Peripheral cyclical expression of GABA-A receptor subunit genes and menstrual cycle affective change: a dimensional, transdiagnostic study Authors: Jordan C. Barone, Raffaele Romano, Anisha Nagpal, Antonio Salerno, Anna Patterson, Elizabeth S. Wenzel, Gabriela Guzman, Viraja Alluri, Pauline Maki, Graziano Pinna, Tory A. Eisenlohr-Moul Published: 2025 ABSTRACT Menstrual cycle affective change (MCAC), including conditions like Premenstrual Dysphoric Disorder (PMDD), is triggered by ovarian hormone fluctuations, but its neurobiological mechanisms remain poorly understood. A leading hypothesis implicates abnormal GABA-A receptor (GABAAR) subunit regulation in altering sensitivity to neuroactive steroids, a theory difficult to test in humans. We examined peripheral mRNA expression of GABRD, GABRG2, and GABRA5 in a transdiagnostic sample of 112 participants (75 psychiatric outpatients with suicidality; 37 controls) to capture a dimensional range of MCAC severity. Participants provided daily affect ratings and five LH-timed blood samples using PAXgene tubes for RT-PCR analysis. Replicating preclinical findings, GABRG2 expression decreased during the midluteal phase. Critically, individual differences in both person-mean expression and perimenstrual change (AUCi) were correlated with cyclical symptom trajectories. Greater perimenstrual increases or higher average expression were associated with classic luteal-phase worsening of anxiety (GABRG2, GABRA5), irritability (GABRG2), and suicidal ideation (GABRG2, GABRA5, GABRD). However, relative to those with moderate or stable expression, perimenstrual decreases or lower average expression were also linked to cyclical worsening of anxiety and suicidal ideation that emerged around menses and persisted into the follicular phase. Overall, more stable subunit expression was associated with less cyclical symptom change. These findings could be consistent with a loss of homeostatic stability in GABAergic plasticity, rather than a simple unidirectional differences in levels or changes, as a potential correlate of MCAC. Peripheral GABAAR gene expression warrants further investigation, particularly in experimental studies, to determine whether there are causal associations between subunit gene expression and MCAC. Introduction Menstrual Cycle Affective Change (MCAC) refers to cyclical changes in mood and behavior triggered by cyclical fluctuations in estradiol (E2) and progesterone (P4). Sensitivity to these hormonal shifts varies widely, with many in the general, healthy population experiencing no hormone-triggered symptoms or mild symptoms (i.e., Premenstrual Syndrome) [1–4]. More severe, clinically significant points on this spectrum include Premenstrual Dysphoric Disorder (PMDD)—mood symptoms that emerge in the luteal phase and remit shortly after menstruation [5]—and Perimenstrual Exacerbation (PME)—cyclical worsening of psychiatric symptoms without complete follicular remission [3, 6, 7]. Indeed, these extreme manifestations of MCAC are linked to substantial functional impairment [2, 8] and risk for suicidal thoughts and behaviors [9–12]. Human experimental models of the menstrual cycle [13, 14], pregnancy [15], and menopause [16] support a hormone sensitivity framework in which typical ovarian hormone fluctuations elicit symptoms via underlying neurobiological vulnerability. A leading mechanistic hypothesis implicates altered GABAergic signaling in response to fluctuations in progesterone-derived neuroactive steroids (NAS) like allopregnanolone (ALLO), which modulate inhibition via the γ-aminobutyric acid type A receptor (GABAAR) [17]. This pentameric chloride channel is assembled from five subunits encoded by 19 genes (α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3) [18, 19]. Subunit composition governs receptor localization and NAS sensitivity: α/β/γ configurations are typically synaptic, mediate phasic inhibition, and are relatively NAS-insensitive, whereas α/β/δ configurations are extrasynaptic, mediate tonic inhibition, and are highly NAS-sensitive [20–22]. In animals, GABAAR subunit expression shifts in response to hormonal state [23–27]. Accordingly, one hypothesis for MCAC pathophysiology is that altered expression of GABAAR subunits—either as stable traits or cyclical trajectories—modifies neurobiological responses to normal hormonal rhythms, triggering MCAC [17, 26, 28]. The present study begins to test this hypothesis in humans by examining cyclical changes in peripheral gene expression for GABAAR δ, γ2, and α5 subunits. We selected these because they are each hormonally regulated and linked to affect. The δ subunit generally appears in extrasynaptic GABAARs and mediates tonic inhibitory current [29, 30]. In preclinical models, fluctuations (increases or decreases) in P4 and ALLO precipitate δ subunit increases in the hippocampus [31], nucleus accumbens [32], and periaqueductal gray matter [33]. This upregulation reduces excitability, protecting against anxiety and seizures during hormonal fluctuations [34, 35]. Further, plasticity in δ subunit expression [36] during peripartum hormonal fluctuations appears crucial in reducing postpartum depressive-like symptoms [28, 37]. GABAARs containing the γ2 subunit are generally synaptic and mediate phasic inhibitory current [38]. γ2 is also regulated by P4 and ALLO, but generally in the reverse direction of the δ subunit: increases in P4 and ALLO predict γ2 subunit decreases in preclinical models in the estrus cycle or pregnancy [28, 34, 39, 40]. While some studies suggest a positive correlation between P4 and the γ2 subunit, this is either not replicated in females [32], or holds for P4 but not ALLO [41], further supporting that increased ALLO and decreased γ2 subunit concentrations may be specific to NAS fluctuation in females. However, evidence is mixed regarding the γ2 subunit and affective symptoms [31, 32, 42, 43]. Although the α4 subunit is known to be hormonally sensitive and to modulate GABAergic sensitivity to NAS [44, 45], we excluded it given technical difficulties in detecting GABRA4 in peripheral blood. Instead, we included the α5 subunit (GABRA5), which shares several functional properties with α4, including NAS sensitivity. Although less is known about α5 hormonal regulation [40, 43, 46, 47], its expression —at both synaptic and extrasynaptic