Brain-derived Neurotrophic Factor and Mood in Perimenopausal Depression

Abstract

Background:

Previous work implicates high pro-inflammatory biomarkers in mood disturbance and low brain-derived neurotrophic factor (BDNF) levels in major depression. However, in hormonally-sensitive premenstrual dysphoric disorder (PMDD), BDNF levels are higher when mood is worse. Perimenopausal depression has not been studied to date. We evaluated whether BDNF and inflammatory cytokines predict mood symptoms across the menstrual cycle in hormonally-sensitive perimenopausal depression symptoms.

Methods:

Data from 49 time points derived from mid-to-late follicular phase [M/L-FP] and perimenstrual assessments of 14 perimenopausal women ages 38–52 with ovulatory menstrual cycles 24–35 days long across 1–2 cycles for mood symptoms, BDNF levels, cytokines, gonadal steroids. Depression was assessed with Montgomery-Åsberg Depression Rating Scale (MADRS), Beck Depression Inventory (BDI); irritability with Kellner Symptom Questionnaire Anger-Hostility subscale (SQ); overall psychological distress with Profile of Mood States (POMS). Mixed models were run on dependent measures of MADRS (primary endpoint) and other mood outcomes (BDI, POMS, SQ) with independent variables of interest (each biomarker, cycle phase), controlling for cycle number and participant.

Results:

After FDR adjustment, BDNF levels showed consistent significant positive relationships to MADRS (β=0.00053; p=0.0028), POMS (β=0.00153; p=0.0394), SQ (β=0.00053; p=0.0067), and BDI (β=0.00039; p=0.0231). Cycle phase did not affect this relationship. No other biomarker consistently predicted affective symptom severity.

Limitations:

Small sample size and large number of comparisons.

Conclusion:

In women with perimenopausal depression symptoms, BDNF is elevated in association with more severe mood symptomatology, resembling the pattern in hormonally-sensitive PMDD and suggesting a hormonally-sensitive mood disorder biomarker profile distinct from that of major depression.

1. Introduction:

Women are at higher risk for major depression than are men. For some women, fluctuating sex steroid hormones during premenstruum, postpartum, and perimenopause trigger mood symptoms, a ‘reproductive subtype’ of depression (Payne et al., 2009). This maladaptive mood response to variation in estradiol and progesterone includes premenstrual dysphoric disorder (PMDD, Wittchen et al., 2003), postpartum depression (Woody et al., 2017), and perimenopausal depression (Gordon et al., 2015). In PMDD, mood is normal during the mid-to-late follicular phase when estradiol is rising toward its mid-cycle peak and progesterone is low, but mood is poor during the late-luteal phase when estradiol and progesterone fall, potentially due to a delayed response to hormonal elevations earlier in the cycle (as demonstrated by Schmidt et al., 1998). Perimenopause is the period preceding menopause when menstrual-cycle length is variable, estradiol fluctuates widely, and ovulation is less frequent, resulting in diminished progesterone production. However, amidst these changes, there are ovulatory menstrual cycles of normal length. Better understanding how fluctuating gonadal steroids and other biological factors contribute to perimenopausal depression risk is critical.

While molecular mechanisms underlying different types of depressive disorders remain under investigation, a leading hypothesis posits a link between depression and chronic immune dysregulation. Elevations of proinflammatory biomarkers like tumor necrosis factor-α (TNF-α) and interleukin-6 (IL-6) are associated with major depression and are reversible with treatment (Haase and Brown, 2015). High levels of proinflammatory cytokines adversely affect neurotrophins, especially brain-derived neurotrophic factor (BDNF), which is involved in synaptic plasticity and maintenance (Haase and Brown, 2015).

