Associations of endogenous 17-β-estradiol with theta amplitude and performance in semantic categorization in young women

Highlights

• •We correlated estradiol with performance in a semantic categorization task.
• •In early follicular women estradiol correlates positively with response time.
• •Response time in semantic categorization correlates negatively with theta amplitude.
• •Estradiol level correlates negatively with theta amplitude.

Abstract

In semantic categorization processes, individuals form a relation between perceived or imagined objects and knowledge about these objects. In the present semantic categorization study, we correlated endogenous 17-β-estradiol levels (E2) with performance as well as amplitude of theta oscillations in young women (age 23.1 ± 3.4 years). The semantic categorization task consisted of nouns representing either living or non-living items. Each item was characterized either by many or by few features. We identified parameters associated or not associated with menstrual cycle phases. Irrespective of the menstrual cycle phase, women (1) responded faster to living items as well as to nouns characterized by many features compared to non-living items and items characterized by few features, (2) showed higher accuracy to non-living items and items having many features, and (3) showed negative correlation between response time (RT) and theta amplitude. RT, accuracy and post-stimulus theta amplitude were not statistically significantly different among early follicular, late follicular or luteal women. In early follicular but not in late follicular or luteal women, we observed (1) a positive correlation between E2 and latency in RT, (2) a negative correlation between E2 and accuracy, and (3) a negative correlation between E2 and post-stimulus theta amplitude. A mosaic of menstrual cycle phase-dependent and -independent associations may indicate that a similar performance in each menstrual cycle phase is related to different modulation of synaptic activity by hormones.

Introduction

Cortical activity monitored by electroencephalogram (EEG) fluctuates during the menstrual cycle (Vogel et al., 1971; Creutzfeldt et al., 1976; Becker et al., 1982; Solís-Ortiz et al., 1994; Brötzner et al., 2014). Frequency within the alpha band decreases in the late follicular phase compared to the early follicular or luteal phase (Becker et al., 1982; Brötzner et al., 2014). Frequency in the theta band increases in the late follicular compared to luteal phase (Becker et al., 1982). At least changes in alpha frequency correlate with fluctuations in ovarian sex hormone levels (Brötzner et al., 2014). In the early follicular phase, E2 and progesterone are at a low level, in the late follicular phase (periovulatory period) E2 level increases whereas progesterone stays at a low level, and in the luteal phase, E2 as well as progesterone is elevated.

Verbal performance fluctuates during the menstrual cycle. In young women, E2 or menstrual cycle phases characterized by elevated E2 correlate positively with performance in verbal memory and verbal fluency tasks (Hampson, 1990; Phillips and Sherwin, 1992; Maki et al., 2002; Rosenberg and Park, 2002). However, Simic and Santini challenge the stereotypic association between elevated E2 and improvement in verbal performance. These authors describe improved performance in a phonemic fluency task in menstrual cycle phases, when E2 and progesterone are low (menstrual phase), as well as the mid-luteal phase, when E2 and progesterone are elevated. In the pre-ovulatory phase, when E2 is elevated, but progesterone is low, phonemic fluency is impaired (Simić and Santini, 2012). Further, Hausmann et al. document in a lexical decision task an association between low levels of E2 and fast response time (RT) (Hausmann et al., 2002). These contrasting findings may relate to different aspects of verbal performance.

An initial process in verbal performance is categorization of objects. During a categorization process, an individual assigns meaning to sensory stimuli or to a word. Accordingly, in a successful categorization process, subjects establish a relation between sensory input and knowledge of properties of perceptual objects. Numbers of features of items are reliable predictors for RTs in subjects. Notably, items having a high number of features provoke faster RTs than items having a low number of features (Klimesch, 1981; Zauner et al., 2014).

Cortical activity changes predictably during semantic categorization. Specifically, changes in P1, a product of evoked synchronized alpha oscillation, and alpha and theta power have been observed (Bastiaansen et al., 2005; Fellinger et al., 2012; Zauner et al., 2014). Visual sematic categorization is associated with alpha desynchronization, e.g., alpha power following presentation of stimulus is smaller than prestimulus alpha power (Fellinger et al., 2012; Zauner et al., 2014). Bastiaansen et al. suggest that an increase in theta power in a language processing task is associated with the retrieval of lexical semantic information (Bastiaansen et al., 2005).

