Genome-wide transcriptional profiling and structural magnetic resonance imaging in the maternal immune activation model of neurodevelopmental disorders

: Prenatal exposure to maternal infection increases the risk of neurodevelopmental disorders, including schizophrenia and autism. The molecular processes underlying this pathological association, however, are only partially understood. Here, we combined unbiased genome-wide transcriptional profiling with follow-up epigenetic analyses and structural magnetic resonance imaging to explore convergent molecular and neuromorphological alterations in corticostriatal areas of adult offspring exposed to prenatal immune activation. Genome-wide transcriptional profiling revealed that prenatal immune activation caused a differential expression of 116 and 251 genes in the medial prefrontal cortex and nucleus accumbens, respectively. A large part of genes that were commonly affected in both brain areas were related to myelin functionality and stability. Subsequent epigenetic analyses indicated that altered DNA methylation of promoter regions might contribute to the differential expression of myelin-related genes. Quantitative relaxometry comparing T1, T2, and myelin water fraction revealed sparse increases in T1 relaxation times and consistent reductions in T2 relaxation times. Together, our multi-system approach demonstrates that prenatal viral-like immune activation causes myelin-related transcriptional and epigenetic changes in corticostriatal areas. Even though these abnormalities do not seem to be associated with overt white matter reduction, they may provide a molecular mechanism whereby prenatal infection can impair myelin functionality and stability. Abstract Prenatal exposure to maternal infection increases the risk of neurodevelopmental disorders, including schizophrenia and autism. The molecular processes underlying this pathological association, however, are only partially understood. Here, we combined unbiased genome-wide transcriptional pro ﬁ ling with follow-up epigenetic analyses and structural magnetic resonance imaging to explore convergent molecular and neuromorphological alterations in corticostriatal areas of adult offspring exposed to prenatal immune activation. Genome-wide transcriptional pro ﬁ ling revealed that prenatal immune activation caused a differential expression of 116 and 251 genes in the medial prefrontal cortex and nucleus accumbens, respectively. A large part of genes that were commonly affected in both brain areas were related to myelin functionality and stability. Subsequent epigenetic analyses indicated that altered DNA methylation of promoter regions might contribute to the differential expression of myelin-related genes. Quantitative relaxometry comparing T 1 , T 2 , and myelin water fraction revealed sparse increases in T 1 relaxation times and consistent reductions in T 2 relaxation times. Together, our multi-system approach demonstrates that prenatal viral-like immune activation causes myelin-related transcriptional and epigenetic changes in corticostriatal areas. Even though these abnormalities do not seem to be associated with overt white matter reduction, they may provide a molecular mechanism whereby prenatal infection can impair myelin functionality and stability.