In major depressive disorder (MDD) unrelated to reproductive hormones, higher BDNF levels are associated with better mood (Brunoni et al, 2008). In MDD, BDNF levels are lower prior to treatment, higher after treatment, and increasing levels correlate with reduced symptom severity (Brunoni et al, 2008). Conversely, higher pro-inflammatory cytokine levels are associated with worse mood in MDD (Haase and Brown, 2015). In contrast, in premenopausal women without premenstrual dysphoria, data conflict on how BDNF varies across the menstrual cycle. Some studies show luteal-phase BDNF levels overall are higher than follicular-phase levels overall in healthy women, including in mid-to-late luteal phase, in comparison to the mid-follicular phase (e.g., Begliuomini et al., 2007) while others show higher mid-follicular-phase levels (Oral et al., 2015).

Systemic inflammation and BDNF are understudied in women with PMDD. Inflammatory markers C-reactive protein (CRP) and interleukin-2 (IL-2) are positively associated with mood symptoms in the peri-menstrual period (Gold et al., 2016; Bertone-Johnson et al., 2014). In PMDD, BDNF levels are higher in the luteal-phase than in mid-follicular-phase, and, unlike MDD, correlate significantly with worse mood (Oral et al., 2015, 2013).

After menopause, BDNF levels are lower but are restored to follicular-phase levels by hormone therapy, suggesting that estradiol or progesterone regulate BDNF expression (Begliuomini, et al., 2007). However, BDNF has not been well-studied in perimenopausal depression, and only one study shows no association of inflammatory cytokines with perimenopausal depression (Karaoulanis et al., 2012).

In this study we evaluated the relationship of perimenopausal depression symptoms to BDNF and cytokines across the menstrual cycle. We hypothesized that, as a reproductive mood disorder, perimenopausal depression should mirror what is observed in PMDD, with BDNF levels and inflammatory cytokines elevated when mood is poor.

2. Methods

2.1. Study design

Forty-nine specimens from 14 perimenopausal women with mild-to-moderate depressive symptoms were derived from a larger cohort that completed a 9-week observational study (Joffe et al., 2020). Participants completed weekly assessments of depressive symptom severity, irritability, and psychological distress, and had gonadal steroids assayed. Detailed data on menstrual-cycle patterns were collected prospectively. Biomarker assays were batched and performed on previously frozen serum samples.

2.2. Menstrual Cycle Characteristics

A total of 49 specimens drawn from 20 menstrual cycles that met inclusion criteria from 14 unique subjects (n=8 with one cycle; n=6 with 2 cycles) contributed to this analysis. Menstrual cycles meeting inclusion criteria were 24–35 days long and had peak serum luteal-phase progesterone levels ≥3.0 ng/mL, consistent with an ovulatory cycle. Ovulatory cycles were selected to enable analysis of follicular and perimenstrual phases separately. All menstrual cycles analyzed had gonadal steroids assayed at least once during the mid-to-late follicular phase (M/L-FP; cycle days 5–15) and once during the peri-menstrual period (within 7 days prior to or during first 3 days of menstruation). Nine of 20 cycles had a second M/L-FP timepoint available from the same cycle (n=29 M/L-FP data points).

Subjects were perimenopausal 38–52 year-old women reporting mild-to-moderate depression symptoms, defined as a score 10–25 on the clinician-rated Montgomery-Åsberg Depression Rating Scale (MADRS), and/or a score >7.7 on the self-rated Kellner Symptom Questionnaire (SQ) Anger-Hostility subscale (Joffe et al., 2020). This study did not recruit based on PMDD criteria. Exclusion criteria included antidepressant or hormone use, current suicidal ideation, psychiatric hospitalization or suicide attempt within past five years, history of bipolar disorder or psychosis, or recent substance use disorder.

2.3. Procedures and Measures

All visits from the parent protocol meeting inclusion/exclusion criteria for this analysis were included. Across 9 weekly study visits, mood symptoms were assessed and serum estradiol (liquid chromatography, tandem mass spectrometry) and progesterone (chemiluminescence) were measured. Depressive symptoms were measured with both the clinician-administered MADRS (primary endpoint) and the self-reported Beck Depression Inventory (BDI), irritability with the SQ, and psychological distress with the Profile of Mood States (POMS).

