Introduction
Postpartum (or puerperal) psychosis (PP) is a severe postpartum disorder characterised by an acute onset typically occurring within the first few weeks after childbirth (Munk-Olsen, Laursen, Pedersen, Mors, & Mortensen, Reference Munk-Olsen, Laursen, Pedersen, Mors and Mortensen2006). Women with an existing diagnosis of bipolar disorder (BD) or schizoaffective disorder, or those who have a personal or family history of PP are at very high risk of suffering a PP episode (Jones & Craddock, Reference Jones and Craddock2001; Jones, Chandra, Dazzan, & Howard, Reference Jones, Chandra, Dazzan and Howard2014; Sit, Rothschild, & Wisner, Reference Sit, Rothschild and Wisner2006).
To date, most literature has focused on the evaluation of genetic markers, clinical presentation, prognosis, and treatment of PP. Despite the fact that cognition and emotional processing are typically identified as key impairments in non-puerperal psychoses and BD (Bora, Yücel, & Pantelis, Reference Bora, Yücel and Pantelis2010; Dickinson, Becerra, & Coombes, Reference Dickinson, Becerra and Coombes2017; Kohler, Walker, Martin, Healey, & Moberg, Reference Kohler, Walker, Martin, Healey and Moberg2010; Malhi et al., Reference Malhi, Ivanovski, Hadzi-Pavlovic, Mitchell, Vieta and Sachdev2007a), little research has been conducted to understand the neurofunctional aspects of these processes in PP. Considering the close relationship between PP and BD and its presentation with both affective and psychotic symptoms, PP would be expected to present with a similar pattern of impairments. Furthermore, previous reports of structural brain changes associated with experiencing a PP episode support the notion that abnormalities of these processes will likely be detectable on a neural level (Fusté et al., Reference Fusté, Pauls, Worker, Reinders, Simmons, Williams and Dazzan2017; Lanczik et al., Reference Lanczik, Fritze, Hofmann, Schulz, Knoche and Becker1998).
In this study, we used functional neuroimaging to investigate brain activation and functional connectivity during working memory and emotional face recognition in women at-risk of PP who did and did not develop a PP episode and compared these groups to a group of healthy postpartum women.
Abnormalities in functional magnetic resonance imaging (fMRI) brain activation during working memory in non-puerperal psychoses such as schizophrenia (Glahn et al., Reference Glahn, Ragland, Abramoff, Barrett, Laird, Bearden and Velligan2005; Henseler, Falkai, & Gruber, Reference Henseler, Falkai and Gruber2009; Minzenberg, Laird, Thelen, Carter, & Glahn, Reference Minzenberg, Laird, Thelen, Carter and Glahn2009), and BD (Chen, Suckling, Lennox, Ooi, & Bullmore, Reference Chen, Suckling, Lennox, Ooi and Bullmore2011; Dell'Osso et al., Reference Dell'Osso, Cinnante, Di Giorgio, Cremaschi, Palazzo, Cristoffanini and Altamura2015; Fernández-Corcuera et al., Reference Fernández-Corcuera, Salvador, Monté, Salvador Sarró, Goikolea, Amann and Pomarol-Clotet2013; Lagopoulos, Ivanovski, & Malhi, Reference Lagopoulos, Ivanovski and Malhi2007; Thermenos et al., Reference Thermenos, Goldstein, Milanovic, Whitfield-Gabrieli, Makris, Laviolette and Seidman2010), consistently involve the dorsolateral prefrontal cortex (DLPFC), although other areas have been implicated (Thermenos et al., Reference Thermenos, Goldstein, Milanovic, Whitfield-Gabrieli, Makris, Laviolette and Seidman2010). Nevertheless, conflicting evidence exists in both disorders regarding the direction of change, with both hypo- (Jogia, Dima, Kumari, & Frangou, Reference Jogia, Dima, Kumari and Frangou2012; Monks et al., Reference Monks, Thompson, Bullmore, Suckling, Brammer, Williams and Curtis2004) and hyper-activations (Adler, Holland, Schmithorst, Tuchfarber, & Strakowski, Reference Adler, Holland, Schmithorst, Tuchfarber and Strakowski2004) being reported in patients in comparison with healthy controls.
