Glucocorticoid ultradian rhythmicity differentially regulates mood and resting state networks in the human brain: A randomised controlled clinical trial

Adrenal glucocorticoid secretion into the systematic circulation is characterised by a complex rhythm, composed of the diurnal variation, formed by changes in pulse amplitude of an underlying ultradian rhythm of short duration hormonal pulses. To elucidate the potential neurobiological significance of glucocorticoid pulsatility in man, we have conducted a randomised, double-blind, placebo-controlled, three-way crossover clinical trial on 15 healthy volunteers, investigating the impact of different glucocorticoid rhythms on measures of mood and neural activity under resting conditions by recruiting functional neuroimaging, computerised behavioural tests and ecological momentary assessments. Endogenous glucocorticoid biosynthesis was pharmacologically suppressed, and plasma levels of corticosteroid restored by hydrocortisone replacement in three different regimes, either mimicking the normal ultradian and circadian profile of the hormone, or retaining the normal circadian but abolishing the ultradian rhythm of the hormone, or by our current best oral replacement regime which results in a suboptimal circadian and ultradian rhythm. Our results indicate that changes in the temporal mode of glucocorticoid replacement impact (i) the morning levels of self-perceived vigour, fatigue and concentration, (ii) the diurnal pattern of mood variation, (iii) the within-network functional connectivity of various large-scale resting state networks of the human brain, (iv) the functional connectivity of the default-mode, salience and executive control networks with glucocorticoid-sensitive nodes of the corticolimbic system, and (v) the functional relationship between mood variation and underlying neural networks. The findings indicate that the pattern of the ultradian glucocorticoid rhythm could affect cognitive psychophysiology under non-stressful conditions and opens new pathways for our understanding on the neuropsychological effects of cortisol pulsatility with relevance to the goal of optimising glucocorticoid replacement strategies.


Timeline of Day 5
Metyrapone administration + Daily hydrocortisone dose: 20 mg

Figure S2
Overview of the type and timeline of the clinical trial. Fifteen subjects have participated in a randomised, double-blind, placebo-controlled crossover study. Each of them underwent three 5-day long periods of hydrocortisone replacement therapy [after concurrent pharmacological blockage of endogenous cortisol biosynthesis via the oral administration of metyrapone, for more details see Kalafatakis et al. (2016b)] with a minimum interval between the treatment periods of 2 weeks. For each treatment period hydrocortisone was substituted in one of the following modes; either subcutaneously, via a pump, delivering pulses of hydrocortisone every 3 hours, with the pulse amplitude varying depending on the time of day (big pulses of 4 mg of hydrocortisone were being infused at 03:00 am, 06:00 am and 09:00 am, intermediate pulses of 2.3 mg of the hormone were being infused at 12:00 pm, 03:00 pm and 06:00 pm, and small pulses of 0.5 mg of the hormone were being infused at 09:00 pm and 12:00 am) (SCP). Or subcutaneously, via a pump, infusing hydrocortisone continuously in a rate-varying manner depending on the time of day (starting at 2 mg/hour at 02:00 am, dropping to 1 mg/hour at 08:00 am, further dropping to 0.4 mg/hour at midday, and again to 0.1 mg/hour at 08:00 pm, before increasing back to 2 mg/hour at 02:00 am of the next day) (SCC). Or orally, three times a day (10 mg at breakfast, 5 mg at lunch and 5 mg at dinner) (PO). In all treatment modes the daily hydrocortisone dose was the same (20 mg). Each participant was allocated in a random order of treatment periods, and this order was unknown to him and the group of researchers; for every treatment period participants were given oral pills (hydrocortisone/placebo) with instructions on when to take them, and connected to a pump of subcutaneous drug delivery (placebo/hydrocortisone).
Starting after midday on day 1 of each treatment period, subjects were participating in an ecological momentary assessment (EMA) experiment mediated by an android phone they were constantly carrying with them. Through the android phone subjects were answering questions on their mood and reactivity either during fixed timepoints per day (in the morning after waking up, i.e. morning report, and in the evening after 19:00, i.e. evening report, blue arrows) or at random timepoints and for a random number of times throughout each day (yellow arrows) (for more details see supplementary figure 2). The EMA experiment was being completed after the subjects answered the questions of the morning report on the fifth study day. Thereafter, subjects were undergoing the rsfMRI experiment, followed by the ECAT and EREC.