A priori biomarkers of the neutrophin BDNF, acute-phase reactant CRP, and cytokines and cytokine receptors IL-6, TNF-α, tumor necrosis factor-receptor 1 (TNF-R1), interferon gamma-induced protein-10 (IP-10), and interleukin-1 receptor antagonist (IL-1RA) were measured using electrochemiluminescence detection technology with the Meso Scale Discovery (MSD) Sector imager S600 platform (MSD, Gaithersburg MD).

All serum samples were assayed neat in duplicate measures with lower limits of detection of 21.8 pg/ml, 2.05 pg/ml, 3.04 pg/ml, 8.6 pg/ml, 0.40 pg/ml, 0.21 pg/ml, and 1.12 pg/ml for BDNF, CRP, TNF-R1, IP-10, TNF-α, IL-6, and IL-1RA, respectively. Quality control pools comprised of split aliquots of the same sample type were tested on each assay plate to assess inter-assay variation, with coefficients of variation of 10–12% for BDNF, CRP, and TNF-R1, and 5–7% for IP-10, TNF-α, IL-6 and IL-1RA.

2.4. Analysis

Mixed random and fixed effects analyses were run separately for the primary dependent variable MADRS and secondary mood disturbance measures (BDI, POMS, SQ). In each case, fixed-effect predictors were the cycle phase (M/L-FP, peri-menstrual), cycle selected (first versus second) within each subject, and one of seven biomarkers, each analyzed separately. In each case, the random effect was the subject intercept. We then ran a backward elimination mixed-effects analysis, beginning with fixed effects of BDNF, cycle selected, cycle phase, and the interaction of BDNF with cycle selected and with cycle phase, and using subject intercepts as the random effect term. For all analyses, examination of model assumptions revealed that distributions of residuals were reasonably normally distributed, given the modest sample size. Corrections for multiple comparisons were performed using the False Discovery Rate (FDR).

3. Results

Subjects were mean age 46.6 years (range 38–52), 64.3% white, had average body mass index 25.3 kg/m², and average cycle length 27.9 days (range 24–35) (Table 1). Mean depression scores did not differ significantly between menstrual-cycle time points (M/L-FP and perimenstrual MADRS 8.8 and 7.6, respectively; Table 1). M/L-FP sampling occurred on cycle day 10.3 (Table 1) and peri-menstrual sampling ranged from 7 days prior to 3 days after onset of menstruation (mean 0.05 days before menses). Cycle phase did not predict mood symptoms. Mean biomarker values averaged across menstrual-cycle time points were: BDNF 9163.6 pg/ml, CRP 1539.2 ng/ml, TNF-R1 3860.0 pg/ml, IP-10 969.2 pg/ml, TNF-α 5.4 pg/ml, IL-6 4.8 pg/ml, and IL-1RA 217.9 pg/ml.

Table 1.

Characteristics by Participant (n=14)

	Mean ± SD, (Range or %)
Demographics
Age (years)	46.6 ± 4.2
Race, N (%)
White	9 (64.3%)
Non-white	5 (35.7%)
Body-mass index (kg/m2)	25.3 ± 5.7
Smoking Status, N (%)
Never smokers	10
Former smokers	2
Unknown	2
Menstrual Cycle Characteristics (n=20)
Number of menstrual cycles contributed per subject, N (%)
1 menstrual cycle	8 (57.1%)
2 menstrual cycles	6 (42.9%)
Cycle length (days)	27.9 ± 2.6
Mid-to-late follicular phase sampling day (of cycle)	10.3 ± 3.1 (5 to 15)
Peri-menstrual sampling days relative to menstruation onset (range)	−0.05 ± 3.2 (−7 to +3)
Hormone levels at time of sampling
Estradiol level (pg/mL) at M/L-FP†	160.2 ± 114.3
Estradiol level (pg/mL) at peri-menstrual phase	99.7 ± 163.0
Progesterone level (ng/mL) at M/L-FP†	0.5 ± 0.9
Progesterone level (ng/mL) at peri-menstrual phase	3.5 ± 6.8
Affective Characteristics
Depression level in M/L-FP†
MADRS	8.8 ± 4.5
BDI	5.9 ± 3.5
Depression level in peri-menstrual phase
MADRS	7.6 ± 4.1
BDI	6.8 ± 4.3
Irritability level (SQ) in mid- to late- follicular phase†	5.4 ± 3.6
Irritability level (SQ) in peri-menstrual phase	5.3 ± 5.2
Psychological distress (POMS) in mid- to late- follicular phasef	24.2 ± 14.6
Psychological distress (POMS) in peri-menstrual phase	23.2 ± 18.9
Biomarker Characteristics
BDNF	9163.6 pg/ml (range 2665.8–16719.8 pg/ml)
CRP	1539.2 ng/ml (range 66.1–12248.6 ng/ml)
TNF-R1	3860.0 pg/ml (range 2161.6–6201.4 pg/ml)
IP-10	969.2 pg/ml (range 225.4–4653.5 pg/ml)
TNF-a	5.4 pg/ml (range 1.5–43.5 pg/ml)
IL-6	4.8 pg/ml (range 0–72.7 pg/ml)
IL-1RA	217.9 pg/ml (range 84.9–706.9 pg/ml)

