Hippocampal subfield volumes are nonspecifically reduced in premature‐born adults

Abstract Reduced global hippocampus volumes have been demonstrated in premature‐born individuals, from newborns to adults; however, it is unknown whether hippocampus subfield (HCSF) volumes are differentially affected by premature birth and how relevant they are for cognitive performance. To address these questions, we investigated magnetic resonance imaging (MRI)‐derived HCSF volumes in very premature‐born adults, and related them with general cognitive performance in adulthood. We assessed 103 very premature‐born (gestational age [GA] <32 weeks and/or birth weight <1,500 g) and 109 term‐born individuals with cognitive testing and structural MRI at 26 years of age. HCSFs were automatically segmented based on three‐dimensional T1‐ and T2‐weighted sequences and studied both individually and grouped into three functional units, namely hippocampus proper (HP), subicular complex (SC), and dentate gyrus (DG). Cognitive performance was measured using the Wechsler‐Adult‐Intelligence‐Scale (full‐scale intelligence quotient [FS‐IQ]) at 26 years. We observed bilateral volume reductions for almost all HCSF volumes in premature‐born adults and associations with GA and neonatal treatment intensity but not birth weight. Left‐sided HP, SC, and DG volumes were associated with adult FS‐IQ. Furthermore, left DG volume was a mediator of the association between GA and adult FS‐IQ in premature‐born individuals. Results demonstrate nonspecifically reduced HCSF volumes in premature‐born adults; but specific associations with cognitive outcome highlight the importance of the left DG. Data suggest that specific interventions toward hippocampus function might be promising to lower adverse cognitive effects of prematurity.

but not birth weight. Left-sided HP, SC, and DG volumes were associated with adult FS-IQ. Furthermore, left DG volume was a mediator of the association between GA and adult FS-IQ in premature-born individuals. Results demonstrate nonspecifically reduced HCSF volumes in premature-born adults; but specific associations with cognitive outcome highlight the importance of the left DG. Data suggest that specific interventions toward hippocampus function might be promising to lower adverse cognitive effects of prematurity.
The hippocampus is a complex structure located in the medial temporal lobe, which is critical for neural pattern formation and its use in interaction with other structures, mainly cortices, for example in the context of memory consolidation, planning, recall, spatial navigation, or more complex concepts such as learning (Buzsaki & Moser, 2013). The function of the hippocampus as a "computational hub" is supported by its extensive connections to cortical and subcortical regions in the mammalian forebrain via the entorhinal and parahippocampal cortices (Sweatt, 2004). The hippocampus formation consists of structural and functional distinct subfields, most importantly the cornu ammonis (CA) subfields 1-4, the dentate gyrus (DG), which has an anatomical overlap with parts of CA4 and the subiculum (Amaral, 1978;Duvernoy, Cattin, Risold, Vannson, & Gaudron, 2013;Witter & Amaral, 2004). The human hippocampus is particularly vulnerable to hypoxic-ischemic events and its subfields exhibit differential vulnerability to hypoxia, ischemia, and hypercortisolemia (Bartsch et al., 2015;Sapolsky, 2000;Schmidt-Kastner & Freund, 1991). For example, it was shown in human patients that the CA1 subfield is selectively vulnerable to a variety of metabolic and cytotoxic insults, in particular ischemia (Bartsch et al., 2015). The DG, as the primary site of adult neurogenesis in the hippocampal formation, has shown to be susceptible to increased stress and stress hormone levels (Schoenfeld & Gould, 2012). Volume reductions of distinct hippocampal subfields (HCSF) have been reported for several neuropsychiatric diseases such as major depressive disorder, posttraumatic stress disorder, and schizophrenia (Hayes et al., 2017;Nakahara, Matsumoto, & van Erp, 2018;Roddy et al., 2019) and also in patients after pediatric brain tumor surgery or microsurgical clipping of intracranial aneurysms (Decker et al., 2017;Hedderich, Reess, et al., 2019).
With respect to premature birth, lower whole hippocampus volumes have been described for infants, children, adolescents and adults and associations with memory functions were found (Aanes, Bjuland, Skranes, & Lohaugen, 2015;Bjuland, Rimol, Løhaugen, & Skranes, 2014;Lodygensky et al., 2008;Meng et al., 2016;Nosarti & Froudist-Walsh, 2016). Further, in a cohort of premature-born children at approximately 9 years of age, lower subfield volumes for the bilateral DG and the right subiculum as well as a positive correlation between right DG volume and working memory testing were reported (Aanes et al., 2019). However, the long-term effect of premature birth on adult hippocampus structure has not been investigated, yet. Thus, it is not known whether lower hippocampus volumes after premature birth are due to selective vulnerability and volume loss of distinct subfields or whether the volume reduction is unspecific, probably because of general developmental impairment of the hippocampus. It is also unknown whether distinct HCSF volume reductions are specifically linked with cognitive function in adulthood after premature birth.
To address these questions, we analyzed HCSF in a large cohort of very premature-born adults and term-born controls using an established and automated pipeline based on T1-and T2-weighted high-resolution magnetic resonance imaging (MRI). We investigated associations with perinatal variables of premature birth and with adult full-scale intelligence quotient (FS-IQ) as a global outcome measure of cognitive performance in order to study their specificity for premature birth and their functional relevance. Furthermore, we investigated whether functional HCSF units mediate the relationship between gestational age (GA) and adult FS-IQ in premature-born adults.