Introduction
The etiology of multifactorial and multi-symptomatic neuropsychiatric disorders likely includes exposures to adverse events during prenatal and early postnatal life, which may disrupt the development and maturation of neural systems and brain functions (Brown 2011). Prenatal exposure to infectious or inflammatory insults is increasingly recognized to play an important role in this context. Indeed, immune-related prenatal adversities have been repeatedly linked with a higher risk of neurodevelopmental psychiatric disorders, including schizophrenia, autism, and bipolar disorder (Brown and Derkits 2010;Patterson 2011;Marangoni et al. 2016). These epidemiological associations are further supported by translational work in animal models demonstrating abnormal brain development and behavioral dysfunctions following prenatal administration of infectious pathogens or immune activating agents (Meyer and Feldon 2010;Harvey and Boksa 2012;Meyer 2014).
The advances in modeling prenatal immune activation effects in animals hold promise for the identification of pathological mechanisms that translate the prenatal insult into longterm brain abnormalities. The majority of these experimental attempts, however, were hypothesis-driven and focused on the role of a specific cellular or molecular mechanism, pathway or system (Eyles et al. 2012;Ibi and Yamada 2015). While this is a laudable and possibly fruitful approach for the examination of a presumed pathophysiological process, hypothesis-driven investigations may mask the discovery of novel disease mechanisms.
The implementation of genome-wide transcriptional profiling is one possible strategy to overcome these limitations. It allows an unbiased screen of gene expression changes in response to prenatal immune activation, which can form the basis for followup investigations that take into account this transcriptomic information (Fatemi et al. 2005(Fatemi et al. , 2008(Fatemi et al. , 2009aConnor et al. 2012;Tebbenkamp et al. 2014;Horvath and Mirnics 2015). The present s t u d yf o l l o w e ds u c ha na p p r o a c h ,i nw h i c hu n b i a s e dt r a n s c r i ptomic profiling formed the basis for subsequent immunohistochemical investigations and epigenetic analyses. Genome-wide transcriptomic profiling was performed using unbiased microarray techniques in the medial prefrontal cortex (mPFC) and nucleus accumbens (NAc), 2 brain regions implicated in neurodevelopmental disorders such as schizophrenia and autism (Richey et al. 2015;Schubert et al. 2015;Selemon and Zecevic 2015). The inclusion of 2 brain regions allowed us to identify region-specific and -overlapping transcriptomic effects of prenatal immune activation using a within-subjects comparison. These investigations were performed in offspring that were first subjected to behavioral and cognitive testing.
We also explored whether microscale molecular alterations reverberate to influence macroscale features that are detectable using clinically comparable magnetic resonance imaging (MRI) (Turkheimer et al. 2015). As an adjunct to cellular and molecular methods, MRI provides a non-invasive global visualization of brain maturation, through the evolution of gray and white matter contrast on T 1 and T 2 relaxation time-weighted images (Ballesteros et al. 1993;Paus et al. 2001;Leppert et al. 2009). We further implemented a novel multicomponent relaxation (MCR) technique termed multicomponent-driven equilibrium single pulse observation of T 1 and T 2 (mcDESPOT) (Deoni et al. 2008a)to provide quantitative information about proton tissue water relaxation times and myelin water fraction (MWF) signals. This technique aims to divide the MR signal resulting from steadystate sequences into 3 pools representing water trapped in the myelin sheath, intra-extracellular water, and free water in cerebrospinal fluid (Deoni et al. 2008b(Deoni et al. , 2013. Whilst applied in human imaging before (Deoni et al. 2011(Deoni et al. , 2012, we have recently validated the mcDESPOT methodology in rodents using a model of cuprizone-induced demyelination (Wood et al. 2016a(Wood et al. , 2016b. All investigations were performed using a well-established mouse model of prenatal viral-like immune activation. The model is based on maternal administration of the viral mimetic poly(I:C) (=polyriboinosinic-polyribocytidilic acid), which induces a cytokine-associated viral-like acute phase response in maternal and fetal compartments, including the fetal brain (Meyer et al. 2009). Prenatal poly(I:C) treatment in rodents has repeatedly been shown to cause multiple behavioral and cognitive disturbances in the offspring, many of which are implicated in developmental psychiatric disorders such as schizophrenia and autism (Meyer et al. 2009;Meyer and Feldon 2010;Harvey and Boksa 2012;Meyer 2014). The poly(I:C) administration model thus offers a unique opportunity to explore genome-wide transcriptomic changes following prenatal exposure to an etiologically relevant risk factor, and to further link such changes with neurobehavioral and MRI-detectable abnormalities.

Animals
C57Bl6/N mice were used throughout the study. Female and male mice were originally obtained from Charles River Laboratories (Germany) and kept in our in-house specific-pathogenfree (SPF) facility until breeding began to generate poly(I:C)exposed and control offspring (see below). All animal breeding and holding rooms were temperature-and humidity-controlled (21 ± 1°C, 55 ± 5%) and kept under a reversed light-dark cycle (lights off: 7:00 A.M. to 7:00 P.M.). All animals had ad libitum access to food (Kliba 3430, Kaiseraugst, Switzerland) and water throughout the entire study. All procedures described in the present study had been previously approved by the Cantonal Veterinarian'sO f fice of Zurich, and all efforts were made to minimize the number of animals used and their suffering.