Furthermore, patients with non-puerperal psychoses show reductions in the connectivity within the fronto-parietal network (Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017) and between the DLPFC and other cortical areas during working memory paradigms (Kyriakopoulos et al., Reference Kyriakopoulos, Dima, Roiser, Corrigall, Barker and Frangou2012). Alterations in working memory-related connectivity are also present in BD, with evidence of reduced cortico-limbic connectivity during different working memory paradigms (Cremaschi et al., Reference Cremaschi, Penzo, Palazzo, Dobrea, Cristoffanini, Dell'Osso and Altamura2013; Stegmayer et al., Reference Stegmayer, Usher, Trost, Henseler, Tost, Rietschel and Gruber2015) and increased fronto-parietal connectivity during the n-back task (Wu et al., Reference Wu, Wang, Mwansisya, Pu, Zhang, Liu and Shan2014).
To date, only one fMRI case study has investigated emotional processing in PP, by comparing a pair of monozygotic twins, one of whom experienced a PP episode. A decreased activation of the orbitofrontal cortex in response to viewing an emotional film was seen in the PP-affected twin compared to the healthy twin (Fahim, Stip, Mancini-Marie, Potvin, & Malaspina, Reference Fahim, Stip, Mancini-Marie, Potvin and Malaspina2007).
Aberrant brain activations during emotional processing in patients with non-puerperal psychoses have been reported in all cortical lobes, amygdala, and insula compared to healthy controls (Li et al., Reference Li, Chan, Gong, Liu, Liu, Shum and Ma2012; Modinos et al., Reference Modinos, Tseng, Falkenberg, Samson, McGuire and Allen2015; Taylor et al., Reference Taylor, Kang, Brege, Tso, Hosanagar and Johnson2012). Similarly for BD, widespread cortical and subcortical abnormalities, centred on the amygdala, have been reported (Chen et al., Reference Chen, Suckling, Lennox, Ooi and Bullmore2011; Houenou et al., Reference Houenou, Frommberger, Carde, Glasbrenner, Diener, Leboyer and Wessa2011; Kupferschmidt & Zakzanis, Reference Kupferschmidt and Zakzanis2011; Malhi et al., Reference Malhi, Lagopoulos, Owen, Ivanovski, Shnier and Sachdev2007b). As in the case of working memory abnormalities, both increases and decreases of brain activation related to emotional processing have been demonstrated in non-puerperal psychoses and BD.
In terms of functional connectivity, patients with non-puerperal psychoses show significantly reduced effective connectivity between the amygdala and parietal regions during emotional face recognition (Mukherjee et al., Reference Mukherjee, Whalley, McKirdy, McIntosh, Johnstone, Lawrie and Hall2012). Further reductions of amygdalar connectivity have been observed with occipital and prefrontal cortices (Das et al., Reference Das, Kemp, Flynn, Harris, Liddell, Whitford and Williams2007). Dysconnectivity between the amygdala and frontal regions has also been observed in BD patients compared to controls during emotion regulation (Townsend et al., Reference Townsend, Torrisi, Lieberman, Sugar, Bookheimer and Altshuler2013; Zhang et al., Reference Zhang, Opmeer, van der Meer, Aleman, Ćurčić-Blake and Ruhé2018).
This is the first study investigating performance and neural correlates of working memory and emotional face recognition in women at-risk of PP. We chose the Ekman faces paradigm with fearful faces to elicit emotion processing as this is frequently used in the fMRI literature and has known sensitivity to both psychoses unrelated to childbirth and BD (Grave, Soares, Martins, & Madeira, Reference Grave, Soares, Martins and Madeira2017), thus allowing comparison with other studies (Fusar-Poli et al., Reference Fusar-Poli, Placentino, Carletti, Landi, Allen, Surguladze and Politi2009). We selected the n-back task to investigate working memory since this is an established paradigm in the neuroimaging psychosis literature, with studies across numerous patient groups available for comparison (Owen, McMillan, Laird, & Bullmore, Reference Owen, McMillan, Laird and Bullmore2005). The overarching goal of this study was to compare women at-risk of PP to controls on tests sensitive to non-puerperal psychosis and BD, in order to understand the potential relationships with these other disorders. In particular, we investigated the relationship between vulnerability to PP and emotional and cognitive function. We hypothesised that: (1) women at-risk of PP would show worse performance than controls in the working memory task; (2) women at-risk of PP would show decreased frontal activation during working memory and increased amygdalar activation during emotional processing; (3) in addition, we predicted decreased functional connectivity with the DLPFC for the working memory task and with the amygdala for the emotion processing task; and finally, we predicted that (4) women who developed a PP episode would have more pronounced performance deficits, brain and functional connectivity abnormalities than women at-risk of PP who did not develop an episode.
