Poor Brain Growth in Extremely Preterm Neonates Long Before the Onset of Autism Spectrum Disorder Symptoms

Abstract

Preterm infants face an increased risk of autism spectrum disorder (ASD). The relationship between autism during childhood and early brain development remains unexplored. We studied 84 preterm children born at <27 weeks of gestation, who underwent neonatal magnetic resonance imaging (MRI) at term and were screened for ASD at 6.5 years. Full-scale intelligence quotient was measured and neonatal morbidities were recorded. Structural brain morphometric studies were performed in 33 infants with high-quality MRI and no evidence of focal brain lesions. Twenty-three (27.4%) of the children tested ASD positive and 61 (72.6%) tested ASD negative. The ASD-positive group had a significantly higher frequency of neonatal complications than the ASD-negative group. In the subgroup of 33 children, the ASD infants had reduced volumes in the temporal, occipital, insular, and limbic regions and in the brain areas involved in social/behavior and salience integration. This study shows that the neonatal MRI scans of extremely preterm children, subsequently diagnosed with ASD at 6.5 years, showed brain structural alterations, localized in the regions that play a key role in the core features of autism. Early detection of these structural alterations may allow the early identification and intervention of children at risk of ASD.

autism, brain development, extreme prematurity, structural MRI

Introduction

When Leo Kanner became the first clinician to identify autism in 1943 (Kanner 1943), he thought it was partially due to “genuine lack of maternal warmth.” This idea has been completely discarded and there is now a consensus that there is a connection between genetic heritability and autism spectrum disorder (ASD; Sandin et al. 2014). Prenatal and perinatal risk factors also appear to play a role independent of genetic susceptibility (D'Onofrio et al. 2013; Ben-Ari 2015). It has actually been proposed that the high rates of autism shared by monozygotic twins might be largely due to them sharing both a prenatal and perinatal environment (Hallmayer et al. 2011).

Children who are born extremely preterm face a dramatically elevated risk of ASD, ranging from 15.8 to 26% (Limperopoulos et al. 2008; Johnson et al. 2010), and the risk is inversely related to the child's gestational age at birth. These infants face high exposure to stressors during a critical period of brain development, which may play a key role in the development of autism (Casanova 2007). In this context, early exposure to an abnormal environment adversely affects brain development (Smith et al. 2011; Brummelte et al. 2012; Pineda et al. 2014) and contributes to impaired socio-emotional maturation and later neurodevelopmental alterations (Rogers et al. 2012).

Cerebellar lesions have been reported in preterm infants at risk of developing autism (Limperopoulos et al. 2008) and studies of adolescents with autistic traits have demonstrated white matter reductions with ventricular enlargement (Skranes et al. 2007). However, the relationship between altered neonatal brain volumes and the risk of ASD in extremely preterm children remains unexplored. This study used 3 methods—automatic segmentation, atlas-based segmentation, and voxel-based morphometry—to provide a comprehensive global and regional characterization of the neonatal brain. Our aim was to test the hypothesis that structural brain alterations could be identified in the neonatal period in extremely preterm infants who later developed ASD and prominent autism traits at 6.5 years and to compare them with a group of extremely preterm infants who showed typical development.

Materials and Methods

Patients

This study was a regional cohort of a national prospective research program on extreme prematurity that involved neonatal assessment and long-term follow-up (Serenius et al. 2013). We recruited 111 preterm children of <27 weeks of gestation born in Stockholm County, Sweden, from 2004 to 2007. All of the subjects successfully underwent magnetic resonance imaging (MRI) at term-equivalent age. Infants with congenital infections and malformations were excluded. When it came to the morphometric studies, we did not include infants with periventricular leukomalacia or intraventricular hemorrhage grade III–IV, focal brain lesions (cysts and malformations), persistent ventricular dilatation, or moderate or severe white matter abnormalities qualitatively defined on MRI examination at term-equivalent age (Skiold et al. 2010) (Fig. 1). Despite these efforts, we are fully aware that microscopic white matter lesions not readily seen by neuroimaging might be present in the subsample (Volpe 2009).

Figure 1.

Flow diagram of study participants. ASD, autism spectrum disorder; SRS, Social Responsiveness Scale (positive screening t-score ≥60); MRI, magnetic resonance imaging.