MADRS = Montgomery-Åberg Depression Rating Scale. BDI = Beck Depression Inventory. SQ = Kellner Symptom Questionnaire Anger-Hostility subscale. POMS = Profile of Mood States. M/L-FP = Mid-late follicular phase.

^†N = 9 subjects contributed >1 M/L-FP measurement.

In the initial mixed-effects analysis, BDNF showed consistently significant relations to mood outcomes, positively predicting POMS (β=0.00143, p=0.0375), SQ (β=0.000513, p=0.0019), and BDI (β=0.000352, p=0.0268). For prediction of MADRS, the mixed-effects model did not converge, requiring that the random effects term be excluded; however in subsequent analyses using the fixed term alone BDNF did significantly predict MADRS. After FDR correction, only the relationship between BDNF and SQ remained significant (p=0.026). TNF-R1 was a significant negative predictor of POMS (β=−0.00746, p=0.0026) but not other mood outcomes.

In follow-up backward elimination mixed effects models involving BDNF, cycle selected, cycle phase, and interactions, BDNF was the only significant predictor of MADRS (β=0.000533, p=0.0004), POMS (β=0.001530, p=0.0225), and SQ (β=0.000528, p=0.0019), accounting for 23%, 2.7% and 20% of their variances, respectively (Figure 1). For MADRS scores, this translates to a 1-point increase in clinician-rated depressive symptoms for every 2000-pg/mL increase in BDNF. BDNF also significantly predicted BDI (β=0.000393, p=0.0099) along with cycle selected (p=0.0003; first>second), which together accounted for 45% of variance in BDI. After FDR correction, p-values for BDNF as a predictor remained significant for MADRS (p=0.0028), SQ (p=0.0067), BDI (p=0.0231), POMS (p=0.0394). Cycle phase (M/L-FP v. perimenstrual) did not affect the relationship between BDNF and mood. The relationship between BDNF and SQ was strengthened in the presence of higher progesterone levels (significant interaction of BDNF X progesterone, p=0.0184), but otherwise gonadal steroid levels did not affect the BDNF/mood relationship. BDNF did not significantly correlate with progesterone, estradiol, or the estradiol/progesterone ratio.

Figure 1.

Effect of serum brain-derived neurotrophic factor (BDNF) on predicted mood symptoms as measured by: A) MADRS, B) POMS, C) SQ, and, for BDI, separately by D) cycle 1, and E) cycle 2 showing a significant positive association of BDNF level with all affective variables after False Discovery Rate (FDR) correction.

Solid lines represent values predicted for depression, irritability, and psychological distress versus BDNF based on mixed-effects longitudinal models. Symbols represent 49 actual data samples for 14 perimenopausal women across the 9-week study; each subject is represented by a unique symbol. Subjects have 1 perimenstrual data point and 1–2 mid-to-late follicular-phase data points in 1–2 menstrual cycles. P-values are FDR-corrected.

MADRS = Montgomery-Åsberg Depression Rating Scale. BDI = Beck Depression Inventory. SQ = Kellner Symptom Questionnaire Anger-Hostility subscale. POMS = Profile of Mood States.