sites —has been shown to increase with environmental adversity [48–51], aggressive behavior [52], and depressive phenotypes [53]. Together, alterations in any of these subunits may shift inhibitory tone [29, 30, 38], influencing affect and behavior. This study is the first to assess peripheral GABAAR subunit gene expression longitudinally across the menstrual cycle, doing so in a large, diverse sample of healthy individuals and psychiatric outpatients with suicidal ideation (SI). We test the hypothesis that altered overall levels or menstrual cycle trajectories of subunit expression in peripheral blood will correspond to dimensionally greater cyclical affective changes (MCAC). Materials and methods This study, including recruitment of healthy controls (HCs), is a supplemental investigation to the “CLEAR-3” RCT (NCT04112368). Hypotheses, power calculations, and planned analyses—including the preregistered shift from categorical to dimensional modeling of MCAC if categorical reliability checks failed—were preregistered at https://osf.io/m9u5r. Participants Patient recruitment – transdiagnostic suicidal ideation Psychiatric outpatients reporting past-month SI were recruited as described previously [8]. Recruitment targeted SI and natural menstrual cycles, not pre-diagnosed MCAC, to ensure a broad MCAC severity range. Healthy control recruitment HCs were recruited via ads and listservs for “individuals with no history of mental health diagnoses for a clinical research study.” Eligibility matched patient criteria except for exclusion of psychiatric care and lifetime SI. This investigation was approved by the University of Illinois Chicago Institutional Review Board (IRB), and all participants provided informed consent. Study design Enrollment included the Structured Clinical Interview for DSM-5 (SCID-5) [54], and the Self-Injurious Thoughts and Behaviors Interview (SITBI-R) [55, 56]. Starting the next day, participants completed the Daily Record of Severity of Problems (DRSP; [57]), which captures DSM-5 PMDD symptoms, and the Adult Suicidal Ideation Questionnaire (ASIQ; [58]). Medication, substance use, and illness were recorded. After one baseline cycle, participants completed at-home urine ovulation tests during expected ovulation windows (luteinizing hormone (LH), ClearBlue®; threshold 40 mIU/mL). After a positive test (day 0), five morning lab visits were scheduled on days +7, +10, +14, +18, and +22. Patient visits occurred during two randomized conditions: active hormone manipulation or placebo (bisected by a washout cycle; full study protocol can be found on clinicaltrials.gov). The present study analyzes only placebo-condition data alongside HC observational visits. RNA extraction and Real-Time Quantitative Polymerase Chain Reaction (RT-PCR) At each visit, 35 ml of blood was drawn, including 5 ml in PaxGene RNA tubes. Whole blood was prepared according to manufacturer’s instructions (PreAnalytiX, Hombrechtikon, Switzerland). After RNA isolation and extraction, total RNA was reversed transcribed into cDNA, followed by RT-PCR with primer pairs for β-actin and the δ, γ2, and α5 subunits of the GABAAR. Full methods and primer pair sequences (Table S1) can be found in Supplemental Materials. Variables and data preparation Menstrual cycle time Visit scheduling targeted midluteal (LH + 7), perimenstrual (LH + 14), and midfollicular (LH + 22) phases, with LH + 10 and LH + 18 as intermediates. For visit-based analyses, lab visits were dummy-coded (visit 1 at LH + 7/midluteal, visit 3 at LH + 14/perimenstrual, and visit 5 at LH + 22/midfollicular, with the perimenstrual visit as the reference group in phase contrasts). Then, we re-scaled from −1– +1 based on individual LH surge and menses onset (at time = 0) to create a standardized, continuous cycle time variable. Time was modeled with two splines (luteal and follicular), scaled −1– +1. Detailed time scaling methods are described in [59]. Gene expression GABRG2, GABRD, and GABRA5 gene expression (coding for the γ2, δ, and α5 subunits of the GABAAR, respectively) was measured as fold-change or 2-ΔCt (Ct: Cycle Threshold, reference β-actin). Assays were run in triplicate, with outliers (2-ΔCt > 0.1) re-assayed in triplicate. For limit of detection (LOD), a two-step approach ensured reliability: 1) an analytical threshold based on the quantification cycle (Cq ≥32) excluded samples with unreliable amplification, 2) we applied a normalized biological LOD, defined as 2^–ΔCt < 0.00001, corresponding to approximately 100,000-fold lower expression than the reference gene. This threshold was empirically derived from standard curves and replicate analyses, reflecting the lowest cDNA concentration consistently detected across replicates. Our approach aligns with MIQE guidelines, which define LOD as the lowest concentration at which ≥95% of replicates yield detectable amplification. Any sample with a non-detectable value in all assays was recorded as 0.00 and included in analyses. Outliers (> 95th percentile) were Winsorized to the 95th percentile value. We then log-transformed the Winsorized value + 1 to approximate normality. Two trait-level indices of gene expression were calculated per gene as level 2 moderators in multilevel models: 1) person-mean, and 2) perimenstrual change (increase or decrease from the midluteal to midfollicular phase) as area under the curve with respect to increase, AUCi). Person-mean is the average fold-change of that gene across all five blood draws per individual, providing one value per person that represents person-to-person “baseline” differences in gene expression. AUCi was calculated across the perimenstrual weeks per [60] (see Equation 1). Briefly, AUCi summarizes repeated measures change (overall increase or decrease) across a timeline into one comparable index between individuals. For each person, AUCi is the area between the lowest value datapoint and all higher values (y-axis) across each independent time observation (x-axis), approximating a trapezoidal shape. In our design, AUCi is the sum of the areas between visit 1 and 2, visit 2 and 3, and so on, where the distance between visits was scaled based on individual cycle phase length, to standardize hormonal meaning. We interpolated missing values as the mean between neighboring visits. If a participant was missing two consecutive values, we removed that participant. Person-mean and the AUCi were z-scored in models.