Methods
Participants
Women were approached through perinatal services in South and West London while pregnant or within the first year of delivery. Forty-six women were recruited to participate in this study. Twenty-five participants were identified as at-risk of PP according to the following criteria: current or previous personal history of BD, schizoaffective disorder, or PP and/or family history of PP. Out of this group, 11 women developed an episode of PP (PE group), while 14 did not suffer any mental illness following the index pregnancy (NPE). For the healthy comparison group, 24 women matched to the at-risk group on age, IQ, ethnicity, and education were recruited from antenatal services. Exclusion criteria for the control group were a current diagnosis of any psychiatric condition, a personal history of BD, schizoaffective disorder, or PP and/or family history of PP, and the use of medication other than nutritional supplements at the time of recruitment. All women were between 18 and 45 years of age, had a good understanding of English, suffered no severe obstetric complications, and had no chronic medical conditions. Two women were excluded at the analysis stage, one due to high motion in the scanner (>2 mm) and one due to acquisition error, leaving N = 44 for analyses (controls = 20, NPE = 13, and PE = 11). A summary of the demographics and symptom levels at the time of assessment is given for the final group in Table 1. All women gave written informed consent. Ethical approval was granted by the Joint South London and Maudsley/Institute of Psychiatry, Psychology & Neuroscience Research Ethics Committee.
Table 1. Sociodemographic and clinical data
HC = Healthy Controls; NPE = No Postpartum Episode; PE = Postpartum Episode; HDRS = Hamilton Depression Rating Scale; MRI = Magnetic Resonance Imaging; PANSS = Positive and Negative Syndrome Scale; YMRS = Young Mania Rating Scale. Post-hoc analysis for weeks after delivery showed that differences were among the NPE and PE groups (p = 0.016), as well as HC and PE groups (p = 0.035), while post-hoc analysis for age showed no significant differences across the groups. Post-hoc analyses for the total scores on PANSS, YMRS, and HDRS showed no differences between the NPE and PE groups (p = 0.225), (p = 0.096), and (p = 1.000) respectively. Three of the women in the NPE group were taking SSRIs.
Image acquisition
Data were acquired on a 3T GE Signa HDx System (General Electric, Milwaukee, Wisconsin) with an 8-channel head-coil at the Centre for Neuroimaging Sciences, King's College London. T1-weighted images were acquired using a Spoiled Gradient Recalled (SPGR) sequence [196 coronal slices, repetition time (TR) = 7 ms, echo time (TE) = 2.812 ms, flip angle = 20°, in-plane voxel size = 1.09 mm, slice thickness = 1.1 mm, slice gap = 1.1 mm]. Functional data for the n-back task comprised 39 interleaved slices acquired bottom to top over 186 acquisition volumes; for fearful faces 39 slices were acquired sequentially descending from top to bottom over 180 acquisition volumes. Data for both functional scans were acquired using a gradient echo echoplanar imaging sequence (TR = 2 s, TE = 30 ms, flip angle = 75°, in-plane voxel size = 3.75 mm, slice thickness = 3.5 mm, slice gap = 0.5 mm).
Procedure and tasks
At the beginning of the session, each participant was familiarised with the scanner environment by lying in a ‘mock-up’ scanner and performing brief versions of both tasks. Tasks and basic MR safety were explained to all women. The testing session consisted of two tasks: a working memory task (n-back) and an emotional processing task (fearful faces). Stimuli were presented to participants on a screen positioned at the rear of the scanner bore visible via a mirror mounted on the head coil.
Working memory task: n-back
In this task, a stream of letters was projected on the screen, one at a time. The participant's task was to identify trials where the letter matched the letter occurring ‘n’ before the current one. 0-, 1-, 2-, and 3-back conditions were used requiring a button press when a correct target was displayed. 0-back was the control condition and required participants to press a button when the letter ‘X’ appeared. The task consisted of 180 trials presented in 12 blocks, each lasting 30 s with an inter-stimulus interval of 2 s. Each block consisted of an instruction (2 s) and 14 letter presentations. This task lasted a total of 6 min and 20 s.
