutic targets for such diseases ^12–14.

Despite decades of research, only a few proteins have been identified from either serum or plasma in studies that included a small number of pre- or postmenopausal women ¹⁵. Single proteins do not act in isolation, and when studied individually in the plasma, they have limited clinical value in predicting chronic diseases ¹². To date, no large epidemiologic studies have comprehensively examined profiles of the plasma proteome during the menopausal transition to uncover protein changes that relate to menopause, independent of chronological age and risk factors for chronic diseases. A recent study profiling the proteome of ovarian cortical tissue from six women found different expression of proteins between premenopausal and postmenopausal women, with 151 proteins upregulated and 65 downregulated ¹⁶. This study did not account for confounding factors that may potentially influence the observed differences in proteins between pre- and postmenopausal women. Furthermore, these findings may not be generalizable to women of other races in light of widely reported racial/ethnic differences in the timing of menopause ¹⁷.

Blood is the vehicle for the accumulative evidence for most pathological insults, however, the large dynamic concentration range of proteins in serum made it a challenging proteome to effectively characterize ¹⁵. Recent advances in high-throughput technology have enhanced the large-scale characterization of circulating proteins ¹⁸. The aim of this exploratory study of women enrolled in the Atherosclerosis Risk in Communities (ARIC) study was twofold: first, to characterize cross-sectional differences and long-term changes in plasma proteins between premenopausal and naturally postmenopausal women accounting for potential confounding factors, and second, to examine the potential effect modification by race, cigarette smoking, or hormone therapy use.

METHODS

Study population

The ARIC study is a multi-center prospective cohort study of 15,792 participants (8710 women) aged 45–64 who were recruited and examined in 1987–1989 from four U.S. communities in North Carolina, Mississippi, Minnesota, and Maryland. Through 2020, six follow up in-person examinations were conducted in 1990–92, 1993–95, 1996–98, 2011–13, 2016–17, and 2018–19. Participants have been contacted by phone annually or semi-annually (since 2012). Institutional review boards from all field centers and the ARIC Coordinating Center approved the ARIC protocol, and all participants provided written informed consent. Details of the study design and methods for ARIC are described elsewhere ^19,20.

Protein profiles were analyzed in 2019 from plasma samples collected in 1993–95 (the baseline for the current study) and 2011–13. Of the 7,170 women who attended the 1993–95 exam, we excluded those reporting surgical menopause, unknown menopausal status, or unknown type of menopause. This left 5162 pre- or peri-menopausal (hereafter referred to as premenopausal) or naturally menopausal women eligible for the current study. From this sample, women who reported races other than Black or White (n=28) and those with no proteins measured in 1993–95 (n= 626) were excluded, resulting in an analytic sample of 4508 women (472 premenopausal and 4036 naturally menopausal women).

Plasma protein assessment

The ARIC protocol for blood sample collection and processing is consistent with recommended practice for epidemiological studies ^21,22. Blood was drawn from participants at each ARIC exam. Protein biomarkers were assessed on blood samples that were stored at −80 °C within 90 min from venipuncture obtained from participants who attended the 1993–95 and 2011–13 ARIC exams. We used SOMAscan (v. 4) multiplexed modified DNA-based aptamer technology (Somalogic, Inc., Boulder, CO, USA). The SOMAscan assay indirectly determines individual protein concentrations by first isolating them and then measuring the identity and number of SOMAmer reagents that bind to the proteins in the blood sample. The protein concentrations are quantified by their fluorescence intensity values, which are reported as relative fluorescent units (RFUs) ²². RFUs reflect relative protein quantity and not absolute protein concentrations, and are not directly comparable across different analytes in the SOMAscan assay ²³. For instance, a twofold increase in RFU values does not indicate a twofold increase in absolute protein quantity ²³. At both ARIC exams, 4,877 aptamers identified 4,697 unique human proteins or protein complexes. Protein analyte measurements underwent standard SOMAscan normalization, calibration and data quality control processes, as described in detail elsewhere ^14,22,24,25. Briefly, hybridization control normalization was undertaken to correct for systematic biases and adjust for nuisance variance on the basis of individual wells on the SOMAscan plates ^22,24. Median signal normalization, an intraplate normalization procedure conducted within wells of the same sample class, was performed to remove sample-to-sample differences in total relative fluorescence unit brightness that may be due to differences in overall protein concentration, pipetting variation, variation in reagent concentrations, assay timing, and other sources of systematic variability within a single plate run ^22,24. Finally, calibration normalization, which is an inter-plate procedure, was performed separately on each SOMAmer reagents and applied to all the samples in a plate ²⁴. Calibrator samples for each SOMAmer reagent were included in each plate and used to correct for plate-to-plate variation based on established global reference standards ²². Eleven wells of the 96-well plates of the SOMAscan assay were allocated for quality control duplicate ARIC samples. These duplicates were used to control for batch effects as well as to estimate the accuracy, precision, and buffer background of the assay over time ^14,26. In ARIC, the SOMAscan assay has been shown to have excellent precision and high reproducibility.²²

Among 42 randomly selected individuals who had proteins measured from previously unthawed plasma collected at exam 2 (1990–1992), exam 5 (2011–2013), and a repeated exam 5 (4–9 weeks after exam 5) where the repeated exam 5 samples were split as 2 vials, with their identities masked to personnel processing the SOMAscan modified aptamer assays, the coefficients of variation averaged less than 5.0%, the Spearman correlation coefficients were greater than 0.89, and the interclass correlation coefficients were greater than 0.96 among the split samples ²². Compared to measures of 9 proteins quantified using clinical assays, 6 proteins were highly correlated across platforms, including long-term changes for 5 proteins. Over the short term (4–9 weeks), only 1 protein analyte had a statistically significant change in the same individuals across platforms ²². Over the long term (approximately 20 years), 284 protein analytes (7.7%) had a significant change, thus 163 (4.4%) increased and 121 (3.3%) decreased. The long-term variation of the protein analytes as captured by principal components was correlated with age (−0.73) and estimated glomerular filtration rate (EGFR) (0.60) ²².