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{AUC}}_{I}=\left(\mathop{\sum }\limits_{i=1}^{n-1}\frac{\left({m}_{\left(i+1\right)}+{m}_{i}\right)\cdot {t}_{i}}{2}\right)-\left({m}_{1}\cdot \mathop{\sum }\limits_{i=1}^{n-1}{t}_{i}\right)$$\end{document}AUCI=∑i=1n−1mi+1+mi⋅ti2−m1⋅∑i=1n−1ti Equation 1. Area Under the Curve with respect to Increase, per Pruessner et al., [60]. Symptoms We examined daily depressed mood, anxiety, and irritability on the DRSP. For patients, we calculated mean SI from eight ASIQ items adapted for daily use. See Tables S2-3 for item text. Covariates Sensitivity analyses included age (linear, z-scored), recruitment group (0 = HC, 1 = Suicidal Patient), history of DSM-5 diagnosis (0 = no lifetime diagnoses, 1 = at least one lifetime DSM-5 diagnosis), SSRI use (0 = no use, 1 = any use during study), and parity (0 = never given birth, 1 = given birth). Analytic plan Analyses examined: 1) how GABAAR subunit expression varies across the cycle, and 2) whether baseline differences in overall levels (the person-mean moderator) or cycle trajectories of subunit expression (the AUCi moderator) associate with cyclical symptom trajectories. Subunits across the cycle Multilevel models with random intercepts by participant tested the relationship between cycle phase and gene expression by comparing the LH + 7 (midluteal), LH + 14 (perimenstrual), and LH + 22 (midfollicular) lab visits. Next, we characterized subunit trajectories across visits using standardized cycle time variables and spline-based multilevel growth models. Splines were specified as linear or quadratic piecewise functions (see Supplemental Methods, Tables S4-S5, for model fit statistics). We tested model fit improvement with random slopes for the luteal and follicular splines. Sensitivity analyses tested the impact of covariates. Subunits and symptoms To each best-fitting growth model (i.e., predicting daily symptoms from luteal and follicular time variables), we then added person-mean gene expression, perimenstrual AUCi gene expression, and the interactions of both of these gene expression moderators with the time splines. These models examined how subunit expression indices interact with cycle time to predict symptom trajectories. Visualizing significant statistical moderation of mcac by gene expression indices When our models indicated significant moderation (i.e., interaction between individual differences in gene expression indices with the cyclical trajectory of symptoms), we describe the shape of these associations by inspecting means plots subgrouped by levels of the gene expression moderator (lowest, middle, and highest 20% observed in the sample). First, person-centered means (with standard errors (SE)) of the symptom across the cycle were plotted in three subgroups of the sample (lowest, middle, and highest 20% of the moderator). Second, raw means across the cycle are provided for the same subgroups to provide a more direct view of the data and any intercept differences emerging across levels of the moderator. Additional details All primary analyses were preregistered (https://osf.io/m9u5r). As the first human test of these links, our approach was theory-driven. Best practices discourage multiple comparison corrections in this context, as they increase false negatives [61, 62], which are of concern during early translational signal detection. All tests were two-sided, α = 0.05. Analyses were completed in R (4.2.1) with packages ‘menstrualcycleR’ [59], ‘lme4’ [63] ‘stats’ [64], ‘interactions’ [65], ‘mgcv’ [66], and ‘performance’ [67]. Results Participants and data Table 1 presents demographics. Patients (vs. HC) were younger, more diverse in race (p < 0.05) and sexual orientation (p = 0.088) and reported lower family income (p < 0.001), aligning with literature linking minority stress to psychiatric risk and SI [68, 69].Table 1Demographic and baseline clinical information for patient and HC groups.HCPatientpn3775Randomized trial sequence (NCT04112368) EP – PBONA33 PBO – EPNA42Age (mean (SD))29.14 (7.08)26.71 (4.99)0.038BMI (mean (SD))23.63 (3.77)24.33 (4.08)0.393Race (%)0.017 >1 Race2 (5.4)6 (8.1) American Indian/Alaska Native0 (0.0)1 (1.4) Asian13 (35.1)6 (8.1) Black/African American3 (8.1)10 (13.5) Don't Know1 (2.7)5 (6.8) Other4 (10.8)4 (5.4) White14 (37.8)42 (56.8)Hispanic = 1 (%)7 (18.9)21 (28.4)0.395Education (%)0.287 Associate's Degree0 (0.0)2 (2.7) Bachelor's Degree17 (45.9)32 (43.8) Doctoral Degree4 (10.8)3 (4.1) High School Diploma0 (0.0)6 (8.2) Master's Degree10 (27.0)16 (21.9) Some College6 (16.2)14 (19.2)Family Income (%)<0.001 Unknown/Did Not Report3 (8.1)0 (0.00) Under $25,0000 (0.00)28 (20.9) $25,000 - $49,9996 (16.2)37 (27.6) $50,000 - $99,99917 (45.9)45 (33.6) $100,000+11 (29.7)24 (17.9)Sexual Orientation (%)0.088 Asexual0 (0.0)1 (1.4) Bisexual7 (18.9)24 (32.9) Gay0 (0.0)2 (2.7) Heterosexual30 (81.1)36 (49.3) Lesbian0 (0.0)4 (5.5) Other0 (0.0)1 (1.4) Polysexual0 (0.0)1 (1.4) Queer0 (0.0)4 (5.5)Gender Orientation (%)0.224 Endorsed >1 Gender Identity0 (0.0)4 (5.5) Non-Binary0 (0.0)3 (4.1) Other0 (0.0)1 (1.4) Woman (cisgender)37 (100.0)65 (89.0)Parity = 1 (%)8 (24.2)12 (16.7)0.516SSRI = 1 (%)0 (0.0)32 (42.7)<0.001Lifetime DSM-5 Diagnosis = 1 (%)10 (27.0)72 (98.6)<0.001Lifetime Depressive Disorder = 1 (%)7 (18.9)66 (90.4)<0.001Lifetime Anxiety Disorder = 1 (%)3 (8.1)53 (72.6)<0.001Lifetime Substance Use Disorder = 1 (%)4 (10.8)35 (47.9)<0.001Lifetime Eating Disorder = 1 (%)2 (5.4)21 (28.8)0.009Lifetime Trauma-Related Disorder = 1 (%)0 (0.0)41 (56.2)<0.001Current Depressive Disorder = 1 (%)0.0 (0.0)43 (58.9)<0.001Current Anxiety Disorder = 1 (%)0.0 (0.0)47 (64.4)<0.001Current Substance Use Disorder = 1 (%)0.0 (0.0)13 (17.8)0.015Lifetime Eating Disorder = 1 (%)0.0 (0.0)5 (5.8)0.252Current Trauma-Related Disorder = 1 (%)0.0 (0.0)17 (23.2)0.004Group differences were measured with t-tests (numeric variables) or chi-square tests (categorical variables); significance (p < 0.05) is denoted with bold text. Table 1Demographic and baseline clinical information for patient and HC groups.HCPatientpn3775Randomized trial sequence (NCT04112368) EP – PBONA33 PBO – EPNA42Age (mean (SD))29.14 (7.08)26.71 (4.99)0.038BMI (mean (SD))23.63 (3.77)24.33 (4.08)0.393Race (%)0.017 >1 Race2 (5.4)6 (8.1) American Indian/Alaska Native0 (0.0)1 (1.4) Asian13 (35.1)6 (8.1) Black/African American3 (8.1)10 (13.5) Don't Know1 (2.7)5 (6.8) Other4 (10.8)4 (5.4) White14 (37.8)42 (56.8)Hispanic = 1 (%)7 (18.9)21 (28.4)0.395Education (%)0.287 Associate's Degree0 (0.0)2 (2.7) Bachelor's Degree17 (45.9)32 (43.8) Doctoral Degree4 (10.8)3 (4.1) High School Diploma0 (0.0)6 (8.2) Master's Degree10 (27.0)16 (21.9) Some College6 (16.2)14 (19.2)Family Income (%)<0.001 Unknown/Did Not Report3 (8.1)0 (0.00) Under $25,0000 (0.00)28 (20.9) $25,000 - $49,9996 (16.2)37 (27.6) $50,000 - $99,99917 (45.9)45 (33.6) $100,000+11 (29.7)24 (17.9)Sexual Orientation (%)0.088 Asexual0 (0.0)1 (1.4) Bisexual7 (18.9)24 (32.9) Gay0 (0.0)2 (2.7) Heterosexual30 (81.1)36 (49.3) Lesbian0 (0.0)4 (5.5) Other0 (0.0)1 (1.4) Polysexual0 (0.0)1 (1.4) Queer0 (0.0)4 (5.5)Gender Orientation (%)0.224 Endorsed >1 Gender Identity0 (0.0)4 (5.5) Non-Binary0 (0.0)3 (4.1) Other0 (0.0)1 (1.4) Woman (cisgender)37 (100.0)65 (89.0)Parity = 1 (%)8 (24.2)12 (16.7)0.516SSRI = 1 (%)0 (0.0)32 (42.7)<0.001Lifetime DSM-5 Diagnosis = 1 (%)10 (27.0)72 (98.6)<0.001Lifetime Depressive Disorder = 1 (%)7 (18.9)66 (90.4)<0.001Lifetime Anxiety Disorder = 1 (%)3 (8.1)53 (72.6)<0.001Lifetime Substance Use Disorder = 1 (%)4 (10.8)35 (47.9)<0.001Lifetime Eating Disorder = 1 (%)2 (5.4)21 (28.8)0.009Lifetime Trauma-Related Disorder = 1 (%)0 (0.0)41 (56.2)<0.001Current Depressive Disorder = 1 (%)0.0 (0.0)43 (58.9)<0.001Current Anxiety Disorder = 1 (%)0.0 (0.0)47 (64.4)<0.001Current Substance Use Disorder = 1 (%)0.0 (0.0)13 (17.8)0.015Lifetime Eating Disorder = 1 (%)0.0 (0.0)5 (5.8)0.252Current Trauma-Related Disorder = 1 (%)0.0 (0.0)17 (23.2)0.004Group differences were measured with t-tests (numeric variables) or chi-square tests (categorical variables); significance (p < 0.05) is denoted with bold text. Demographic and baseline clinical information for patient and HC groups. Group differences were measured with t-tests (numeric variables) or chi-square tests (categorical variables); significance (p < 0.05) is denoted with bold text. Of N = 83 with at least one placebo condition visit, we removed N = 8 for abnormal cycle phase lengths and N = 10 visits (across participants) for illness, resulting in N = 328 visits (N = 75 patients). Of N = 40 HC participants with at least one blood draw, we removed N = 1 for C-PASS-defined menstrually-related mood disorder, N = 1 for abnormal cycle phase length, and N = 1 for illness; we removed N = 1 additional visit for illness, resulting in N = 179 samples in N = 37 HCs. Altogether, analyses predicting gene expression from cycle time variables included N = 507 visits in N = 112 participants. AUCi interpolation: ~25% of participants required one value interpolated, and 1–3% required interpolation of two. The maximum N was N = 2320 observations in N = 95 participants. Simulations [70] indicate that our design (~25 observations/participant) yields ~80% power to detect cross-level interactions of f ≈ 0.33. There were no statistically significant differences between HCs and patients on AUCi or person-mean subunit values. Thus, participants were pooled into one sample. Because models included random intercepts for subject it was unnecessary to covary for group. MCAC symptoms MCAC is often conceptualized categorically, assuming HCs show little cyclical symptom change, while some psychiatric patients experience clear PMDD/PME. However, extant work supports a continuous view: cyclical affective changes range from absent to severe, with no obvious threshold for demarcating binary diagnosis [2]. Accordingly, we hypothesized (https://osf.io/m9u5r) that MCAC would be minimal in most HCs and variable among patients. As expected, MCAC did not sort into reliable subgroups (for distributions of percent-cyclical-change across cycles, see Fig. 1). We assessed reliability of categorical MCAC using the commonly-used 30% pre/perimenstrual symptom elevation threshold for daily ratings [5]. 49 (65%) of patients with SI showed 30% elevation of at least one symptom in any cycle, but only 16 (21%) in both cycles. Surprisingly, 18 (48%) of the HC group also met this 30% threshold for at least one symptom in one cycle, despite being recruited for (and, via the SCID-5, confirmed to have) no distressing/impairing pattern of psychiatric symptoms; however, only 3 HCs (8%) showed this pattern repeatedly. Further details are preprinted (https://osf.io/preprints/osf/tpsbf_v1).Fig. 1Distribution of Menstrual Cycle-Associated Cyclicity (MCAC) Across Groups and Cycles.This figure visually compares the magnitude and variability of symptom cyclicity between Patients with SI and Healthy Controls (HC) across baseline and observational cycles (N = 105). Each dot represents the calculated percentage of cyclical change for an individual participant, and the spread of these dots illustrates the overall distribution of cyclicity within each group. To quantify MCAC, the lowest mean symptom rating from a low-risk phase was subtracted from the highest mean rating in a high-risk phase. This difference was then divided by the participant’s individual range of scale use and multiplied by 100 to yield a percent change. High-risk phases include the Midluteal (days −9 to −5), Premenstrual (days −7 to −1), and Perimenstrual (days −3 to +2). Low-risk phases include the Post menstrual (days +4 to +10) and Midfollicular (days +6 to +10), with day 1 defined as the onset of menses. Distribution of Menstrual Cycle-Associated Cyclicity (MCAC) Across Groups and Cycles. This figure visually compares the magnitude and variability of symptom cyclicity between Patients with SI and Healthy Controls (HC) across baseline and observational cycles (N = 105). Each dot represents the calculated percentage of cyclical change for an individual participant, and the spread of these dots illustrates the overall distribution of cyclicity within each group. To quantify MCAC, the lowest mean symptom rating from a low-risk phase was subtracted from the highest mean rating in a high-risk phase. This difference was then divided by the participant’s individual range of scale use and multiplied by 100 to yield a percent change. High-risk phases include the Midluteal (days −9 to −5), Premenstrual (days −7 to −1), and Perimenstrual (days −3 to +2). Low-risk phases include the Post menstrual (days +4 to +10) and Midfollicular (days +6 to +10), with day 1 defined as the onset of menses. These findings demonstrate the unacceptable reliability of categorical MCAC cutoffs in this sample and highlight the continuous nature of MCAC. Accordingly, we modeled cycle-affect associations as linear outcomes across participants, consistent with prior work showing that categorical MCAC variables, like many psychiatric diagnoses, have poor reliability and specificity. It also aligns with our work showing dimensional variability in MCAC degree and timing [8, 12], even in prospectively-confirmed PMDD samples [2]. As preregistered, we therefore modeled MCAC as a continuous variable, utilizing our sample’s wide range of MCAC severity. Cross-level interactions tested whether person-level factors (e.g., person-mean subunit expression or AUCi) were associated with within-person symptom trajectories across cycle time [71] regardless of recruitment group. Subunit data: basics Detailed descriptive statistics for raw and transformed subunit expression can be found in Table S6. ICCs were 0.18 for GABRD, 0.30 for GABRG2, and 0.15 for GABRA5. Low ICCs indicate that most variance occurred at the visit level, not between person. Between-person correlations were significant between person-mean GABRD and GABRG2 (r = 0.26, p < 0.01) and GABRD and GABRA5 (r = 0.92, p < 0.001); there was a nonsignificant association between GABRG2 and GABRA5 (r = 0.17, p = 0.07). Repeated measures correlations were significant between GABRD and GABRG2 (rmcorr = 0.34, p < 0.001), GABRD and GABRA5 (rmcorr = 0.88, p < 0.001), and GABRG2 and GABRA5 (rmcorr = 0.28, p < 0.001). Full correlation matrices are in Table S7. In sum, correlations among subunit expression were generally significant—but particularly strong between GABRD and GABRA5. GABAAR subunit gene expression across the cycle Cycle time as categorical visits (Table S8, panel A). GABRG2 was lower at the midluteal visit compared to the perimenstrual visit (Est = −0.009, p = 0.05, Fig. 2). There were no significant phase differences in GABRD or GABRA5 expression.Fig. 2Person-centered Gene Expression by Cycle-Timed Visit.Data are person-centered for each gene. Person-centering involves subtracting the person-mean across all visits from the raw value at that visit; therefore, zero indicates the participant’s average level across visits. Positive values indicate higher-than-usual values, and negative values indicate lower-than-usual values. X-axis represents menstrual cycle time as a categorical predictor: MIDLUT = midluteal (LH + 7), PERIMENS = perimenstrual (LH + 14), MIDFOL = midfollicular (LH + 22). Person-centered Gene Expression by Cycle-Timed Visit. Data are person-centered for each gene. Person-centering involves subtracting the person-mean across all visits from the raw value at that visit; therefore, zero indicates the participant’s average level across visits. Positive values indicate higher-than-usual values, and negative values indicate lower-than-usual values. X-axis represents menstrual cycle time as a categorical predictor: MIDLUT = midluteal (LH + 7), PERIMENS = perimenstrual (LH + 14), MIDFOL = midfollicular (LH + 22). Cycle time re-scaled and standardized to a continuous variable (Table S8, Panel B). For GABRD and GABRA5, random effects for both luteal and follicular time splines significantly improved model fit, indicating significant individual differences in subunit expression trajectories across phases. There were no significant fixed relationships of luteal or follicular phase time, or patient versus HC groups, on expression. See Figures S2-S4 for individual differences in gene expression across the cycle. Covariates. In sensitivity analyses (where covariates were added as independent predictors to each model predicting a gene from cycle time), there were no significant effects of age, BMI, parity, history of DSM-5 diagnosis, or SSRI use on expression (all p’s > 0.05). GABAAR subunit gene expression as a biological correlate of MCAC ICCs for symptoms ranged from 0.47–0.73, indicating that 47–73% of variance reflected between-person differences, while 27–53% reflected within-person fluctuations (i.e., that could potentially be explained by the menstrual cycle). Recruitment groups (Patients with SI vs. HC), used to broaden variability in MCAC, did not significantly differ on mean or AUCi gene expression indices, although nonsignificant trends suggest somewhat greater detectability, person-mean levels, and perimenstrual change (absolute value of AUCi) in patients with SI that may be observable in larger samples (Figure S1; Table S9). Across all moderation models, trajectories were best fit with a quadratic luteal phase and rectilinear follicular phase patterns (Tables S10-S12). GABRG2 expression as a moderator of cyclical symptom trajectories A participant’s average GABRG2 expression (person-mean) was not associated with any symptom trajectories. However, degree of perimenstrual change in GABRG2 expression (AUCi) was associated with anxiety and irritability trajectory (but not depression or SI, ps > 0.05; all results in Table S10). With respect to anxiety, a significant interaction emerged between GABRG2-AUCi and follicular cycle time (Linear Interaction Est = −0.10, 95% CI = −0.19–0.01, p = 0.034); this marker correlated with stronger premenstrual increases in anxiety. Figure 3A, B illustrates anxiety means across the cycle. Greater perimenstrual increases in GABRG2 expression (upper 20%) were associated with a luteal rise in anxiety that peaked premenstrually and dropped off rapidly during menses, with further reductions leading up to ovulation (Fig. 3A). For those in the lowest 20% of GABRG2-AUCi, representing perimenstrual reductions in GABRG2 expression, there was a moderate, later perimenstrual increase in anxiety that dropped rapidly at menses but spiked again leading to ovulation. Cyclical change was less robust for those in the middle 20%. Thus, both greater increases and decreases in GABRG2 were associated with cyclical symptom change, with a robust premenstrual pattern in those with the highest GABRG2 increases and a more complex dual-peak pattern in those with perimenstrual decreases in GABRG2. Of note, raw means (Fig. 3B) indicated that participants with the largest perimenstrual changes (increases or