Longitudinal neuroanatomical increases from early to one‑year postpartum

Abstract

Preclinical studies have provided causal evidence that the postpartum period involves regional neuroanatomical changes in ‘maternal’ brain regions to support the transition to offspring caregiving. Few studies, in humans, have examined neuroanatomical changes from early to one-year postpartum with longitudinal neuroimaging data and their association with postpartum mood changes. In the present study, we examined longitudinal changes in surface morphometry (cortical thickness and surface area) in regions previously implicated in the transition to parenthood. We also examined longitudinal volumetric neuroanatomical changes in three subcortical regions of the maternal brain: the hippocampus, amygdala, and ventral diencephalon. Twenty-four participants underwent longitudinal structural magnetic resonance imaging at 1–4 weeks and 1 year postpartum. Cortical thickness increased from early to one-year postpartum in the left (p = .003, Bonferroni corrected) and right (p = .02, Bonferroni corrected) superior frontal gyrus. No significant increases (or decreases) were observed in these regions for surface area. Volumetric increases, across the postpartum period, were found in the left amygdala (p = .001, Bonferroni corrected) and right ventral diencephalon (p = .01, Bonferroni corrected). An exploratory analysis of depressive symptoms found reductions in depressive symptoms from early postpartum to one-year postpartum were associated with greater cortical thickness in the superior frontal gyrus for both the left (p = .02) and right (p = .02) hemispheres. The findings expand our evidence of the neuroanatomical changes that occur across the postpartum period in humans and motivate future studies to examine how mood changes across this period are associated with cortical thickness of the superior frontal gyrus.

Introduction

Evidence from both animal and human studies suggest that neuroanatomical changes occur across the postpartum period to support the demands of parenthood (Kim et al. 2016; Kim et al. 2010; Luders et al. 2022; Barrière et al. 2021; Barba-Müller et al. 2019). Several studies have used structural magnetic resonance imaging (MRI) to quantify these neuroanatomical changes; these studies have primarily focused on neuroanatomical changes across pregnancy and into the early postpartum period (Barba-Müller et al. 2019; Nehls et al. 2024; Martínez-García et al. 2021a; Carmona et al. 2019; Hoekzema et al. 2017; Paternina-Die 2024; Servin-Barthet et al. 2023). However, there is evidence of long-lasting neuroanatomical changes occurring in the postpartum period in response to the initiation and maintenance of the parent–off-spring relationship (Barba-Müller et al. 2019; Hoekzema et al. 2022; Orchard et al. 2020). This evidence suggests that neuroanatomical changes across pregnancy may be hormonally driven and primarily result in decreases in brain structural metrics, while changes across the postpartum may be driven by and to support parent–offspring interactions via experience-dependent plasticity and result in increases (Luders et al. 2022; Barba-Müller et al. 2019). In the present study, we examined the longitudinal neuroanatomical changes from early to one-year postpartum using structural MRI. We focus on changes from early to one-year postpartum given the first year postpartum is when birthing parents are most vulnerable to psychopathology (Serretti et al. 2006; Florio et al. 2014) and it is during this time that long-term parent–child emotional bonds are established (Takács et al. 2020; Kinsey et al. 2014). Regarding postpartum mood, few studies that have examined perinatal longitudinal brain changes, especially in the postpartum, have examined how brain structure is associated with changes in mood symptoms across the postpartum period. Here, we conduct an exploratory analysis of the association between the change in depressive symptoms from early to one-year postpartum and brain structure at one-year postpartum.

Human studies of the neuroanatomical changes across the perinatal period have examined both gray matter volume and surface morphometry metrics (primarily cortical thickness and surface area) from structural MRI (Luders et al. 2022). Gray matter volume is the product of cortical thickness (CT) and surface area (SA) (Fischl and Dale 2000); however, it is critical to examine CT and SA in addition to gray matter volume as these measures have unique developmental trajectories (Wierenga et al. 2014), associations with behavior (Winkler et al. 2010), and environmental/genetic contributions (Jha et al. 2019; Panizzon et al. 2009). Human studies have examined changes in cortical volume, CT, and SA by comparing these measures between pregnant/postpartum participants and nulliparous ‘control’ participants. One study examined brain structural changes in 110 pregnant participants from late pregnancy (mean of 36.23 weeks) to the early postpartum (mean of 22 days) and compared the changes to nulliparous participants (Paternina-Die 2024). At the postpartum session, the pregnancy/postpartum group had significantly lower gray matter volume across all the brain networks examined. Changes from late pregnancy to early postpartum primarily occurred for gray matter volume in the posterior cingulate, precuneus, paracentral gyrus, precentral gyri, supramarginal gyri, and superior temporal gyrus. These increases were also observed for CT and SA but to a lesser degree. The greater percentage of postpartum time between the two sessions was associated with greater increases in cortical volume, CT, and SA.

Using a cross-sectional study, we previously found a correlation between postpartum timing and CT for primiparous individuals (Kim et al. 2018). The study found a correlation such that individuals who had been a parent longer (further into the postpartum period or greater age of the child) had higher CT in the superior frontal, caudal middle frontal, lateral occipital, and precentral gyri. These findings of positive correlations between months postpartum (child age) and CT in the postpartum may at first seem contradictory with several studies finding regional cortical thickness decreases of the medial frontal cortex, precuneus, posterior cingulate, inferior frontal gyri, and superior temporal sulci during pregnancy. However, these regions having little anatomical overlap and the underlying processes for pregnancy-related decreases and postpartum-related increases are hypothesized to be unique (Luders et al. 2022; Martínez-García et al. 2021b).