Other measures

At each ARIC exam, standardized protocols were used to collect information on demographics, anthropometrics, lifestyle and behavioral factors, medical history, cardiovascular risk factors, biomarkers, and medications. Age, sex, race, education, smoking status, age at menarche, parity, and the use of oral contraceptives and postmenopausal hormone therapy were all self-reported. Women were considered to be postmenopausal if they reported not having had a menstrual period within the two years before the ARIC exam. This definition is conservative and ensures that perimenopausal women were not classified as postmenopausal. Postmenopausal women who reported cessation of menstruation that was not preceded by bilateral oophorectomy or medical procedures that influence menstrual cessation such as radiation therapy were classified as having natural menopause. Participants brought to each exam all medications taken in the two weeks prior to the exam; medication names were transcribed and coded. Blood pressure was measured three times using a random-zero sphygmomanometer with participants seated and after five minutes of rest. The average of the second and third consecutive measurements was used for the current analysis. Body mass index was calculated by dividing weight in kilograms by height in meters squared. Waist circumference was measured in centimeters at the level of the umbilicus. Sports-related physical activity was assessed by the Baecke questionnaire ²⁷. Glucose, total cholesterol and high-density lipoprotein (HDL) cholesterol were all determined by enzymatic methods. Diabetes was defined as a fasting blood glucose level ≥126 mg/dl, non–fasting blood glucose ≥200 mg/dL, a self-reported physician diagnosis of diabetes, or current use of antidiabetic medication. EGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration formula ²⁸. Prevalent cardiovascular disease at baseline was identified using self-reported information enrollment (1987–89) or adjudicated incident events occurring between enrollment and the 1993–95 exam.

Statistical analysis

Characteristics of participants were described using proportions and means (standard deviations). T tests and Chi square tests were used to test for differences in continuous and categorical characteristics, respectively, between premenopausal and postmenopausal women. All proteins were log transformed with base 2 to normalize distributions. Linear regression models were used to compare differences in proteins at baseline (cross-sectional analyses) and change in proteins from baseline to the end of follow up (longitudinal analyses of 18-year change in protein levels). Due to some proteins having changes that were negative or zero that limits log transformation, untransformed values of proteins were used. To ensure that the change in proteins conformed to a normal distribution, participants who had change in proteins that were below the first percentile or above the 99^th percentile of the distribution of the change in proteins were excluded from longitudinal change analyses. For the main analyses, adjustment was made for baseline age, race, ARIC center, education, estimated glomerular filtration rate, smoking, systolic blood pressure, anti-hypertensive medication use, cholesterol medication use, total cholesterol, physical activity, body mass index and hormone therapy use. All p values were adjusted to control for false discovery rate at 5% using the Benjamini-Hochberg method ²⁹. Due to their known associations with the timing of natural menopause ¹, possible effect modifications of postmenopausal status with race, smoking and hormone therapy use were explored for proteins that were significantly different between premenopausal and postmenopausal women in cross-sectional and longitudinal analyses by introducing an interaction term in the multivariable models (one term at a time). All analyses were performed using the SAS version 9.4 software.

Sensitivity analyses were performed applying propensity score methods in an effort to reduce the possibility of bias due to differences in covariate distributions by menopausal status. Premenopausal women were matched to postmenopausal women with 1:2 ratio using optimal variable matching algorithm that minimize the total absolute difference in propensity score across all matches. Propensity scores were generated from a multivariable logistic regression model that included the following baseline characteristics (age, race-center, education, estimated glomerular filtration rate, smoking, systolic blood pressure, anti-hypertensive medication use, cholesterol medication use, total cholesterol, physical activity, body mass index and hormone therapy use (never, former or current user)). The total matched sample of 1383 women included 461 premenopausal and 922 postmenopausal women.

Ingenuity Pathway Analysis (IPA)

To identify and explain biological processes, pathways, and networks of the proteins that were found to be statistically different between pre- and postmenopausal women, Ingenuity Pathway Analysis (QIAGEN IPA, Aarhus, Denmark) was performed. For these analyses, proteins that could not be mapped to the IPA database were excluded. Log beta coefficients from the linear regression models from cross-sectional and longitudinal analyses, and false discovery rate-corrected p values derived from age and multivariable adjusted analyses were uploaded for each protein. IPA core analysis uses Fisher’s exact tests to quantify the probability of overlap between a set of proteins found to be different between pre- and postmenopausal women identified in the current study and a set of proteins known to exist within a specific pathway or process due to random chance ^18,30. P values from the exact test were used to determine the probability that the association between the proteins and the canonical (well-understood biological) pathways and upstream regulators were explained by chance. To determine canonical pathways and upstream regulators, the IPA software calculates a z score which quantifies the likelihood and directionality of the protein expression. The activation Z-score is used to infer likely activation states of biological functions based on comparison with a model that assigns random regulation directions. The activation z score is then used to find likely regulating molecules based on a statistically significant pattern match of up- and down-regulation, and also to predict the activation state (either activated or inhibited) of a putative regulator. A z score less than minus 2 or greater than 2 was used as the threshold for significance when interpreting directionality ³⁰. For these analyses, the default settings in the IPA core analysis were used.