Maternal Immune Activation during Pregnancy
Female C57Bl6/N mice were subjected to a timed mating procedure as described previously (Meyer et al. 2005). Pregnant dams on gestation day (GD) 17 were randomly assigned to receiving either a single injection of poly(I:C) (potassium salt; Sigma-Aldrich, Buchs, St Gallen, Switzerland) or vehicle. Poly(I:C) (5 mg/kg; calculated based on the pure form of poly(I:C)) was dissolved in sterile pyrogen-free 0.9% NaCl (vehicle) solution to yield a final concentration of 1 mg/mL and was administered intravenously (i.v.) into the tail vein under mild physical constraint. The dose of poly(I:C) was selected based on previous dose-response studies (Meyer et al. 2005). 10 pregnant dams were injected with poly(I:C), and another 10 with vehicle solution.
GD 17 in the mouse roughly corresponds to human gestational weeks 28 to 29 in terms of cortical neurogenesis (http:// translatingtime.net/translate). It was selected based on previous studies showing that prenatal poly(I:C) exposure during this gestational period causes adult behavioral, cognitive, and neuroanatomical abnormalities relevant to neurodevelopmental brain disorders, including schizophrenia and autism (Meyer et al. 2006(Meyer et al. , 2008Bitanihirwe et al. 2010aBitanihirwe et al. , 2010bRichetto et al. 2013Richetto et al. , 2014Richetto et al. , 2015. We previously verified that poly(I:C) administration on GD 17 is effective in terms of eliciting cytokine-associated inflammatory response in maternal and fetal tissues (Meyer et al. 2006).

Allocation of Offspring and Group Sizes
Offspring of poly(I:C)-treated dams (POL) and vehicle-exposed control offspring (CON) were weaned and sexed on postnatal day (PND) 21. Littermates of the same sex were caged separately and maintained in groups of 3 to 4 animals per cage. Only male animals were included in all experiments because our previous research using the mouse prenatal poly(I:C) administration model did not reveal sex-dependent effects on the behavioral and cognitive functions of primary interest, including social interaction and working memory (Meyer et al. 2008;Bitanihirwe et al. 2010a;Richetto et al. 2013). A first cohort of CON and POL offspring was used for the assessment of behavioral and cognitive functions, which was followed by postmortem microarray analyses, validation of gene expression, and DNA methylation analyses (see below). A second cohort of offspring was used for ex vivo MRI imaging and subsequent postmortem immunohistochemical analyses (see below). In both cohorts, 1 male offspring per litter was randomly selected for the investigations of interest in order to avoid litter effects (Zorrilla 1997). This led to a group size of N = 10 offspring per prenatal treatment condition in each cohort. The DNA methylation analyses were performed using a subset of offspring, with N = 5 offspring per prenatal treatment condition. A graphical representation of the experimental design is provided in Supplementary Figure 1.

Behavioral Testing
CON and POL offspring of cohort 1 were subjected to cognitive and behavioral testing when they reached early adulthood (12 weeks of age). The tests included paradigms assessing social approach behavior, and social recognition, and short-term spatial recognition memory working. These tests were selected based on their relevance to neurodevelopmental disorders with infectious and inflammatory components, including schizophrenia and autism (Meyer et al. 2009;Peleg-Raibstein et al. 2012). A detailed description of the test apparatuses and procedures is provided in the Supplementary Information. Each animal underwent all behavioral tests in the following order: (1) spatial recognition memory test and (2) social approach and recognition test. A testfree resting period of 2 days was imposed between the 2 tests.