Regarding longitudinal brain volumetric changes in the postpartum period in humans, gray matter volume increases have been reported globally (Oatridge et al. 2002; Luders et al. 2020), regionally (Martínez-García et al. 2021a; Luders 2021), and locally (Kim et al. 2010; Lisofsky et al. 2019). In one study, gray matter volume at 2–4 weeks was compared to volume at 3–4 months postpartum. Longitudinal gray matter volume increases were observed in subcortical regions including the hypothalamus, caudate, substantia nigra, putamen, globus pallidus, mamillary body, and thalamus (Kim et al. 2010). The increases in the hypothalamus, substantia nigra, and amygdala were associated with positive perceptions of the participant’s infant. Cortical volume increases were also found in the medial frontal, postcentral, cingulate, inferior frontal, precentral, middle frontal, superior temporal, and parahippocampal gyri, in addition to the insula and cerebellum. Luders et al., found voxel-wise gray matter volume increases from 1–2 days postpartum to 4–6 weeks postpartum. The regions found were widespread and included both cortical (pre- and postcentral gyri, inferior frontal gyrus, and frontal operculum) and subcortical (thalamus and caudate) regions (Luders et al. 2020). Using the same dataset, longitudinal increases were found for gray matter volume in the amygdala, particularly in the superficial subregion of the amygdala (Luders et al. 2021). The superficial subregion of the amygdala has particularly strong structural connectivity with other regions of the limbic system, has direct sensory inputs including from the olfactory system, and is involved in responsive and expressive affective processing (Luders et al. 2021; Yilmazer-Hanke et al. 2016; Lei et al. 2015). Another study with more dense longitudinal sampling examined gray matter volume changes for 3 week intervals for 12 postpartum weeks starting at 1 week postpartum (Nehls et al. 2024). Changes were also compared to a control group of 20 nulliparous age-matched controls. The study found the greatest gray matter volume increases occurred in the first 3 weeks postpartum for the bilateral parietal, temporal, and occipital cortices and insula. Comparing 9-week postpartum gray matter volume to 6 weeks postpartum, increases were found for the bilateral superior medial gyri, left middle orbital gyrus, left anterior cingulate gyrus, right amygdala, and right hippocampus. A data-driven classification analysis, classifying participants as being members of the postpartum versus nulliparous group, indicated left amygdala volume had the greatest predictive power. Across these studies, the evidence suggests that several subcortical and cortical regions increase in volume in the early postpartum; however, little is known about changes beyond the first months postpartum.

Non-human animal studies have also provided evidence of associations between depressive-like behaviors and brain structural changes across the postpartum (Barba-Müller et al. 2019; Servin-Barthet et al. 2023; Haim et al. 2014; Mir et al. 2022). Understanding the associations between postpartum mood and postpartum brain changes is critical as one out of eight individuals experience elevated depressive symptoms in the postpartum period (Cox et al. 1993). However, in humans, the neurobiological mechanisms underlying postpartum depression remain elusive. A meta-analysis of neuroimaging studies of postpartum depression conducted in 2015 did not find any studies reporting associations between postpartum depression and brain structure (Fiorelli et al. 2015). A study of 157 euthymic participants underwent neuroimaging 6 days after birth with a 12 week follow up (Schnakenberg et al. 2021). At birth, no structural brain differences were observed between groups later identified (at 12 weeks) as meeting diagnostic criteria for postpartum depression or adjustment disorder in comparison with to a group without postpartum mood disorders. Regarding associations between mood symptoms (more broadly) and neuroanatomical change in the postpartum period, Luders et al., found that increases in the right superficial subregion of the amygdala from 1–2 days after childbirth to 4–6 weeks after childbirth, were associated with decreases in state anxiety symptoms (Luders et al. 2021). As mentioned, the superficial subregion of the amygdala has been hypothesized to be involved in both receptive and expressive affective processing in the postpartum period (Luders et al. 2021). Luders et al., provides some preliminary evidence of associations between postpartum neuroanatomical changes and postpartum mood, but whether changes in mood symptoms, particularly depressive symptoms, across the postpartum period are prospectively associated with postpartum brain structure is unclear. Examining these associations is critical as postpartum depressive symptoms may impact the neuroanatomical changes that support parenting as evidence suggests higher postpartum depressive symptoms are associated with reduced emotion regulation of the parent (Cardoso and Fonseca 2023; Marques et al. 2018) and lower parental sensitivity (Brummelte and Galea 2016; Bernard et al. 2018).

The present study aims to further characterize neuroanatomical changes (for both surface morphometry and subcortical gray matter volume) across the postpartum period. To expand on previous longitudinal neuroimaging studies, we conducted the first MRI session in the early postpartum period (between 0.03 and 0.46 months postpartum, Early Postpartum) and approximately one year later (One-Year Postpartum). For Kim et al., the second scan timepoint was 3–4 months postpartum, therefore longitudinal neuroanatomical change across the first year postpartum is unclear. Further, few studies have examined longitudinal changes in CT and SA from early to one-year postpartum. Based upon the regions in which the CT of primiparous mothers was associated with postpartum months for the first 6 months in the previous cross-sectional study (Kim et al. 2018), we hypothesized longitudinal increases of CT from early to one-year postpartum in the superior frontal gyrus, lateral occipital gyrus, caudal middle frontal gyrus, and precentral gyrus. As SA across the postpartum period is unclear, we analyzed SA only in the regions examined for the CT analysis. Further, we conducted a post-hoc test in which we examined the neuroanatomical change in regions that have been previously found to decrease in CT across pregnancy including the right middle temporal gyrus, left superior temporal sulcus, right posterior cingulate cortex, bilateral fusiform gyrus, left lingual gyrus, left inferior parietal cortex, and right lateral occipital cortex (Hoekzema et al. 2022).