Emotional processing task: fearful faces
To investigate emotional processing an implicit Ekman faces paradigm was used. Images of five men and five women were chosen from the Pictures of Facial Affect set (Ekman & Friesen, Reference Ekman and Friesen1976). Each face was presented six times with two presentations at three different levels of emotion intensity – neutral, lower fear (50%), and standard fear (100%). Additionally, 12 null trials showing only a fixation cross were included. This was an event-related paradigm with a total of 72 trials of 2 s durations with a variable inter-trial interval (3.33–8.64 s, mean = 4.99 s). Participants were instructed to indicate the model's gender using a button press (left: female, right: male) and were not informed of the true purpose of the task until after completing the scanning session. This task lasted approximately 6 min.
Performance data analysis
Performance data were analysed using SPSS version 24 (IBM Corp., 2016). The n-back data were analysed using 2 × 4 mixed analysis of variances (ANOVAs), with group (at-risk and control) and memory load (0-, 1-, 2-, 3-back) as factors. For the faces task 2 × 3 mixed ANOVAs were used, with factors of group (at-risk and control) and fear intensity (neutral, lower fear, and standard fear). For both tasks, the percentage accuracy and reaction times were chosen as main performance measures. Omissions were treated as incorrect responses. Non-normally distributed data were transformed using square root transformation for accuracy and logarithmic transformation for reaction times.
Additionally, mixed ANOVA models were used to investigate potential differences between the NPE and the PE groups. n-back was analysed with a 3 × 4 mixed ANOVA, with group (control, NPE, and PE) and memory load (0-, 1-, 2-, 3-back) as factors. Fearful faces task was analysed with a 3 × 3 mixed ANOVA, with group (control, NPE, and PE) and fear intensity (neutral, lower fear, and standard fear) as factors.
fMRI data analysis
Preprocessing
Structural and functional data were processed using FSL version 5.0.9 (FMRIB Software Library). Brain extraction was performed on structural images using Brain Extraction Tool (BET). The preprocessing of functional data included motion correction, spatial smoothing with a 6 mm FWHM Gaussian process, high-pass filtering (100 s), and linear registration to a template image (MNI152) using the default two-stage process via the subject's skull-stripped structural image.
General linear model
Statistical modelling of both tasks was performed using a general linear model (GLM) implemented in FSL FEAT. Individual regressors were created for each task condition based on task onsets and a standard gamma function was added to model the haemodynamic response function (s.d. = 3 s, lag = 6 s). Default head motion parameters were included as regressors of no interest. For the n-back task, four primary contrasts were created to explore the effect of each condition alone. Additional contrasts were generated to investigate the impact of increasing task difficulty, i.e. 0-back < 1-back, 0-back < 2-back, 0-back < 3-back.
For the fearful faces task contrasts explored each experimental condition (i.e. neutral, lower fear, and standard fear trials) relative to the null trials. Both fear levels were also individually compared to the neutral condition and against each other.
For both tasks, the first level analyses were combined into group level models. The at-risk group was compared to controls using two-sample t tests. An exploratory analysis was performed to test for potential effects of a PP episode occurrence. F-tests were used to investigate group effects. Additional two-sample t tests were performed to determine the direction of the statistically significant effects. Randomise (Winkler, Ridgway, Webster, Smith, & Nichols, Reference Winkler, Ridgway, Webster, Smith and Nichols2014) with threshold-free cluster enhancement (TFCE) was used for non-parametric permutation-based tests to investigate group differences (with 5000 permutations) and a corrected value of p < 0.05 was considered significant.
Psychophysiological interaction
We conducted psychophysiological interaction (PPI) analyses to examine functional connectivity in both tasks. For the n-back, the left and right DLPFC (l-DLPFC, MNI coordinates [x, y, z]: −44, 18, 22 and r-DLPFC, MNI coordinates: 40, 32, 30) were defined as primary anatomical seed regions based on strong responses to the n-back paradigm (Owen et al., Reference Owen, McMillan, Laird and Bullmore2005). The seeds were defined as 10 mm radius spheres in standard space. These masks were subsequently transformed into each participant's native space using FSL FLIRT and used to extract the time-series for these regions.