In our study, participants performed a semantic categorization task, where they had to categorize words representing either living or non-living items. Given (1) the findings that cortical activity as well as semantic processing fluctuates during the menstrual cycle and (2) E2 has been associated with verbal performance, we analyzed correlations between E2 and performance in early semantic categorization as well as cortical activity in young women during the menstrual cycle.

Experimental procedures

Participants

Twenty healthy women gave written consent to participate in this study, which was approved by the local ethics committee at the University of Salzburg. Two women were excluded from analysis. One subject was excluded because she did not complete the entire study and one subject because her RT was more than two standard deviations higher than the mean. The remaining 18 subjects (19–30 years, age 23.06 ± 3.35 years) did not use hormonal contraceptives and had a regular menstrual cycle (mean cycle length: 28.08 ± 2.34 days). Women did not report a history of neurological disorders and were free of medications. Participants were recruited mainly from the Department of Biology or Department of Psychology from the University of Salzburg. Students were compensated for participating in the study with 30 Euro.

Women completed the semantic categorization paradigms three times during the menstrual cycle (early follicular, late follicular and luteal phases). About a third of the participants started in the early follicular phase, about a third in the late follicular phase and about a third in the luteal phase. The early follicular phase was defined from onset of menstruation plus 5 days, the late follicular phase approximated by counting back 14 days from the next predicted onset of menstruation, evaluated using a commercial ovulation test (Pregnafix® Ovulations test), and verified by verbal reports of the actual onset of menstruation, and the mid-luteal phase was defined by three days post-ovulation and five days before onset of the next expected menstruation.

Quantification of 17-β-estradiol (E2)

E2 was quantified in saliva using a Salivary Estradiol ELISA kit according to the recommendations of the provider (Demeditec Diagnostics GmbH, Kiel, Germany). Saliva samples were collected in sterile polypropylene centrifuge tubes and stored in a freezer at −20 °C. Particles in saliva samples were removed by centrifugation (2355g for 15 min) before sex hormone quantification. The mean and standard deviation of E2 levels were n(18) 5.91 ± 2.59 pg/ml in the early follicular phase, n(18) 8.19 ± 2.88 pg/ml in the late follicular phase and n(18) 7.51 ± 3.13 pg/ml in the luteal phase.

Experimental design

Women performed a verbal semantic categorization task (McRae et al., 2005; Zauner et al., 2014). In the McRae study, participants were asked to list features for each of the 541 concept words (McRae et al., 2005). Zauner et al. selected living items with a high number of features (NOF+) or a low number of features (NOF−) as well as non-living NOF+ and NOF−. Each category contained 70 items (total 280 items). “Living items” included animals and fruits and “non-living items” included tools and everyday objects. Living NOF+ are characterized by more than four features, living NOF− by less than four features, non-living NOF+ by more than five features, and non-living NOF− by less than five features (see Appendix in Zauner et al., 2014). For example, the living NOF+, “pony”, has been attributed with nine features, whereas the living NOF−, “bison”, has been attributed with one feature (McRae et al., 2005; Zauner et al., 2014). Nouns were presented using Presentation Software (version 0.71, 2009, Neurobehavioral Systems Inc., Albany, CA, USA) on a computer screen (75-Hz refresh rate). Between the presentation of nouns, a gray fixation cross “+” was presented at the center of a black screen for 400–600 ms. To minimize expectations, 50-ms intervals were used between 400- and 600-ms fixation cross presentation. Fixation cross was followed immediately by 1000-ms noun presentation. Nouns (capital letters, horizontal angle: 2.8–4.3°; vertical angle = 0.66°) were presented in a bright gray box (9.7° × 2.6°). Following word presentation, women had to decide between living and non-living items. Nouns were presented in a randomized order. Women were instructed to focus on a fixation cross and to respond as quickly as possible by pressing one of two buttons. Presentation of nouns representing living items required button pressing with the right index finger and nouns representing non-living items required button pressing with right middle finger. Women practiced a version of the task prior to monitoring of RT and accuracy as well as EEG.