Discussion

This study evaluated the effects of biomarkers (BDNF, inflammatory cytokines) and menstrual-cycle phase on mood symptoms in women with perimenopausal depression symptoms. In line with hormonally-sensitive mood disorder PMDD, higher BDNF levels were consistently associated with depressive symptomatology, irritability, and psychological distress. As perimenopausal depression symptoms do not track with cycle phase as in PMDD, it is unsurprising that cycle phase was not associated with symptom burden. There was a consistent pattern of significant associations with BDNF as a positive predictor for worse mood, a finding present across multiple mood measures that remained significant after multiple comparisons correction, making it unlikely the findings were due to chance. The association of higher neurotrophin levels with worse mood in cycling women parallels that seen in women with PMDD (Oral et al, 2013; Oral et al, 2015) and contrasts with that seen in men and women with MDD (Brunoni et al, 2008).

In women with PMDD, the high luteal-phase BDNF seen correlates with a marker of acute cellular stress, HSP70, suggesting BDNF elevations may be a response to cellular stress and represent a compensatory process (Oral et al., 2013). Since BDNF can be associated with improvement in mood, higher BDNF in the dysphoric luteal phase of PMDD may indicate a corrective process leading to normalization of mood in the euthymic follicular phase. In perimenopausal depressive symptoms, higher BDNF may suggest a recent cellular stressor; due to less-efficient repair systems with aging the corrective effects of BDNF may not yet be reflected in concurrent mood assessments.

Hormonal milieu may also inform the relationship of higher BDNF to worse mood in reproductive mood disorders. The BDNF gene contains a regulatory sequence that estrogen-receptor complexes can bind to, and evidence from animal studies indicates that estrogens alter BDNF gene expression and increase BDNF levels (Sohrabji & Lewis, 2006). Progesterone also regulates BDNF expression, although the direction of that relationship may vary with the presence of estradiol (Pluchino et al., 2013). In perimenopause, estradiol levels are highly variable while progesterone levels decline; increased estradiol/progesterone ratios may enhance BDNF production in response to stressors that also worsen mood. In our study, ovarian hormone levels did not correlate with BDNF levels. The lack of E2 correlation with BDNF was an unexpected finding which may reflect the erratic hormonal fluctuations of perimenopause as well as declines in circulating BDNF associated with aging (Lommatzsch et al., 2005). Our finding that the BDNF/irritability relationship was strengthened in the presence of progesterone is interesting given the observed concentration-dependent relationship between the progesterone metabolite allopregnanolone and PMDD symptoms, including its core symptom irritability (Bixo et al., 2018). The parallels observed between progesterone-dependent susceptibility to irritability in PMDD and in our perimenopausal population underscores the hormonal sensitivity that characterizes both mood disorders and sets them apart from non-hormonally responsive mood disorders.

Our study did not find consistent associations between cytokine levels and mood symptoms, confirming and extending previous findings in perimenopausal depression (Karaoulanis et al., 2012) to include additional biomarkers.

Study strengths include concurrent assessment of mood, neurotrophins, inflammatory biomarkers, and gonadal steroids at distinct timepoints in the menstrual cycle. Limitations include the small sample size and large number of comparisons performed, but consistency of the relationship between BDNF and mood variables mitigates these concerns. Ovulatory cycles were selected and severe depression was excluded; possibly different relationships would be found if anovulatory cycles or severe depression were included.

4. Conclusions

In perimenopausal depression symptoms, BDNF is elevated in association with more severe mood symptomatology, resembling the pattern in PMDD, suggesting that the neurotrophin profile of hormonally-sensitive mood disorders may be distinct from that of MDD.

Highlights.

• BDNF levels are elevated in association with low mood in perimenopausal depression

• Perimenopausal depression BDNF pattern differs from that in major depression (MDD)

• BDNF in perimenopausal depression resembles premenstrual dysphoric disorder pattern

• Inflammation and cycle phase do not predict mood in perimenopausal depression

• Hormonally-linked mood disorders may have different biomarker profiles than MDD