For the fearful faces task, left and right amygdalae were chosen as seed regions. Amygdalae were extracted from each participant's anatomical image using FSL command run_first_all. To allow for better warping, functional images were transformed into anatomical space. Those images were used to extract the time-series for left, right, and bilateral amygdala.
First level models included regressors modelling task onsets for each condition, seed region's time-course, and the interaction between the seed region and each task condition. Additionally, regressors of no interest modelling the white matter and cerebrospinal fluid signals were included along with standard head-motion parameters. Contrasts were designed to explore the interaction of the seed region with each task condition and the differences between PPIs of different conditions.
Group comparisons were performed using the GLM and TFCE for statistical inference as described above.
Results
We conducted two sets of analyses for each comparison. First, all women at-risk of PP were compared to controls. Second, to investigate the effects of having a PP episode, the at-risk group was divided into two subgroups, one comprising women who experienced a PP episode following the current pregnancy (PE) and the other consisting of women who did not develop a PP episode following the current pregnancy (NPE). These two groups were compared to each other and to the control group.
Performance results
No differences were observed between the at-risk and control groups, as well as across the NPE, PE, and control groups for both the n-back and the fearful faces task. Details of the results are presented in online Supplementary Materials.
fMRI results
Motion
Independent sample t tests showed no significant differences in motion parameters of both tasks between the controls and the at-risk group [n-back, t(42) = 0.55, p = 0.585; fearful faces, t(42) = 1.96, p = 0.057]. Subsequent ANOVAs were conducted to test differences in motion across the three groups: NPE, PE, and control. No significant effect of motion was found for the n-back task (F 2, 41 = 0.28, p = 0.759); however, for the fearful faces task a significant group effect of motion was observed (F 2, 41 = 3.80, p = 0.031). This result was driven by the higher average motion of the PE group compared to controls (p = 0.027; meancontrol = 0.18, s.d.control = 0.08; meanPE = 0.42, s.d.PE = 0.42).
Brain activation
n-back. Comparison of the at-risk group and controls revealed significantly higher brain activations in the at-risk participants across task conditions of various difficulties. During 0-back and 1-back, the at-risk group showed hyperactivity of the right lateral occipital cortex. A significant increase in brain activity of the PE group relative to controls during 0-back was also observed in a larger bilateral cluster encompassing the cuneus, supracalcarine, and intracalcarine cortices, and lingual gyrus, while the 1-back condition was associated with additional increases in the precuneus, occipital fusiform, supramarginal, inferior lateral occipital, and the occipital pole. The 3-back hyperactivation in the at-risk group was centred around the precuneus and the posterior cingulate. In the PE group, the same hyperactivation was observed with additional increases in the supracalcarine and intracalcarine cortices, and in the lingual gyrus (see online Supplementary Table S3). Importantly, there were no differences for the contrasts of the 1-, 2-, and 3-back with 0-back.
The results observed during high working memory load (i.e. 2-back) remained statistically significant after covarying for age and motion.
Fearful faces. A significant hyperactivation of the cingulate, frontal, and subcortical regions was observed in the at-risk group compared to control when contrasting standard fear stimulus with the null condition. There were no differences for the NPE and PE groups compared to the controls and as such an isolated result it is difficult to interpret and will, therefore, not be discussed further.
This result remained statistically significant after covarying for motion but not when age was included in the model.
Psychophysiological interaction
n-back. Increased r-DLPFC connectivity with a number of regions was observed in the at-risk group relative to controls across the task conditions and these are depicted in Fig. 1 and listed in online Supplementary Table S5. During the 0-back condition there was hyperconnectivity with the left cerebellum, superior temporal, postcentral, and central opercular regions. Further regions were included in the network for 1-back and 2-back (see Fig. 1). Similar hyperconnectivity with the r-DLPFC was observed for the NPE and PE groups compared to controls separately, although the significant clusters were reduced in number. Additionally, the PE group had increased functional connectivity of the r-DLPFC with the right middle temporal gyrus compared to the NPE group during 2-back. Differences between NPE and PE groups remained significant after covarying for daily antipsychotic medication dose and age, and significance was at p = 0.06 after covarying for motion.
