Characterizing within-person variance in, and menstrual cycle associations with, event-related potentials associated with positive and negative valence systems: The Reward Positivity and the Error-Related Negativity

Abstract

Event-related potentials (ERPs) are widely employed as measures of transdiagnostic cognitive processes that are thought to underlie various clinical disorders (Hajcak et al., 2019). Despite their prevalent use as individual difference measures, the effects of within-person processes, such as the human menstrual cycle, on a broad range of ERPs are poorly understood. The present study leveraged a within-subject design to characterize between- and within-person variance in ERPs as well as effects of the menstrual cycle in two frequently studied ERPs associated with positive and negative valence systems underlying psychopathology—the Reward Positivity (RewP) and the Error- Related Negativity (ERN). Seventy-one naturally-cycling participants completed repeated EEG and ecological momentary assessments of positive and negative affect in the menstrual cycle’s early follicular, periovulatory, and mid-luteal phases. We examined the mean degree of change between cycle phases in both ERPs, the between-person variability in the degree of change in both ERPs, and whether an individual’s degree of cyclical change in these ERPs show coherence with their degree of cyclical change in positive and negative affect recorded across the cycle. Results revealed no significant changes in positive and negative affect across the cycle and rather small changes in ERP amplitudes. Significant random slopes in our model revealed larger individual differences in trajectories of change in ERP amplitudes and affect, in agreement with prior evidence of heterogeneity in dimensional hormone sensitivity. Additionally, state-variance in these ERPs correlated with positive and negative affect changes across the cycle, suggesting that cycle-mediated ERP changes may have relevance for affect and behavior. Finally, exploratory latent class growth mixture modeling revealed subgroups of individuals that display disparate patterns of change in ERPs that should be further investigated.

Introduction

Event-related potentials (ERPs), measured using electroencephalography (EEG), are widely used to study transdiagnostic cognitive processes thought to underlie psychiatric symptoms and disorders (Hajcak et al., 2019) and are ideal tools for conducting research in an RDoC framework as they are direct indices of neural activity that are moderately trait-like and have good psychometric properties (Hajcak & Patrick, 2015). While numerous ERP components demonstrate relationships with psychiatric disorders and symptoms (Hajcak et al., 2019; A. Meyer, 2016; Proudfit, 2015), longitudinal change in ERPs as a function of time-varying processes like the menstrual cycle are poorly understood. Additionally, there is a dearth of prior work characterizing between- vs. within-person variance in ERPs. The present study leveraged a within-subject design to characterize change across the menstrual cycle in two frequently-studied ERP components associated with positive and negative valence systems underlying psychopathology—the reward positivity (RewP) and error-related negativity (ERN).

The reward positivity (RewP)

The RewP is an ERP index of neural responsiveness to rewards (Proudfit, 2015). Feedback indicating rewards results in a relative positivity observed in the ERP about 300 milliseconds (ms) later. Negative feedback results in an apparent negativity (Hajcak et al., 2007; Holroyd et al., 2006). The relative difference in ERP amplitudes in response to reward as compared to loss is referred to as the Reward Positivity (RewP; Baker & Holroyd, 2011; Holroyd et al., 2008; Proudfit, 2015). A smaller RewP is linked to major depression (Belden et al., 2016; Bress et al., 2013; Liu et al., 2014; Nelson et al., 2016), more severe depressive symptoms (Bress et al., 2012; Foti & Hajcak, 2009), and substance use disorders (Joyner et al., 2019). While previous meta-analyses have shown heterogeneous effect sizes for the RewP-depression association (Moran et al., 2017), a recent p-curve analysis showed evidential value for a relationship between the RewP and depression (Clayson et al., 2020). In contrast, the RewP is enhanced in hypomania (Glazer et al., 2019; Mason, O’Sullivan, Bentall, et al., 2012; Mason, O’Sullivan, Blackburn, et al., 2012). Studies suggest that the RewP is source-localized to the striatum (Carlson et al., 2011; Foti et al., 2011) and reflects phasic increases in dopamine signaling in the frontostriatal reward circuit (Proudfit, 2015). Within the RDoC Matrix, the RewP is categorized under reward learning in positive valence systems.

The error-related negativity (ERN)

The ERN is an ERP index of sensitivity to errors (A. Meyer, 2016; Weinberg, Kotov, et al., 2015). The ERN is a negativity that peaks approximately 50ms after a participant makes an incorrect response (Hajcak & Foti, 2008). It is localized to the anterior cingulate cortex (ACC; Herrmann et al., 2004; Holroyd et al., 1998) and reflects early stages of error processing for optimizing goal-directed behavior (A. Meyer, 2016; Olvet & Hajcak, 2008). Importantly, the ERN is accentuated (i.e., is more negative) in individuals with anxiety disorders (see Meyer, 2016 for a review). While there is evidence to suggest that individual differences in the ERN reflect differences in early threat sensitivity that drives vigilance and increased defensive motivational responses (see Proudfit et al., 2013 for a review), additional theories of the ERN exist, such as the compensatory error monitoring hypothesis, which posits that an increased ERN reflects a transient increase in effort to compensate for the distracting effects of worry (Moser et al., 2013). The ERN is categorized in the negative valence systems, cognitive systems, and positive valence systems domains in the RDoC Framework (see Weinberg, Dieterich, et al., 2015 for a review).

