Decreased amygdala volume in adults after premature birth

Premature-born infants have impaired amygdala structure, presumably due to increased stress levels of premature birth mediated by the amygdala. However, accounting for lifelong plasticity of amygdala, it is unclear whether such structural changes persist into adulthood. To address this problem, we stated the following questions: first, are whole amygdala volumes reduced in premature-born adults? And second, as adult anxiety traits are often increased after prematurity and linked with amygdala structure, are alterations in amygdala associated with adults’ anxiety traits after premature birth? We addressed these questions by automated amygdala segmentation of MRI volumes in 101 very premature-born adults (< 32 weeks of gestation and/or birth weight below 1500 g) and 108 full-term controls at 26 years of age of a prospectively and longitudinally collected cohort. We found significantly lower whole amygdala volumes in premature-born adults. While premature-born adults had significantly higher T score for avoidant personality reflecting increased social anxiety trait, this trait was not correlated with amygdala volume alterations. Results demonstrate reduced amygdala volumes in premature born adults. Data suggest lasting effects of prematurity on amygdala structure.

Premature-born infants have impaired amygdala structure, presumably due to increased stress levels of premature birth mediated by the amygdala. However, accounting for lifelong plasticity of amygdala, it is unclear whether such structural changes persist into adulthood. To address this problem, we stated the following questions: first, are whole amygdala volumes reduced in prematureborn adults? And second, as adult anxiety traits are often increased after prematurity and linked with amygdala structure, are alterations in amygdala associated with adults' anxiety traits after premature birth? We addressed these questions by automated amygdala segmentation of MRI volumes in 101 very premature-born adults (< 32 weeks of gestation and/or birth weight below 1500 g) and 108 full-term controls at 26 years of age of a prospectively and longitudinally collected cohort. We found significantly lower whole amygdala volumes in premature-born adults. While premature-born adults had significantly higher T score for avoidant personality reflecting increased social anxiety trait, this trait was not correlated with amygdala volume alterations. Results demonstrate reduced amygdala volumes in premature born adults. Data suggest lasting effects of prematurity on amygdala structure.

Results
Sample characteristics. Table 1 presents group demographic and clinical background variables. The VP/VLBW group and the FT group did not differ significantly regarding sex (p = 0.894) and age at scanning (p = 0.165). By design of the study, VP/VLBW subjects had significantly lower GA (p < 0.001) and lower birth weight (BW) (p < 0.001) compared to FT controls. Furthermore, TIV was significantly smaller in VP/VLBW individuals compared to controls (p = 0.001).