Regarding subcortical volume, we hypothesized, based upon the previous studies (Luders et al. 2022; Barba-Müller et al. 2019; Servin-Barthet et al. 2023; Pawluski et al. 2022), that gray matter volume of the hippocampus, amygdala, and ventral diencephalon will increase from the early to one-year postpartum period. The amygdala and ventral diencephalon were chosen as regions of interest due to increases in gray matter volume found in these regions from 2–4 weeks postpartum to 3–4 months postpartum in a human study (Kim et al. 2010). Additional justification comes from additional studies finding increases in the amygdala in the early postpartum (Nehls et al. 2024; Luders et al. 2021) and studies elucidating the central role of regions of the ventral diencephalon suggesting the hypothalamus (particularly the medial preoptic area) to be an integral component of the neural bases of parenting behavior (Numan et al. 1977; Numan 1974; Lonstein et al. 2015; Dobolyi et al. 2014). While the hippocampus was not found to increase in gray matter volume in Kim et al., 2010, the hippocampus has also been found to undergo cellular changes in the postpartum period in animal studies (Servin-Barthet et al. 2023; Pawluski and Galea 2007, 2006) and be the only region in which gray matter volume decreases associated with pregnancy remained at 2 years post-pregnancy in humans (Hoekzema et al. 2022). The analysis focused on the gray matter volume of the ventral diencephalon as longitudinal automatic segmentations of the hypothalamus have yet to be developed (Billot 2020). However, Freesurfer’s ‘recon-all’ pipeline provides a segmentation referred to as the ventral diencephalon which includes the hypothalamus with mamillary body, subthalamic, lateral geniculate, medial geniculate and red nuclei, substantia nigra and the surrounding white matter (Fischl et al. 2002). We did not include any subgenual or prefrontal regions in the gray matter volume analysis as these regions are more appropriately measured with surface morphometry (Essen et al. 1998). These regions were not included as surface morphometric regions of interest as they were not found to be positively correlated with postpartum months (child age) in our previous study (Kim et al. 2010). Lastly, we conducted an exploratory analysis to examine if changes in depressive symptoms from early to one-year postpartum were associated with brain structural metrics (CT for cortical regions and volume for subcortical regions) at one-year postpartum.

Materials and methods

Participants

Participants were recruited during pregnancy from the Department of Obstetrics and Gynecology at hospitals in the Denver metro area. Eligibility criteria were: 1) over 18 years of age; 2) singleton intrauterine pregnancy; 3) prior to 16 weeks of gestation; 4) fluency in English; and 5) a family income-to-needs ratio below 8 (based upon income information gathered from an initial phone screening with participants). Family income-to-needs ratio is calculated by dividing the total family income by the poverty threshold adjusted for the number of individuals living in the household as specified by the United States Census Bureau. It is a precise measure of family socioeconomic status as it adjusts for the number of people living in a household. Exclusion criteria include: 1) current psychotropic medication use; 2) current or lifetime psychiatric/neurological illness other than depression and anxiety diagnosis; 3) maternal substance use during pregnancy except for occasional use of alcohol, cigarettes, or cannabis (assessed using maternal reports and urine toxicology, “occasional use” is defined as who use more than 10 cigarettes per day or drink more than 1 glass of alcohol per day); 4) obstetric risk conditions such as systemic maternal disease, placental or cord abnormalities, uterine anomalies, infection, chromosomal abnormalities; 5) corticosteroid medication usage during their pregnancy; or 6) nonremovable ferromagnetic metal in or on the body (for safety in the magnetic resonance imaging MRI scanner). As recruitment aimed at a representative community sample, individuals with a history of depression and/or anxiety disorders were included as these are the two most common mental disorder diagnoses of the perinatal period (Meltzer-Brody and Rubinow 2021; Shorey et al. 2018; Dennis et al. 2017). As this data was taken from a larger study focused on stress exposure in the postpartum period, the sample was primarily comprised of low- and middle-income participants and excluded participants that were currently experiencing high-income.

Procedures

Data for the present study was taken from a larger study examining stress exposure across the perinatal period. Participants in the study participated in 7 research visits across the perinatal period, with 5 home visits (12–16, 22, 32 weeks of pregnancy, in the first 1–4 weeks postpartum, and approximately one-year postpartum) and 2 neuroimaging visits (denoted at Early Postpartum and One-Year Postpartum) at 1–4 weeks postpartum (mean months postpartum 0.14, SD = 0.09, range = 0.0–0.46 weeks) and approximately one-year postpartum (mean months postpartum was 13.55, SD = 0.99, range = 12.29–15.61 months). In response to the COVID-19 pandemic, home visits occurring after 03/13/2020 transitioned to remote visits conducted via video conferencing or lab visits based upon university regulations at the time of the visit. At the first home visit, participants were given information about the study and provide informed consent. The neuroimaging visits occurred at the Center for Innovation & Creativity at the University of Colorado – Boulder. Childcare support was provided to each family in addition to monetary compensation for the participant’s time at the end of each visit. All procedures were approved by the University of Denver Institutional Review Board.

Edinburgh Postnatal Depression Scale (EPDS)

The EPDS (Cox et al. 1993) was administered to participants at the home visit that occurred at 2–4 weeks postpartum and again at the home visit that occurred at approximately one-year postpartum. The EPDS is a 10-item screening questionnaire used to examine postpartum depressive symptoms in both research and clinical settings. For each item, the respondent chooses from a range from 0 “not at all” to 3 “yes, most of the time/as much as I always could”. Higher total scores indicated higher depressive symptoms and scores above 10 may indicate ‘elevated’ depressive symptoms (Murray and Carothers 1990; El-Hachem 2014). Three participants were missing EPDS data for the study; this reduced the sample size for the exploratory analysis to 21 participants.

Parity

There is evidence from animal studies of associations between parity and brain structure in the postpartum period (Martínez-García et al. 2021b). Regarding parity in humans, of participants that were scanned at 1–4 days postpartum, those with multiple births were found to have lower gray matter volume across several regions of the brain including the cerebellum, fusiform gyrus, basal ganglia, hippocampus, and amygdala compared to nulliparous participants (Chechko et al. 2022). No significant differences were found when contrasting multiparous and primiparous participants. For cortical thickness, multiparous participants had less cortical thickness compared to primiparous participants in the superior parietal gyrus and inferior parietal sulcus. Another human study also found associations between parity and ‘brain age’ which measures the aging levels of an individual’s brain in comparison to their peers (Lange et al. 2019; Lange et al. 2020). Therefore, upon study entry, parity was measured by asking each participant how many live births they have had before the current pregnancy.