Previous studies on the association between ERPs and psychiatric symptoms have been limited by ignoring the role of within-person variance and time-varying biological processes like the menstrual cycle. Associations between psychiatric symptoms and the RewP and ERN may be exclusive to individuals assigned female at birth (AFAB; Harold et al., 2023; Ip et al., 2019; Moran et al., 2012; Moser et al., 2016). Thus, it is crucial to investigate whether hormone fluctuations across the menstrual cycle may affect the RewP and ERN.

The menstrual cycle and reward sensitivity

Prior work suggests that reward responsivity exhibits intra-individual variation across the menstrual cycle. Observational human studies suggest the mid-luteal phase—characterized by high progesterone (P4) and moderate levels of estradiol (E2)—is associated with reduced reward responsivity, while the late follicular/periovulatory phase—with low P4 and high E2—is associated with enhanced reward responsivity (see Sakaki & Mather, 2012 for a review). One study found a blunted RewP in the mid-luteal phase compared to the mid-follicular phase in those with increased trait-level depressive symptoms (Mulligan et al., 2018). In rodent and human experiments, the effects of exogenous hormones on reward sensitivity have yielded mixed findings. In humans, E2 and P4 appear to have opposing effects on reward sensitivity, with E2 enhancing reward responsiveness and P4 reducing response to rewards (Evans et al., 2002; Evans & Foltin, 2006; Hu et al., 2004; Justice & de Wit, 2000; Russo et al., 2003). In rodent studies, both E2 and P4 enhance dopaminergic activity and motivated behaviors (Becker, 1990; Becker & Rudick, 1999; Dluzen & Ramirez, 2008; Yoest et al., 2018). Thus, while prior work suggests enhancing effects of E2 and high-E2 phases on reward sensitivity, findings are mixed with regard to whether P4 enhances or dampens reward sensitivity.

The menstrual cycle, anxiety, and the error-related negativity

In prior studies, intra-individual increases in E2 have been linked to decreased reactions to negative emotional stimuli and reductions in anxious behavior (see Sakaki & Mather, 2012 for a review). P4 may increase reactions to negative stimuli in naturally-cycling humans (Sakaki & Mather, 2012). In line with these findings, the premenstrual phase has also been linked to the exacerbation of OCD symptoms in some patients (Labad et al., 2005; Vulink et al., 2006). One previous study found that while ERN amplitude did not differ between the mid-follicular and mid-luteal phases, changes in self-reported checking of symptoms between phases were associated with phase-related changes in the ERN, suggesting meaningful individual differences in ERN change between phases (Mulligan et al., 2019).

Hormonal Sensitivity

A limitation of most previous studies examining effects of the cycle on ERPs is that they have not examined individual differences in hormonal sensitivity. Studies suggest prominent individual differences in cycle effects on mood, cognition, and behavior; in the general population, most cycling individuals do not experience cyclical changes, while a minority show significant changes ranging from mild to severe (Eisenlohr-Moul, 2019). Experimental studies point to an abnormal neural sensitivity to typical flux in hormone levels across the cycle (see Eisenlohr-Moul, 2019 for a review; Schmidt et al., 1998, 2017). Thus, it is important to test whether there is heterogeneity in trajectories of change in ERPs across cycle phases.

Present Study

The present study examined: 1) the proportion of within- vs. between-person variance in the RewP and ERN, 2) the mean degree of cyclical change in the RewP and ERN 3) the between-person variability in the degree of cyclical change in the RewP and ERN, and 4) whether an individual’s degree of cyclical change in these ERPs shows coherence with their degree of change in positive (PA) and negative affect (NA) across the cycle. We conducted exploratory growth mixture models to examine latent classes of subjects displaying distinct trajectories of ERP change across the cycle. Subjects participated in repeated EEG sessions and ecological momentary assessments (EMA) of PA and NA in the early follicular, periovulatory, and mid-luteal phases.

Our hypotheses were as follows: (1) We predicted small significant fixed effects of cycle phase on RewP amplitude, ERN amplitude, and PA and NA. We anticipated that the RewP would be enhanced in the periovulatory phase vs. the early follicular and mid-luteal phases due to prior studies that have shown effects of E2 in bolstering reward sensitivity (Frank et al., 2010; Yoest et al., 2018) and that the RewP would be enhanced in the early follicular phase vs. the mid-luteal phase due to dampening effects of P4 on reward sensitivity (Dreher et al., 2007; Sakaki & Mather, 2012). We anticipated that the fixed effects of cycle phase on PA would show the same patterns. (2) We predicted the ERN amplitude would be enhanced (i.e., more negative) in the mid-luteal phase vs. the early follicular and periovulatory phases. We expected that the fixed effects of cycle phase on NA would show the same pattern. (3) Finally, we expected large, significant random effects for each of the above-hypothesized fixed effects, consistent with evidence for individual differences in affective hormone sensitivity (Schiller et al., 2016; Schmidt et al., 2017).

Method

A preregistration is on the Open Science Framework (https://osf.io/8uz7s). The project was approved by the Institutional Review Board at Florida State University.