Fig. 1. Functional connectivity with the r-DLPFC during the n-back working memory task. (a) Regions showing significantly higher functional connectivity in the at-risk group compared to healthy controls during 0-back (purple), 1-back (cyan), 2-back (blue), and when contrasting 2-back with 0-back (red). (b) Hyperconnectivity in the NPE group compared to controls during 1-back (cyan) and 2-back (blue). (c) Hyperconnectivity in the PE group compared to controls during 1-back (cyan) and 2-back (blue). (d) Greater functional connectivity of the right middle temporal gyrus during 2-back in the PE group compared to NPE. NPE, no postpartum episode; PE, postpartum episode.
When the 2-back was contrasted with the 0-back, hyperconnectivity of r-DLPFC with bilateral cerebellum, occipital, temporal, superior, and middle frontal regions, frontal pole, hippocampus, parahippocampal gyrus, precuneus, and posterior cingulate, as well as left anterior cingulate, precentral and central opercular areas, and right postcentral and superior parietal regions was observed in the at-risk group relative to controls. These results were not observed when comparing the NPE and PE subgroups to controls. The main pattern of differences between the at-risk group and controls remained significant after covarying for age and motion.
Fearful faces. Reduced functional connectivity between the left amygdala and ipsilateral occipital, angular, precentral, postcentral, supramarginal, and parietal regions was observed in the at-risk group compared to controls during lower fear contrasted with the null condition. These results did not reach statistical significance when age and motion were included in the model as covariates. No differences were observed in the analysis of the NPE and PE subgroups.
Discussion
This is the first study to investigate brain activation and functional connectivity related to working memory and emotional face recognition in women at-risk of developing PP. Our main finding is that the at-risk group as a whole, as well as each of the PE and NPE subgroups, showed hyperconnectivity of the r-DLPFC with various, primarily parieto-occipito-temporal and cerebellar regions compared to controls. Importantly, the PE group had increased connectivity between the r-DLPFC and ipsilateral middle temporal gyrus during the 2-back condition compared to the NPE group. Furthermore, the at-risk group as a whole, and the women who developed PP showed hyperactivity of lateral visual areas, precuneus, and posterior cingulate throughout the working memory task compared to controls. During the emotional face recognition task, the at-risk group showed hyperactivation of the cingulo-frontal and subcortical regions compared to controls in response to the standard fear condition. Hypoconnectivity between the left amygdala and ipsilateral occipito-parietal regions was observed in the at-risk group compared to controls during the lower fear condition. No differences between the PE and NPE groups or relative to controls were observed in the analysis of brain activation and functional connectivity during emotional face processing.
Our main finding is of an increased fronto-temporo-parieto-cerebellar functional connectivity during the working memory task in women at-risk of PP compared to healthy controls. While the increased connectivity was present across the 0-, 1-, and 2-back conditions, the 2-back condition showed the most widespread increases. These changes differ from the fronto-parietal (Deserno, Sterzer, Wüstenberg, Heinz, & Schlagenhauf, Reference Deserno, Sterzer, Wüstenberg, Heinz and Schlagenhauf2012; Loeb et al., Reference Loeb, Zhou, Craddock, Shora, Broadnax, Gochman and Liu2018; Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017), fronto-hippocampal (Bähner & Meyer-Lindenberg, Reference Bähner and Meyer-Lindenberg2017), and fronto-occipito-parietal (Kyriakopoulos et al., Reference Kyriakopoulos, Dima, Roiser, Corrigall, Barker and Frangou2012) reductions in connectivity usually observed in non-puerperal psychoses, and from the reduced cortico-limbic connectivity reported during working memory in BD (Cremaschi et al., Reference Cremaschi, Penzo, Palazzo, Dobrea, Cristoffanini, Dell'Osso and Altamura2013; Stegmayer et al., Reference Stegmayer, Usher, Trost, Henseler, Tost, Rietschel and Gruber2015). One study in BD patients showed increased fronto-parietal connectivity during the n-back task, but unlike the current study, their analysis focused only on the left posterior cingulate and medial prefrontal cortex circuit (Wu et al., Reference Wu, Wang, Mwansisya, Pu, Zhang, Liu and Shan2014). Nonetheless, it is worth noting that when our comparison focused only on women at-risk of PP with an existing diagnosis of BD, the increases in connectivity during 2-back relative to healthy controls remained present (see online Supplementary Materials). While task performance in schizophrenia may modulate connectivity, with only poor performers showing reduced fronto-cortical connectivity (Quidé, Morris, Shepherd, Rowland, & Green, Reference Quidé, Morris, Shepherd, Rowland and Green2013; Wu et al., Reference Wu, Wang, Chen, Zou, Huang, Li and Wang2017), there is evidence of reduced fronto-posterior connectivity in the absence of performance deficits (Nielsen et al., Reference Nielsen, Madsen, Wang, Liu, Friston and Zhou2017). Consequently, our results of increases in fronto-temporo-parieto-cerebellar connectivity during working memory performance in PP relative to controls, in the absence of performance impairment, provide initial evidence that neural abnormalities in PP may be of a different nature compared to those seen in non-puerperal disorders.