decreases) appeared to have higher anxiety symptoms overall.Fig. 3Individual Differences in GABRG2 Person-Mean Expression and Perimenstrual Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Anxiety, Irritability, and Suicidal Ideation.Daily symptoms were rated on a 1–6 scale (1 = Not at All, 6 = Extremely). The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on the area under the curve with respect to increase (AUCi), an index of the magnitude and direction of perimenstrual change in expression (upper 20% = clear midluteal-to-midfollicular increases; lower 20% = decreases). Person-mean-centered outcomes (A, C, and E) are calculated by subtracting each participant’s mean across visits from observed values, such that 0 = average/“usual,” positive = higher-than-usual, and negative = lower-than-usual. Raw values (B, D, and F) combine between- and within-person variance. SI was measured and modeled only in patients recruited for past-month SI; SI was exclusionary for HC participants. Individual Differences in GABRG2 Person-Mean Expression and Perimenstrual Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Anxiety, Irritability, and Suicidal Ideation. Daily symptoms were rated on a 1–6 scale (1 = Not at All, 6 = Extremely). The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on the area under the curve with respect to increase (AUCi), an index of the magnitude and direction of perimenstrual change in expression (upper 20% = clear midluteal-to-midfollicular increases; lower 20% = decreases). Person-mean-centered outcomes (A, C, and E) are calculated by subtracting each participant’s mean across visits from observed values, such that 0 = average/“usual,” positive = higher-than-usual, and negative = lower-than-usual. Raw values (B, D, and F) combine between- and within-person variance. SI was measured and modeled only in patients recruited for past-month SI; SI was exclusionary for HC participants. With respect to irritability, there was a significant interaction between GABRG2-AUCi and cycle time (Linear Interaction Est = 0.11, 95% CI = 0.03–0.18, p = 0.008). Figure 3C, D visualizes irritability means across the cycle at highest, middle and lowest 20% of GABRG2-AUCi. Again, greater perimenstrual increases in GABRG2 expression were associated with the greatest cyclical irritability, showing robust luteal and premenstrual surges in irritability that rapidly resolved during menses (Fig. 3C). Mean cyclical changes in irritability were much milder and less robust among those in the middle and lowest 20% of GABRG2-AUCi. Raw means (Fig. 3D) indicate that those with highest perimenstrual increases in expression showed greater overall irritability. With respect to SI, there was a significant interaction between person-mean levels of GABRG2 expression and cycle time (Linear Interaction Est = −0.10, 95% CI = −0.19–−0.01, p = 0.044). Figure 3E, F illustrates of SI means across the cycle at the highest, middle and lowest 20% of average GABRG2 expression. Those with the highest overall expression of GABRG2 demonstrated the most robust perimenstrual increases in SI, with symptoms emerging rapidly midluteally, peaking at menses onset, and falling to an elevated baseline until after next ovulation (Fig. 3E). Those with the lowest GABRG2 expression also demonstrated cyclical change in SI, with symptoms below baseline in the luteal phase that rose rapidly in the perimenstrual and follicular phases to a peak around ovulation. Those with moderate levels of GABRG2 expression showed less cyclical change. Inspection of raw means (Fig. 3F) suggested that those with highest GABRG2 expression experienced elevated overall SI, driven by perimenstrual worsening. GABRA5 expression as a moderator of cyclical symptom trajectories While interactions predicting cyclical depression and irritability from GABRA5 expression were not significant, GABRA5 expression moderated trajectories of anxiety and SI (Table S11). With respect to anxiety, person-mean GABRA5 significantly interacted with cycle time, with this marker correlating with a stronger quadratic trajectory of anxiety across the luteal phase (Quadratic Interaction Est = −0.28, 95% CI = −0.53–−0.03, p = 0.028). Figure 4 provides visualizations of symptom means across the cycle at the highest, middle and lowest 20% of person-mean GABRA5. Person-mean-centered values revealed a complex picture (Fig. 4A). Those with high person-mean GABRA5 expression demonstrated an average luteal phase increase in anxiety that mirrored the typical shape of P4. Among those with moderate GABRA5 expression, the anxiety trajectory was reversed, remaining low and stable across the luteal phase, then rising at menses and continuing to rise across the follicular phase, peaking at ovulation. In those with the lowest average GABRA5 expression, anxiety transiently peaked postovulation followed by a perimenstrual rise that persisted until next ovulation. Regardless of group, cyclical change appeared to cover an average of ~ 0.3–0.4 scale points (out of a range of 5) across the cycle. Raw mean trajectories (Fig. 4B) indicate those with both high and low GABRA5 expression demonstrated greater anxiety overall than those with moderate expression.Fig. 4Individual Differences in GABRA5 Person-Mean Expression and Perimenstrual Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Anxiety and Suicidal Ideation.Daily symptoms were rated on a 1–6 scale. The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on either person-mean GABRA5 expression (A, B) or perimenstrual change in GABRA5 expression indexed by AUCi (C, D; upper 20% = increases, lower 20% = decreases). A shows person-mean-centered anxiety, (B) raw mean anxiety, (C) person-mean-centered suicidal ideation, and (D) raw mean suicidal ideation. Person-mean-centered outcomes are calculated as in Fig. 2, and raw values reflect both between- and within-person variance. Suicidal ideation was measured and modeled only in patients recruited for past-month SI; SI was exclusionary for HC participants. Individual Differences in GABRA5 Person-Mean Expression and Perimenstrual Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Anxiety and Suicidal Ideation. Daily symptoms were rated on a 1–6 scale. The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on either person-mean GABRA5 expression (A, B) or perimenstrual change in GABRA5 