Data were collected prospectively from the neonatal period and at the age of 6.5 years (Table 1). The regional ethical review board in Stockholm granted ethical approval for the study and written consent was obtained from the parents of the children.

Table 1

Characteristics of the groups

Characteristics	Whole sample		P-value	Good quality MRI		P-value
	ASD positive (N = 23)	ASD negative (N = 61)		ASD positive (N = 11)	ASD negative (N = 22)
Gestational age at birth, mean (SD)	25.1 (1.1)	25.5 (0.1)	NS	25.2 (1.1)	25.6 (0.1)	NS
Birth weight [g, mean (SD)]	718.8 (148.5)	839.97 (152.2)	0.002	728.3 (129.3)	871.7 (160.15)	NS
z-score	−0.87 (0.10)	−0.79 (0.10)	0.002	−0.49 (0.08)	−0.43 (0.10)	NS
HC at birth, mean (SD)	22.7 (1.4)	24.0 (1.5)	0.001	23.0 (1.2)	24.1 (1.6)	NS
z-score	−0.52 ± 0.91	0.25 ± 0.96	0.001	−0.52 (0.25)	−0.34 (0.31)	0.11
Gender (male), n (%)	12 (52.5)	34 (55.7)	NS	8 (72.7)	10 (45.4)	NS
Small for gestational age, n (%)	3 (13.0)	4 (6.6)	NS	1 (9.1)	—	NS
Prenatal steroids, n (%)	21 (91.3)	55 (91.8)	NS	10 (90.9)	22 (100)	NS
Cesarean delivery, n (%)	12 (52.8)	27 (44.3)	NS	6 (54.5)	8 (36.4)	NS
Premature rupture of membranes, n (%)	4 (17.4)	18 (29.5)	NS	3 (27.3)	5 (22.7)	NS
Maternal signs of infection, n (%)	5 (21.7)	24 (39.3)	NS	4 (36.4)	6 (27.3)	NS
CRIB score, mean (SD)	7.09 (3.3)	5.38 (3.3)	NS	6.2 (4.1)	4.1 (3.1)	NS
Inotropic support, n (%)	13 (56.5)	19 (31.1)	0.04	6 (54.5)	9 (40.9)	NS
Duration of ventilation, n (%)	23.2 (17.7)	10.2 (10.5)	0.003	16.5 (15.6)	8.2 (10.7)	0.08
Postnatal steroids, n (%)	6 (26.1)	8 (13.1)	NS	2 (18.2)	2 (9.1)	NS
Full enteral nutrition [days, mean (SD)]	34.5 (25.2)	22.3 (10.3)	0.04	33.3 (21.5)	18.3 (9.5)	0.01
Bronchopulmonary dysplasiaa, n (%)	16 (69.6)	20 (32.8)	0.003	6 (54.5)	7 (31.8)	NS
Necrotizing enterocolitis, any grade, n (%)	7 (30.4)	5 (8.2)	0.01	3 (27.3)	—	0.03
Retinopathy of prematurity (ROP) >2, n (%)	13 (56.5)	19 (31.1)	0.04	6 (54.5)	3 (13.6)	0.03
ROP, laser-treated, n (%)	7 (30.4)	7 (11.5)	0.05	4 (36.3)	—	0.008
Patent ductus arteriosus (PDA), ibuprofen, n (%)	15 (65.2)	40 (65.5)	NS	9 (81.8)	18 (81.8)	NS
PDA, surgical ligation, n (%)	9 (39.1)	16 (26.2)	NS	4 (36.4)	7 (31.8)	NS
PDA surgical ligation and ROP laser-treated	7 (30.4)	3 (4.9)	0.004	4 (36.4)	—	0.008
Intraventricular hemorrhage grade 1–2	8 (34.8)	20 (32.8)	NS	3 (27.3)	8 (36.4)	NS
Intraventricular hemorrhage grade 3–4	2 (8.7)	7 (11.5)	NS	—	—	—
Periventricular leukomalacia, n (%)	2 (8.7)	—	0.07	—	—	—
Cerebellar alterations, n (%)	2 (8.7)	4 (6.6)	NS	1 (10.0)	2 (9.1)	NS
Severe white matter abnormalities, n (%)	2 (8.7)	4 (6.6)	NS	—	—	—
Term-equivalent age
Age at scan [weeks, mean (SD)]	40.9 (1.5)	40.7 (1.2)	NS	40.8 (0.1)	40.9 (0.1)	NS
Weight at scan [g, mean (SD)]	2715.3 (456.5)	2952.6 (447.2)	NS	2688.0 (399.3)	3023 (445.8)	NS
HC at scan [cm, mean (SD)]	33.8 (1.8)	34.8 (1.5)	0.02	34.9 (1.96)	35.2 (1.55)	NS
6.5 years
Age at assessment, [months, mean (SD)]	77.5 (1.3)	77.9 (2.0)	NS	76.9 (1.31)	77.5 (1.59)	NS
Weight [g, mean (SD)]	20.5 (3.7)	20.4 (3.5)	NS	22.0 (4.5)	21.5 (3.4)	NS
Head circumference, mean (SD)	49.9 (1.7)	51.4 (1.6)	0.001	50.9 (1.8)	51.6 (1.7)	NS
Full-scale intelligence	75.6 (24.8)	85.3 (16.7)	0.07	69.1 (24.2)	87.0 (14.7)	0.09
Full-scale IQ <70, n (%)	8/20 (40.0)	1/60 (1.8)	<0.001	7/11 (63.6)	0/22	<0.001