RESULTS

Baseline characteristics

Of the 7,170 women who attended the 1993–95 exam, 4508 women (472 premenopausal and 4036 naturally menopausal women) were included in the current study. Compared to women with protein measures included in the current study, those without protein measures and excluded were younger (59.6 vs. 60.5 years) and a greater proportion of them reported Black race (33% vs 19%). There were no significant differences between these two groups with regard to education, menopausal status, cigarette smoking, body mass index, lipids or the prevalence of cancer or cardiovascular disease. Characteristics of the 4,508 women included in the analytic sample stratified by menopausal status are presented in Table 1. Almost 11% of the women reported being premenopausal with their mean age (standard deviation) being 52.3 (1.12) years compared to the mean age of 61.4 (5.1) years for naturally postmenopausal women. Postmenopausal women had higher waist girth, systolic blood pressure and total cholesterol, with a greater proportion of them being Black women, using hypertensive medications, having had three or more births, or having had diabetes or cardiovascular disease. Hot flashes were more common among premenopausal women than postmenopausal women. There were significant differences in formulation of hormone therapy use between pre-and postmenopausal women. While a greater proportion of postmenopausal women were current users of estrogen alone therapy (14.8 vs. 4.5%), a greater proportion of premenopausal women reported using estrogen and progestin (42.9 vs. 15.4%). There were no significant differences between groups with regard to smoking status, physical activity, age at menarche, or HDL cholesterol.

Table 1.

Baseline characteristics of the 4508 women according to menopausal status, ARIC study 1993–95

Characteristics a	Premenopausal (n = 472)	Postmenopausal (n= 4036)	P value
Age, years	52.3 (1.12)	61.4 (5.1)	<0.001
Race, Black, %	14.0	19.3	0.005
Education (> High school), %	47.0	31.8	<0.001
Smoking status, %			0.634
Former	30.3	31.9
Current	15.7	16.4
Systolic blood pressure, mm Hg	117.6 (17.1)	125.0 (20.2)	<0.001
Hypertension medication use, %	27.1	38.4	<0.001
Diabetes, %	8.9	14.8	0.001
Body mass index, kg/m2	28.4 (6.5)	28.6 (6.1)	0.523
Waist circumference, cm	96.6 (16.7	99.7 (16.0)	<0.001
Sports index	2.4 (0.8)	2.4 (0.8)	0.730
Parity, %			<0.001
None	8.7	8.0
One	11.9	11.0
Two	34.1	24.0
Three or more	45.3	57.0
Age at menarche, years	12.8 (1.6)	12.9 (1.6)	0.244
Currently having hot flashes, %	32.4	6.3	<0.001
Current use of hormone therapy, %			<0.001
Estrogen alone	4.5	14.8
Estrogen and Progestin	42.9	15.4
Lipid-lowering medication use, %	3.4	10.0	<0.001
HDL cholesterol, mg/dL	58.8 (18.7)	57.7 (18.5)	0.234
Total cholesterol, mg/dL	202.7 (37.9)	215.4 (37.6)	<0.001
EGFR, mL/min/1.73m2	95.5 (13.3)	87.6 (15.2)	<0.001
Anticoagulant use, %	0.0	0.8	0.043
Prevalent cardiovascular disease, %	1.5	5.2	<0.001

EGFR: Estimated glomerular filtration rate; HDL: High density lipoproteins

^aValues are mean (standard deviation) for continuous variables and percentages for categorical variables

Cross-sectional analysis

In unadjusted analysis, of the 4,697 proteins tested, 1107 unique proteins were observed to have significantly different levels between premenopausal and postmenopausal women. This reduced to 166 proteins when adjustment for chronological age was made (Supplemental Figure 1, Supplemental Digital Content 9). After controlling for age, race, center, education, estimated glomerular filtration rate, smoking, systolic blood pressure, anti-hypertensive medication use, cholesterol medication use, total cholesterol, physical activity, body mass index and hormone therapy use, 38 proteins levels remained significantly different between premenopausal and post-menopausal women (Table 2). Of these, 17 were higher in premenopausal women than postmenopausal women. These included chordin-like proteins 2 (27.3% higher), prolactin (12.0%) and secreted frizzled-related protein 4 (9.4%), The proteins that were higher among postmenopausal women than premenopausal women included human chorionic gonadotropin (hCG) (63%), hepcidin (59%), ferritin light chain (56%), follicle stimulating hormone (FSH) (55%), ferritin (55%), luteinizing hormone (36%), lutropin subunit beta (11%) and chondroadherin (10%).

Table 2.