Anatomical MRI acquisition

All MRI data were collected using a Siemens 3T MAGNETOM Prisma scanner with a 32-channel head coil. High-resolution, T1-weighted structural images (3D magnetization-prepared rapid acquisition gradient-recalled echo sequence, MP-RAGE) were acquired for both timepoints (early and one-year postpartum) with the following parameters: TR = 2400 ms, TE = 2.22 ms, inversion time = 1000 ms, voxel size = 0.8 × 0.8 × 0.8 mm.

Anatomical MRI quality control procedure

The anatomical MRI quality control procedure (QC) used a combined approach that involved both visual and image-derived QC steps. First, each T1-weighted image was visually inspected, slice-by-slice, and assigned a rating from 1 to 4 based upon its overall quality (Blumenthal et al. 2002). Based on the visual QC step, all images passed and were included in the image-derived QC procedure. For the image-derived QC, T1-weighted images were processed with the anatomical workflow in MRIQC (Esteban et al. 2017). This workflow calculates image quality metrics (IQMs) at the participant level and provides group-level summaries and plots of the IQMs across the sample. Based upon both image coefficient of join variation (cjv) (Ganzetti et al. 2016) and contrast-to-noise ratio (cnr) (Magnotta et al. 2006), no anatomical image was determined to be an outlier in terms of these metrics.

Anatomical MRI analysis

Results included in this manuscript come from preprocessing performed using sMRIPprep 0.12.0 (Esteban et al. 2021) (RRID:SCR_016216), which is based on Nipype 1.8.6 (Gorgolewski et al. 2011). Each T1-weighted (T1w) image was corrected for intensity non-uniformity (INU) with N4BiasFieldCorrection (Tustison et al. 2010), distributed with ANTs 2.3.3 (Avants et al. 2009) and used as T1w-reference throughout the workflow. The T1w-reference was skull-stripped with a Nipype implementation of the antsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as target template. Brain tissue segmentation of cerebrospinal fluid (CSF), white-matter (WM) and gray-matter (GM) was performed on the brain-extracted T1w using fast (FSL 6.0.5.1:57b01774, RRID:SCR_002823), CO, USA 80210 (Zhang et al. 2001). Brain surfaces were reconstructed using recon-all (FreeSurfer 7.3.2, RRID:SCR_001847) (Dale et al. 1999), and the brain mask estimated previously was refined with a custom variation of the method to reconcile ANTs-derived and FreeSurfer-derived segmentations of the cortical graymatter of Mindboggle (RRID:SCR_002438) (Klein et al. 2017). For more details of the pipeline, see the section corresponding to workflows in sMRIPrep’s documentation. After each image was processed with sMRIPrep, visual inspection of the segmentations and surface reconstructions were conducted following previously developed procedures (Raamana, et al. 2020).

Longitudinal anatomical data preprocessing

After each individual session was processed with recon-all cross-sectionally, we used Freesurfer-7.3.2’s longitudinal processing pipeline (Reuter et al. 2012). This pipeline has been extensively described elsewhere (Reuter et al. 2012). The pipeline creates a template or ‘base’ for each participant from all their timepoints and estimates the average anatomy (also referred to as the within-subject template). An unbiased median image is used as the template and segmentation and surface reconstruction were performed. The scans were processed ‘longitudinally’ in which information from the within-subject template and from each of the time points is used to initialize the algorithms used for recon-all. For the subcortical segmentation, the pipeline creates a fused segmentation for each time point using an intensity-based probabilistic voting scheme. To QC the longitudinal processing steps, we conducted visual QC of tissue segmentation and subcortical segmentation accuracy by overlaying the segmentations on the T1 image for the within-subject template (base).

As with the cross-sectional processing, all segmentations passed QC procedures. To conduct the region of interest analysis, subcortical segmentation volumes (from the ‘aseg’ output) were compiled for the bilateral hippocampus, bilateral amygdala, and bilateral ventral diencephalon (using the asegstats2table command). Regional mean cortical thickness and surface area values (from the ‘aparc’ output) were also compiled for the statistical analysis (using the aparcstats2table command). Analysis focused on the Desikan-Killiany Atlas parcellation in Freesurfer to be in alignment with the regions analyzed in Kim, Dufford, & Tribble, 2016.

Linear mixed effects models

To examine the longitudinal change of the regions of interest, we used Linear Mixed Effects modeling implemented by the “lme4” (Bates, et al. 2015; Bates 2010) built under R version 4.2.3. We tested random intercept models for each region of interest. For each of these models, the scan timepoint, age at the early postpartum scan, parity, and global structural measure (for CT it is appropriate to covary for global effects using mean thickness for the hemisphere, for SA the sum SA for the hemisphere, and for volume, the total intracranial volume for both hemispheres) were included as fixed effects. For each model ‘Participant’ was modelled as a random effect. For each model p-values were computed using the “lmerTest” (Kuznetsova et al. 2017, 2015) package in R as these evaluations of significance have been shown to have Type I error rates closest to p = 0.05 (Luke 2017). Uncorrected p-values were corrected for multiple testing using Bonferroni correction implemented in the R function “p.adjust”. We also conducted a post-hoc test, using identical models as described above, but tested regions of interest (ROI) from a previous study (Hoekzema et al. 2017) that have found pregnancy-related decreases in structural volume and thickness including in the fusiform gyrus, inferior frontal gyrus, precuneus, and superior temporal gyrus. These results are presented in the Supplementary Information.

Exploratory analysis of depressive symptoms

For the exploratory analysis, we used Pearson correlations to examine the associations between the change in depressive symptoms from early to one-year and CT at one-year postpartum. We focused this analysis to only regions that had a significant change from early postpartum to one-year postpartum from the ROIs. Identical exploratory tests were also tested for the left amygdala and right ventral diencephalon.

