Altered task-dependent functional connectivity patterns during subjective recollection experiences of episodic retrieval in postpartum women

Highlights

• We investigated recognition memory processes in postpartum women (PP).

• PP endorsed old items with “Remember” responses less often than control women.

• PP showed decreased neural activity across the subjective recollection network.

• We applied iterative functional connectivity multivariate pattern analysis.

• Large-scale functional connectivity patterns across the network were altered in PP.

Abstract

Numerous studies have suggested that postpartum women show a decline in cognitive abilities. However, to date, no study has investigated the presence of qualitative alterations in recognition memory processes in postpartum women that may lead to a decline in cognitive ability. To address this issue, we employed the Remember/Know procedure and functional magnetic resonance imaging (fMRI). Behavioral results demonstrated that compared with the matched control (CTRL) group, the postpartum (PP) group endorsed “Remember” less and “Know” more to old items. A univariate analysis of fMRI data indicated lower neural activity of the subjective recollection network in the PP group than in the CTRL group. We also performed a large-scale functional connectivity multivariate pattern analysis (fcMVPA) using task-dependent time-series to detect differences in functional connectivity patterns and neural interactivity between the PP and CTRL groups. The fcMVPA results revealed that the PP group exhibited altered functional connectivity patterns from which machine learning algorithms could discriminate group membership with 94% accuracy. Collectively, these findings demonstrated that altered subjective recollection processes in the PP group during episodic memory decisions are associated with diminished neural activity and abnormal interactivity across the subjective recollection network. We believe that this is one of the first studies demonstrating qualitative alterations in recognition memory processes in postpartum women.

Introduction

Many pregnant or postpartum women have reported experiencing discomfort in the functioning of their daily lives and a subjective decline in cognitive abilities (Brett and Baxendale, 2001, Crawley et al., 2003, Logan et al., 2014, Parsons and Redman, 1991). This phenomenon, colloquially termed as “momnesia” or “pregnant-brain,” may be a source of distress among pregnant/postpartum women. Indeed, several studies have reported objective changes in the cognitive abilities of pregnant/postpartum women in broad cognitive domains, such as visuo-spatial memory (Piccardi et al., 2014), prospective memory (Rendell & Henry, 2008), and attentional processes (Stark, 2000). In addition, pregnancy-related affective changes, particularly depressive symptoms, and cognitive impairment associated with these symptoms have been consistently reported (de Almeida et al., 2012, Josefsson et al., 2001). Although it has been assumed that psychophysiological factors may influence cognitive and affective changes, including changes in ovarian hormone levels (Glynn, 2010, Workman et al., 2012) and stressful circumstances (Groer et al., 2002, Parcells, 2010), in pregnant/postpartum women, how the alteration of underlying neural mechanisms affects one’s behavioral responses is yet to be investigated.

Among the cognitive domains affected in the pregnancy/postpartum period, memory impairments, particularly deficits in episodic memory, which involves storage of autobiographical events of a person’s life, can create severe discomfort in one’s daily life (see Tulving, 2002, for a review). Although numerous studies have investigated objective memory impairment in pregnant/postpartum women using multiple approaches, these studies have yielded inconsistent results. For example, several studies have suggested that pregnant/postpartum women have memory declines by assessing their memory abilities in a laboratory environment (Brindle et al., 1991, de Groot et al., 2003, de Groot et al., 2006, Glynn, 2010, Keenan et al., 1998, Sharp et al., 1993, Wilson et al., 2011); however, there is evidence that the memory decline in pregnant/postpartum women can be better detected using naturalistic measures rather than using laboratory measures (Cuttler et al., 2011, Rendell and Henry, 2008). Recently, Henry and Rendell (2007) conducted a meta-analysis of 14 studies in which the memory functions of pregnant/postpartum women were compared with those of controls. In their review, they have suggested that inconsistent results of memory decline in pregnant/postpartum women were due to differing measures, methodologies, and the small sample sizes of individual studies. Nonetheless, the authors have suggested that impaired memory performance in pregnant/postpartum women is clearly observed in tasks with relatively high cognitive demand, such as the free recall paradigm.

In contrast, other studies have reported no memory decline in pregnant/postpartum women (Casey, 2000, Casey et al., 1999, Crawley et al., 2003, Logan et al., 2014). Indeed, intact performance of recognition memory in pregnant/postpartum women compared with that in control women is a more consistent finding. Brindle et al. (1991) have found that recognition memory does not differ statistically between pregnant and control women, and similar results have been reported by Sharp et al. (1993). More recently, Mickes, Wixted, Shapiro, and Scarff (2009) directly compared recognition performance (d′) between pregnant and postpartum women and found that both groups showed near optimal performance, suggesting a ceiling effect. Although multiple studies have found intact recognition performance in pregnant/postpartum women, the qualitative aspects of retrieved memory have not been thoroughly investigated. Indeed, considering the subjective discomfort due to broad cognitive dysfunction reported by pregnant/postpartum women, study items correctly recognized by pregnant/postpartum women may yield qualitatively less specific details of episodic memory compared with those correctly recognized by control women.