CRIB, clinical risk index for babies; IQ, full-scale intelligence quotient; ASD, autism spectrum disorder; MRI, magnetic resonance imaging; HC, head circumference.

^aOxygen at age of 36 weeks.

Developmental Assessment

Autistic traits were assessed at 6.5 years of age using the Social Responsiveness Scale (Constantino et al. 2003). Clinical diagnosis of ASD was based on the diagnostic criteria applied in the Diagnostic and Statistical Manual of Mental Disorders—4th edition, and/or the International Classification of Diseases—10th revision. Full-scale intelligence quotient (IQ) was measured using the Wechsler Intelligence Scale for Children—4th edition. A child was considered ASD positive if he or she had a score of at least 60 on the Social Responsiveness Scale and/or a clinical diagnosis of ASD. The examiners were blinded to the diagnosis of the participants.

MRI Data Acquisition

MRI was performed at term-equivalent age without sedation with T₁- and T₂-weighted images using a Philips Intera 1.5-T MRI system (Philips International, Amsterdam, The Netherlands). The conventional MRI protocol consisted of a sagittal T₁-weighted turbo spin-echo sequence, an axial inversion recovery sequence, and an axial T₂-weighted sequence. Details of the sequence parameters have previously been published (Skiold et al. 2010). All the MRI scans were examined for anatomical defects by an experienced neuroradiologist and the structural scans were visually assessed for qualitative white matter abnormalities (Skiold et al. 2010). Cerebellar abnormalities, namely the presence of hemosiderin or abnormal signal intensity, were evaluated. The imaging data were checked for quality as previously described (Padilla et al. 2015). The reasons for excluding MRI scans from the analyses are presented in Figure 1. Clinical reasons for exclusion in both the ASD-positive and ASD-negative subjects were: focal lesions (2/12 and 3/39), hydrocephaly (1/12 and 2/39), severe white matter abnormalities (1/12 and 4/39), and intraventricular hemorrhage III–IV (1/12 and 5/39). A total of 84 children, 23 ASD-positive and 61 ASD-negative, were evaluated. Of these, 11 ASD-positive children and 22 ASD-negative children were considered for further analysis after applying the quality control criteria outlined above, in order to ensure reliable structural MRI data. While this was a relatively small sample size, quality assurance was critical to enable us to obtain accurate data. The children that we included showed no evidence of brain abnormalities on their MRIs. The neonatal characteristics of both groups based on the ASD diagnosis and the quality of the MRI studies were similar (Table 2).

Table 2

Characteristics of the groups based on the ASD diagnosis and quality of the MRI studies