Results

Demographic variables

Demographic information for the sample is presented in Table 1. For the EPDS measured in the early postpartum, 4 participants had a total score > 10, a commonly used cutoff score indicating ‘mild depression’ (McCabe-Beane et al. 2016). For the one-year postpartum scan, 3 participants had a total score > 10. The mean change in depressive symptoms (EPDS) from early to one-year postpartum was 1.19 (3.59) suggesting that on average, depressive symptoms increased slightly from early to one-year postpartum.

Table 1.

Demographic information for the sample

Variable	Sample (N = 24)
Maternal age at early postpartum scan (years) Maternal age at one-year postpartum scan (years) Maternal education (years) Maternal income-to-needs ratio Maternal ethnicity (Hispanic) Maternal race  Black/African American  White/Caucasian  American Indian/Alaska Native  Other Number of live births upon study entry Months postpartum at early postpartum scan Months postpartum at one-year scan Interval between scans (months) EPDS total score at early postpartum home visit EPDS total score at one-year postpartum home visit Change in EPDS from early to one year postpartum	28.04 (4.92) 29.08 (4.88) 14.83 (2.79) 2.94 (2.52) 11 (47%) 2 (8%) 12 (52%) 3 (13%) 6 (26%) 1.16 (1.55) 0.14 (0.09) 13.55 (0.99) 15.31 (5.5) 5.52 (4.15) 6.71 (4.69) 1.19 (3.59)

EPDS Edinburgh Postnatal Depression Scale

N = 3 missing EPDS Total Score at Early Postpartum Scan. N = 3 missing EPDS Total Score at One-Year Postpartum Scan. N = 3 missing Change in EPDS from Early to One Year Postpartum. N = 1 missing Maternal Education (years). N = 2 missing Maternal Income-to-needs Ratio. N = 1 missing Maternal Ethnicity (Hispanic). N = 1 missing Maternal Race

Linear mixed effects models for surface morphometry

We fitted a linear mixed model (estimated using REML and nloptwrap optimizer) for left superior frontal gyrus CT with scan timepoint (Early Postpartum or One-Year Postpartum, age at the early postpartum scan, parity, and mean cortical thickness of the left hemisphere (formula: left superior frontal gyrus cortical thickness ~ 1 + scan timepoint + age at the early postpartum scan + parity + mean cortical thickness of the left hemisphere). The model included participant as random effect (formula: ~ 1 | Participant). The model’s total explanatory power is substantial (conditional R² = 0.89) and the part related to the fixed effects alone (marginal R²) is 0.45. The model’s intercept was nonsignificant: 0.64 (95% CI [− 0.18, 1.47], t(41) = 1.57, p = 0.123). For this model, the effect of scan timepoint was statistically significant and positive (beta = 0.10, 95% CI [0.05, 0.15], t(41) = 4.14, p < 0.001; Std. beta = 0.03, 95% CI [0.01, 0.04], see Fig. 1 and Table 2). An identical model was fit using the right superior frontal gyrus values and mean cortical thickness of the right hemisphere as a covariate (in addition to scan timepoint, age at the early postpartum scan, parity). This model’s total explanatory power is substantial (conditional R² = 0.89) and the part related to the fixed effects alone (marginal R²) is 0.69. The model’s intercept was nonsignificant at 0.10 (95% CI [− 0.63, 0.82], t(41) = 0.27, p = 0.790). Within this model, the effect of scan timepoint was statistically significant and positive (beta = 0.04, 95% CI [0.01, 0.06], t(41) = 3.16, p = 0.003; Std. beta = 0.02, 95% CI [7.19e-03, 0.03]). Results for the caudal middle frontal gyrus CT are shown in Table 3. Results for the precentral gyrus and lateral occipital gyrus CT are shown in Table 4 and 5 respectively. For the SA analysis, none of the regions tested (bilateral superior frontal, lateral occipital, caudal middle frontal, or precentral gyri) were significant after Bonferroni correction (ps > 0.05). The results for the surface area regions of interest are presented in the Supplemental Information in Supplementary Table 1.

Fig. 1.

a Regions of interest for the cortical thickness (CT) analysis. Plots show the changes in CT for the superior frontal gyrus (SFG) from early to one-year postpartum. b Regions of interest for the subcortical volume analysis. Plots show the changes in the left amygdala and ventral diencephalon from early to one-year postpartum

Table 2.

Results from the linear mixed models for superior frontal gyrus cortical thickness

Predictors	Left Superior Frontal Gyrus CT			Right Superior Frontal Gyrus CT
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept)	0.69	− 0.14 – 1.52	0.1	0.1	− 0.63 – 0.82	0.79
Scan timepoint	0.05	0.03 – 0.08	< 0.001, 0.003	0.04	0.01 – 0.06	0.003, 0.02
Maternal age at the early post- partum scan	0	− 0.01 – 0.00	0.367	0	− 0.01 – 0.00	0.752
Live births upon study entry	0.01	− 0.02 – 0.03	0.478	0.01	− 0.01 – 0.02	0.405
Mean thickness	0.85	0.52 – 1.17	< 0.001	1.07	0.79 – 1.35	< 0.001

P-value corr. is the Bonferroni corrected p-value

Table 3.

Results from the linear mixed models for caudal middle frontal gyrus cortical thickness

Predictors	Left caudal middle frontal gyrus CT			Right caudal middle frontal gyrus CT
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept) Scan Timepoint Maternal Age at the Early Post-partum Scan Live Births Upon Study Entry Mean Thickness	0.21 0.03 0 0 1.03	− 0.68 – 1.11 0.00 – 0.06 − 0.01 – 0.00 − 0.02 – 0.03 0.68 – 1.38	0.631 0.044, 0.38 0.234 0.791 < −0.001	− 0.82 0.01 0 − 0.01 1.37	− 1.71 – 0.07 − 0.01 – 0.04 − 0.00 – 0.01 − 0.03 – 0.02 1.02 – 1.72	0.071 0.356, 0.99 0.481 0.637 < 0.001

P-value corr. is the Bonferroni corrected p-value

Table 4.