Collection of Brain Samples for Molecular Analyses
CON and POL offspring of cohort 1 were killed by decapitation 10 days after completion of behavioral testing for the subsequent molecular analyses. The brains were rapidly extracted from the skull (within < 20 s) and placed on an ice-chilled plate. This was followed by preparing 1-mm coronal brain sections using razorblade cuts and subsequent micro-dissection of the brain areas of interest. We dissected the medial prefrontal cortex (mPFC, including anterior cingulate, prelimbic and dorsal parts of the infralimbic cortices; bregma: +2.3 to +1.3 mm) and the nucleus accumbens (NAc, including core and shell subregions; bregma +1.5 to +0.5 mm) as previously described (Bitanihirwe et al. 2010a). Brain specimens were collected in 96-well microtiter plates kept on dry ice and allowed to freeze before storage at −80°C until further use.

DNA and RNA Isolation
Total DNA and RNA were isolated using the Qiagen AllPrep DNA and RNA Mini kit (Qiagen, Italy) as described in the Supplementary Information.

Microarray Analyses
Genome-wide gene expression analyses were performed using Affymetrix microarray assays (Mouse Gene 1.1 ST Array Strips on GeneAtlas platform), following the 3'IVT one cycle labeling as fully described in the Supplementary Information. All the raw data are accessible at the NCBI GEO depository (GEO accession number GSE77973). The data analysis was performed with Partek Genomics Suite (Partek, USA), version 6.6 (for details, see Supplementary information). Differentially expressed genes (DEGs) in POL offspring relative to CON offspring were identified by performing a linear contrast (POL versus CON). In this comparison, a maximum filter of P < 0.05 and a minimum absolute fold change cut-off of 1.2 were applied. The DEGs were loaded into Partek Genomics Suite and were clustered according to the Hierarchical Clustering function of Partek Genomics Suite. For this purpose, the data was normalized with standardization (each column mean is 0, and the standard deviation is scaled to 1), and then multidimensional scaling with a Euclidian distance metric was performed on the normalized samples to allow visualization of the distance between them. To confirm that the overlapping changes were not due to random distribution, we performed the hyper-geometric test in R (P = 5.765284e14). Gene Ontology enrichment analyses were performed using Partek Genomics Suite (Partek, USA). Subsequent validation of selected DEGs was performed by quantitative real-time PCR as described below.
Quantitative Real-Time RT-PCR Analyses mRNA levels were quantified by TaqMan qRT-PCR (CFX384 realtime system, Bio-Rad Laboratories) using the iScript 1-step RT-PCR kit for probes (Bio-Rad Laboratories) (see Supplementary  Information). Relative target gene expression was calculated according to the 2(-Delta Delta C(T)) method. Probe and primer sequences of Claudin11 (Assay: Mm00500915_m1) were purchased from Life Technologies (Switzerland), while the custom designed probe and primer sequences used for MOBP, MOG, MAL and MAG are summarized in Supplementary Table 1 and were purchased from Eurofins Genomics GmbH (Germany).

DNA Methylation Analysis
DNA methylation levels of the α-myelin-associated oligodendrocytic basic protein (MOBP) promoter region were quantified using the EpiTYPER assay. This technique detects and quantifies DNA methylation using base-specific cleavage and Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) mass spectrometry (Suchiman et al. 2015). Genomic DNA was treated and analysed as fully described in the Supplementary Information.

Brain Sample Preparation for MRI
At 12 weeks of age, CON and POL offspring from the second cohort were deeply anaesthetized with an overdose of Nembutal (Abbott Laboratories) and perfused transcardially with 0.9% NaCl, followed by 4% phosphate-buffered paraformaldehyde (PFA) solution containing 15% picric acid (Giovanoli et al. 2013). After perfusion, the animals were decapitated and the skin, lower jaw and ears were removed. The brain within the skull was incubated in 4% PFA overnight at +4°C and then shipped to King's College London the next day. Upon arrival, brain samples were transferred to 0.01 M phosphate-buffered saline containing and 0.05% sodium azide for at least 7 days prior to MR imaging.