Behavioral data

RT and accuracy were determined at an individual subject level. We extracted the median of RT and the accuracy collectively for each experimental condition (living NOF−, living NOF+, non-living NOF−, non-living NOF+) only for correct trials. PASW Statistics 18 (SPSS) was used for statistical analysis. To test for living/non-living or NOF or cycle phase-related differences in RT or accuracy a 2x2x3 ANOVA was calculated with factors living/non-living, NOF(±) and cycle phase. To specify the contribution of E2 on RTs and accuracy, we correlated E2 levels for each menstrual cycle phase and for four trial conditions: living NOF−, living NOF+, non-living NOF−, non-living NOF+. E2 levels were associated with RTs or accuracy using the Pearson correlation coefficient (2-tailed). RTs, E2 level and event-related potential (ERP) amplitudes were normally distributed (Kolmogorov–Smirnov Test).

Data acquisition and analysis

Data were recorded during one testing lasting approximately 14 min, yielding 70 epochs of data per item category (living with low number of features, living with high number of features, non-living with low number of features, non-living with high number of features). All electrophysiological signals were recorded with a 64-channel acquisition system, sampled at a 1000-Hz rate, and amplified with BrainAmp amplifier (Brain Products, Inc., Gilching, Germany). Electrodes were referenced against a nose electrode with a grounding electrode being placed to the forehead. Recording bandwidth was between 0.016 and 100 Hz. The 10–20-system was used for electrode positioning (Jasper, 1958). Impedance of electrodes was kept below 8 kΩ. To exclude interference from local net current, a 50-Hz notch filter was applied. Eye artifacts due to blinks and eye movements were detected using electrodes positioned at vertical and horizontal positions near the right eye (electrooculogram).

Analysis of EEG data was done with BrainVisionAnalyzer 2.0 (Brain Products, Inc., Gilching, Germany). Raw EEG data were re-referenced to earlobe-electrodes and filtered with an IIR bandpass filter between 0.5 and 70 Hz. Eye artifacts were removed by excluding EEG components occurring in parallel to eye movements detected in the electrooculogram based on Gratton and Coles (Gratton et al., 1983). Remaining artifacts were excluded following visual inspection. For ERP-analyses trials including artifacts and trials with RT below 400 ms or above 1000 ms were excluded. We only analyzed correct trials. The raw EEG data were filtered in the theta frequency band adjusted to the individual alpha frequency (IAF) band, because (1) theta co-varies with the alpha oscillation (Klimesch et al., 1994; Doppelmayr et al., 1998; Klimesch, 1999) and (2) alpha is the dominant frequency in the human EEG (Klimesch, 2012). IAF was the spectral component with the largest power in the alpha frequency band between 7 and 14 Hz. Since individual alpha frequency depends on menstrual cycle phase (Brötzner et al., 2014) the frequency range for theta extraction differs among menstrual cycle phases. We filtered the raw EEG data in a frequency range between 4.6 and 6.6 Hz for early follicular and luteal phases and between 4.3 and 6.3 Hz for the late follicular phase.

Analysis of ERPs

Data were segmented into epochs starting 600 ms preceding visual target presentation and ending 1000 ms after target onset. ERPs for four experimental conditions (living with low number of features, living with high number of features, non-living with low number of features, non-living with high number of features) were obtained by averaging over trials. Then, individual ERP-components were semi-automatically detected. The first-positive theta component was detected as the largest positive amplitude between 50 and 150 ms and the first-negative theta component as the largest negative amplitude between 150 and 250 ms. The amplitudes of the three posterior-occipital electrodes on the left hemisphere (Po3, Po7, O1) and right hemisphere (Po4, Po8, O2), respectively, were averaged. Figures were created using MATLAB (R2010b). An ANOVA was used to evaluate EEG data as well as behavioral and endocrinological data. A 2 × 2 × 3 ANOVA was calculated with factors living/non-living, NOF(±) and cycle phase separately for P1 or N1 amplitude and left or right hemisphere. Additionally a 2 × 3 ANOVA was calculated with factors living/non-living and cycle phase separately for P1 or N1 amplitude and for high or low number of features. E2 levels were associated with theta amplitude using the Pearson correlation coefficient (2-tailed). Calculations were done for all four trial conditions (living NOF−, living NOF+, non-living NOF−, non-living NOF+) and separately for all cycle phases. Theta P1 or N1 amplitudes were correlated with RTs or accuracy using the Pearson correlation coefficient (2-tailed) in each trial condition (living NOF−, living NOF+, non-living NOF−, non-living NOF+) and for each menstrual cycle phase. PASW Statistics 18 (SPSS) was used for statistical analysis.