Reduced amygdala volumes in very premature-born adults. Automated FreeSurfer-based segmen-
tation of the amygdala in structural MRI data is visualized in Fig. 1. First, we tested whether whole amygdala volumes are altered in VP/VLBW adults compared to FT controls using general linear models. Total intracranial volume (TIV), sex and scanner were entered as covariates. We found significantly lower left (p < 0.001) and right whole amygdala volumes (p < 0.001) in VP/VLBW individuals compared to controls. Figure 2a and Table S1 show estimated marginal means and p-values. In order to control for an impact of intraventricular hemorrhage on our results, we removed these subjects (see Table S3) and repeated general linear model analyses for left and right whole amygdala volumes between the remaining subjects of the VP/VLBW group and the FT group. We found significantly lower left (p < 0.001) and right whole amygdala volume (p = 0.001) in VP/VLBW subjects without intraventricular hemorrhage compared to FT controls (see Table S4), verifying that amygdala volume is lastingly reduced after premature birth. These results indicate that our main findings of reduced amygdala volume were not affected by effects of intraventricular hemorrhage.
Second, as linked brain regions may develop in concert for their structural features and as such related structural development is altered in human prematurity for selected groups of brain regions as indicated by altered structural covariance 50-52 , we investigated related volumetric features among the left and right amygdalae by a structural covariance approach. www.nature.com/scientificreports/ In order to investigate structural covariance across hemispheres, we tested the correlation between left and right whole amygdala volume. The correlation between left and right whole amygdala volume was significant in both the VP/VLBW group (r = 0.865, p < 0.001) and the FT controls (r = 0.778, p < 0.001), indicating related whole amygdala development across hemispheres. The correlation did not differ significantly between the two groups (p = 0.052). However, it approximated statistical significance towards stronger correlation in the VP/VLBW group compared to FT controls, possibly suggesting that related whole amygdala development across hemispheres could be affected by prematurity. Further studies are needed to shed more light on structural covariance within the amygdalae. Table S5 presents correlation coefficients and 95% confidence intervals from the partial correlation analyses, and the p-value from comparing correlation coefficients between VP/VLBW subjects and FT controls.
Third, in order to verify that amygdala volume changes were indeed related to premature birth, we investigated whether group differences of amygdala volumes between VP/VLBW subjects and FT controls are specifically related to premature birth using partial correlation analyses with TIV, sex and scanner as covariates. After FDR correction for multiple comparisons, we observed a significant positive correlation between GA and both left (r = 0.367, p < 0.001) and right whole amygdala volume (r = 0.317, p = 0.002). While there was no significant relationship between BW and left whole amygdala volume (r = 0.106, p = 0.304), BW and right whole amygdala volume showed a significant positive correlation (r = 0.220, p = 0.031). We found significant negative correlations between duration of ventilation and both left (r = -0.428, p < 0.001) and right whole amygdala volumes (r = -0.413, p < 0.001). Figure 2b shows the relationship between whole amygdala volumes and variables of premature birth, and Table S6 present correlation coefficients, 95% confidence intervals and p-values from the partial correlation analysis between whole amygdala volumes and variables of premature birth.
In summary, our findings support the hypothesis that amygdala volumes are altered in VP/VLBW subjects compared to FT controls and that these volume reductions are specifically related to premature birth.
Increased social anxiety is not associated with reduced amygdala volumes in premature-born adults. To investigate whether premature-born adults had increased anxiety traits, we studied group differences of YASR anxiety and avoidant personality scores by the use of two-samples t-tests. Anxiety scores did not differ significantly between VP/VLBW individuals and FT controls (p = 0.097). Avoidant personality scores were significantly higher in VP/VLBW individuals compared to FT controls (p = 0.001), indicating increased social anxiety traits in premature-born adults. Figure 3a shows mean T scores as bar charts and Table 1 presents T scores and p-values.
In order to study whether reduced amygdala volumes are relevant for increased social anxiety in prematureborn adults, we investigated the relationship between amygdala volumes and the YASR avoidant personality score using partial correlation analyses with TIV, sex and scanner as covariates. There was no significant correlation between the T score for avoidant personality and left (r = 0.120, p = 0.245) or right whole amygdala volume (r = 0.064, p = 0.537). Figure 3b shows the relationship between whole amygdala volume and avoidant personality score and Table S7 presents correlation coefficients, 95% confidence intervals and p-values.
These results indicate that VP/VLBW individuals have increased avoidant personality traits, which reflect increased social anxiety; but it seems unlikely that altered amygdala volumes contribute to this trait.