MRI acquisition, Processing and Analysis
A 7 T horizontal small bore magnet and (Agilent Technologies Inc. Santa Clara, USA) and a quadrature volume radiofrequency coil (39 mm internal diameter, Rapid Biomedical GmbH) were used for all MRI acquisitions. Fixed brain samples were placed securely up to 4 at a time in an MR-compatible holder and immersed in proton-free susceptibility matching fluid (Fluorinert TM FC-70; Sigma-Aldrich, UK). The following MR images were acquired: T 2 -weighted 3D Fast Spin-Echo (FSE) and a multicomponent Driven Equilibrium Single Pulse Observation of T 1 and T 2 (mcDEPSOT) protocol with B1 correction. The latter consists of a Spoiled Gradient echo (SPGR), balanced Steady State Free Procession (bSSFP) and Actual Flip-angle imaging (AFI) scans (Deoni et al. 2013). The mcDESPOT protocol generates data to calculate parametric maps of T 1 , T 2 , and the myelin water fraction (MWF) for each animal (Supplementary Figure 1). Parameters for each scan are summarized in Supplementary  Table 3. MR image processing and analysis were performed as fully described in the Supplementary information.
Group-level differences in MRI parameters (volume, T 1 , T 2 , and MWF) between CON and POL offspring were analysed voxel-wise across the whole-brain using permutation testing and threshold free cluster enhancement (TFCE) implemented in FSL Randomize (Smith and Nichols 2009) and corrected for multiple comparisons using the false discovery rate (FDR) (Genovese et al. 2002)a tq = 0.05. For volume, comparisons were made between CON (N = 10) and POL (N = 10) offspring. For mcDESPOT data, comparisons were made between CON (N = 6) and POL (N = 8) offspring, as 4 scans from the CON group and 2 scans from the POL group had to be discarded due to artefacts in the data (see Supplementary Information).

Immunohistochemistry
After completion of all MRI, fixed brain tissues were processed for immunohistochemical analyses as described in the Supplementary information. Standard immunohistochemical procedures were implemented to stain for α-myelin basic protein (MBP; rabbit anti-MBP; Abcam, Cat no. ab7349; diluted 1:1000) and MOBP (rabbit anti-MOBP; Abcam, Cat no. ab203388; diluted 1:500) as fully described in the Supplementary information.

Threshold Image Analysis for Myelin Staining
Quantitative analyses of MBP and MOBP-positive staining were performed in the mPFC using unbiased threshold image analysis as described in the Supplementary Information. All post-processing and analysis was performed using ImageJ software (http://imagej.nih.gov/ij/). The percentage area of immunopositive pixels in each acquired image of the mPFC, from 4 consecutive sections, were averaged to give a single value per animal.

Statistical Analyses
All behavioral and RT-PCR were analysed using independent Student's t-tests (2-tailed). Immunohistochemical analysis was conducted using a 1-tailed Student's T-test, given the a-priori hypothesis based on the strong gene expression data. Statistical significance was set at P < 0.05 for these analyses. Microarray and MRI data were analysed as described above. DNA methylation levels measured using EpiTYPER were analysed using repeated-measures analysis of variance (RM-ANOVA) followed by Fisher's least significant difference (FLSD) post hoc comparisons whenever appropriate. Correlative analyses between the dependent measures of primary interest (behavioral and cognitive data, mRNA and DNA methylation data, MRI data, and immunohistochemical data) were performed using first-order partial correlations partialling for the 2 prenatal treatment conditions. Hence, partial correlations were used to control for the effects of the independent variable "prenatal immune activation". All statistical analyses were performed using the statistical software SPSS (v22.0; IBM Corporation, Armonk, New York, USA).