expression indexed by AUCi (C, D; upper 20% = increases, lower 20% = decreases). A shows person-mean-centered anxiety, (B) raw mean anxiety, (C) person-mean-centered suicidal ideation, and (D) raw mean suicidal ideation. Person-mean-centered outcomes are calculated as in Fig. 2, and raw values reflect both between- and within-person variance. Suicidal ideation was measured and modeled only in patients recruited for past-month SI; SI was exclusionary for HC participants. Finally, a participant’s degree of perimenstrual change in GABRA5 expression (AUCi) significantly moderated SI trajectories (Quadratic Interaction Est = −0.27, 95% CI = −0.49–−0.06, p = 0.013). Inspection of the person-centered plot (Fig. 4C ) revealed a prominent luteal phase pattern among those with greater GABRA5 increases, with SI rising post-ovulation and falling rapidly with menses onset. However, participants with stable expression (middle 20%) or large perimenstrual decreases in GABRA5 expression (lowest 20%) showed an opposite pattern, with perimenstrual increases in SI that persisted through the follicular phase. Raw mean trajectories (Fig. 4D) showed similar mean SI levels across groups.Fig. 5Individual Differences in Perimenstrual GABRD Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Suicidal Ideation (Patients Only).Suicidal ideation was measured daily in on a 1–6 scale by in patients recruited for past-month SI; SI was exclusionary for HC participants. The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on the area under the curve with respect to increase (AUCi) for GABRD expression, indexing the magnitude and direction of perimenstrual change (upper 20% = increases, lower 20% = decreases). A shows person-mean-centered anxiety, and (B) shows raw mean suicidal ideation. Person-mean-centered outcomes are calculated as in Fig. 2, and raw values combine between- and within-person variance. Individual Differences in Perimenstrual GABRD Change (AUCi) Statistically Moderate the Menstrual Cycle Trajectory of Suicidal Ideation (Patients Only). Suicidal ideation was measured daily in on a 1–6 scale by in patients recruited for past-month SI; SI was exclusionary for HC participants. The first LH surge was confirmed with urine testing; the second was estimated. L = luteal phase; F = follicular phase. Shaded areas represent SE. To probe the interactions, participants were stratified into upper, middle, and lower 20% groups based on the area under the curve with respect to increase (AUCi) for GABRD expression, indexing the magnitude and direction of perimenstrual change (upper 20% = increases, lower 20% = decreases). A shows person-mean-centered anxiety, and (B) shows raw mean suicidal ideation. Person-mean-centered outcomes are calculated as in Fig. 2, and raw values combine between- and within-person variance. GABRD expression as a moderator of cyclical symptom trajectory All but one GABRD × cycle-time interactions predicting depression, anxiety, irritability, or SI failed to reach significance (all ps > 0.05; Table S12). There was a nominally significant interaction between an individual’s degree of perimenstrual increase in GABRD expression (GABRD-AUCi) and cycle time, which correlated with a stronger quadratic shape of SI across the luteal phase (Quadratic Interaction Est = −0.28, 95% CI = −0.54–0.01, p = 0.057). Inspection of the person-centered means (Fig. 5A) suggests nearly identical results to those reported above for GABRA5-AUCi. Examination of raw means (Fig. 5B) suggests that individuals with less perimenstrual change in GABRD expression had lower overall SI than patients with greatest perimenstrual increases (highest 20% GABRD-AUCi) or decreases (lowest 20% GABRD-AUCi). Discussion This preregistered study offers the first direct human test of the longstanding hypothesis that MCAC is linked to altered GABAAR subunit gene expression in response to NAS fluctuations. While decades of preclinical studies have implicated altered GABAAR regulation by hormones as a core mechanism for hormone-sensitive mood symptoms [17, 72–76], translation to humans remains limited. Here, we show for the first time that peripheral expression of GABRG2 fluctuates across the cycle and that both mean levels and cyclical dynamics of GABAAR subunit expression are associated with cycle-related changes in psychiatric symptoms. Despite the known efficacy of SSRIs for PMDD [77], we found that SSRI use did not correspond to GABAAR subunit expression at baseline or across the cycle. Subunit genes and the menstrual cycle We found that GABAAR subunit expression varied more within individuals across the cycle than between them, with heterogeneity in expression trajectories, suggesting dynamic hormonal regulation. While causality and concordance with central expression remain to be demonstrated, these preliminary findings are consistent with animal studies showing that GABAAR composition shifts with steroid fluctuations, producing downstream changes in affect and behavior [17, 73–75]. GABRA5 and GABRD expression were strongly correlated both within and between individuals. Postmortem studies in suicide victims have shown that these genes tend to be co-upregulated or co-downregulated [78], supporting the potential for coordinated regulation of these subunits across brain and peripheral systems and highlighting the value of examining GABAAR subunit co-expression more broadly. Results also suggested reduced GABRG2 expression during the high-P4 midluteal phase, consistent with many preclinical studies linking P4 and its metabolite ALLO to γ2 downregulation. This may represent an adaptive mechanism reducing receptor sensitivity to ALLO’s sedating effects ([28, 34, 35, 39, 40]. A prior human study similarly found lower GABRG2 expression in peripheral monocytes during late pregnancy—another high-P4 state—compared to non-pregnant women [72]. Peripheral GABAAR subunit gene expression may be a biological correlate of MCAC Peripheral GABAAR subunit expression was associated with cycle-related fluctuations in anxiety, irritability, and SI, though not depression. Pending replication, this specificity of findings to higher-arousal symptoms is theoretically consistent with the recent Dimensional Affective Sensitivity to Hormones across the Menstrual Cycle (DASH-MC) framework [1], which posits that luteal phase NAS change