In this study, women who developed a PP episode had specific increases in the connectivity between the r-DLPFC and ipsilateral middle temporal gyrus compared to at-risk women who did not develop PP. The middle temporal gyrus has been shown to be reduced in volume in patients with non-puerperal psychoses (Kuroki et al., Reference Kuroki, Shenton, Salisbury, Hirayasu, Onitsuka, Ersner-Hershfield and McCarley2006; Onitsuka et al., Reference Onitsuka, Shenton, Salisbury, Dickey, Kasai, Toner and McCarley2004), and more notably, in individuals at high risk of psychosis unrelated to childbirth (Fusar-Poli et al., Reference Fusar-Poli, Borgwardt, Crescini, Deste, Kempton, Lawrie and Sacchetti2011). Furthermore, structural and functional abnormalities of other temporal regions have been reported in BD (Hanford, Nazarov, Hall, & Sassi, Reference Hanford, Nazarov, Hall and Sassi2016; Phillips & Swartz, Reference Phillips and Swartz2014). While no structural changes of the right middle temporal lobe were observed in women with PP so far (Fusté et al., Reference Fusté, Pauls, Worker, Reinders, Simmons, Williams and Dazzan2017; Lanczik et al., Reference Lanczik, Fritze, Hofmann, Schulz, Knoche and Becker1998), functionally, the temporal lobe is involved in several cognitive processes in healthy controls (Cabeza & Nyberg, Reference Cabeza and Nyberg2000). Studies in schizophrenia further suggest that patients may produce compensatory hyperactivation of temporal areas during high demand tasks to achieve performance comparable to that of healthy controls (Kim et al., Reference Kim, Tura, Potkin, Fallon, Manoach, Calhoun VD and Turner2010; Ragland et al., Reference Ragland, Moelter, Bhati, Valdez, Kohler, Siegel and Gur2008). Consequently, increased fronto-temporal connectivity in women who develop a PP episode might represent a compensatory brain mechanism required for PP patients to perform at the same level as women at-risk of PP who remain well and as healthy controls. Despite this potential explanation for our findings, it is not straightforward to directly compare our results with studies in psychoses unrelated to childbirth for two reasons. First, our group was tested outside of the most acute period of psychosis and second, few studies examine or demonstrate connectivity with frontal and temporal lobe regions during task phases. However, PFC–hippocampal dysconnectivity is a replicated finding, with some authors suggesting it could serve as a biomarker in schizophrenia (Bähner & Meyer-Lindenberg, Reference Bähner and Meyer-Lindenberg2017).
The observed hyperactivity in posterior cortical areas during the n-back task was unexpected and, notably, limited to the task conditions compared to the implicit baseline. We recognise the difficulty in defining a baseline state in the brain. Here, we are using the ‘implicit baseline’ which refers to the modelling approach rather than assuming an actual baseline. The contrast of the working memory conditions with the visuo-motor control condition (0-back) showed no difference for any group comparison suggesting that general attentional processes were over-recruited during the task, not specific to the working memory load. The signal during individual conditions is rarely examined in other studies, so it is difficult to compare our findings with those of patients with non-puerperal psychoses and BD. Nevertheless, this finding is unlike the primarily frontal hypoactivation seen in non-puerperal psychoses and BD, that is more typical for the working memory contrasts (Chen et al., Reference Chen, Suckling, Lennox, Ooi and Bullmore2011; Henseler et al., Reference Henseler, Falkai and Gruber2009; Lagopoulos et al., Reference Lagopoulos, Ivanovski and Malhi2007; Minzenberg et al., Reference Minzenberg, Laird, Thelen, Carter and Glahn2009; Schneider et al., Reference Schneider, Habel, Reske, Kellermann, Stöcker, Shah and Gaebel2007; Thermenos et al., Reference Thermenos, Goldstein, Milanovic, Whitfield-Gabrieli, Makris, Laviolette and Seidman2010).