Results

Low endogenous E2 associates with fast RT and high accuracy

Table 1 summarizes mean and SD for RTs and accuracy in a semantic categorization task in women during early follicular, late follicular and luteal phases. Irrespective of the menstrual cycle phase, women responded significantly faster (1) to living compared to non-living items (F_(1,17) = 30.783, p < .001, η² = .644) and (2) to items characterized by a high number of features compared to a low number of features (F_(1,17) = 35.404, p < .001, η² = .676). RTs did not differ significantly among menstrual cycle phases. Correct responses ranged between 80 to 100%, with an overall mean percentage of correct responses of 96%. Accuracy was higher for non-living items compared to living items (F_(1,17) = 14.026, p = .002, η² = .452), and higher for NOF+ compared to NOF – (F_(1,17) = 6.393, p = .022, η² = .273). Accuracy was not statistically different among menstrual cycle phases (p > .05).

Table 1.

Performance in semantic categorization of visually presented nouns representing either living or non-living items in early follicular (EFP), late follicular (LFP) and luteal women (LP). RT (in ms) and accuracy (in %) do not differ among menstrual cycle phases

	Living NOF+	Living NOF−	Non-Living NOF+	Non-Living NOF−	Mean
EFP
RT	635 ± 54	657 ± 51	674 ± 69	692 ± 65	664 ± 60
Accuracy	95 ± 3	96 ± 3	97 ± 3	96 ± 3	96 ± 3
LFP
RT	645 ± 66	655 ± 63	683 ± 84	691 ± 81	669 ± 74
Accuracy	95 ± 3	95 ± 5	98 ± 2	96 ± 4	96 ± 4
LP
RT	635 ± 69	660 ± 68	665 ± 84	680 ± 95	660 ± 79
Accuracy	95 ± 3	95 ± 4	98 ± 2	96 ± 3	96 ± 3

In early follicular women, E2 correlated positively with RTs (low E2 level is associated with fast RT), irrespective of the item. Late follicular or luteal women did not reveal significant associations between E2 and RTs (Table 2). E2 correlated negatively with accuracy for non-living items in early follicular women. With the exception of non-living NOF− in luteal women, significant correlations between E2 and accuracy were not identified in late follicular or luteal women. Thus, performance in categorization of items was associated with E2 only in the early follicular phase.

Table 2.

‘Pearson’s R describing correlations between E2 and RT and accuracy, respectively

	RT				Accuracy
	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−
EFP estradiol	.636⁎⁎	.574⁎	.531⁎	.514⁎	−.363	−.343	−.579⁎	−.505⁎
LFP Estradiol	.070	.098	.258	.164	.048	−.437	−.448	−.443
LP estradiol	.371	.317	.410	.310	−.339	−.209	−.116	−.517⁎

^⁎p < .05.

^⁎⁎p < .01.

Theta amplitude correlates with RT and accuracy

Raw EEG was filtered in the theta frequency band and post-stimulus theta amplitude was quantified. Women were segregated in fast and slow responders by a median split. Women responding fast to nouns showed a significantly larger theta amplitude compared to women responding slowly to nouns (Fig. 1). Especially for non-living items, we observed significant correlations between RTs and theta amplitude in early follicular, late follicular and luteal women (Table 3). Furthermore we found significant correlations between accuracy and theta amplitude but only in early follicular and luteal phases for non-living items.

Fig. 1.

Illustration of cortical theta oscillations in response to non-living NOF− items. (A) Woman showing the shortest latency in response to a noun (red) and woman showing the longest latency in response to a noun (black). (B) Average of theta oscillations of women with shortest (below median split of RTs) (red) and longest (above the median split) latencies. Line at 0 ms indicates onset of noun presentation. ERP, event-related potential. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 3.