Comparison of proteins (in relative fluorescent units) among women who were premenopausal and postmenopausal at baseline, n =4508, ARIC study 1993–95

Proteins a	Menopausal status b		Difference (%) c	P value
	Premenopause, Mean (95% CI)	Postmenopause, Mean (95% CI)
Higher in premenopausal women
Chordin-like protein 2	3529 (3385–3679)	2771 (2699–2847)	27.3	<0.001
Prolactin	9495 (8983–10044)	8481 (8181–8786)	12.0	0.001
Secreted frizzled-related protein 4	8198 (7929–8481)	7496 (7337–7659)	9.4	<0.001
R-spondin-1	2708 (2618–2803)	2507 (2452–2564)	8.0	<0.001
WNT1-inducible-signaling pathway protein 2	11952 (11569–12357)	11098 (10870–11339)	7.7	<0.001
Probable carboxypeptidase X1	1662 (1587–1740)	1547 (1502–1594)	7.4	0.035
N-acetylated-alpha-linked acidic dipeptidase 2	1947 (1865–2034)	1822 (1772–1873)	6.9	0.037
Microfibrillar-associated protein 2	532 (518–545)	498 (489–506)	6.8	0.000
Proactivator polypeptide-like 1	2295 (2200–2394)	2150 (2092–2210)	6.7	0.041
Stanniocalcin-1	1686 (1633–1740)	1598 (1565–1630)	5.5	0.013
UDP-glucuronic acid decarboxylase 1	2114 (2057–2175)	2006 (1970–2042)	5.5	0.002
Secreted frizzled-related protein 1	4116 (3978–4258)	3907 (3824–3995)	5.3	0.041
SPARC-related modular calcium-binding protein 2	1062 (1033–1091)	1011 (993–1029)	5.0	0.005
ADP-dependent glucokinase	1628 (1590–1668)	1554 (1531–1578)	4.8	0.001
Hemojuvelin	817 (800–836)	787 (777–799)	3.8	0.015
Repulsive guidance molecule A	9449 (9268–9628)	9178 (9071–9293)	2.9	0.045
Putative hydrolase RBBP9	1453 (1430–1478)	1417 (1404–1432)	2.5	0.036
Higher in postmenopausal women
Human Chorionic Gonadotropin	1556 (1377–1758)	2534 (2343–2740)	62.9	<0.001
Hepcidin	1515 (1376–1668)	2409 (2265–2562)	59.0	<0.001
Ferritin light chain	5650 (5081–6282)	8792 (8215–9410)	55.7	<0.001
Follicle stimulating hormone	1881 (1708–2072)	2914 (2740–3100)	55.0	<0.001
Ferritin	4417 (3976–4908)	6836 (6392–7317)	54.8	<0.001
Luteinizing hormone	1838 (1688–2003)	2497 (2364–2639)	35.9	<0.001
Lutropin subunit beta	481 (461–501)	534 (520–548)	11.0	<0.001
Chondroadherin	6370 (6093–6663)	7038 (6841–7246)	10.5	<0.001
Prostate-associated microseminoprotein	1150 (1099–1205)	1267 (1230–1304)	10.1	<0.001
Collagen alpha-2(XI) chain	5737 (5473–6009)	6260 (6076–6450)	9.2	0.002
Fatty acid-binding protein, adipocyte	42201 (40510–43993)	45545 (44361–46760)	7.9	0.003
Fatty acid-binding protein, heart	18613 (17830–19430)	19990 (19443–20552)	7.39	0.015
Netrin-4	5757 (5580–5939)	6127 (6005–6252)	6.42	0.001
Guanylate-binding protein 1	16464 (15804–17151)	17511 (17056–17979)	6.37	0.045
Granzyme A	2502 (2422–2585)	2661 (2607–2717)	6.36	0.002
Neogenin	3490 (3423–3561)	3679 (3633–3728)	5.43	0.000
Neuroplastin	5064 (4918–5213)	5334 (5235–5435)	5.36	0.004
BMP-binding endothelial regulator protein	2034 (1974–2097)	2138 (2097–2181)	5.14	0.015
Tissue factor pathway inhibitor	30215 (29369–31108)	31718 (31130–32317)	4.93	0.014
Carbohydrate sulfotransferase 15	4087 (4006–4170)	4258 (4202–4312)	4.14	0.001
WD repeat-containing protein 5	1441 (1405–1479)	1498 (1473–1523)	3.99	0.041

^aThirty-eight (38) proteins were significantly different between pre- and postmenopausal women after controlling for false discovery rate. Model adjusted for age, race center, education, estimated glomerular filtration rate, smoking, systolic blood pressure, anti-hypertensive medication use, cholesterol medication use, total cholesterol, physical activity, body mass index and hormone therapy use

^bBack transformed log₂ means and 95% confidence intervals measured in relative fluorescent units (RFUs).

^cDifference (%) estimated using log₂ transformed beta coefficients from linear regression models

ADP: adenosine diphosphate; BMP: bone morphogenetic protein; RBBP9: retinoblastoma binding protein 9; SPARC: secreted protein acidic and cysteine rich; UDP: Uridine diphosphate; WNT1: Wingless-type.

Longitudinal analysis

There were 1,880 women who had protein measures at both baseline and 2011–13 of whom 324 were premenopausal and 1,556 postmenopausal at baseline. All 324 women who were premenopausal at baseline transitioned to postmenopausal status by 2011–13. Of the 38 proteins that were significantly different between pre- and postmenopausal women at baseline, 29 also showed significantly different changes between pre and postmenopausal women over the approximately 18-year period in multivariable adjusted analysis. The top ten proteins that changed based on statistical significance were putative hydrolase RBBP9, collagen alpha-2(XI) chain, chondroadherin, ferritin light chain, guanylate-binding protein 1, R-spondin-1, secreted frizzled-related protein 4, WD repeat-containing protein 5, luteinizing hormone, and follicle- stimulating hormone (Figure 1, Supplemental Table 1, Supplemental Digital Content 1). Compared with women who were postmenopausal at baseline, premenopausal women at baseline (who transitioned from pre- to postmenopausal status during follow up) had significantly greater positive changes over time in 19 proteins, with the changes ranging from 102% for hCG, 82% for follicle stimulating hormone, 36% for heart fatty acid-binding protein, 33% for R-spondin-1, 28% for luteinizing hormone through to 1.3% for neogenin. Additionally, compared with women who were postmenopausal at baseline, premenopausal women at baseline (who transitioned from pre- to postmenopausal status during follow up) had significant reductions in 10 proteins with changes ranging from −17% for chordin-like protein 2, −15% for proactivator polypeptide-like 1, −13.9% for ferritin light chain, −13.8% for ferritin, −9% for putative hydrolase RBBP9 through to −2% for Netrin-4.