Discussion
Based on structural MRI, we demonstrated that whole amygdala volumes are decreased in VP/VLBW subjects compared to FT controls at 26 years of age. Furthermore, while VP/VLBW individuals had higher avoidant personality T scores, i.e., they had stronger social anxiety trait, there was no significant association between altered amygdala volumes and social anxiety. To the best of our knowledge, these results demonstrate for the first time decreased amygdala volume in adults of premature birth. Data suggest that prematurity has lasting effects on amygdala volume.
Reduced amygdala volumes in premature-born adults. We found significantly reduced whole amygdala volumes in VP/VLBW subjects compared to FT controls. Findings in neonates and children after premature birth are mostly in line with our results. At term-equivalent age, reduced amygdala volumes compared to FT controls were reported by Cismaru et al. 37 . While late preterm children showed no significant difference in amygdala volume 53 , children born between 26 and 33 weeks of GA had smaller amygdala volume compared to term controls at school age 36 . www.nature.com/scientificreports/ Furthermore, coordinated developmental processes may be reflected by structural covariance [50][51][52]54 . More specifically, after premature birth, reduced as well as increased covariance has been reported between brain volumes including cortical areas, thalamus and cerebellum 51,52 . We found significant correlation between left and right whole amygdala volume in both the VP/VLBW group and the FT controls, indicating related whole amygdala development. While the correlation did not differ significantly between the two groups, it approximated statistical significance towards stronger correlation in the VP/VLBW group compared to FT controls (p = 0.052). This result might suggest that related whole amygdala development across hemispheres could be affected by prematurity. Further studies are needed to shed more light on structural covariance within the amygdalae.
To test whether differences in amygdala volumes between VP/VLBW subjects and FT controls are specifically related to premature birth, we correlated amygdala volumes with variables of premature birth. In general, amygdala volume was consistently related to GA and duration of ventilation, and partly with BW. Our results are in line with findings from other studies at term equivalent age and in children, which found significant correlation between whole amygdala volumes and GA 38,55 . Furthermore, Chau et al. 38 reported that lower amygdala volume at 8 years was associated with more neonatal invasive procedures and more days of ventilation. Consistent with these findings, our data also showed strong negative correlation of amygdala volume in early adulthood with days of ventilation. This relationship might reflect vulnerability of amygdala to stress exposure induced by premature birth. Scheinost et al. 56 reporting that also prenatal stress altered amygdala functional connectivity in preterm neonates furthermore supports that stress exposure modulates amygdala development.
In conclusion, our data suggest that amygdala volume is lastingly affected by premature birth.
Increased social anxiety is not associated with reduced amygdala volumes in premature-born adults. VP/VLBW individuals showed stronger trait social anxiety, as avoidant personality T scores were significantly higher in the VP/VLBW group compared to the FT group. However, there was no significant association between altered amygdala volumes and social anxiety. These results are in line with Johnson et al. 57 who found higher scores on avoidant personality scales in extremely preterm adults. Furthermore, they are in line with findings that VP/VLBW are less extraverted and more withdrawn 47,48 . On the other hand, we found no support for the hypothesis that observed reductions in www.nature.com/scientificreports/ amygdala volumes after premature birth may be associated with anxiety traits. There are several possible reasons why we found no association between increased social anxiety and amygdala volume. First, results about the relevance of amygdala volume for anxiety are heterogeneous. For example, in patients with panic disorder, amygdala volume was significantly smaller compared to controls which was associated with anxiety 58 . Also, in healthy adults an inverse correlation between anxiety measures and volume of the amygdala has been described 46 .
Conversely, in children with generalized anxiety disorder, amygdala volume was significantly larger compared to controls 59 . More specifically, enlargement of basolateral amygdala was associated with high childhood anxiety 60 . Likewise, adults with generalized anxiety disorder had larger amygdala volumes 61 . Furthermore, a number of studies found no relationships between personality traits and amygdala volume in relatively healthy adults 62,63 . Second, the amygdala is a region in which highly complex integration takes place 64 and different aspects of amygdala structure and function contribute to social anxiety. For example, altered structural and functional connectivity of the amygdala have been linked to anxiety 60,65-68 . In particular, social impairments in preterm-born adolescents were associated with functional connectivity of the amygdala, supporting a possible relationship between prematurity, social anxiety and the amygdala 49 . Third, there are several systems contributing to anxiety besides the amygdala [69][70][71] . For example, apart from the relationship with amygdala volume, Spampinato et al. 46 found an inverse correlation between anxiety measures and volumes of other regions of the limbic system as well as prefrontal cortex in healthy adults. Lastly, while the 95% confidence intervals reported in Table S7 suggest that we can reject the hypothesis that there is a medium to strong correlation between amygdala volume and social anxiety, there is a possibility that there is only weak correlation which we were not able to detect.
In conclusion, further investigations exploring other modalities and other systems as well as their connectivity patterns are necessary to elucidate neural correlates of social anxiety after premature birth.

Strengths and limitations.
One of the strengths of our study is a large sample size (101 VP/VLBW and 108 FT adults) that enhances the generalizability of our findings. We showed that the sample mean of our data is likely to be close to the 'true' population mean and that uncertainty in the estimation of amygdala volumes is relatively low as indicated by small standard errors and narrow 95% confidence intervals (see Tables S1, and S2).
Another strength of our study is that VP/VLBW subjects and FT controls had the same mean age of 26 years. Hence, a relevant impact of patient age on amygdala volumes at the time of the MRI scan is excluded.
Third, segmentation quality was improved as both high-resolution T1-weighted and T2-weighted images were used for amygdala segmentation.
One limitation of this study is that structural MRI of this dataset has previously been analyzed (see methods section). We are aware of the proposed problem of dataset decay-analog to open datasets-increasing the risk www.nature.com/scientificreports/ of false-positives 72 . However, we analyze this large, prospectively collected cohort of premature-born adults and healthy controls to investigate biologically motivated questions and focused hypotheses. With multiple brain systems affected in human prematurity, it is a first step to focus on distinct brain regions, and a second step to then integrate previous findings, hopefully improving the understanding of brain development after premature birth. Second, brain structure and its relationship with variables of premature birth as well as social anxiety trait is influenced by multiple other individual, social and environmental factors. Therefore, results have to be interpreted with care.
Third, since individuals with more birth complications in the initial Bavarian Longitudinal Study sample were more likely to be excluded in the initial screening for MRI due to exclusion criteria for MRI, the current sample is biased to VP/VLBW adults with less severe neonatal complications. Therefore, the observed differences in amygdala volumes between VP/VLBW subjects and FT controls are conservative estimates of true differences. Nevertheless, the sample with MRI data was still representative of the full cohort in terms of GA and BW. Mean GA and BW were not significantly different in VP/VLBW subjects with MRI data compared to subjects without MRI data (see Table S8). There were few subjects with intraventricular hemorrhage in the neonatal period (see Table S3). To investigate whether removing subjects with intraventricular hemorrhage impacts the results, we repeated general linear model analyses for left and right whole amygdala volumes (see Table S4) between the remaining subjects of the VP/VLBW group and the FT group. These results indicate that our main findings of reduced amygdala volume were not affected by effects of intraventricular hemorrhage.