Figure S3
Overview of the ecological momentary assessment (EMA) study. The questionnaires delivered to the subjects via android phones contained two types of questions; (ICFS) 29 statements from the multi-dimensional identity-consequence fatigue scale, and (VASQoM) 9 questions on their self-perceived emotional state. Typically, five components were being constructed by the 29 statements of each ICFS, and the value of each component was being derived by the mean value from the answers of the participants in the corresponding statements (VIGOR was being constructed by answers in the statements 3/5/7/14, DISTRACTION by answers in the statements 9/15/16/17/18, FATIGUE by answers in the statements 1/2/4/6/10/12, MOTIVATION by answers in the statements 8/11/13, and ACTIVITY by answers in the statements 19-29). Subjects had to complete a morning report (after waking up), containing both the ICFS and VASQoM questionnaires (blue boxes), as well as an evening report, which was available from 19:00 until midnight and containing only the VASQoM questions (orange boxes). In between, subjects had to complete at random timepoints a random number of VASQoM questionnaires (red boxes). The EMA was starting around midday of the first treatment day (per study arm) and finishing after completing the morning report of the fifth treatment day (per study arm).
(A) An example of how each statement of the ICFS was displayed through the android phone to participants. The latter needed to choose the most appropriate answer from the following five: (value=1) I strongly agree, (value=2) I agree, (value=3) I neither agree nor disagree, (value=4) I disagree, (value=5) I strongly disagree.
(B) An example of how each statement of the VASQoM was displayed through the android phone to participants. The latter needed to move the slider up or down, to the appropriate level (in a 0-100 scale, with 0 being the absolute negative response and 100 the absolute positive).  Additional information on the processing pipeline followed for performing a ROI-dependent analysis of the resting state functional neuroimaging data. Two independent approaches have been used; in the first of them (blue arrows) the timeseries of each of the 10 preselected ROIs have been correlated with the timeseries of each of the 3 main RSNs corresponding to each subject and treatment mode. A mean value has been calculated from the resulting correlation coefficients per ROI and treatment group, and subsequently these mean values have been compared at treatment group-level by applying Fisher z-transformation. The 3 main RSNs have been isolated from the group-level ICA of the resting state functional neuroimaging data across all subject and treatment group sessions. The second approach (red and green arrows) consisted of a validation step, as well as the main analysis step. We implemented the seed-based functional connectivity analysis proposed by Di Martino et al. (2008), and validated that it created sensible outputs when applied to our dataset; indeed, when using PCC as a seed for that type of analysis, and averaging the output across all subjects and sessions, we were able to show that PCC, precisely as expected from the literature [9], was positively correlating with brain regions like precuneous, medial prefrontal cortex and vACC (corresponding to DMN) and negatively correlating with brain regions like insular cortices, sACC, frontal and parietal regions (corresponding to the salience and executive control networks). Finally, we used RAmy, RNA and ROFC to perform a seed-based functional connectivity analysis at individual level, and subsequently perform an analysis of covariance at each treatment group-level between the seed-related networks and the corresponding degree of positive affect of the participants (as captured by the ecological momentary assessment morning report a few minutes prior the resting state neuroimaging experiment).

Figure S6
A different perspective on presenting the correlation coefficient values between the timeseries of the three main large-scale resting state networks of the human brain (default-mode network/ DMN, salience network/ SN and executive control network/ ECN) and the preselected regions of interest (ROIs, see Supplementary  (Ι) Exploratory factor analysis was performed using principal component analysis to reduce the 9 items of the VASQoM (see supplementary Figure 3) to a lower number of variables and to identify empirically related groups of variables. We extracted two factors, positive affect and negative affect, based on the examination of the eigenvalues, the scree plot and the interpretability of the factors. We applied a varimax rotation to the factor loading matrix to achieve a simpler loading pattern. Only rotated factor loadings with a magnitude of 0.4 or greater were retained for the computation of the factor scores. The factor scores are a weighted sum of the loaded factors for each participant.
(II) An individual analysis of each treatment condition over the course of the 24-hour cycle revealed a decrease of positive affect ratings on the SCC, an increase in positive affect ratings on SCP and no considerable change of positive affect ratings on PO. The opposite pattern was found for negative affect ratings. The diurnal variation of mood [positive affect minus negative affect] is plotted in Figure 2.
(III) For SCC, no individual effect of day was found on negative or positive affect ratings. For SCP, negative affect started decreasing on day 3 and continued to do so on day 4, while positive affect ratings increased on day 4. For PO, there was some indication for the increase of negative affect ratings on days 3 and a decrease on day 4. There was also an increase of positive affect ratings on day 4.