Participants

Eighty-six naturally-cycling AFAB individuals were recruited from the Florida State University Psychology Department research subject pool. Eight participants withdrew from the study following the baseline assessment, and seven were excluded during retrospective cycle phase validation (see below), resulting in a final sample size of 71. Inclusion criteria were age 18–30 years and endorsement of having a regular menstrual cycle (i.e., able to predict menses within 3 days of accuracy). Exclusion criteria were use of hormonal birth control within the past 4 months, endorsement of irregular menstruation, and pregnancy or lactation within the past 12 months. The sample had a mean age of 19.99 (SD=1.86) and was composed of 4.2% Asian, 9.9% Black, 21.1% Hispanic, 9.9% More than one race, and 54.9% White. While sexual orientation was not collected for 32.4% of the sample, of those with data, 85% were heterosexual, 8.33% were lesbian/gay, 4.17% were asexual, and 2.08% were bisexual. Participants were compensated with research participation credit or $100.

Ecological Momentary Assessment (EMA) Measures

Multidimensional Emotion Questionnaire:

Participants completed the Multidimensional Emotion Questionnaire (MEQ; Klonsky et al., 2019), four times per day for three days within each cycle phase. Prompts were sent at random time points within the morning (9am-12pm), afternoon (12pm-4pm), evening (4pm-8pm), and night (8pm-11pm) through a smartphone application from LifeData LLC. Participants rated current intensity of 10 emotions on a 5-point Likert scale ranging from very low to very high. PA and NA scores were computed by averaging positive and negative emotions separately. Multilevel reliability of PA and NA change over time was high (PA: 0.94; NA: 0.83; Revelle, 2018), indicating that PA and NA items could be combined and utilized as summary scores.

Baseline Session Measures

For clinical characterization, additional self-report measures were collected at the baseline session: the Quick Inventory of Depressive Symptomatology 16-Item Self-Report (QIDS-SR16; Rush et al., 2000), the Anhedonia Subscale from the Personality Inventory for DSM-5 (PID-5 Anhedonia; Krueger et al., 2012), and the Penn State Worry Questionnaire (PSWQ; T. J. Meyer et al., 1990).

Procedure

Subjects participated in five data collection sessions. During the baseline session, participants provided informed consent, completed questionnaires, and reported their cycle length and the most recent start of menses. The remaining four laboratory visits were scheduled during different menstrual cycle phases (i.e., early follicular, late follicular, periovulatory, mid-luteal). The cycle phase during which participants began their EEG and EMA assessments was counterbalanced. Each lab visit included self-report questionnaires and EEG recordings during which a monetary reward task (i.e., Doors Task) and cognitive control task (i.e., Flankers Task) were completed. Finally, participants completed at-home EMA measures (see above) and daily urine-based ovulation testing to detect the mid-cycle luteinizing hormone surge (test sensitivity = 25 mIU/ml).

EEG Recording

Continuous EEG was recorded using an elastic cap with 32 electrode sites placed according to the 10/20 system. Electrooculogram (EOG) was recorded using four additional facial electrodes: two placed approximately 1 cm outside of the right and left eyes, and two placed approximately 1 cm above and below the left eye. All electrodes were sintered Ag/AgCl electrodes. Data was recorded using an acti-CHamp system (Brain Products GmbH, Gilching, Germany). The EEG was digitized with a sampling rate of 1000 Hz using a low-pass online filter set at 100 Hz. EEG data was analyzed using Brain Vision Analyzer (Brain Products, Gilching, Germany). During the EEG recordings, the Doors Task and Flanker Task were administered using Presentation software (Neurobehavioral Systems, Inc., Albany, CA, USA). They were similar to the versions used in previous studies (Hajcak & Foti, 2008; Proudfit, 2015) and are described in detail in the supplement.

ERP Processing and Scoring

Data was referenced offline to the average of left and right mastoids, band-pass filtered (0.01 to 30 Hz with a 24 dB/oct roll-off). For the RewP, feedback-locked epochs were extracted with a duration of 1,200ms, including a 200ms pre-stimulus and 1,000ms post-stimulus interval. For the ERN, response-locked epochs were extracted with a duration of 1,300ms, including a 500ms pre-response and 800ms post-response interval. These segments were then corrected for eye movement artifacts using a regression-based approach (Gratton et al., 1983). Epochs containing a voltage greater than 50μV between sample points, a voltage difference of 175μV within a 400ms interval, or a maximum voltage difference of less than 0.50μV within 100ms intervals were automatically rejected on a channel-specific basis. The 200ms pre-stimulus interval and the −500 to −300ms pre-response interval were used as the baseline for the RewP and the ERN, respectively. Averaged ERP waveforms were derived by averaging the epochs for each condition for each participant. A 225–325ms time window for RewP was determined by visual inspection of grand-averaged waveforms, and the averaged amplitude of the RewP in that time window was extracted. A 0–100ms time window was used to compute the averaged amplitude of the ERN within each condition for each participant.

Additionally, due to the use of multiple cycle phase contrasts in our models and our data structure of up to three observations per person, we found there were insufficient degrees of freedom to examine random intercepts along with the two planned random cycle phase slopes. Therefore, we shifted our approach to examine the trial-level (i.e., non-averaged) ERP data over time, which provided sufficient degrees of freedom to model random intercept and slopes as planned. Using the same time windows described above, averaged amplitudes were exported per trial while excluding trials with artifacts, and these were used for the multilevel models (trials nested within visits nested within person) examining phase effects on ERP amplitudes.