sensitivity may trigger high-arousal symptoms confined to the luteal phase [79], whereas lower-arousal symptoms like depression may be more closely linked to later-onset perimenstrual sensitivity to E2 withdrawal [80, 81]. DASH-MC, informed by both experimental studies [14] and trajectory subgrouping studies in patients with prospectively-confirmed PMDD [82], may be a useful lens for understanding the timing and content of symptom correlations in the current study. Consistent with their co-fluctuation, GABRD and GABRA5 showed similar associations with SI, mirrored in anxiety for GABRA5. Across models, moderate levels and changes in subunit expression appeared protective, whereas individuals with more extreme levels or changes demonstrated greater MCAC. Higher overall expression in GABRA5 and greater perimenstrual increases in GABRD expression were associated with luteal elevations in anxiety and SI that resolved rapidly with menses onset. Conversely, lower levels and larger perimenstrual reductions in GABRD and GABRA5 expression were associated with increased cyclicity of anxiety and SI. Lower levels or greater decreases in expression of these subunits were linked with a late luteal anxiety onset or SI that often persisted through the follicular phase, failing to resolve until the next luteal phase, aligning with prior work linking GABRD to anxiety in animals [34, 35], and to suicidality in humans [78, 83], and suggesting that expression differences may causally shape perimenstrual affective trajectories. Similarly, these findings converge with evidence that altered α5 subunit expression, particularly in hippocampal circuits, regulates cognition and emotion [84, 85], associating in complex ways with adverse affective and behavioral outcomes [48–50, 53]—including behavioral anxiety across the estrous cycle [86]. Future studies should replicate these results and clarify whether there may be causal relationships between 1) increased GABRA5/GABRD expression and luteally-confined affective symptoms or 2) decreased GABRA5/GABRD expression and perimenstrual- or follicular affective symptoms. GABRG2 showed the strongest and most consistent associations with symptoms across the cycle, correlating with anxiety, irritability, and SI trajectories. Similar to the results for GABRD and GABRA5, more extreme levels or changes in GABRG2 expression were associated with greater cyclicity of anxiety and SI; cyclical irritability change appeared to be solely linked to perimenstrual increases. Given irritability’s prominence in PMDD [87] and the limitations of animal models for midluteal onset irritability [88], further validation of peripheral correlates like GABRG2 is important. Participants with marked GABRG2 decreases also demonstrated cyclical increases in anxiety and SI (though not irritability), with symptoms again emerging later in the perimenstrual phase and persisting into the follicular phase. These findings echo prior studies linking lower γ2 expression to early life adversity and chronic stress [50, 52, 89]. Of note, the dynamic pattern of GABRG2 regulation in our sample (reduced in the midluteal phase compared to perimenstrual and midfollicular phases) mirrored the timing of suicidal risk observed in hormone-sensitive samples [12]. Given the lack of validated time-varying biomarkers for SI, future work is needed to replicate and extend these findings in causal paradigms. Cyclical depression change was not significantly associated with either mean expression or perimenstrual change in GABRD, GABRG2, or GABRA5, possibly reflecting the lower arousal profile of depression compared to anxiety and irritability—symptoms that have shown stronger links to ovarian hormone fluctuations in both PMDD [90, 91] and peripartum mood disorders [91]. Our prior work in an experimental perinatal model similarly found that high-arousal symptoms, particularly anger and anxiety, responded rapidly to steroid surges, while depressive symptoms emerged weeks later [91]. These findings suggest that hormone sensitivity may initially manifest through high-arousal dimensions like irritability and anxiety, with depression reflecting downstream or cumulative effects. Greater focus on irritability in PMDD and related models is warranted, given its luteal specifity and functional impact [87, 92]. Limitations and future directions This translational study has several limitations. It remains uncertain whether peripheral GABAAR subunit expression reflects central receptor dynamics, though peripheral expression has been linked to symptoms in migraine [93], Tourette Syndrome [94], and depression [95]. Our spline-based model with a knot at menses captured cyclical inflections, but generalized additive modeling could offer complementary insights. While our preregistered approach focused on three subunits and four symptoms to reduce false positives, exploratory analyses may reveal additional candidate genes or symptoms. We were unable to reliably measure α4, a key subunit in NAS sensitivity [44, 45]. Additionally, observational HCs were compared to placebo-phase data from patients. While no placebo effects on GABAAR expression are expected, this design choice limits interpretability. Finally, unmeasured factors such as stress or trauma—which can alter expression of γ2 and δ subunits [96]— should be addressed in future studies. Conclusions This study is the first to test whether peripheral expression of GABAAR subunit genes varies across the menstrual cycle and relates to hormone-sensitive mood symptoms. In a naturally-cycling transdiagnostic sample of female psychiatric outpatients with suicidality alongside HCs, we found that GABRG2 expression fluctuated across the cycle and that interindividual differences in cyclicity of GABRD, GABRA5, and GABRG2 were linked to menstrual cycle changes in anxiety, irritability, and SI. Broadly, larger absolute perimenstrual levels and changes correlated with greater MCAC, highlighting the possibility that lower reactivity of subunit expression to hormone fluctuations could be adaptive. While these findings require replication and cannot be assumed to reflect central processes, they offer preliminary support that peripheral GABAAR gene expression could represent a potential mechanistic correlate of MCAC. Supplementary information Supplemental Materials Supplemental Materials