Characteristics	ASD positive		P-value	ASD negative		P-value
	HQ MRI (N = 11)	LQ MRI (N = 12)		HQ MRI (N = 22)	LQ MRI (N = 39)
Gestational age at birth, mean (SD)	25.2 (1.1)	25.0 (1.1)	NS	25.6 (0.1)	25.4 (1.0)	NS
Birth weight [g, mean (SD)]	728.3 (129.3)	664.3 (148.8)	NS	871.7 (160.15)	832.5 (154.85)	NS
z-score	−0.19 (1.07)	−0.30 (1.14)	NS	0.21 (0.95)	−0.04 (0.95)	NS
HC at birth, mean (SD)	23.0 (1.2)	22.4 (1.5)	0.05	24.1 (1.6)	23.9 (1.4)	NS
Gender (male), n (%)	8 (72.7)	4 (33.3)	NS	10 (45.4)	16 (41.0)	NS
Small for gestational age, n (%)	1 (9.1)	2 (16.7)	NS	—	4 (10.3)	NS
Prenatal steroids, n (%)	10 (90.9)	11 (91.7)	NS	22 (100)	33 (84.6)	NS
Cesarean delivery, n (%)	6 (54.5)	6 (50)	NS	8 (36.4)	18 (46.2)	NS
Premature rupture of membranes, n (%)	3 (27.3)	1 (8.3)	NS	5 (22.7)	14 (35.9)	NS
Maternal signs of infection, n (%)	4 (36.4)	1 (8.3)	NS	6 (27.3)	18/37 (48.6)	NS
CRIB score, mean (SD)	6.2 (4.1)	7.9 (2.3)	NS	4.1 (3.1)	6.1 (3.1)	0.02
Inotropic support, n (%)	6 (54.5)	10 (83.3)	NS	9 (40.9)	19 (48.7)	NS
Duration of ventilation, n (%)	16.5 (15.6)	29.4 (17.8)	NS	8.2 (10.7)	11.4 (10.3)	NS
Postnatal steroids, n (%)	2 (18.2)	4 (33.3)	NS	2 (9.1)	6 (15.4)	NS
Full enteral nutrition [days, mean (SD)]	33.3 (21.5)	35.5 (29.2)	NS	18.3 (9.5)	24.17 (10.4)	NS
Bronchopulmonary dysplasiaa, n (%)	6 (54.5)	10 (83.3)	NS	7 (31.8)	13/38 (34.2)	NS
Necrotizing enterocolitis, any grade, n (%)	3 (27.3)	4 (33.3)	NS	—	5 (12.8)	NS
Retinopathy of prematurity (ROP) >2, n (%)	6 (54.5)	7 (58.3)	NS	3 (13.6)	16 (41.0)	0.04
ROP, laser-treated, n (%)	4 (36.3)	3 (25.0)	NS	—	7/38 (18.4)	0.04
Patent ductus arteriosus (PDA), ibuprofen, n (%)	9 (81.8)	6 (50.0)	NS	18 (81.8)	21 (53.8)	0.02
PDA, surgical ligation, n (%)	4 (36.4)	5 (41.7)	NS	7 (31.8)	9 (23.1)	NS
PDA surgical ligation and ROP laser-treated	4 (36.4)	3 (25)	NS	—	3 (7.7)	NS
Intraventricular hemorrhage grade 1–2	3 (27.3)	5 (41.7)	NS	8 (36.4)	13 (33.3)	NS
Intraventricular hemorrhage grade 3–4	—	2 (16.7)	NS	—	6 (15.4)	NS
Periventricular leukomalacia, n (%)	—	2 (16.7)	NS	—	—	—
Cerebellar alterations, n (%)	1 (10.0)	1 (8.3)	NS	2 (9.1)	2 (5.1)	NS
Severe white matter abnormalities, n (%)	—	1 (8.3)	NS	—	4 (10.3)	NS
Term-equivalent age
Age at scan [weeks, mean (SD)]	40.8 (0.1)	41.0 (1.9)	NS	40.9 (0.1)	40.7 (1.3)	NS
Weight at scan, [g, mean (SD)]	2688.0 (399.3)	2738.1 (55.8)	NS	3023 (445.8)	2918.9 (448.1)	NS
HC at scan [cm, mean (SD)]	34.9 (1.96)	33.3 (1.5)	NS	35.2 (1.55)	34.6 (1.4)	NS
6.5 years
Age at assessment [months, mean (SD)]	76.9 (1.31)	77.9 (1.1)	NS	77.5 (1.59)	78.2 (2.35)	NS
Weight [g, mean (SD)]	22.0 (4.5)	19.4 (2.9)	NS	21.5 (3.4)	20.0 (3.3)	NS
Head circumference, mean (SD)	50.9 (1.8)	49.2 (1.3)	0.05	51.6 (1.7)	51.4 (1.5)	NS
Full-scale Intelligence	69.1 (24.2)	80.5 (25.6)	NS	87.0 (14.7)	83.6 (17.9)	NS
Full-scale IQ <70, n (%)	7/11 (63.6)	1/9 (11.1)	0.03	0/22	1/39	NS

CRIB, clinical risk index for babies; IQ, full-scale intelligence quotient; HQ, high quality; LQ, low quality; ASD, autism spectrum disorder; MRI, magnetic resonance imaging; HC, head circumference.