Menstrual cycle phase staging

Based on the subsequent onset of menses and ovulation testing, the cycle phase during which lab visits had occurred were retrospectively validated and recategorized if needed (see supplement for full details). This process resulted in a final sample size of 71 participants contributing 48 early follicular, 52 periovulatory, and 56 mid-luteal lab visits for analysis.

Statistical Analyses and Power

Based on simulations from Arend & Schäfer, 2019 and taking into account our proposed interaction terms, with moderate clustering of the outcomes within a participant (ICCs = .41-.77) and approximately 50 participants providing 3 observations apiece, our minimal detectable effect size was r = .26, a conventionally small effect size, accounting for 6.7% of the variance in the outcome.

All analyses were done in R version 4.2.1. For the two EEG outcome variables (i.e., ERP response amplitudes in the Doors and Flanker Tasks), two sets of three-level models were conducted (one set for each outcome variable) in which we covaried for fixed effects of phase order (level 2), trial condition (level 1), and trial number (level 1) and examined fixed and random effects of cycle phase (level 2) while including a random intercept for ID and random slopes for the cycle phase contrasts. Specifically, random slope terms consisted of cross-level interactions between phase and trial condition (e.g., win or loss; error or correct) to facilitate observing random effects of phase on the difference between conditions (i.e., because the RewP and ERN are both measured as the difference between conditions). Two models were included in each set, one with the periovulatory phase as the reference and one with the mid-luteal phase as the reference, allowing examination of all three cycle phase contrasts.

For the self-report outcome variables (i.e., PA and NA), two sets of three-level models were similarly conducted (one set for each outcome variable) in which we covaried for fixed effects of EMA day count (i.e., the number of days elapsed since the first EMA assessment; level 1) and time of day (i.e., morning, afternoon, evening, night; level 1), and examined fixed and random effects of cycle phase (level 2) while including a random intercept for ID and random slopes for the cycle phase contrasts. Again, there were two models within each set with rotated cycle phase reference to examine all three cycle phase contrasts.

Four additional multilevel models were conducted to examine whether trait ERP (i.e., person mean across assessments) and state ERP (i.e., person-centered values of ERPs at each assessment) predict affect. In these models, we covaried for EMA day count and time of day, and included trait and state ERPs as fixed effect predictors (see supplement for more detail). Next, we conducted post-hoc repeated measures correlations to examine associations more closely between individual MEQ affect items, and state and trait ERPs (see supplementary materials). Finally, given the observed heterogeneity in ERP change across the cycle, we conducted post-hoc exploratory analyses to determine if data-driven latent classes in RewP and ERN trajectories would emerge. Growth mixture models were conducted in the R package OpenMx (Neale et al., 2016) and additional information regarding these statistical methods can be found in the supplement. We also examined differences between the resulting classes in depression, worry, and anhedonia scores using independent samples t-tests.

Results

Task Effects

As a manipulation check, we examined if task conditions affected ERPs as anticipated and results are found in Table 1. All Doors Task models showed more positive ERP response to monetary wins than losses. All Flanker Task models showed more positive ERP response on correct trials than on error trials.

Table 1.

Multilevel models predicting ERP amplitudes

	Doors Task ERPs Reference Phase: Periovulatory			Doors Task ERPs Reference Phase: Mid-luteal			Flanker Task ERPs Reference Phase: Periovulatory			Flanker Task ERPs Reference Phase: Mid-luteal
Predictors	Est	SE	p	Est	SE	p	Est	SE	p	Est	SE	p
Intercept	10.03	0.59	<0.001	10.73	0.71	<0.001	7.56	0.60	<0.001	6.41	0.59	<0.001
Time point	−1.22	0.13	<0.001	−1.16	0.16	<0.001	−0.39	0.01	<0.001	−0.16	0.01	<0.001
Trial Number	0.03	0.01	<0.001	0.03	0.00	<0.001	0.01	0.01	<0.001	0.01	0.01	<0.001
Condition: Loss/Correct (0) vs. Win/Error (1)	5.04	0.28	<0.001	5.80	0.28	<0.001	−7.38	0.54	<0.001	−6.70	0.39	<0.001
Phase: midluteal (vs periovulatory)	0.86	0.41	0.037	-	-	-	−0.58	0.35	0.100	-	-	-
Phase: early follicular (vs periovulatory)	1.08	0.43	0.013	-	-	-	−0.63	0.35	0.072	-	-	-
Condition*phase: 1*midluteal	0.75	0.16	<0.001	-	-	-	0.41	0.34	0.233	-	-	-
Condition*phase: 1*early follicular	−0.67	0.16	<0.001	-	-	-	1.05	0.34	0.002	-	-	-
Phase: early follicular (vs midluteal)	-	-	-	0.19	0.38	0.620	-	-	-	0.12	0.38	0.758
Phase: periovulatory (vs midluteal)	-	-	-	−0.75	0.38	0.047	-	-	-	0.56	0.38	0.139
Condition*phase: 1*early follicular	-	-	-	−1.47	0.08	<0.001	-	-	-	−0.02	0.34	0.952
Condition*phase: 1*periovulatory	-	-	-	−0.75	0.08	<0.001	-	-	-	−0.66	0.34	0.052
Random effects
σ2	1.00			1.00			1.00			1.00
τ00	15.71ID			20.51 ID			22.45ID			22.12ID
	1.58 ID:Visit			3.18 ID:Visit
τ11	5.28 ID.Phase (POv vs. others)			5.09 ID.Condition			5.48 ID.Phase (POv vs. others)			7.10 ID.Phase (MLu vs. others)
	5.16 ID.Condition						18.19 ID.Condition			9.61 ID.Condition
	1.01 ID.Phase (POv vs. others):Condition						5.25 ID.Phase (POv vs. others):Condition			6.59 ID.Phase (MLu vs. others):Condition
ρ01	0.22 ID.Phase (POv vs. others)			0.14 ID			−0.53			−0.47
	0.41 ID.Condition						0.03			−0.21
	−0.95 ID.Phase (POv vs. others):Condition						−0.01			0.78
N	71ID			71ID			69ID			69ID
	158 Visit			158 Visit
Observations	9,417			9,417			46,492			46,492
Marginal R2 / Conditional R2	0.256 / 0.970			0.224 / 0.973			0.161 / 0.967			0.168 / 0.965