| Participants
The participants examined in this study are part of the Bavarian Longitudinal Study (BLS), a geographically defined, whole-population sample of neonatal at-risk children and healthy full term controls who were followed from birth into adulthood (Riegel, Orth, Wolke, & Österlund, 1995;Wolke & Meyer, 1999). Of the initial 682 infants born very preterm (VP; <32 weeks) and/or with very low birth weight (VLBW) <1,500 g), 411 were eligible for the 26-year follow-up assessment, and 260 (63.3%) participated in psychological assessments (Breeman, Jaekel, Baumann, Bartmann, & Wolke, 2015). Of the initial 916 full-term born infants from the same obstetric hospitals that were alive at 6 years, 350 were randomly selected as control subjects within the stratification variables of sex and family socioeconomic status (SES) in order to be comparable with the VP/VLBW group. Of these, 308 were eligible for the 26-year follow-up assessment, and 229 (74.4%) participated in psychological assessments. All of the 260 subjects from the VP/VLBW group underwent an initial screening for MR-related exclusion criteria, which included: (self-reported) claustrophobia, inability to lie still for >30 min, unstable medical conditions (e.g., severe asthma), epilepsy, tinnitus, pregnancy, nonremovable, MRIincompatible metal implants and a history of severe CNS trauma or disease that would impair further analysis of the data. The most frequent reason not to perform the MRI exam, however, was a lack of motivation. The remaining eligible, 103 VP/VLBW and 109 FT individuals underwent MRI at 26 years of age.
The MRI examinations took place at two sites: The Department of Neuroradiology, Klinikum rechts der Isar, Technical University of Munich (n = 144) and the Department of Radiology, University Hospital of Bonn (n = 68). The study was carried out in accordance with the Declaration of Helsinki and was approved by the local institutional review boards. Written consent was obtained from all participants. All study participants received travel expenses and a small payment for attendance. A more detailed description of participants, including incidental brain MRI findings can be found in a previous publication .

| Birth-related variables
GA was calculated from maternal reports on the first day of the last menstrual period and serial ultrasounds during pregnancy. In cases where the two measures differed by more than 2 weeks, clinical assessment at birth with the Dubowitz method was applied (Dubowitz, Dubowitz, & Goldberg, 1970). Maternal age, BW, and duration of hospitalization were obtained from obstetric records (Gutbrod, Wolke, Soehne, Ohrt, & Riegel, 2000;Riegel et al., 1995).
Intensity of Neonatal Treatment Index (INTI) was determined by daily assessments of care level, respiratory support, feeding dependency and neurological status (mobility, muscle tone, and neurological excitability). Each of the six variables was scored on a 4-point rating scale (0-3) by the method of Casaer and Eggermont (Casaer & Eggermont, 1985) (see Table S1 for a description of the variables).
The INTI was computed as the mean score of daily ratings during the first 10 days of life or until a stable clinical state was reached (total daily scores <3 for 3 consecutive days), depending on which occurred first, ranging from 0 (best state) to 18 (worst state). Family SES was assessed through structured parental interviews within 10 days of childbirth. SES was computed as a weighted composite score based on the profession of the self-identified head of each family together with the highest educational qualification held by either parent (Bauer, 1988).