Results from the linear mixed models for lateral occipital gyrus cortical thickness

Predictors	Left lateral occipital gyrus CT			Right lateral occipital gyrus CT
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept) Scan Timepoint Maternal Age at the Early Postpartum Scan Live Births Upon Study Entry Mean Thickness	− 0.84 − 0.05 0.01 − 0.02 1.12	− 1.81 – 0.13 − 0.09 – 0.016 0.00 – 0.01 − 0.04 – 0.01 0.74 – 1.51	0.09 0.01, 0.08 0.038 0.135 < 0.001	− 0.18 − 0.04 0.01 0 0.88	− 1.32 – 0.96 − 0.08 – − 0.01 − 0.00 – 0.02 − 0.03 – 0.02 0.44 – 1.33	0.751 0.023, 0.2 0.133 0.721 < 0.001

P-value corr. is the Bonferroni corrected p-value

Table 5.

Results from the linear mixed models for precentral gyrus cortical thickness

Predictors	Left precentral gyrus CT			Right precentral gyrus CT
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept)	− 1.21	− 2.12 – − 0.30	0.01	− 0.39	− 1.47 – 0.68	0.465
Scan Timepoint	− 0.02	− 0.04 – 0.01	0.225, 0.99	− 0.02	− 0.05 – 0.01	0.127, 0.99
Maternal Age at the Early Post- partum Scan	0	− 0.01 – 0.01	0.731	0	− 0.02 – 0.01	0.665
Live Births Upon Study Entry	0	− 0.03 – 0.02	0.823	0.01	− 0.03 – 0.05	0.703
Mean Thickness	1.57	1.21 – 1.92	< 0.001	1.24	0.84 – 1.64	< 0.001

P-value corr. is the Bonferroni corrected p-value

Linear mixed effects models for subcortical volumes

A linear mixed model was fit for the left amygdala and right ventral diencephalon. For the left amygdala, the model’s conditional R² was 0.98 and marginal R² was 0.07. The model’s intercept was significant at 1086.01 (95% CI [418.07, 1753.94], t(41) = 3.28, p = 0.002). The effect of scan timepoint was statistically significant and positive (beta = 25.80, 95% CI [13.53, 38.07], t(41) = 4.25, p < 0.001; Std. beta = 0.10, 95% CI [0.05, 0.15], see Fig. 1 and Supplementary Table 5). For the right diencephalon, the conditional R² was 0.95 with a marginal R² of 0.26. The model’s intercept was significant: 1951.54 (95% CI [456.32, 3446.76], t(41) = 2.64, p = 0.012). The effect of scan timepoint was statistically significant and positive (beta = 73.85, 95% CI [29.00, 118.69], t(41) = 3.33, p = 0.002; Std. beta = 0.12, 95% CI [0.05, 0.19], see Fig. 1 and Supplementary Table 6). Results for the hippocampus that were not significant are presented in Supplementary Table 4.

As they were separate structural metric analyses, Bonferroni correction was conducted across all eight CT regions (left and right superior frontal gyrus, left and right caudal middle frontal gyrus, left and right lateral occipital gyrus, left and right precentral gyrus), all eight SA regions (same regions as CT), and six subcortical volume regions (left hippocampus, right hippocampus, left amygdala, right amygdala, left ventral diencephalon, right diencephalon) separately. After Bonferroni correction, longitudinal increases were found for the left (p = 0.003, corrected) and right (p = 0.02, corrected) superior frontal gyrus CT. No significant increases or decreases were observed for the bilateral caudal middle frontal gyrus (Table 3), lateral occipital gyrus (Table 4), or precentral gyrus (Table 5). There was no significant increase or decrease for the regions of interest for SA (ps > 0.05, corrected). After Bonferroni correction, significant increases in gray matter volume were observed for the left amygdala (p = 0.001, corrected) and right ventral diencephalon (p = 0.01, corrected). No significant increases or decreases were found for the right amygdala (Table 6), left ventral diencephalon (Table 7), or bilateral hippocampus (Table 8).

Table 6.

Results from the linear mixed models for amygdala volume

Predictors	Left amygdala volume			Right amygdala volume
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept) 1086.01 Scan timepoint 25.8 Maternal age at the early post-partum scan Live births upon study entry Total intracranial volume	1086.01 25.8 1.27 − 18.08 0	396.11 – 1775.90 13.23 – 38.37 − 11.69 – 14.23 − 57.66 – 21.50 − 0.00 – 0.00	0.004 < 0.001, 0.001 0.84 0.352 0.627	1202.34 22.13 1.72 − 27.84 0	433.54 – 1971.13 1.14 – 43.11 − 12.72 – 16.17 − 71.95 – 16.26 − 0.00 – 0.00	0.004 0.04, 0.22 0.806 0.203 0.465

P-value corr. is the Bonferroni corrected p-value

Table 7.

Results from the linear mixed models for ventral diencephalon volume

Predictors	Left ventral diencephalon volume			Right ventral diencephalon volume
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept)	2046.16	583.03 – 3509.29	0.009	1951.54	407.19 – 3495.89	0.016
Scan Timepoint	50.92	− 21.33 – 123.18	0.158, 0.94	73.85	27.91 – 119.78	0.003, 0.017
Maternal Age at the Early Post-partum Scan	1.81	− 25.67 – 29.29	0.892	7.26	− 21.74 – 36.27	0.607
Live Births Upon Study Entry	− 32.49	− 116.41 – 51.44	0.429	37.6	− 126.19 – 50.9	0.387
Total Intracranial Volume	0	0.00 – 0.00	0.03	0	− 0.00 – 0.00	0.053

P-value corr. is the Bonferroni corrected p-value

Table 8.