Note. POv = periovulatory. MLu = mid-luteal. Significant results are bolded (p<0.05).

Impact of covariates on ERPs and affect

Table 1 depicts effects of covariates on ERP amplitudes. As trial number increased, ERP responses (to both conditions) in the Doors Task increased. Also, as assessment number increased, ERP responses in the Doors Task decreased. As trial number increased, ERP responses (to both conditions) in the Flanker Task increased. As assessment number increased, ERP responses in the Flanker Task decreased. These effects were covaried in final models.

Table S5 in the supplement depicts effects of covariates on affect. There were no fixed effects of day count (i.e., number of days since EMA assessments began) on PA. PA was significantly reduced in the morning and night, and increased in the evening, as compared to the afternoon. As day count increased, NA increased. NA did not vary between mornings and afternoons, but NA levels were higher at night vs. afternoons and lower in the evening vs. afternoons. These effects were covaried in final models.

Hypothesis tests: Does cycle phase influence ERPs?

ICCs for the RewP and ERN were .41 and .55, respectively. Table 1 presents Doors and Flanker ERP models.

Average effects of cycle phase on Doors Task ERPs

For the model predicting Doors Task ERPs with the periovulatory phase as the reference phase, the cross-level interaction term between trial condition and a dichotomous periovulatory vs. other phases variable was entered as a random slope term to facilitate model convergence (as recommended in Brauer & Curtin, 2018). For the mid-luteal reference model, the model did not converge when including phase contrast random slopes, and therefore, only random slopes for condition (i.e., win vs. loss) were included. There were significant fixed effects of cycle phase on ERP responses elicited by the feedback stimulus in the doors task (i.e., collapsed across win and loss conditions). Specifically, the ERP response was enhanced in the mid-luteal and early follicular phases (vs. periovulatory) and did not differ between the mid-luteal and early follicular phases, suggesting a more fundamental cyclical regulation of ERP responses that was independent of win or loss context.

There were also significant fixed effects of cycle phase on the RewP (i.e., the phase × condition interaction term). The RewP was enhanced in the mid-luteal phase vs. the periovulatory and early follicular phases. The RewP was reduced in the early follicular phase vs. the periovulatory phase. These sample-level findings differed from our hypotheses the RewP would be enhanced in the periovulatory phase vs. the early follicular and mid-luteal phases and that the RewP would be enhanced in the early follicular phase vs. the mid-luteal phase. These effects were small—the largest phase-related difference spanned approximately one microvolt.

Do people differ in cycle-related RewP changes?

Finally, there was a significant random effect of cycle phase on the RewP contrast between the periovulatory phase and the other two phases, suggesting that there were significant individual differences in RewP amplitude change between them. Figure 3 depicts the marked individual differences in the influence of the cycle on ERPs, even after accounting for visit order effects—consistent with our hypotheses and with evidence for individual differences in affective hormone sensitivity (Schiller et al., 2016; Schmidt et al., 2017).

Figure 3.

Grand mean (black) and person-centered individual mean (multicolored) values illustrating effects of menstrual cycle phases on the RewP (Panel A), ERN (Panel B), positive affect scores (Panel C), and negative affect scores (Panel D). Effects of phase order have been regressed out of the plot values.

Average effects of cycle phase on Flanker Task ERPs

For Flanker Task models, the cross-level interaction term between condition and a dichotomous reference phase vs. other phases variable was entered as a random slope term to facilitate model convergence. Additionally, level 2 (visit-level) random intercepts were not significant according to likelihood ratio tests (i.e., they did not improve model fit) and were excluded from the models. There were no fixed effects of cycle phase on response-locked ERPs in the Flanker Task (i.e., collapsed across correct and error conditions). There were significant fixed effects of cycle phase on the ERN (i.e., the phase × condition interaction term). The ERN was larger in the periovulatory phase (vs. follicular) but did not differ between the mid-luteal and periovulatory phases, nor between the mid-luteal and early follicular phases. These sample-level findings differed from our hypotheses that the ERN would be enhanced (i.e., more negative) in the mid-luteal phase as compared to the early follicular and periovulatory phases.