| Neurocognitive assessment
At 26 years of age, study participants were assessed using a short version of the German Wechsler Adults Intelligence Scale, Third edition (von Aster, Neubauer, & Horn, 2006): The assessment took place prior to and independent of the MRI scan and was carried out by trained psychologists who were blinded to group membership. Subsequently, an FS-IQ was computed. Across all scanners, sequence parameters were kept identical. Scanners were checked regularly to provide optimal scanning conditions. MRI physicists at the University Hospital Bonn and Klinikum rechts der Isar regularly scanned imaging phantoms, to ensure withinscanner signal stability over time. Signal-to-noise ratio was not significantly different between scanners (one-way analysis of variance with factor "scanner-ID" (Bonn 1, Bonn 2, Munich 1, Munich 2); All images were visually inspected for artifacts.

| HCSF volumetry
Image analysis was performed with the FreeSurfer image analysis suite (version 6.0), which is documented and freely available for download online (http://surfer.nmr.mgh.harvard.edu/) (Dale, Fischl, & Sereno, 1999;Fischl et al., 2002;Segonne et al., 2004). Within that framework, the cross-sectional, multispectral pipeline was chosen and both T1-weighted and T2-weighted scans were used as input for HCSF volumetry. In this FreeSurfer release (v6.0), an HCSF segmentation tool has been implemented which is based on a Bayesian model with Markov random field priors (Iglesias et al., 2015). Briefly, the applied parametric segmentation algorithm was developed based on high-resolution (0.13 mm) ex vivo MRI scans of the human hippocampus from 15 autopsy samples. These ex vivo MRI samples were manually segmented and integrated with in vivo T1-weighted images

| Statistical analysis
Statistical analyses were carried out using SPSS (IBM SPSS Statistics, version 25). General linear models were used to determine whether premature birth status is a significant factor for different HCSF volumes (dependent variable: HCSF; fixed factors: history of premature birth; covariates: sex, scanner, total intracranial volume [TIV]). This analysis was repeated using a similar general linear model, correcting for left/ right hippocampus volumes instead of TIV. Partial correlations restricted to the VP/VLBW group and corrected for scanner, TIV, and sex were used to investigate the associations between HCSF volumes and variables of premature birth: GA, BW, and INTI. A similar partial correlation analysis was used to investigate FS-IQ and HCSFs.
Differences between VP/VLBW and FT individuals were tested using chi-square tests (sex, SES) or two-sample t tests (age, GA, BW, INTI, maternal age, FS-IQ). Statistical significance was set at p < .05; all tests are two-sided. p-Values of post hoc tests were FDR-corrected for multiple comparisons according to the Benjamini-Hochberg method within every discrete set of analyses (Benjamini & Hochberg, 1995). In order to test whether functional HCSF units mediate the association between low GA and FS-IQ in adulthood, a mediation analysis restricted to the VP/VLBW group was performed using the PROCESS toolbox (version 3.0) (Hayes, 2017). In the mediation model, GA was entered as causal variable, adult FS-IQ as the outcome variable, left and right HP, SC and DG volumes were introduced simultaneously as mediator variables, and MRI scanner, TIV and sex as covariates of no interest. Path coefficients for total effect, direct effect and indirect effect were estimated using (unstandardized) regression coefficients from multiple regression analyses, and statistical significance of the indirect effect was tested using a nonparametric bootstrap approach (with 5.000 repetitions) to obtain 95% confidence intervals. We calculated p-values for indirect effects based on 95% confidence intervals, SE and estimated effect as described by Altman and Bland (Altman & Bland, 2011 19 years of age, these aberrations were mostly limited to subiculum and CA1 (Cole et al., 2015). In contrast, our results suggest a rather nonspecific long-term effect of premature birth on bilateral HCSF volumes. We know from longitudinal MRI studies that HCSF volumes in fact grow uniformly, showing a steep volume increase until 13-15 years, followed by a phase with only small age-related changes (Krogsrud et al., 2014). However, this growth pattern may be altered after premature birth by selectively injured HCSF that show catch-up growth during late adolescence and early adulthood, while a global shift of the developmental brain trajectory persists.

| Functional relevance of HCSF volumes for general cognitive performance
The hippocampus serves as a computational hub with vast connectivity to subcortical and cortical regions of the mammalian forebrain (Buzsaki, 2011;Sweatt, 2004). Specific hippocampal computational functions are sequential neural patterning including pattern completion and pattern separation (Buzsaki, 2011;Buzsaki & Moser, 2013).
These are very basic processes needed for distinct cognitive functions such as planning, learning, and navigation, so that a central role for the development of general cognitive performance can be proposed. Note: Correlation coefficients from partial correlation analyses in the VP/VLBW sample are given. TIV, scanner, and sex served as covariates.