Conclusions
Amygdala volume is lastingly reduced after premature birth. Furthermore, premature-born individuals showed stronger trait social anxiety, however, there seems to be no association between altered whole amygdala volumes and social anxiety.

Methods
Participants. Our study sample was previously described in 12,73-75 : All subjects were part of the Bavarian Longitudinal Study (BLS), a geographically defined, whole-population sample of neonatal at-risk children and healthy FT controls who were followed from birth, between January 1985 and March 1986, into adulthood [73][74][75] . 682 infants were born VP (< 32 weeks of gestation) and/or with very low birth weight (VLBW, birth weight < 1500 g). Informed consent from a parent and/or legal guardian was obtained. From the initial 916 FT born infants born at 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 in order to be comparable with the VP/VLBW sample. Of these, 411 VP/VLBW individuals and 308 controls were eligible for the 26-year follow-up assessment. 260 from the VP/VLBW group and 229 controls participated in psychological assessments 13 . All subjects were screened for MR-related exclusion criteria including (self-reported): claustrophobia, inability to lie still for > 30 min, unstable medical conditions (e.g. severe asthma), epilepsy, tinnitus, pregnancy, non-removable MRI-incompatible metal implants and a history of severe CNS trauma or disease that would impair further analysis of the data. However, the most frequent reason not to perform the MRI exam was a lack of motivation. Finally, 101 VP/VLBW subjects and 111 FT controls underwent MRI at 26 years of age (see Fig. S1). The MRI examinations took place at two sites: The Department of Neuroradiology, Klinikum rechts der Isar, Technische Universität München, (n = 145) and the Department of Radiology, University Hospital of Bonn (n = 67). The study was carried out in accordance with the Declaration of Helsinki and was approved by the local ethics committee of the Klinikum rechts der Isar, Technische Universität München and the University Hospital Bonn. All study participants gave written informed consent. They received travel expenses and a small payment for participation.
Birth variables. GA in weeks was estimated from maternal reports on the first day of the last menstrual period and serial ultrasounds during pregnancy. In cases in which the two measures differed by more than 2 weeks, clinical assessment at birth with the Dubowitz method was applied 77 . BW in grams was obtained from obstetric records. Duration of mechanical ventilation in days was computed from daily records by research nurses.
Variables related to anxiety. To assess behavioral and emotional outcome related to anxiety, we used the German version of the Young Adult Self Report (YASR) 78 . We chose two (Anxiety and Avoidant personality) of six (Depressive, Anxiety, Somatic, Avoidant personality, Attention deficit/hyperactivity problems, and Antisocial personality) included DSM-IV-oriented scales, which are related to anxiety and social anxiety. Avoidant personality disorder and social anxiety disorder share similar symptoms, such as anxiety in social situations and avoidance of these situations up to self-isolation, and diagnostic criteria largely overlap 79 . We used T scores of the scales as outcome measures. A T score in psychometric testing is a standardized measurement with an average score of 50 and a standard deviation of 10.
MRI data acquisition. MRI data acquisition was previously described in 9,12 : At both sites, Bonn and Munich, MRI data acquisition was performed on Philips Achieva 3T TX systems or Philips Ingenia 3T system using an 8 www.nature.com/scientificreports/ as covariates of no interest. Across all scanners, sequence parameters were kept identical. Scanners were checked regularly to provide optimal scanning conditions and MRI physicists at the University Hospital Bonn and Klinikum rechts der Isar regularly scanned imaging phantoms, to ensure within-scanner signal stability over time.
MRI processing and amygdala segmentation. Images saved as DICOMs were converted to Nifti-format using dcm2nii 80 . MRI data were segmented using the FreeSurfer image analysis suite, version 6.0 (http:// surfe r.nmr.mgh.harva rd.edu/). It includes a tool to produce an automated segmentation of the amygdala 81 . This atlas was created based on Bayesian inference using ultra high-resolution ex-vivo MRI data and evaluated by applying it to in-vivo MRI data of patients with Alzheimer's Disease and patients with Autism Spectrum disorders as well as healthy controls 81 . Using both high-resolution T1-weighted and T2-weighted images, the function segmentHA_T2.sh generates whole amygdala volumes for the left and right hemisphere, respectively. Output segmentations were inspected visually. Amygdala segmentation is shown in Fig. 1. Successful segmentation of the amygdala was available in 101 VP/VLBW subjects and 109 FT subjects (see Fig. S1). Estimation of TIV was performed with the CAT12 toolbox, version r1364 (http://www.neuro .uni-jena.de/ cat/) 82 within SPM12 (https ://www.fil.ion.ucl.ac.uk/spm/softw are/spm12 /).