Figure 1.

Percent change in plasma proteins from baseline to end of follow-up among women who were postmenopausal at baseline (n=1556) and those who were premenopausal at baseline and transitioned to postmenopause by the end of follow-up (n=324), ARIC study 1993–2013

At the false discovery rate of 5%, we observed significant interactions of race, smoking status and hormone therapy use with menopausal status at baseline on 18-year changes in proteins. Of the 24 proteins that changed significantly by race, 14 increased as women transitioned from premenopause to postmenopause. Overall, Black women had higher concentrations of follicle stimulating hormone, chondroadherin, collagen alpha-2(XI) chain, hepcidin, heart fatty acid-binding protein, and WNT1-inducible-signaling pathway protein 2. Among the 6 proteins that had negative 18-year changes among women who transitioned from premenopause to postmenopause, Black women had a greater change in chordin-like protein 2, while White women had greater negative changes in ferritin, ferritin light chain, N-acetylated-alpha-linked acidic dipeptidase 2, putative hydrolase RBBP9, and Hemojuvelin (Supplemental Table 2, Supplemental Digital Content 2). With regard to smoking status, 25 proteins changed significantly as women transitioned from premenopausal to postmenopausal status. Focusing on the top 10 proteins by means of their statistical significance, women who reported having ever smoked at baseline had lower changes in follicle stimulating hormone, luteinizing hormone, hCG, chondroadherin, collagen alpha-2(XI) chain, R-spondin-1, hepcidin and neogenin than women who were never smokers at baseline (Supplemental Table 3, Supplemental Digital Content 3). Finally, hormone therapy interacted significantly with menopausal status for 27 of the 29 proteins that changed over the 18 years of follow-up. Among the top 10 proteins by statistical significance, premenopausal women who were ever users of hormone therapy at baseline had higher changes in follicle stimulating hormone, hCG, luteinizing hormone, chondroadherin, collagen alpha-2(XI) chain, R-spondin-1, and neuroplastin but lower changes in chordin-like protein 2 and hepcidin compared to premenopausal women who were never users of hormone therapy at baseline (Supplemental Table 4, Supplemental Digital Content 4)

In sensitivity analysis, the baseline characteristics of the 461 women who were premenopausal and matched to 922 postmenopausal women by propensity scores are shown in Supplemental Table 5, (Supplemental Digital Content 5). Overall, 26 proteins were observed to be different between pre-and postmenopausal women at baseline in multivariable adjusted models (Supplemental Table 6, Supplemental Digital Content 6). Twenty of those 26 proteins replicated findings of the primary analysis. These 20 proteins were arechordin-like protein 2, hepcidin, ferritin light chain, follicle-stimulating hormone, ferritin, hCG, chondroadherin, neogenin, luteinizing hormone, lutropin subunit beta, secreted frizzled-related protein 4, netrin-4, carbohydrate sulfotransferase 15, WNT1-inducible-signaling pathway protein 2, Granzyme A, Prostate-associated microseminoprotein, R-spondin-1, secreted frizzled-related protein 1, BMP-binding endothelial regulator protein, and fatty acid-binding protein (Supplemental Figure 2, Supplemental Digital Content 10).

Ingenuity Pathway Analysis

For the 166 plasma proteins that were significantly different at baseline after age adjustment at a false discovery rate threshold of 0.05, 160 were matched to proteins in the IPA database. An overview of the most significant canonical pathways, upstream regulators, diseases and biological functions predicted in the IPA core analysis are shown in Figure 2. The top enriched canonical signaling pathways were acute phase response signaling, agranulocyte adhesion and diapedesis, FXR/RXR activation, granulocyte adhesion and diapedesis, LXR/RXR activation, atherosclerosis signaling, apelin liver signaling, ferroptosis signaling, inhibition of matrix metalloproteases, iron homeostasis signaling, intrinsic prothrombin activation and osteoarthritis pathways (Supplemental Figure 3, Supplemental Digital Content 11). The proteins that were found in these pathways are shown in Supplemental Table 7, Supplemental Digital Content 7. Based on z-scores, acute-phase response signaling, natural killer cell signaling, osteoarthritis pathways, IL-17 signaling and leukocyte extravasation signaling were enhanced while LXR/RXR activation was attenuated among postmenopausal women. The networks for these pathways are shown in Supplemental Figures 4–7, Supplemental Digital Content 12– 15. Other pathways identified did not show activation or attenuation. The predicted upstream and downstream regulators of the proteins that were significantly different between pre-and postmenopausal women are presented in Supplemental Table 8, Supplemental Digital Content 8. Several transcription regulators (lysine methyltransferase 2D, estrogen related receptor alpha, FOS-related antigen 1), inflammatory cytokines (interleukin 1 alpha, interleukin 22, interleukin 4, oncostatin M) chemical drug (lipopolysaccharide), enzyme (KRAS proto-oncogene GTPase) and growth factors (vascular endothelial growth factor A) were predicted to be among the top upstream regulators of the proteins that differed significantly between pre-and postmenopausal women. For the 38 plasma proteins that were significantly different at baseline in multivariable adjusted models, 34 matched to proteins in the IPA database. The canonical pathways, diseases and disorders found to be associated with these proteins are shown in Table 3; Supplemental Figures 8–9, Supplemental Digital Content 16– 17. These plasma proteins were associated with various molecular/cellular functions (cellular development, growth, proliferation and maintenance), physiological system development (connective tissue development and function, skeletal and muscular system development, and cardiovascular system development and function), and diseases and disorders (hematological and metabolic diseases and developmental disorders). Upstream regulators of these 34 proteins included DS cell adhesion molecule, period circadian regulator 3, peroxisome proliferator-activating receptor gamma coactivator 1 alpha, beta-estradiol and angiotensinogen.