‘Pearson’s R describing correlations between first-positive or first-negative amplitude of theta oscillation with RT or accuracy (average of electrodes Po4, Po8, O2 in right hemisphere). For calculation of correlations, absolute values were used

	First-positive amplitude				First-negative amplitude
	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−
EFP_RT	−.550⁎	−.449	−.624⁎⁎	−.637⁎⁎	−.529⁎	−.448	−.618⁎⁎	−.651⁎⁎
EFP_accuracy	.082	.005	.414	.489⁎	−.033	−.026	.348	.403
LFP_RT	−.432	−.428	−.607⁎⁎	−.563⁎	−.422	−.425	−.633⁎⁎	−.537⁎
LFP_accuracy	−.171	.092	.090	.343	−.191	.053	.095	.302
LP_RT	−.474⁎	−.549⁎	−.537⁎	−.478⁎	−.530⁎	−.501⁎	−.563∗	−.443
LP_accuracy	.067	.176	.217	.591⁎⁎	−.040	.138	.134	.508⁎

^⁎p < .05.

^⁎⁎p < .01.

E2 correlates negatively with post-stimulus theta amplitude in early follicular women

Correlations between E2 and theta amplitude revealed significantly larger amplitude in women low in E2 compared to women high in E2 in early follicular and luteal phases (Fig. 2). Table 4 summarizes correlations between E2 and theta amplitude for early follicular, late follicular, and luteal women. Significant correlations between E2 level and theta amplitudes were identified for living and non-living items in early follicular women, but not in late follicular or luteal women. In luteal women, correlations reached significance only for NOF− items. Thus, in early follicular women, good performers were characterized by low E2 and a large theta amplitude.

Fig. 2.

Cortical theta amplitudes in response to presentation of non-living NOF− items for women during early follicular phase (EFP), late follicular phase (LFP) and luteal phase (LP). Theta activity is shown for women below (red) or above (black) the median split of 17-β-estradiol. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 4.

‘Pearson’s R describing correlations between E2 levels and theta amplitude. Note: Significant negative associations between E2 levels and the first-positive amplitude in early follicular phase. For correlational analyses average signals from Po3, Po7, O1 electrodes (left hemisphere) and from Po4, Po8, O2 electrodes (right hemisphere) were used

	Left hemisphere				Right hemisphere
	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−
EFP estradiol	−.580⁎	−.542⁎	−.538⁎	−.619⁎⁎	−.523⁎	−.457	−.416	−.579⁎
LFP Estradiol	−.120	−.211	−.256	−.223	−.117	−.257	−.219	−.181
LP estradiol	−.270	−.236	−.285	−.292	−.387	−.465	−.357	−.527⁎

^⁎p < .05.

^⁎⁎p < .01.

Theta amplitude changes by tendency within the menstrual cycle. Notably, we identified hemisphere-specific differences. Theta amplitude in the right hemisphere was by tendency larger in early follicular compared to late follicular or luteal women (F_(2,34) = 2.233, p = .123, η² = .116) (Table 5). Specifically for a low number of feature items (F_(2,34) = 2.819, p = .087, η² = .142) compared to a high number of feature items (F_(2,34) = 1.009, p = .372, η² = .056), theta amplitude did not differ in the left hemisphere among menstrual cycle phases (p > .6).

Table 5.

Theta amplitude in μV (mean + SD) in the right hemisphere in early follicular (EFP), late follicular (LFP) and luteal women (LP). Theta amplitude in the right hemisphere was by tendency larger in early follicular compared to later follicular or luteal women (p = .087) for a low number of feature items. EEG signals from the posterior electrodes Po4, Po8, and O2 were combined