^aOxygen at age of 36 weeks.

Atlas-Based Segmentation

The whole brains of the included infants were divided into 90 anatomical regions, by using the automated anatomical labeling neonatal atlas (Shi et al. 2011) and adapting it to our sample. First, the intensity image, from which the neonatal atlas was generated, was registered to the T₁-weighted image of each infant. Second, the generated deformation field was used to transform the label map with 90 regions from atlas space to subject space. Visual inspection was performed for each subject (see Supplementary Fig. 1). The brain regions that we studied were: the frontal, temporal, parietal, occipital, insular, cingulate, limbic cortices, subcortical gray matter, and central regions (precentral and postcentral gyri, rolandic operculum bilaterally). The volume of each brain region was determined by summing the volume of their components. We also assessed the volumes of the anatomical regions implicated in autism, namely social communication and behavior (Amaral et al. 2008). We complemented this by studying the volume of the brain regions that are part of the salience network (Seeley et al. 2007) and theory of mind (Kana et al. 2014; see Supplementary Table 1).

Automatic Segmentation

We have recently described the segmentation process. A quantitative evaluation of the segmentations was performed, and the Dice coefficient (Dice 1945) was calculated showing an agreement of 0.87 ± 0.02 for cortical gray matter, 0.86 ± 0.02 for white matter, 0.79 ± 0.03 for deep gray matter, 0.84 ± 0.02 for cerebellum, and 0.83 ± 0.01 for brainstem (Padilla et al. 2015). Automatic segmentation of brain tissues was performed using SPM8 software (http://www.fil.ion.ucl.ac.uk/spm), running in MATLAB version 7.5 (MathWorks, Natrick, MA, USA), with specific neonatal priors including white matter, cortical and deep gray matter, cerebrospinal fluid (CSF), cerebellum, and brainstem. We applied a Diffeomorphic Anatomical Registration Through Exponential Lie Algebra (DARTEL) algorithm to improve the intersubject registration (Ashburner 2007). Finally, all images were modulated and smoothed with a 3-mm Gaussian kernel. Volumes were extracted from the segmented, normalized, and modulated images of each subject. Voxel-based morphometry was used to compare the ASD-positive and ASD-negative groups.

Statistical Analysis

The variables we studied were tested for normality and homogeneity before each analysis. All statistical analyses were computed using SPSS version 20.0 (SPSS, Inc., Chicago, IL, USA). The significant threshold was set at two-sided P < 0.05. For voxel-based morphometry analyses, a corrected, family-wise error of P < 0.05 and an uncorrected P < 0.001 at cluster level were reported. Descriptive statistics were calculated on all measures to define the characteristics of the sample. Logistic regression models were applied to determine independent risk neonatal factors related to ASD outcome. A general linear model analysis was performed to assess differences in global volumetric measures between the groups, with all brain volumes as dependent variables, group status as the independent factor, and the binary IQ variable and the total volume of cortical gray matter (Gilmore et al. 2012) as covariates. T-test group comparisons were performed for group voxel-based morphometry analyses.

Volumetric analyses were adjusted for potential confounders, including IQ cutoffs of <70 and 70 or more, and total volume of the cerebral parenchyma, including all of the brain tissues, but excluding CSF. Since birth weight was significantly different between groups, all analyses were also run with birth weight z-score and total cerebral parenchyma as covariates. Gender did not account for significant variations in the model. The continuous variable full-scale IQ was missing for a portion of the ASD-positive group (4/11). It was not possible to assess these 4 infants with the full-test battery. However, these children were assessed clinically and had diagnoses of mental retardation, which by definition means that their IQ was below 70. Since general lineal models need to be undertaken on cases with complete data for all covariates, we chose to dichotomize the IQ variable.