Figure 2.

A graphical summary of the most significant canonical pathways, upstream regulators, diseases and biological functions predicted in the IPA core analysis for the 160 proteins that were significantly different between pre-and postmenopausal women at baseline in age-adjusted analysis.

Table 3.

Summary of the IPA analysis of proteins that were significantly different between pre-and postmenopausal women at baseline.

	P-value	Number of proteins included in network
Top Diseases and Biofunctions
Diseases and Disorders
Developmental Disorder	<0.001	13
Hematological Disease	<0.001	7
Hereditary Disorder	<0.001	10
Metabolic Disease	<0.001	10
Molecular and cellular functions
Cellular development	<0.001	21
Cellular function and maintenance	<0.001	16
Molecular transport	<0.001	13
Small molecule biochemistry	<0.001	15
Cellular growth and proliferation	<0.001	18
Physiological System Development and Function
Connective tissue development & function	<0.001	18
Tissue development	<0.001	22
Skeletal & muscular system development & function	<0.001	18
Cardiovascular System Development and Function	<0.001	13
Organismal Development	<0.001	17
	P-value	% of proteins in our analysis that make up the pathway
Top canonical pathways
Iron homeostasis signaling pathway	<0.001	2.2 %
UDP-D-xylose and UDP-D-glucuronate Biosynthesis	<0.001	50.0 %
LPS/IL-1 Mediated Inhibition of RXR Function	<0.001	1.2 %
Extrinsic Prothrombin Activation Pathway	<0.001	6.2 %
Wnt/-catenin Signaling	<0.001	1.2 %

LPS: lipopolysaccharide; RXR: retinoid X receptors; UDP: Uridine diphosphate; WNT1: Wingless-type.

DISCUSSION

In this large-scale, biracial cohort study characterizing the plasma proteome, 38 plasma proteins differed at baseline between premenopausal and postmenopausal women and 29 exhibited different patterns of change between these two groups over 18 years of follow-up. With the exception of FSH, LH, guanylate-binding protein 1, WNT1-inducible-signaling pathway protein 2, Ferritin, and Ferritin light chain, the remaining 23 of the 29 proteins observed to change significantly over time between premenopausal and postmenopausal women were novel and may reflect biological processes that contribute to elevated risks of chronic diseases after menopause.

Numerous physiologic changes accelerate during the menopausal transition and persist through to the postmenopausal years ⁹. These changes are modulated by genetic and protein-related regulatory processes within the cells ^10,11. ^{31,32,33,31,34,35}The aberrant methylation, phosphorylation, and acetylation of proteins play important roles in the development of chronic diseases ¹⁶. However, no large epidemiologic studies have comprehensively compared the plasma proteome between pre- and postmenopausal women. In the current exploratory study of 4508 women enrolled in the ARIC study which used multiplexed modified DNA-based aptamer technology to measured 4,697 plasma proteins, 1107 proteins had significantly different levels between pre-and postmenopausal women in unadjusted analyses. Adjusting for age reduced the significant proteins to 166. After adjusting further for socio-demographic, behavior/lifestyle and anthropometric factors, kidney function, medical conditions, and hormone therapy use, 38 proteins still showed significantly different levels between pre-and postmenopausal women. Similar to the current study, prior studies that mostly included small samples of women, and quantified proteins on a relatively smaller scale by either micro single radial immunodiffusion ³² or liquid chromatography with tandem mass spectrometry reported cross-sectional differences in protein expressions between pre- and postmenopausal women ^16,36. However, none of these prior studies accounted for factors that may differ between pre- and postmenopausal women and could confound the observed associations. With adjustment for baseline characteristics substantially reducing the number of proteins differing between pre– and postmenopausal women in either cross-sectional or longitudinal analyses in the current study, the critical importance of proper adjustment for potential confounding factors in proteomics analysis of menopausal status cannot be overstated.

A few of the proteins identified in this study have been previously reported in other studies using different proteomic methods. Yi et al identified five of the 38 significant proteins in the current study (probable carboxypeptidase X1, WNT1-inducible-signaling pathway protein 2, secreted frizzled-related protein 1, guanylate-binding protein 1 and ferritin light chain) to be different between pre-and postmenopausal women ¹⁶, while Laakkonen et al identified one (BMP-binding endothelial regulator protein) of the 38 proteins ³⁶. Laakkonen et al also reported some of the canonical pathways reported in the current study ³⁶. These include acute phase response signaling, atherosclerosis signaling, gluconeogenesis I, glycolysis I, LXR/RXR activation and methylglyoxal degradation I. Differences in the overlap of proteins and canonical pathways among studies may be due to the type of tissues from which proteins were quantified and differences in the assays used. Although the modified aptamer technology used in this study has the ability to measure more than 5000 proteins, it has been reported to have preferential measurement for secreted proteins in the plasma ¹⁸. Yet, in comparison with other protein quantification methods, the multiplexed aptamer technology may not accurately capture plasma concentrations for a number of proteins ^23,37. However, aptamer platforms such as the SOMAscan assay overcome the multiplex limitations of immunoassays ^14,38.