T A B L E 4 Hippocampus subfield volumes and cognitive function
Results were FDR-corrected for multiple comparisons (6)  that left-sided hippocampus volume was associated with verbal IQ and right-sided hippocampus volume with performance IQ (Amat et al., 2008). However, in the light of other studies investigating structural correlates of cognitive performance in the general population, it seems that-in case of an undamaged and well-developed hippocampus-its volume is not sufficient to describe the variance in FS-IQ scores (Gregory et al., 2016). The link between hippocampus and cognitive performance seems clearer in the context of premature birth, where prior evidence exists regarding the correlation of hippocampus structure and cognitive performance (Aanes et al., 2019;Sølsnes et al., 2015;Strahle et al., 2019;Thompson et al., 2013). In line with these findings, we found correlations between all left-sided functional hippocampus units (i.e., SC, HP, and DG) and FS-IQ. In contrast to these left-sided findings, the associations between right-sided functional hippocampus units and FS-IQ were not statistically significant. This argues for a lateralization of hippocampus function in our cohort of VP/VLBW adults and is in line with a previous report on premature-born young adults . In this study, whole hippocampus volumes of the left side correlated with a higher number of memory test scores than right whole hippocampus volumes . The differential pattern of significant correlations was explained by different functional properties of left and right hippocampi known from the general population with higher involvement of the left hippocampus in verbal and more general episodic memory, while the right hippocampus is more involved in spatial memory tasks (Burgess, Maguire, & O'Keefe, 2002;Suthana, Ekstrom, Moshirvaziri, Knowlton, & Bookheimer, 2011). This possible explanation may also apply to the lateralized finding in our study, assuming that typically "left hippocampal capacities" contribute more to general IQ than "right hippocampal capacities." Another explanation for lateralized hippocampus function in our cohort may be that the left hippocampus is more severely affected by premature birth to a degree that volume loss becomes relevant for cognitive functioning. Although we observe bilateral hippocampal volume reductions, there is a trend toward increased volume reduction of the left hippocampus compared to the right side.

| The DG mediates the impact of GA on adult FS-IQ
It is known that premature-born individuals are at increased risk for low cognitive performance throughout the life course which has been shown to be associated with several aspects of impaired brain development after premature birth (Ball et al., 2013;Hedderich, Bäuml, et al., 2019;Jaekel, Baumann, & Wolke, 2013;Lohaugen et al., 2010;Meng et al., 2016;Wolke, Johnson, & Mendonça, 2019 and contains a unique cell type, so-called granule cells, which show very sparse firing activity and outnumber cells in other compartments such as the entorhinal cortex (Amaral, Scharfman, & Lavenex, 2007;Hasselmo, 2011). In computational models, the DG serves as a "pattern separator" by diversifying consolidated input from the entorhinal cortex (Hasselmo, 2011). Moreover, the DG is a site of adult neurogenesis, a fact that is thought to help its pattern separation task in the context of learning (Aimone, Deng, & Gage, 2010;Becker, 2005). Specific functional significance of the DG has been also shown in the context of aging by mediating the impact of aging on pattern separation performance (Dillon et al., 2017). Moreover, in patients with amnestic mild cognitive impairment, pharmacological regulation of DG activity improved cognitive performance at a memory task (Bakker et al., 2012). However, in our cohort, that includes young adults with impaired neurodevelopment, the impact of DG on FS-IQ might be explained through an association with learning, since impaired learning processes might translate into lower global cognitive performance later on.