Results

We found that 23 (27.4%) of the 84 children were ASD positive and the remaining 61 (70.11%) were ASD negative. The overall IQ scores in the ASD-positive group were lower than in the ASD-negative group (Table 1). Although the groups had similar gestational ages, the ASD-positive group had a lower mean birth weight than the ASD-negative group, and the head circumferences at birth, term age, and 6.5 years were significantly smaller. The ASD-positive infants also had a higher overall frequency of neonatal complications (Table 1).

We examined a number of perinatal and neonatal risk factors in the big sample using a logistic regression model. Univariate analyses revealed that lower birth weight was associated with an ASD-positive outcome, with an odds ratio (OR) of 0.99 and 95% confidence interval (CI) of 0.991–0.998 (P = 0.003). Other factors associated with a positive ASD outcome were: head circumference at birth (OR 0.55, 95% CI 0.38–0.80, P = 0.002), duration of ventilation (OR 1.05, 95% CI 1.02–1.09, P = 0.002), necrotizing enterocolitis (OR 4.12, 95% CI 1.17–14.5, P = 0.02), retinopathy of prematurity (ROP) > stage 2 (OR 0.34, 95% CI 0.13–0.87, P = 0.02), surgical treatment for patent ductus arteriosus (PDA) and ROP (OR 0.14, 95% CI 0.03–0.59, P = 0.008), and bronchopulmonary dysplasia (OR 0.23, 95% CI 0.09–0.63, P = 0.004). When all of these variables were entered into the stepwise logistic regression model, birth weight (OR 0.99, 95% CI 0.99–1.0, P = 0.03) and surgical treatment for PDA and ROP (OR 7.38, 95% CI 1.72–31.5, P = 0.007) continued to show a significant association with being ASD positive.

We found that 11 of the children who were in the subgroup of 33 children with high-quality MRI scans were ASD positive and the remaining 22 were ASD negative. The subgroup was representative of the larger sample in terms of perinatal characteristics and neonatal variables. The ASD-positive children in this subgroup displayed a significantly higher frequency of neonatal complications and this finding was also observed in the larger study sample (Table 1). After adjustment for total brain volume and binary IQ variable, global volumetric studies showed significantly smaller total volumes of temporal, occipital, insular, and limbic cortices in the ASD-positive group than the ASD-negative group (Table 3). Notably, these differences disappeared when birth weight z-score was included as a covariate. The ASD-positive group tended to have smaller mean volumes, specifically found in the gray matter and cerebellum, but the differences did not achieve statistical significance. They also tended to have higher mean volumes of CSF (see Supplementary Table 2). These results did not change after adjusting for covariates.

Table 3

Atlas-based segmentation: brain volume data

Brain volumes (cm3) mean (SD)	ASD positive (N = 11)	ASD negative (N = 22)	P-value
			Unadjusted	Adjusted*	Adjusted†
Total cortical gray matter	217.5 (25.1)	233.0 (22.7)	0.09
Frontal	59.1 (6.8)	63.3 (6.2)	0.09	0.82	0.29
Temporal	39.7 (4.5)	42.6 (4.1)	0.09	0.04	0.27
Parietal	33.9 (4.0)	36.3 (3.5)	0.10	0.69	0.73
Occipital	34.9 (4.0)	37.5 (3.6)	0.08	0.02	0.09
Insula	4.3 (0.5)	4.64 (0.4)	0.10	0.007	0.55
Limbic lobe	16.1 (1.8)	17.2 (1.7)	0.10	0.02	0.11
Central regions	20.6 (2.3)	22.0 (2.1)	0.09	0.09	0.97
Subcortical gray matter	13.4 (1.5)	14.4 (1.4)	0.09	0.25	0.33
Domains
Social	31.6 (3.6)	33.8 (3.3)	0.10	0.02	0.03
Communication	29.9 (3.4)	32.1 (3.1)	0.09	0.66	0.77
Behavior	22.7 (2.6)	24.3 (2.4)	0.10	0.005	0.009
Salience network	22.4 (2.5)	23.9 (2.3)	0.10	0.003	0.02
Theory of mind	17.2 (1.9)	18.4 (1.8)	0.09	0.52	0.73

*P-value adjusted for total brain volume and IQ binary variable.

^†P-value adjusted for total brain volume and birth weight z-score.