Some of the proteins that we identified to be different and/or that changed significantly between pre- and postmenopausal women have important relevance to the pathophysiology of several disease processes. The hypothalamic-pituitary-gonadal axis plays important roles in the regulation of reproductive activity through the actions of important proteins such as FSH and LH ³⁹. Therefore, the higher levels of FSH and luteinizing hormone in postmenopausal compared to premenopausal women observed in the current study were expected as they are direct markers of the menopause transition ³⁹. A few reports, mostly cross-sectional studies suggest that higher levels of FSH are inversely associated with metabolic syndrome ^39,40. However, other studies reported a positive association between increasing trajectories of FSH and subclinical atherosclerosis, but inconsistent relations of FSH with overt CVD ^39,40. hCG is produced in higher quantities during pregnancy⁴¹. Compared with premenopausal women, hCG was previously observed to be higher in peri- and postmenopausal women, with no detectable placental source ^41,42. This observation has been hypothesized to be due to declining ovarian function that results in pituitary hyperstimulation ^41,42. Elevated hCG levels in postmenopausal women have been reported to be associated with gynecological malignancies and nongynecological malignancies including cancers of the bladder, kidney, prostate, breast, and lung ⁴¹.

Other proteins identified to change significantly between pre-and postmenopausal women in the current study are involved in iron metabolism, cell development and other disease processes. Ferritin light chains is important for iron homeostasis and the delivery of iron to cells, and it has been reported to be associated with the survival of glioblastoma multiforme ⁴³. Systemic hepcidin is the central regulator of systemic iron homeostasis, but local hepcidin has been reported to influence the physiology of other organs ⁴⁴. For instance, several studies confirmed the role of local hepcidin in the lungs, stomach, and prostate influencing diseases in these organs ⁴⁴. The loss of cardiac hepcidin has been reported to result in a significant increase in left ventricle mass and apoptosis while cardiac stress has been associated with increased hepcidin expression ⁴⁴. Similarly, heart-type fatty acid binding protein, which exists inside cardiomyocytes, and are responsible for the transportation of fatty acids and lipophilic materials into or out of cells, has been reported to be a biomarker for acute myocardial injury and long-term post-ischemic prognosis ³⁷. Overexpression of this protein also promotes inflammation as well as growth and migration of vascular smooth muscle cells ³⁷. Neogenin is a multifunctional transmembrane receptor that regulates cell adhesion in many diverse developmental processes ⁴⁵. It has been reported to play important roles in the development of cancer, with elevated expression of neogenin detected in breast cancer cells, but this may be a consequence of malignancy and not causal ⁴⁶. Finally, chordin-like protein 2, which was observed to be low in postmenopausal women compared to premenopausal women, is involved in the promotion of osteogenic differentiation of bone marrow mesenchymal stem cells ⁴⁷. All these functions are consistent with our IPA analysis showing developmental disorders, hematological and metabolic diseases being the top disease networks reflected by our protein discovery.

While race has been reported to influence the timing of menopause ¹, studies profiling the plasma proteome of either pre or postmenopausal women have primarily focused on White women. In the current study, 86% of the proteins that changed significantly as women transitioned to menopause had changes that differed by race, with some of these proteomic changes representing difference in some biological processes or disease occurrence between Black and White women. For instance, in the current study, Black women, overall, had a higher positive change in follicle-stimulating hormone levels than White women. This finding is corroborated by evidence from midlife women transitioning to menopause in the Study of Women’s Health Across the Nation that suggests a potential racial differences in the pituitary-gonadal relationship during the menopausal transition ⁴⁸. The higher concentrations of hepcidin and lower changes in ferritin and ferritin light chain among Black women compared to White women reflects the substantial racial differences in iron homeostasis reported in some studies, and this has been postulated to play a major role in the cardiovascular health disparity between Black and White populations ⁴⁹. Similarly, the concentration of heart-type fatty acid binding protein, a potential biomarker for acute myocardial injury and long-term post-ischemic prognosis ⁵⁰, was twice as high among Black women who transitioned to menopause compared to White women. With emerging evidence suggesting that chondroadherin is a regulator of osteoclast motility and counteracting bone loss ⁵¹, the two-fold increase in chondroadherin among Black women observed in the current study may explain to some extent the low prevalence of osteoporosis and low bone mass in Black women compared to White women ⁵².

Furthermore, we found that several other biologic and behavior/lifestyle factors such as smoking and hormone therapy use influenced proteins concentrations, as well as changes in protein levels as women transitioned from pre- to postmenopausal status. Cigarette smoke contains compounds that are known to be associated with both hormone levels in women, and the timing of menopause ⁵³. Findings from the current study where premenopausal women who were ever smokers had lower negative changes in follicle-stimulating hormone and luteinizing hormone, representing higher levels of these hormones in ever smokers than never smokers, are consistent with several report of cigarette smoking being related to higher levels of follicle-stimulating hormone ^53,54. Hormone therapy is known to affect large numbers of proteins and metabolites ⁵⁵. Accordingly, about 93% of the changes in proteins observed in the current study were associated with the use of hormone therapy before menopause. While no prior studies have evaluated the role of hormone therapy in the plasma proteome of women during the menopausal transition, some of the proteins found in the current study such as collagen alpha-2(XI) chain ⁵⁶ and chordin-like protein 2⁵⁷ were also identified in other studies among a small sample of postmenopausal women using either equine estrogens alone ^56–58 or with progesterone ^57–59.