Results from the linear mixed models for hippocampus volume

Predictors	Left Hippocampus Volume			Right Hippocampus Volume
	Estimates	CI	p (p value corr.)	Estimates	CI	p (p value corr.)
(Intercept)	2524.94	1327.26 – 3722.63	< 0.001	2218.88	777.5 – 3660.1	0.004
Scan timepoint	33.81	2.26 – 65.37	0.037, 0.22 corrected	34.51	− 6.38 – 75.40	0.094, 0.56 corrected
Maternal age at the early postpartum scan	16.6	− 5.90 – 39.10	0.139	13.25	− 13.83 – 40.32	0.32
Live births upon study entry	− 51.69	− 120.40– 17.02	0.132	21.75	− 104.43 – 60.9	0.589
Total intracranial volume	0	− 0.00 – 0.00	0.125	0	− 0.00 – 0.00	0.069

P-value corr. is the Bonferroni corrected p-value

Exploratory analyses of depressive symptoms

For the exploratory analyses we examined the associations between longitudinal change in depressive symptoms and brain structure. First, we determined if there were any statistical outliers in the change in depressive symptoms by examining the z-scored values. The participant that had a change in EPDS score of ‘− 8’ was not considered an outlier as it had a z-score value of − 2.84, and − 3 to 3 is the traditionally used cutoff (Seo 2006). The correlation between change in depressive symptoms (from early to one-year postpartum) and left superior frontal gyrus CT was negative and statistically significant (r = − 0.50, 95% CI [− 0.77, − 0.09], t(19) = − 2.54, p = 0.020, see Fig. 2a). Again, a similar pattern was observed for the association between changes in depressive symptoms and right superior frontal gyrus CT (r = − 0.50, 95% CI [− 0.77, − 0.09], t(19) = − 2.54, p = 0.020, see Fig. 2b). However, if the participant whose change in symptom score had a z-score of − 2.84 was removed from the analysis, these correlations were no longer significant and at a trend level (p = 0.09 for the left superior frontal gyrus and p = 0.11 for the right superior frontal gyrus). All correlations between left amygdala volume, right ventral diencephalon volume, and depressive symptoms (and depressive symptom change) were nonsignificant (ps > 0.05).

Fig. 2.

Scatter plots for the exploratory analyses examining change in depressive symptoms from early to one-year postpartum and left (a) and right (b) superior frontal gyrus thickness

Discussion

Human neuroimaging has provided a lens to understand the neuroanatomical changes that occur across the perinatal period (Luders et al. 2022; Barba-Müller et al. 2019; Martínez-García et al. 2021b; Pawluski et al. 2022). Understanding these brain changes is critical to further our understanding of the neural underpinnings of the onset and maintenance of parental behaviors. The present findings expand upon our knowledge of postpartum brain changes by quantifying structural changes from the early postpartum period to one-year postpartum. Further, we examined both CT and subcortical volumes. The surface morphometry analysis focused on four regions of interest where we previously found CT was associated with postpartum timing (Kim et al. 2018). Of these four regions, longitudinal neuroanatomical increases in CT were found for the left and right superior frontal gyrus. For the subcortical regions examined, we found increases in the left amygdala and right diencephalon. As the association between postpartum brain structure and maternal mood is unclear, we conducted an exploratory analysis in which we tested the correlation between the regions that significantly increased in the postpartum and depressive symptoms in the postpartum period. In an exploratory analysis, we found that reductions of depressive symptoms from early postpartum to one-year postpartum were associated with greater bilateral superior frontal gyrus CT at one-year postpartum. We provide evidence for future studies to examine and attempt to replicate the association between changes in depressive symptoms across the postpartum period and superior frontal gyrus CT using a larger, clinically-enriched sample.

The present study was motivated by our previous study that found positive associations (and no negative associations) between postpartum timing and CT in the super frontal, caudal middle frontal, lateral occipital, and precentral gyri (Kim et al. 2018). As this study was cross-sectional, we could not examine if the CT in these regions ‘changed’ across the postpartum. Here, we found that of the regions correlated with postpartum timing in our previous study, only the superior frontal gyrus had significant increases in CT from early postpartum to one-year postpartum. This finding is aligned with a previous study that found the volume of the right superior frontal gyrus increased from 2–4 weeks to 4–6 weeks postpartum (Kim et al. 2010). The mechanism of these structural increases in the prefrontal cortex remains unclear, although it is hypothesized they are due to experience-dependent plasticity (Barba-Müller et al. 2019; Kim et al. 2018; Martínez-García et al. 2021b). This is hypothesized to include responding to the increased regulatory demands of parenting (Kim et al. 2016; Barba-Müller et al. 2019), affective regulation for both parent and child (Barba-Müller et al. 2019; Grande et al. 2021), and needing to recognize and respond appropriately/flexibly to the offspring’s needs (Leuner and Gould 2010). Animal studies support these hypotheses and have found increased neuronal spine density from the early to late postpartum (Hillerer et al. 2018) as well as postweaning (Leuner and Gould 2010). Whether these neuronal increases underlie the CT increases observed with structural MRI remains to be confirmed. The superior frontal gyrus is a large region of cortex, therefore more precise localization of these increases is required with larger studies that can utilize vertex-wise analytic methods.