During emotional face recognition, women at-risk of PP showed hypoconnectivity between the left amygdala and ipsilateral occipital, angular, precentral, postcentral, supramarginal, and parietal regions relative to controls during lower fear contrasted with the null condition. This result could not be attributed to PP specifically, as the comparison of women with an existing diagnosis of BD in our at-risk of PP sample to controls showed a similar difference (see online Supplementary Materials). In addition, the findings did not remain significant after covarying for age. This emphasises the importance of recruiting a larger sample of patients and sufficiently large numbers of participants with PE who do not have a diagnosis of BD to understand the unique contribution of PP to emotional processing.
A limitation of this study is the relatively small sample size, which has limited the power of the NPE and PE subgroups comparisons. Furthermore, while we included BD as one of our inclusion criteria based on the extremely high incidence of PP in BD (Jones et al., Reference Jones, Chandra, Dazzan and Howard2014; Jones & Craddock, Reference Jones and Craddock2001; Sit et al., Reference Sit, Rothschild and Wisner2006), the majority of women with BD in this sample did not develop a PP episode in the current pregnancy (although some of them experienced PP in the past and might develop PP in future pregnancies). This highlights the challenge of accurately defining the risk groups in these studies. Furthermore, women in the at-risk group were not medication-naïve, with approximately three-quarters of them taking medication at the time of the scan. This is an important factor to consider as antipsychotic drugs have been shown to affect brain structure (Vita, De Peri, Deste, Barlati, & Sacchetti, Reference Vita, De Peri, Deste, Barlati and Sacchetti2015). While a meta-analysis suggests no common effects of antipsychotics on functional neuroimaging (Röder, Dieleman, van der Veen, & Linden, Reference Röder, Dieleman, van der Veen and Linden2013), we cannot fully exclude the possibility that drug treatment could have influenced our results. We have, nevertheless, controlled for daily antipsychotic use in the comparison of NPE and PE groups and observed no effects of antipsychotic medication on the differences in brain connectivity between those groups, suggesting that medication was unlikely to explain the observed differences. Nonetheless, as in many other studies in BD and psychoses unrelated to childbirth, it is difficult to assess and control completely for the potential effects of medications, because of the variable length and heterogeneity of exposure, and the variability in dose (Chen et al., Reference Chen, Suckling, Lennox, Ooi and Bullmore2011). In the current sample, while we had concurrent medication details, we lacked full medication history for all participants.
Another important limitation lies in the differential time between delivery and fMRI acquisition in the three groups, with PE women having the largest interval, as they needed time to recover prior to undergoing the scan. While there is recent evidence suggesting that during the first weeks after delivery women's pattern of brain activity undergoes significant changes (Gingnell et al., Reference Gingnell, Bannbers, Moes, Engman, Sylvén, Skalkidou and Sundström-Poromaa2015), it remains unclear how and whether these changes progress past 6-weeks postpartum. Although some effects of PE compared to controls overlapped with results from the NPE group compared to controls (who did not differ in time since delivery), we cannot exclude the possibility that such differences could have affected our results. Consequently, future studies should employ longitudinal designs to investigate functional brain changes related to working memory, emotional processing, and other cognitive domains in PP throughout early motherhood.
In conclusion, we provide initial evidence of differential brain activation and functional connectivity, in the absence of cognitive impairment, in women at-risk of PP compared to healthy postpartum women. These results require replication and extension into other cognitive domains but suggest that functional abnormalities in PP differ from those present in non-puerperal psychosis. Consequently, determining these distinct functional abnormalities might, in the future, help to identify women who are at a particularly high risk of developing PP. Furthermore, more in-depth understanding of these processes may help in guiding and developing targeted interventions, thus lessening the burden of PP.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291720000471.
Acknowledgements
The research was supported by a NARSAD Independent Investigator Award to Dr P. Dazzan. The research was also in part financially supported by the UK Medical Research Council, Medical Research Foundation, the Psychiatry Research Trust and the National Institute for Health Research (NIHR) Mental Health Biomedical Research Centre (BRC) at South London and Maudsley NHS Foundation Trust and King's College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. We thank the Perinatal Research Team and most of all the women who shared their early lives as mothers with us.