This study has notable strengths that include the use of a large population-based biracial cohort with repeated measurement of plasma proteins about 18 years apart and the extensive assessment of cardiometabolic factors. Potential limitations of this study warrant consideration. First, menopausal status was self-reported and may be subject to recall error. This error should be random with respect to protein levels and thus would typically attenuate measured associations between menopause status and proteins. Second, the average age of women who reported being premenopausal was 52.3 years, which suggest that a considerable proportion of these women were perimenopausal. This misclassification might have led to underestimation of true differences between pre- and postmenopausal women. Third, while the proteomic profiling conducted in this study is the most extensive to date among pre-and postmenopausal women, it does not cover the entire human proteome as only a small proportion of the approximately 30,000 proteins in humans enter the bloodstream by purposeful secretion or leakage from cell damage and cell death ¹⁴. Some studies report that the protein content of the blood is quite reflective of the overall profile of the human organism ¹⁵. Fourth, as mentioned previously, protein concentrations were quantified by RFUs which reflect relative and not absolute protein concentrations. Regardless, protein quantification by RFU has been shown to be more robust indicators of protein expression compared to absolute concentrations in terms of reproducibility, and for statistical differential analysis ²³. Fifth, blood samples were in long-term storage, and it is possible that this could have resulted in degradation of some proteins. Therefore, the longitudinal changes observed in the current study may reflect either true changes over time or may reflect artifacts due to protein degradation and batch effects. The low proportion of protein analytes (7.7%) observed to significantly change in variance over 20 years among ARIC participants does not support widespread protein degradation ²². Finally, protein data used in this study were not validated by methods like liquid chromatography and tandem mass spectroscopy or western blot analyses, and the results of the current study are not externally validated owing to the lack of a large biracial cohort with long follow-up of pre-and postmenopausal women who have extensive characterization of their proteome. However, SOMAscan assay has been applied successfully to biomarker discovery and validation studies, and has been found to have similar ability to correctly identify proteins as other orthogonal proteomic platforms, including liquid chromatography with tandem-mass spectrometry ⁶⁰.

CONCLUSIONS

In summary, this study which to date is most extensive characterization of plasma proteins in the largest sample of pre- and postmenopausal women identified several proteins that differ significantly between pre-and postmenopausal women. Many of these proteins have not been previously recognized to differ in level or expression between pre-and postmenopausal women. These discovered proteins are associated with several important cellular and molecular processes that may play a role in cardiometabolic diseases. Findings from this study warrant external validation. While exploratory, findings from this current study offers valuable data to perform additional confirmatory research of the association of the discovered proteins with menopause-related chronic disease.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)_1

Supplemental Digital Content 1. Table that illustrates the comparison of mean protein levels at baseline and end-of-follow-up, and changes in proteins across these two time points among women who were postmenopausal at baseline and those who were premenopausal at baseline and transitioned to postmenopause by the end of follow-up.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_2

Supplemental Digital Content 2. Table that illustrates the interaction of race and menopausal status on changes in proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_3

Supplemental Digital Content 3. Table that illustrates the interaction of smoke and menopausal status on changes in proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_4

Supplemental Digital Content 4. Table that illustrates the interaction of hormone therapy (HT) and menopausal status on changes in proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_5

Supplemental Digital Content 5. Table that illustrates the characteristics of women who had proteins measured and at baseline and reported being either premenopausal or reached natural menopause in propensity scores matched analysis.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_6

Supplemental Digital Content 6. Table that illustrates the comparison of proteins among women who were premenopausal and postmenopausal women at baseline in propensity scores matched analysis.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_7

Supplemental Digital Content 7. Table that illustrates the top canonical pathways and related differential proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_8

Supplemental Digital Content 8. Table that illustrates the upstream analysis based on 160 proteins that were different between pre- and postmenopausal women at baseline.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_9

Supplemental Digital Content 9. Figure that illustrates the volcano plots of difference in plasma proteins between pre- and postmenopausal women at baseline in unadjusted and age-adjusted models.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_10

Supplemental Digital Content 10. Figure that illustrates the proteins that were significantly different between pre-and postmenopausal women at baseline in multivariable adjusted models for the primary analysis and the propensity score matched analysis.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_11

Supplemental Digital Content 11. Figure that illustrates the canonical signaling pathways of the proteins that were significantly different between pre- and postmenopausal women.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_12

Supplemental Digital Content 12. Figure that illustrates the acute-phase response signaling pathway network with participated proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_13

Supplemental Digital Content 13. Figure that illustrates the osteoarthritis signaling pathway network with participated proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_14

Supplemental Digital Content 14. Figure that illustrates the IL-17 signaling pathway network with participated proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_15

Supplemental Digital Content 15. Figure that illustrates the LXR/RXR activation network with participated proteins.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_16

Supplemental Digital Content 16. Figure that illustrates the canonical signaling pathways of proteins that were significantly different between pre- and postmenopausal women in multivariable analysis.pdf

Supplemental Data File (.doc, .tif, pdf, etc.)_17

Supplemental Digital Content 17. Figure that illustrates the functional analysis of specific differentially expressed proteins associated with cardiovascular cystem development and function.pdf