Do people differ in cycle-related ERN changes?

There were significant random effects of cycle phase on the ERN cycle phase contrasts. Specifically, there was a significant random effect of cycle phase on the ERN contrast between the periovulatory phase and the other two phases, between the mid-luteal phase and the other two phases, and between the early follicular phase and the other two phases, suggesting that there were significant individual differences in ERN amplitude change between each of the phases (Figure 3)—consistent with our hypotheses and with evidence for individual differences in affective hormone sensitivity (Schiller et al., 2016; Schmidt et al., 2017).

Hypothesis tests: Does cycle phase influence affect?

Average Effects of Cycle Phase on Affect

ICCs for PA and NA were .77 and .48, respectively. Table S5 presents PA and NA models. There were no significant fixed effects of cycle phase on PA and NA [all p’s > 0.27], consistent with previous findings that a majority of naturally-cycling individuals do not show significant changes in mood throughout the cycle (Eisenlohr-Moul, 2019; Hengartner et al., 2017; Leeners et al., 2017).

Do people differ in cycle-related affect changes?

There were significant random effects of cycle phase on PA and NA changes between all three cycle phase contrasts (i.e., periovulatory vs. mid-luteal, mid-luteal vs. early follicular, early follicular vs. periovulatory) as shown in Table S5. This suggests that there were significant individual differences in PA and NA changes between each of the phases, as is also shown by spaghetti plots in Figure 3.

Hypothesis tests: Are state and trait variance in ERPs and affect concordant across the cycle?

Next, we examined whether state-related variation in the RewP and ERN exhibited fixed effects on PA or NA. First, trait (i.e., person-means) and state (i.e., person-mean centered deviations) RewP were entered in multilevel models as fixed effect predictors of PA and NA. Model results in Table S6. State RewP was significantly and positively associated with state PA and was not associated with NA. Trait RewP was not associated with trait PA but was significantly and negatively associated with trait NA.

ERN model results are depicted in Table S7. Neither state nor trait variance in ERN were associated with PA. An enhanced (i.e., more negative) state ERN was associated with increased state NA. On the other hand, a blunted (i.e., less negative) trait ERN was associated with increased NA overall.

Exploratory Analyses

Associations between ERPs and individual affect items

To clarify whether the counter-intuitive association between trait ERN and NA was due to a lack of granularity in our outcomes (e.g., maybe anxiety would show an opposite association, but that would be obscured in an NA total outcome), we include post-hoc repeated measures correlations examining associations between individual MEQ affect items, and state and trait ERPs (Tables S3 and S4 in the supplement).

Are there latent classes of cycle-related ERP changes?

Additionally, we conducted exploratory latent trajectory modeling using growth mixture models to examine latent classes of ERP change between cycle phases. Latent trajectory modeling revealed two-class solutions for both the RewP and the ERN. Figure 4 presents a depiction of the latent classes for the RewP, as well as plots of PA and NA change across the cycle by RewP class. Fit statistics and parameter estimates are reported in Tables S8 and S9. Class 1 represented 90% of the sample and appeared to have an enhanced RewP in the mid-luteal phase vs. the other two phases. In comparison to Class 1, Class 2 (10% of the sample) had a relatively lower/blunted RewP overall; this class also had a blunted RewP in the mid-luteal and early follicular phases vs. the periovulatory phase.

Figure 4.

Line plots depicting mean RewP amplitudes by latent class (Panel A), as well as averaged positive (Panel B) and negative (Panel C) affect scores plotted by menstrual cycle phase and RewP latent class.

Figure 5 provides a depiction of the latent classes for the ERN, as well as plots of PA and NA change across the cycle by ERN class. Fit statistics and parameter estimates are provided in Tables S10 and S11. Class 1 represented 78.3% of the sample and demonstrated an enhanced (i.e., more negative) ERN in the periovulatory phase (vs. all other phases). In comparison to Class 1, Class 2 (21.7% of the sample) had an enhanced ERN overall. Class 2’s trajectory showed a blunted ERN in the periovulatory and a most enhanced ERN in the early follicular phase.

Figure 5.

Line plots depicting mean ERN amplitudes by latent class (Panel A), as well as averaged positive (Panel B) and negative (Panel C) affect scores plotted by menstrual cycle phase and ERN latent class.

Finally, to provide clinical characterization of the RewP and ERN classes derived by the latent trajectory modeling analyses, we examined how QIDS-SR16, PID-5 Anhedonia, and PSWQ scores varied by class. First, mean questionnaire scores for the full sample were as follows: QIDS-SR16=7.09 (SD=4.2); PID-5 Anhedonia=6.31 (SD=5.27); PSWQ=53.11 (SD=13.66). As reported in Table S12, we found that RewP Class 2 individuals showed increased scores on the QIDS-SR16, the PID-5 Anhedonia subscale, and the PSWQ as compared to RewP Class 1, suggesting that RewP Class 2 consists of individuals with increased symptoms of depression, anhedonia, and worry. On the other hand, the two ERN classes did not differ in their scores on any of these measures.