According to dual-process models of recognition memory, recognition judgments are based on two distinct processes, recollection and familiarity (Atkinson and Juola, 1974, Gardiner, 1988, Jacoby, 1991, Mandler, 1980, Yonelinas, 2002, Yonelinas and Jacoby, 1995). Recollection refers to one’s ability to retrieve specific details of a past event, whereas familiarity refers to the feeling of knowing that an event has been experienced before but lacks any mental reinstatement or detail. One of the most widely used paradigms for assessing these two processes is the “Remember/Know (R/K)” procedure (Tulving, 1985), which measures subjective recognition judgments of participants. Accordingly, participants are required to make a subjective judgment as to whether they consciously recollect specific details of an item presented in a study run (“Remember”) or whether they are only familiar (“Know”) with it in the absence of specific information on the study run episode (Henson et al., 1999, Migo et al., 2012, Rajaram, 1993, Yonelinas, 2001, Yonelinas and Jacoby, 1995). Studies on the underlying neural substrates have consistently reported that recollection and familiarity evoke distinct activation patterns across different brain regions (Daselaar et al., 2006, Evans and Wilding, 2012, Henson et al., 2005, Henson et al., 1999, Montaldi et al., 2006, Vilberg and Rugg, 2007, Woodruff et al., 2006, Yonelinas et al., 2005). In addition, the contrast imaging of Remember-hit versus Know-hit responses, the so-called “subjective recollection effect,” yields core recollection-related brain regions, including the medial prefrontal cortex, hippocampus, parahippocampal gyrus, precuneus, and left ventral posterior parietal cortex (Kim, 2010, Spaniol et al., 2009, for reviews). Since participants are required to subjectively decide qualitative aspects of their memory in the R/K procedure, it has been suggested that recollection success effect observed from this paradigm reflects the neural substrates underlying one’s introspective judgments. Therefore, this effect is distinct from the objective recollection effect assessed by the source memory paradigm (see Spaniol et al., 2009, for a review).

The R/K procedure has been widely used to investigate altered recollection processes in various clinical groups, such as patients with chronic schizophrenia (van Erp et al., 2008), amnestic mild cognitive impairment and probable Alzheimer’s disease (Hudon, Belleville, & Gauthier, 2009), and temporal lobe damage (Yonelinas et al., 2002). Moreover, the R/K procedure has also been utilized in non-clinical groups, such as aging populations (Duarte et al., 2008, Parks et al., 2010, Wang et al., 2016), to find qualitative differences in recollected experiences, which are difficult to detect with simple recognition tasks. To date, however, neither behavioral nor neuroimaging studies have been performed by applying the R/K procedure to investigate the qualitative differences in recognition memory processes between postpartum and control women.

Recently, substantial research interest has focused on using functional connectivity patterns across the entire brain to decode brain activity. In particular, multivariate pattern analysis using machine learning algorithms has been successfully applied to resting-state functional magnetic resonance imaging (fMRI) data for discriminating clinical populations, such as patients with mild cognitive impairment (Chen et al., 2011), schizophrenia (Shen, Wang, Liu, & Hu, 2010), and depression (Craddock et al., 2009, Zeng et al., 2012), from normal healthy groups. The classification of non-clinical populations based on their intrinsic functional connectivity has also been conducted to discriminate age-related brain status (Dosenbach et al., 2010, Meier et al., 2012), sex-associated differences in cognition (Satterthwaite et al., 2015), and different cognitive states of individuals (Shirer, Ryali, Rykhlevskaia, Menon, & Greicius, 2012). However, the functional connectivity of large-scale brain network in postpartum women, especially during episodic retrieval decisions, has not yet been investigated. Therefore, a multivariate pattern approach using task-related fMRI data (i.e., data obtained from the test run of the R/K procedure) would be useful for advancing our understanding of altered functional connectivity patterns during subjective recollection processes in postpartum women compared with those in control women.

In the present study, we utilized the R/K procedure combined with fMRI to investigate the presence of qualitative alterations in recognition memory processes between postpartum women (the PP group) and matched controls (the CTRL group) at the behavioral and neural levels. During the test run of the R/K procedure, participants were required to make subjective recognition judgments (“Remember,” “Know,” or “New”) when presented word items (previously presented or new) by introspectively searching their memory traces. We hypothesized that although overall recognition performance of the PP group may be equivalent to that of the CTRL group as reported previously (Brindle et al., 1991, Mickes et al., 2009, Sharp et al., 1993), the PP group would fail to retrieve qualitative contents for encoded items, responding to old items as “Remember” relatively less frequently and “Know” more frequently than the CTRL group. Furthermore, we predicted that altered recollective experiences of the PP group would be associated with altered neural connectivity, particularly within the subjective recollection network, including the medial prefrontal cortex, hippocampus, parahippocampal gyrus, precuneus, and left ventral posterior parietal cortex (Kim, 2010, Spaniol et al., 2009). In addition to investigating differences in qualitative recollection between groups, we further assessed how different brain regions within the subjective recollection network differentially interact during recollection processes. For this purpose, we applied large-scale functional connectivity multivariate pattern analysis (fcMVPA) for group classification using task-dependent time-series data extracted in a voxel-wise manner from recollection-related regions. Motivated by recently established novel approaches for finding the most discriminating network for group classification (Pantazatos et al., 2012a, Pantazatos et al., 2012b, Pantazatos et al., 2014), we iterated the group classification process, based on machine learning algorithms, as a function of the number of input data elements (i.e., features) included to detect any abnormal changes in functional connectivity in the PP group.