Regional volumetric studies showed gray matter reductions in the ASD-positive group, compared with the ASD-negative group, and these predominantly involved the left parietal cortex (voxels = 96, t = 4.83, corrected P = 0.03). The right cortex of the inferior frontal gyrus (voxels = 71, t = 4.68, uncorrected P = 0.008) and the left lateral occipital cortex (voxels = 88, t = 4.53, corrected P = 0.04) were also affected. The left lateral occipital cortex remained significant at an uncorrected level after adjustment for covariates (voxels = 95, t = 4.45, uncorrected P < 0.001; Fig. 2). Decreased white matter volumes were found in the genu of the corpus callosum (voxels = 31, t = 4.09, uncorrected P < 0.001), but these differences were not evident after adjustments for covariates.

Figure 2.

Regional volumetric differences in gray matter in the ASD-positive group compared with the ASD-negative group. Coronal (A), sagittal (B), axial (C), and render (D) views showing regions of decreased gray matter (warm colors) in the ASD-positive group compared with the ASD-negative group. Volumetric differences are shown before (warm colors) and after adjustment for covariates (green colors). The crosshairs are in the left lateral occipital cortex. The color bars represent the t-score. Right = right.

Using atlas-based segmentation, the composite volume of brain regions implicated in social communication, behavioral domains, and the salience network were significantly smaller in the ASD-positive group than in the ASD-negative group after adjustment for covariates (Table 2). The most affected regions were the insula, anterior cingulate cortex, and orbitofrontal cortex (see Supplementary Table 1). There were no volumetric differences in the brain regions implicated in theory of mind.

Discussion

The major finding of this study was that children, who were born extremely preterm and were subsequently diagnosed with ASD, displayed altered neonatal brain development, with volume reductions in the core regions involved in autism, long before the onset of symptoms.

The groups we studied did not differ in terms of the global mean volume of gray and white matter across the entire brain, in contrast with findings of brain overgrowth in children with idiopathic autism (Schumann et al. 2010; Hazlett et al. 2011; Shen et al. 2013). One possible explanation is that the infants we included were evaluated at term-equivalent age, whereas the studies on overgrowth were carried out in toddlers.

Regional analyses showed structural brain differences in the ASD-positive group, compared with the ASD-negative group, mostly in terms of reduced gray matter. The largest gray matter cluster included the left angular gyri, a heteromodal association region involved in complex language functions and known to be affected in ASD (Li et al. 2014). After adjustment for covariates, we found a significant volume reduction in the gray matter in the lateral occipital cortex. White matter was less affected and involved reductions in the corpus callosum. With regard to this, studies in young children with idiopathic autism have demonstrated gray matter and white matter enlargements (Hazlett et al. 2005) where the occipital cortex was usually unaffected by the volume and growth trajectory (Schumann et al. 2010). However, consistent structural abnormalities have been reported in older populations in the lateral occipital cortex and these play a critical role in visual processing in ASD subjects (Nickl-Jockschat et al. 2012).

Regional reductions in the corpus callosum in our ASD-positive group support the hypothesis of impaired interhemispheric communication, particularly involving the frontal and parietal regions (Just et al. 2007). Overall, our results showed that gray and white matter displayed early deviations from the normal growth trajectory in subjects who were ASD positive and this may result in potentially disruptive patterns of brain connectivity during development.

Our study found restricted growth in the brain regions that play a key role in the core features of autism, including social/interaction and behavioral domains. Volumetric alterations were also observed in regions belonging to the salience network, which integrates external sensory stimuli with internal states through the integration of sensory, visceral, autonomic, and hedonic signals (Seeley et al. 2007). Regions within the salience network are also implicated in attention and awareness. It has recently been shown that dysfunction of this network may provide a biomarker for identifying children with ASD (Uddin et al. 2013). We identified volume reductions in the orbitofrontal, anterior cingulate, and insular cortices, which are brain regions that are critically important for efficient neuronal signaling and communication. Notably, the insula has important connections with the cingulate and orbitofrontal cortices among other structures, and constitutes one of the main structural and functional hubs in neonates. It has an essential role for multiple functions, including sensory perception, cognition, socio-emotional processing, and stress (Alcauter et al. 2015). In this regard, although we grouped brain areas according to the core regions affected in ASD, the same regions appear to be involved in multiple cognitive functions. How structural changes in brain regions relate to cognitive changes in our study is an issue that clearly needs to be explored by future studies. We could not find any differences between the groups in the main structures of the brain that participate in the theory of mind. Nevertheless, this does not rule out the presence of structural differences in areas such as white matter organization, which was not examined by the present study.