Discussion

The present study examined whether there are normative changes (i.e., fixed effects) in the RewP and ERN components as a function of menstrual cycle phase, and also whether there are individual differences (i.e., random effects) in trajectories of ERP change across the cycle. Furthermore, we examined whether state and trait variance in these ERPs showed coherence with PA and NA changes across the cycle. Finally, we conducted exploratory latent trajectory modeling to identify and characterize subgroups of individuals that show disparate patterns of change in ERPs across the cycle.

For RewP, we found small fixed effects of cycle phase on mean RewP. The RewP was most enhanced in the mid-luteal phase (vs. all other phases), and the periovulatory RewP was enhanced (vs. follicular). This contrasts with our hypothesis and previous human studies (Dreher et al., 2007; Sakaki & Mather, 2012) which suggest highest reward sensitivity in the periovulatory phase wherein E2 is elevated and unopposed by P4. It also conflicts with past rodent studies that have found dopamine-agonistic properties of E2 in core reward structures (Becker & Rudick, 1999; Diekhof, 2018). However, our findings are consistent with prior animal studies indicating that E2 enhances dopaminergic activity, which is further increased by P4 (Becker, 1990; Becker & Rudick, 1999; Dluzen & Ramirez, 2008; Yoest et al., 2018). Additionally, new evidence suggests emotional symptoms during the perimenstrual phase (which includes the early follicular phase) due to sensitivity to E2 withdrawal effects on dopaminergic and serotonergic networks (Eisenlohr-Moul et al., 2023; Peters et al., 2024). Our finding of lowest RewP amplitudes in the early follicular phase align with this pattern. Future experimental studies should examine whether this reflects a reduction in dopaminergic activity due to perimenstrual withdrawal from E2.

Importantly, our random effects revealed significant variability in the slope of RewP change between phases—particularly between the periovulatory phase and the other two phases—suggesting substantial between-person variability in trajectories of RewP change between phases. Moreover, latent trajectory modeling of RewP change across the cycle revealed a two-class solution wherein 90% of the sample (Class 1) had an enhanced RewP in the mid-luteal phase (vs. all other phases), while the remaining 10% (Class 2) had a blunted RewP in the mid-luteal and early follicular phases (vs. the periovulatory phase)—a pattern aligned with our initial hypotheses. Class 2 also had a relatively blunted RewP overall and significantly higher scores on depression, anhedonia, and worry self-report measures. Thus, this finding resembles a prior study wherein the RewP was blunted in the mid-luteal phase as compared to the mid-follicular phase only in individuals with heightened depressive symptoms (Mulligan et al., 2018). Additionally, these latent classes reveal that it could be those with blunted RewP amplitudes and elevated internalizing symptoms that demonstrate the hypothesized cyclical patterns in RewP amplitudes.

We also saw effects of cycle phase on overall ERP responses in the Doors Task with ERP amplitudes being lower in the periovulatory phase (vs. all other phases). As these effects were observed across both the win and loss conditions, these effects may be mediated by some other cognitive process modulated by the menstrual cycle. For instance, the P3 is an ERP component that typically occurs 300–600ms after a task-relevant stimulus, and that is linked to attention allocation, context updating, inhibitory control, and processing of motivationally-relevant stimuli (Polich, 2012). Since this component partially coincides with the time window of the RewP, these effects may be due to effects of the cycle on broader executive functioning processes (Sundström Poromaa & Gingnell, 2014).

Regarding the ERN, we found a small fixed effect of cycle phase on mean ERN amplitude with enhanced ERN (i.e., more negative) in the periovulatory (vs. early follicular) phase. However, it is important to note that this effect was especially small and not apparent in the grand average ERN waveforms. Future studies should investigate if this effect is replicable. Additionally, random effects again revealed significant variability in the slope of ERN change between phases, indicating varied ERN change trajectories.

Overall, this set of findings aligns with a prior ERN study that found no normative differences in the ERN between the mid-follicular and mid-luteal phases (Mulligan et al., 2019), but it was contrary to our hypothesis that the ERN would be enhanced in the mid-luteal phase (vs. all other phases), informed by previous research that has found that reactivity to aversive stimuli and obsessive-compulsive symptoms are enhanced in the luteal phase (Labad et al., 2005; Sakaki & Mather, 2012; Vulink et al., 2006). However, latent trajectory modeling revealed a latent ERN class in line with this hypothesis. Specifically, while 78.3% of the sample (Class 1) demonstrated an enhanced ERN in the periovulatory phase (vs. all other phases), Class 2’s trajectory showed a blunted ERN in the periovulatory and a most enhanced ERN in the early follicular phase—a pattern more closely aligned with our original hypotheses. Class 2 also had an enhanced ERN overall, which is often observed in individuals prone to anxiety or obsessive-compulsive symptoms; however, Class 2’s scores on depression, anxiety, and anhedonia self-report measures did not differ from Class 1. Taken together, it may be those with aberrant ERN amplitudes that demonstrate the hypothesized cyclical patterns in ERN amplitudes. Future studies with clinical samples are necessary to examine this possibility.

While there were no fixed effects of cycle phases on PA and NA, there were significant individual differences in the slope of PA and NA change between phases. These results dovetail with the preponderant evidence that there is substantial between-person variability in neural sensitivity to hormone flux (Eisenlohr-Moul, 2019).