| The hippocampus and its subfields-A possible target for therapeutic interventions?
In our opinion, the impact of the DG as a pattern separator and site of adult neurogenesis on the association between GA and FS-IQ is of special importance. Given the multidimensional problem of impaired brain development, it is clear that HCSFs will only contribute a small part to the whole picture. Previous studies have found other structural brain aberrations correlating with adult cognitive performance, for example, global gray and white volumes or gyrification (Hedderich, Bäuml, et al., 2019;Nosarti et al., 2008). However, this differential contribution by HCSF is potentially significant since the hippocampus is involved in central cognitive-behavioral functions of the brain. Furthermore, the hippocampus could be an attractive treatment target.
First, its identification and volumetric measuring can be reliably performed across the lifespan, also in younger cohorts (Nosarti & Froudist-Walsh, 2016 shown that environmental enrichment (e.g., increasing stimuli for spatial navigation) had an effect on the hippocampus in mice that showed an increased number of neurons in the granule-cell layer of the DG (Kempermann, Kuhn, & Gage, 1997). Also, learning of associative tasks (e.g., navigation in a water-maze under different circumstances) which requires the hippocampus, increased adult neurogenesis in rats (Gould, Beylin, Tanapat, Reeves, & Shors, 1999). In humans, pattern separation, which is mainly carried out by the DG, can be tested explicitly by mnemonic similarity task (MST) which is used as a proxy for hippocampal integrity (Stark, Kirwan, & Stark, 2019). MST performance has been shown to improve with light aerobic exercise as well as through virtual environmental enrichment through specific exposure to 3D video games in young and older adults (Clemenson & Stark, 2015;Suwabe et al., 2017Suwabe et al., , 2018. Responsiveness of the human hippocampus to training interventions has been also demonstrated in humans using MRI-based volumetry. Erickson et al. have shown that physical aerobic exercise in healthy older individuals was associated with an increase of hippocampal volume, increased levels of brain-derived neurotrophic factor (which is thought to promote adult neurogenesis in the hippocampus), and improved performance in spatial short-term memory tasks (Erickson et al., 2011). Another study found an effect of cognitive training on left hippocampus activation and performance during verbal memory tasks in patients with mild cognitive impairment (Rosen, Sugiura, Kramer, Whitfield-Gabrieli, & Gabrieli, 2011). Taken together, it seems attractive to design treatment interventions targeted at hippocampus functions in premature-born individuals, in order to promote their development.

| Strengths and limitations
Some points should be carefully considered when interpreting our results. The current sample is biased to VP/VLBW adults with less severe neonatal complications, less functional impairments, and higher IQ. Individuals with stronger birth complications and/or severe lasting impairments in the initial BLS sample were more likely to be excluded for MRI screening due to specific MRI exclusion criteria (e.g., infantile cerebral palsy, the inability to lie still hypothetically due to underlying ADHD, or epilepsy) and to reject MRI assessment due to their stronger impairments and related lower level of general activity. Thus, differences in HCSF volumes between VP/VLBW and term-born control adults reported here are conservative estimates of true differences. Classically, neurocognitive tests focusing on declarative memory are used in order to test hippocampus function clinically. However, no dedicated memory tests such as the MST were acquired in our study and we used FS-IQ scores to correlate HCSF volumes with general cognitive performance. In our opinion, this is justified due to the large amount of hippocampus functions apart from memory encoding and retrieval such as spatial navigation or learning. It seems reasonable that especially impaired learning mechanisms might lead to impaired FS-IQ in adulthood after premature birth. However, this rather indirect connection must be noted as a limitation. One of the strengths of our study is its large sample size (103 VP/VLBW and 109 FT adults), which enhances the generalizability of our findings.
Moreover, the homogeneous and small age range of the study population in adulthood excludes effects of chronological age differences. Another strength of our study is that high-quality brain imaging was performed so that both high-resolution T1-weighted and T2-weighted input images could be used for HCSF segmentation. Advances in automated segmentation pipelines have made this study possible because the high workload of manual delineations severely limits sample size in brain imaging studies. However, several concurrent automated segmentation protocols have been published with slight differences with regard to subfield definition (Mueller et al., 2018).

| Conclusion
Our results show a general reduction of HCSF volumes in adulthood after premature birth, associated with low GA and high neonatal treatment intensity, not BW. We propose a potential dominating effect of a shifted developmental trajectory after premature birth over a special vulnerability of HCSF. Moreover, we demonstrated a significant mediation effect of left DG on the relationship between low GA and adult FS-IQ as well as for long-term cognitive development, underlining the functional relevance of this HCSF that is important for pattern separation and adult neurogenesis. In our opinion, this underlines that the hippocampus may constitute a potential treatment target for interventions aiming at improving long-term cognitive outcomes after premature birth. Future studies will be needed in order to integrate several modules of impaired brain development after premature birth into a more complete framework with respect to social, biological, and psychological factors.

ACKNOWLEDGMENTS
The authors thank all current and former members of the Bavarian

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
Patient data used in this study are not publicly available but stored by the principal investigators of the Bavarian Longitudinal Study.