Most of the cortical regions that were affected in the ASD-positive infants were cortical hubs in structural networks that have previously been described in extremely preterm infants (Fransson et al. 2011) and that play an important role in neural communication and integrating information. It has been proposed that cortical hubs may be particularly vulnerable to diverse disease processes (Crossley et al. 2014). Thus, one can speculate that numerous neonatal complications in preterm infants might be expected to have a disproportionate impact on these vulnerable cortical regions.

In our study, 64% of the ASD-positive children with high-quality MRI had cognitive impairment (IQ <70) whereas none in the ASD-negative group. Although individuals with ASD do not necessarily have low IQ scores, in extremely preterm populations this association is described (Johnson et al. 2010; Perrone-McGovern et al. 2015). A previous study in extremely preterm children reported a mean IQ difference between ASD and no ASD groups of 14 points (mean IQ 85 vs. 71) with almost all children with ASD having a cognitive impairment (Johnson et al. 2010). In this case, brain abnormalities may depend on autism itself or could be related to intellectual disability associated with ASD. Future studies are clearly needed to explore this issue in greater depth.

The ASD-positive group in this study experienced more neonatal clinical complications. In this regard, we identified several independent neonatal risk factors associated with a greater likelihood of being diagnosed with ASD and the most important were birth weight and surgical treatment for PDA and ROP. Our results confirm previous findings that suggested that birth weight and neonatal complications increased the risk of ASD (Limperopoulos et al. 2008; Buchmayer et al. 2009). The current findings also showed that infants in the ASD-positive group were smaller at birth and showed smaller growth in head circumference and weight between birth and term-equivalent age. In this regard, several potentially adverse factors may influence brain growth; however, low birth weight seems to be one of the most important perinatal risk factors for autism (Gardener et al. 2009). After controlling for birth weight z-score global brain differences between groups were no longer of significance, while regional and network volumetric differences remained significant. This suggests that whereas low birth weight was related with a lower scaling of the brain, ASD was related with volume reductions in specific brain regions. Our results highlight the importance of characterizing the population under study with regard to the presence or absence of suboptimal growth.

The strengths of this study included the unique design, with a carefully selected, well-defined, and prospectively followed cohort of extremely preterm infants. Moreover, the imaging was conducted during the early neonatal period, long before the diagnosis of symptoms of ASD. A possible limitation of this study was the number of participants who could not be included in the morphometric analyses. This was due to rigorous entry and data quality criteria and implicit methodological difficulties related to MRI processing in preterm infants at term-equivalent age (Padilla et al. 2015). Additionally, inclusion criteria were restricted to extremely preterm neonates presenting with normal findings on conventional MRI scans and having no severe motion artifacts on the series of high-resolution structural and quantitative volumes. However, the MRI sample was representative of the whole sample in terms of perinatal characteristics and general outcomes. Despite the modest sample size, we believe that combining multiple structural MRI methods provided an excellent opportunity to elucidate the early underlying structural substrates of ASD in preterm children.

In conclusion, these preliminary results provide evidence that impaired brain growth can be detected as early as the neonatal period in preterm children diagnosed with ASD and prominent autistic traits at the age of 6.5 years. The impaired brain development that we found in the ASD-positive group may be linked to early exposure to the extrauterine environment, with a possible primary effect on specific brain regions participating in sensory integration and socio-emotional regulation. Larger studies, which take neonatal complications into account, and apply multimodal imaging techniques, are needed to accurately characterize the brains and the neonatal risks of preterm infants who go on to develop ASD.

Supplementary Material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/online.

Funding

This work was supported by the Swedish Medical Research Council (grant nos 523-2011-3981, 2009-4250, and 523-2009-7054); the Regional Agreement on Medical Training and Clinical Research (grant no. ALF-20140316) between Stockholm County Council and the Karolinska Institutet; the Marianne and Marcus Wallenberg Foundation (grant no. 2011.0085); the European Union Seventh Framework Project (grant no. 223767); the Swedish Order of Freemasons in Stockholm; the Swedish Medical Society; the Swedish Brain Foundation (grant no. FP2014-0135); Sällskapet Barnavård; and the Swedish Research Council, in partnership with FAS, FORMAS, and VINNOVA (grant no. 259-2012-24).

Notes

We thank the families and infants who took part in the study and the Express study group. Conflict of Interest: None declared.