Moreover, we found expected associations between the RewP and affect, and some unexpected associations between the ERN and affect. Enhanced trait (person-level means) RewP was associated with reduced NA overall, while state (within-person-centered phase-level means) RewP was enhanced in tandem with enhanced PA throughout the cycle. These findings are interesting as they suggest that variance in the RewP associated with NA may be more trait-like, and PA-associated RewP variance may be more state-dependent. Furthermore, these findings reveal that changes in PA tracked with changes in RewP across the cycle, suggesting that the RewP is a correlate of variation in positive affect across the cycle.

Furthermore, neither state nor trait variance in the ERN predicted PA, but state ERN was enhanced (i.e., more negative) at visits in which state NA was increased, suggesting that the ERN may be a correlate of cycle-related variation in negative affect. On the other hand, an enhanced trait ERN was associated with less trait negative affect overall. Repeated measures correlations between trait ERN and single MEQ items further suggested that this association was most driven by the “anxious”, “afraid”, “angry”, and “sad” items (see supplemental materials). The direction of this association was surprising, as we would have expected individuals with an elevated trait ERN to report increased negative affect broadly. Overall, these models reveal that state-related variance in these ERPs was correlated with positive and negative affect changes across the cycle, suggesting that cycle-mediated ERP changes may have relevance for affect and behavior.

There are limitations of this study that warrant consideration. First, the growth mixture model exploratory analysis is limited by the relatively small sample size used. While the resultant models successfully converged, the minimum sample size for latent class analyses as estimated through Monte Carlo simulation is 200 participants (Kim, 2012) making the models potentially unstable. However, this is as a common concern in latent class trajectory modeling (Kim, 2012); there are many studies that employ latent class analyses with smaller sample sizes (Claus et al., 2023; Velkoff & Smith, 2019) and the growth mixture modeling approach is superior to other clustering approaches when utilized with smaller sample sizes (Martin & von Oertzen, 2015). These analyses should be interpreted as exploratory descriptive analyses intended to provide insights about the most basic individual differences in RewP and ERN change across the cycle that should be tested again in larger studies.

The study exhibited several strengths, including its repeated measures design and hierarchical linear modeling to model longitudinal change in ERPs across the cycle. The present study revealed that there may be small normative changes in the RewP and ERN between cycle phases. In most cases, these changes were quite small, suggesting cycle phases may not be a major confounder in studies of the RewP and ERN in healthy samples. However, our latent class analyses suggest that individuals with increased general levels of depression or anxiety might exhibit greater cycle-related ERP amplitude changes. For instance, in Figure 4, which depicts plots of the RewP latent classes, it appears that Class 2 (which exhibited higher scores on depression and anhedonia measures), demonstrated slightly larger changes in RewP amplitudes between phases as compared to Class 1. This dovetails with our previous study wherein the RewP was blunted in the mid-luteal phase as compared to the mid-follicular phase only in individuals with heightened depressive symptoms (Mulligan et al., 2018). Thus, cycle phases may have bigger impacts of ERP amplitudes for clinical samples. Future studies should also directly examine the effects of these phases on the associations between these ERPs and symptoms of internalizing disorders.

Another strength of this study is its examination of individual differences in cyclical changes in ERPs and affect. Future studies should build on this work by continuing to examine whether there are phenotypic subgroups displaying disparate patterns of change in ERPs across the human menstrual cycle. This work will bring about a more idiosyncratic and accurate understanding of hormonal effects on mechanisms underlying psychopathology and will help to identify not only which mechanisms to target with interventions, but also in whom.

In conclusion, we found no normative changes in affect across the cycle and small normative changes in ERPs. This is the first study to have acknowledged and aimed to characterize individual differences in effects of the cycle on ERPs, and we found significant random slopes in our model that reveal large individual differences in trajectories of change in ERP amplitudes and affect. Additionally, state-related variance in these ERPs correlated with affect changes across the cycle, suggesting that state-level ERP changes could have relevance for affect and behavior. Finally, exploratory latent class modeling revealed subgroups displaying disparate patterns of cyclical change in ERPs. Future work should continue to explore the heterogeneity in menstrual cycle effects on these neural measures while leveraging longitudinal and experimental designs.

Figure 1.

Feedback-locked ERPs (left) and topographic maps (right) elicited in the Doors Task in the early follicular (Panel A), periovulatory (Panel B), and mid-luteal (Panel C) phases of the menstrual cycle. Panel D depicts the feedback-locked gain-loss difference waveforms for each cycle phase.

Figure 2.

Response-locked ERPs (left) and topographic maps (right) elicited in the Flanker Task in the early follicular (Panel A), periovulatory (Panel B), and mid-luteal (Panel C) phases of the menstrual cycle. Panel D depicts the feedback-locked error-correct difference waveforms for each cycle phase.

Highlights.

• Examined ERPs and affect in early follicular, periovulatory, and mid-luteal phases

• RewP and ERN amplitudes showed small group-level changes by cycle phase

• RewP, ERN, and affect showed individual differences in trajectories of change

• State variance in ERPs was associated with affect change across the cycle

• Latent class modeling revealed subgroups of ERP change across the cycle