For the subcortical regions of interest, we have replicated previous studies that have found structural increases in amygdala volume across the postpartum period (Kim et al. 2010; Luders et al. 2020; Luders et al. 2021). It has been found in a previous study that amygdala volume increases from 1–2 days after childbirth to 4–6 weeks after childbirth (Luders et al. 2020). These findings were further examined in terms of amygdala subareas in which the most pronounced increases were found in the superificial region (Luders et al. 2021). Further, amygdala volume was also found to increase, using a voxel-wise approach, from 2–4 weeks to 3–4 months postpartum (Kim et al. 2010). We extend this knowledge by replicating this finding from 2–4 weeks to one-year postpartum. The amygdala is recognized to be a core region of the ‘maternal’ brain for its role in salience detection (Kim et al. 2016; Grande et al. 2021; Strathearn and Kim 2013), parent–child interaction (Barrett et al. 2012), and affective processing (Strathearn and Kim 2013). Regarding potential neuronal mechanisms for the observed increases in the amygdala, animal studies suggest that dendritic spine concentrations increase post-parturition in the anterodorsal medial amygdala (Rasia-Filho et al. 2004). The anterodorsal medial amygdala is part of four main subnuclei involved in the regulation of social behaviors (Bolhuis et al. 1984), affective stimuli interpretation (Dielenberg et al. 2001), and sensory information processing (Adamec and Morgan 1994). In addition to the amygdala, we replicate findings of postpartum structural increases in regions that include the hypothalamus (Kim et al. 2010). We found structural increases of the ventral diencephalon which includes the hypothalamus, mamillary bodies, subthalamic, lateral geniculate, medial geniculate and red nuclei, substantia nigra and surrounding white matter (Fischl et al. 2002). Therefore, future studies will need to use recent advances in image segmentation for the smaller structures included in the ventral diencephalon segmentation (such as the mamillary bodies) as they have been implicated to be involved in infant caregiving (Leckman and Herman 2002). Specifically, neuronal dendritic branching of the medial preoptic area of the hypothalamus has been found to increase across the postpartum period in animal studies and impact the onset of parenting behaviors (Numan et al. 1977; Numan 1974; Shams et al. 2012).

We did not find evidence of a significant increase in gray matter volume for the hippocampus. However, this may be due to lack of power given a significant p-value for the left hippocampus before multiple comparisons correction. Animal studies indicate decreases in neuronal proliferation in the middle and late postpartum periods (Darnaudéry et al. 2007; Leuner et al. 2007). Studies have also found dendritic remodeling across the perinatal period for the CA1 and CA3 regions of the hippocampus (Pawluski and Galea 2006). In a small sample of humans (n = 11) pregnancy-related decreases had returned to pre-pregnancy baseline except for in the left hippocampus, which had a partial recovery (Hoekzema et al. 2017). Therefore, future studies with prepregnancy, prenatal, and postnatal scanning may be needed to examine the complex trajectories of hippocampal structure across the perinatal period.

A major strength of the present study was the ability to examine if change in depressive symptoms from early to one-year postpartum were associated with brain structure at one-year postpartum. As several animal studies have provided causal evidence of postpartum structural changes impacting depressive-like behaviors (Haim et al. 2014), we conducted an exploratory analysis to examine if changes in depressive symptoms across the postpartum were associated with brain structural metrics. We found reductions in depressive symptoms from early to one-year postpartum were associated with greater CT in both the left and right superior frontal gyrus. Previous studies have implicated the superior frontal gyrus in studies of participants diagnosed with postpartum depression including a positive correlation observed with EPDS scores (Schnakenberg et al. 2021) and reduction of functional connectivity between the anterior cingulate cortex and bilateral superior frontal gyrus (Deligiannidis et al. 2013). As mentioned, more precise localization of these potential brain-symptom correlations is needed given the large anatomical region defined as the superior frontal gyrus. Other studies of perinatal brain changes have not found associations between brain structure and depressive symptoms (Barba-Müller et al. 2019). This may be due to several factors. First, few studies have examined mood symptoms longitudinally in the postpartum period and therefore could not examine changes in symptoms. Further, it is worth noting the present sample, in comparison to previous perinatal neuroimaging studies, is diverse in terms of the socioeconomic and ethnic backgrounds of the participants.

The findings from the present study should be interpreted considering the following limitations. The sample size was modest for longitudinal neuroimaging and will need replication in larger samples. The modest sample size motivated the analytic approach of focusing on regions of interest found in previous studies, rather than mass-univariate approaches (vertex-wise or voxel-wise). Future studies, with larger samples, will be able to harness advances in more computationally efficient linear mixed modeling approaches for neuroimaging data (Parekh, et al. 2024). Further investigations will use deep-learning based segmentations of these regions and examine specific hypothalamic subunits as ventral diencephalon is comprised of multiple unique brain regions and with the segmentation that was used, we could not delineate which structures were contributing most to the longitudinal increase in volume. Future studies will be needed to replicate the exploratory findings of the superior frontal gyrus correlation with depressive symptom change found in a community-based sample; this relation will need replication in a clinically-enriched sample with higher levels of depressive symptoms. Last, expansions of this work will include the use of dense longitudinal sampling (Pritschet et al. 2021; Pritschet et al. 2020; Taylor et al. 2021; Taylor et al. 2020) across the perinatal transition, which is still critical to establish prenatal to postnatal trajectories for brain structure.

Conclusions

Despite the troubling lack of neuroimaging studies focusing on women’s health issues, consistency is being found across longitudinal neuroimaging studies of the perinatal period (Luders et al. 2022). Here, we provide additional support for the hypothesis that the postpartum period entails increase in brain structural metrics. Specifically, we found increases from the early postpartum to one-year postpartum in the bilateral superior frontal gyrus in terms of CT, and no significant decreases. We did not find any increases or decreases in cortical SA for the regions of interest (the superior frontal, caudal middle frontal, lateral occipital, and precentral gyri). For subcortical gray matter volume, longitudinal increases were found for the left amygdala and right ventral diencephalon. Lastly, we found that reductions in depressive scores from early postpartum to one-year postpartum were associated with greater thickness of the bilateral superior frontal gyrus. Here, we add evidence to the emerging understanding of longitudinal neuroanatomical changes across the perinatal period by focusing on the early postpartum to one-year postpartum. Further, we find exploratory evidence that changes in depressive symptoms across the postpartum period may play a role in the neuroanatomical increases observed from early to one-year postpartum. This finding can provide a foundation for future longitudinal studies of the postpartum period to further identify and/or replicate brain regions that may confer risk and resilience to changes in postpartum mood.

Supplementary Material

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s00429-024-02852-x.

Data availability

The data that support the findings of this study are available from the senior author upon reasonable request.