Statistical analysis.
To ensure statistical power, we conducted an a priori power analysis using G*Power 83 .
A previous study investigating amygdala volumes after preterm birth at school age found smaller amygdala volume compared to term controls observing large effect sizes 36 . However, since this sample is only partly comparable to our sample, we chose a smaller, medium effect size to calculate sample size. Aiming at obtaining statistical power of 90% at 0.05 level of significance. a sample size of 171 would have sufficed to detect statistically significant results. Hence, our sample size is a very conservative estimate.
All statistical analyses were performed using IBM SPSS Version 26 (IBM Corp., Armonk, NY, USA). First, we checked the data for outliers using a method proposed by Hoaglin and Iglewicz 84 . The interquartile range was multiplied by the factor 2.2. One FT subject was excluded from analyses because left whole amygdala volume was an outlier value. Finally, 101 VP/VLBW subjects and 108 FT subjects were included in the analyses (see Fig. S1).
Age was not included as a covariate in our analyses, as VP/VLBW subjects and FT controls had the same mean age of 26 years (p = 0.165).

Group comparisons of amygdala volumes.
To test whether amygdala volumes are significantly altered in VP/VLBW individuals compared to FT controls, general linear models were used. We entered left and right whole amygdala volumes as dependent variables, group membership as fixed factor and TIV, sex and scanner as covariates.
There were few subjects with intraventricular hemorrhage in the neonatal period (see Table S3). To investigate whether removing these subjects impacts the results, we repeated general linear model analyses for left and right whole amygdala between the remaining subjects of the VP/VLBW group and the FT group. To test whether amygdala volumes are significantly altered in VP/VLBW individuals without intraventricular hemorrhage compared to FT controls, we entered left and right whole amygdala volumes as dependent variables, group membership as fixed factor and TIV, sex and scanner as covariates.
To study interrelated development among left and right amygdala, we investigated structural covariance. This approach is motivated by the general notion that linked brain regions may develop in concert, including volumetric features, and that such concerted development seems to be affected by prematurity [50][51][52] . To investigate structural covariance in the amygdala, we tested the correlation between amygdala volumes. We entered left whole amygdala volume and right whole amygdala volume as variables of interest into a two-tailed partial correlation analysis in each group (VP/VLBW and FT) separately. TIV, sex and scanner were entered as covariates.
Furthermore, we investigated difference in structural covariance between VP/VLBW subjects and FT controls. Using Fisher r-to-z transformation, we calculated the z-score and p-value to assess the significance of the difference 85 .
To test whether differences in amygdala volumes between VP/VLBW subjects and FT controls are specifically related to premature birth, we correlated amygdala volumes with GA, BW and duration of ventilation as variables of premature birth. Specifically, in the VP/VLBW group, we entered left and right whole amygdala volumes, respectively, and GA, BW and duration of ventilation as variables of interest and TIV, sex and scanner as covariates into a two-tailed partial correlation analysis.
All analyses were FDR corrected for multiple comparisons using the Benjamini-Hochberg procedure 86 . Statistical significance was defined as p < 0.05. 95% confidence intervals for partial correlation analyses were obtained using a bootstrap approach (with 5000 repetitions) in SPSS.

Correlation of amygdala volumes and anxiety.
To investigate the relationship between altered amygdala volumes and anxiety, we used the anxiety and avoidant personality subscales of the YASR DSM-IV-oriented scales. First, we compared YASR anxiety and avoidant personality scores between the VP/VLBW group and the FT control group using a two-sample t-test. www.nature.com/scientificreports/ Second, in the VP/VLBW group, we entered left and right whole amygdala volumes, respectively, and anxiety scores (i.e., the avoidant personality T scores) as variables of interest and TIV, sex and scanner as covariates into a two-tailed partial correlation analysis. The partial correlation analysis was FDR corrected for multiple comparisons using the Benjamini-Hochberg procedure 86 . Statistical significance was defined as p < 0.05. 95% confidence intervals were obtained using a bootstrap approach (with 5000 repetitions) in SPSS.

Data availability
Patient data used in this study are not publicly available but stored by the principal investigators of the Bavarian Longitudinal Study. www.nature.com/scientificreports/