	Living NOF+	Living NOF−	Non-living NOF+	Non-living NOF−	Mean
EFP
Positive amplitude	2.09 ± 1.01	2.21 ± 1.12	2.08 ± 1.13	2.14 ± 1.03	2.13 ± 1.07
Negative amplitude	−2.22 ± 1.07	−2.33 ± 1.12	−2.33 ± 1.16	−2.31 ± 1.15	−2.30 ± 1.13
LFP
Positive amplitude	1.89 ± 1.13	1.86 ± 1.35	1.98 ± 1.31	1.95 ± 1.16	1.92 ± 1.24
Negative amplitude	−2.05 ± 1.28	−2.05 ± 1.43	−2.23 ± 1.48	−2.20 ± 1.27	−2.13 ± 1.37
LP
Positive amplitude	2.07 ± 1.09	1.94 ± 1.16	2.01 ± 1.10	1.99 ± 1.11	2.00 ± 1.12
Negative amplitude	−2.15 ± 1.21	−2.05 ± 1.15	−2.21 ± 1.17	−2.23 ± 1.25	−2.16 ± 1.20

Discussion

The present behavioral and EEG findings indicate that performance in semantic categorization arises from factors, which correlate or do not correlate with menstrual cycle phase. Factors not associated with menstrual cycle phase include (1) higher latency in RT for non-living and NOF− items compared to living and NOF+ items, (2) higher accuracy for non-living and NOF+ items compared to living and NOF− items and (3) negative correlation between latency of RT and post-stimulus theta amplitude. Latency in RT, accuracy, and amplitude in post-stimulus theta are not significantly different among early follicular, late follicular and luteal women. Factors correlating with the menstrual cycle phase are restricted to the early follicular phase. In early follicular, but not late follicular or luteal women (1) E2 correlates positively with latency in RT, (2) E2 correlates negatively with accuracy, and (3) E2 correlates negatively with post-stimulus theta amplitude. Thus, to maintain a similar performance in each menstrual cycle phase, activity of neural networks involved in semantic categorization is modulated differentially in each menstrual cycle phase.

Although verbal performance is superior in women compared to men, surprisingly few studies focus on verbal performance in relation to the menstrual cycle phase or sex hormone levels in young women. Speeded articulation, verbal fluency, and verbal memory are improved in luteal compared to follicular women (Hampson, 1990; Phillips and Sherwin, 1992; Rosenberg and Park, 2002). Maki et al. (2002) found a positive correlation between E2 levels and verbal fluency. However, their correlation included E2 samples collected in different menstrual cycle phases and, accordingly, ignored menstrual cycle-specific functions of sex hormones. In contrast to the findings showing an improved verbal performance in women having elevated E2, participants in a phonemic fluency task performed best during menstrual and mid-luteal phases compared to the pre-ovulation phase (Simić and Santini, 2012). Furthermore, in a lexical decision task, E2 associates positively with RTs (Hausmann et al., 2002). In the present verbal semantic categorization study, RTs and accuracy are not statistically different between early follicular, late follicular and luteal women. However, an association between E2 and RT or accuracy is detectable in early follicular women. A similar association between E2 and RT in a lexical decision task has been described by Hausmann et al. (2002). This may indicate a similar neural mechanism in processing of categorical and lexical tasks in contrast to verbal memory or verbal fluency tasks.

Synchronization in the theta range is associated with memory formation and retrieval (Hanslmayr et al., 2012). Experimental studies reveal that theta burst stimulation triggers cytoskeletal rearrangements required for changes in spine morphology in long-term potentiation (Lynch et al., 2007). Interestingly, acute application of E2 facilitates synaptic transmission and enhances LTP (Kramár et al., 2013). In addition, endogenous fluctuations of E2 in rats are associated with the number of spines on dendrites. The maximum number of spines is detectable when E2 reaches its maximum (McEwen et al., 2012). These correlative and experimental studies indicate that synaptic environment changes considerably during an ovulatory cycle. Accordingly, menstrual cycle phase-selective correlation between E2 and theta amplitude, like in the present study, may be due to changes in the synaptic properties.

Previous EEG studies have shown that semantic categorization evokes widespread activity across frontal and posterior cortical areas (Mainy et al., 2008; Fellinger et al., 2012; Yvert et al., 2012; Zauner et al., 2014). The present study focused on the association between E2 and posterior cortical activity. Further studies comparing functional connectivity between frontal and posterior cortical areas will be required to disentangle the contribution of E2 to coherent activity in distinct cortical areas. Furthermore, characterization of associations between E2 and neural equivalents of semantic processing might provide a model for analyzing the impact of ethinyl estradiol, a synthetic estrogen in oral contraceptives, on cognition in women.