Prenatal immune activation induces deficits in shortterm spatial recognition memory and social interaction
First, we aimed to ascertain the effects of prenatal immune activation on adult behavioral and cognitive functions. We found that POL offspring displayed impaired performance in a Y-maze spatial recognition memory test. (Fig. 1A). In this test, the critical measure of spatial recognition memory is the relative time spent in the novel (previously unexplored) arm during the choice phase. CON offspring displayed a noticeable preference towards the novel arm, indicating intact spatial recognition memory in these groups (Fig. 1A). In contrast, POL offspring exhibited a marked reduction (P < 0.05) in this measure and performed only at chance level ( Fig. 1A). There were no group differences with respect to the total distance moved ( Fig. 1A), indicating that the negative effects of prenatal immune activation on spatial recognition memory are not confounded by possible differences in basal locomotor activity.
POL offspring also displayed impaired marked deficits in the social interaction test, in which they were first allowed to concomitantly explore an inanimate dummy object and an unfamiliar live mouse (Fig. 1B). During this phase of the test, CON offspring showed a clear preference (>65%) for the live mouse versus the inanimate dummy object (Fig. 1B). By contrast, POL offspring did not display such a preference (Fig. 1B), indicating reduced sociability towards unfamiliar conspecifics. This led to a significant (P < 0.001) group difference in the percent time spent with the live mouse (Fig. 1B).
To test social recognition memory, the inanimate dummy object was then replaced by another unfamiliar live mouse, and the relative exploration time between the previously explored and novel live mouse was measured. During this phase of the social interaction test, CON offspring showed a clear preference (>65%) for the novel mouse (Fig. 1B). POL offspring did not display such a preference, leading to a significant (P < 0.05) group difference in the percent time spent with the novel mouse ( Fig. 1B). In both phases of the social interaction test, there were no group differences with respect to the total distance moved. Hence, prenatal immune activation leads to genuine deficits in social approach behavior and short-term spatial recognition memory without concomitant effects on general exploratory behavior.
Prenatal immune activation alters the long-term transcription profile of the medial prefrontal cortex and nucleus accumbens As shown in Figure 2A, which represents the hierarchical clustering of expression changes induced by prenatal infection with poly(I:C), 116 genes were differentially expressed in the mPFC (fold change cut-off: ± 1.2; P < 0.05), while 251 were differentially expressed in the NAc. Of the 116 genes that were differentially expressed in the mPFC, 55 were downregulated and 61 were upregulated (Tables 1 and 2), while in the NAc, 126 were downregulated and 125 upregulated (Tables 3 and 4). Interestingly, many of these have been already associated with schizophrenia, such as adenosine 2a receptor (ADORA2a), apolipoprotein D (APOD), the dopamine receptors DRD2 and DRD3, forkhead box P2 (FOXP2), glutaminase (GLS), the glutamate receptor subunit GRIN2A, histone cluster 1 (HIST1H2BC), 5-hydroxytryptamine receptors (HTR1A, HTR2A, HTR4), oxytocin (OXT), solute carriers (SLC17A7), and, among others, vescicle-associated membrane protein 4 (VAMP4) (Butler et al. 2016). A Gene Ontology (GO) enrichment analysis computing the DEGs in the mPFC and NAc is presented in Supplementary Table 2. Next, the genes affected in both brain areas were compared. As shown in Figure 2B, the Venn diagram of the 2 brain areas highlighted 14 common genes, 11 of which are modulated in the same direction in both brain areas (Fig. 2B). In particular, (POL), or they were exposed to prenatal control (CON) treatment. (A) Percent time spent in the novel arm during the Y-maze working memory test. *P < 0.05 based on independent Student's t tests (2-tailed). (B) Percent time spent with an unfamiliar live mouse, relative to an inanimate dummy object, during the social interaction test; and percent time spent with a novel live mouse, relative to a familiar one, during the social recognition test. *P < 0.05 and ***P < 0.001 based on independent Student's t tests (2-tailed). All data are based on N (CON) = 10, N(POL) = 10 and represent means ± s.e.m.

Figure 2.
Unique and common gene expression differences following late prenatal immune activation in the mPFC and NAc, revealed by microarray. Mice were subjected to prenatal poly(I:C) treatment on gestation day 17 (POL), or they were exposed to prenatal control (CON) treatment. (A) Hierarchical clustering of differentially expressed genes in POL offspring relative to CON offspring in the mPFC and NAc. Down-and upregulated genes are represented in purple and yellow color, respectively. (B) Venn Diagram depicting the number of genes that are uniquely and commonly affected in the mPFC and NAc of POL offspring. The commonly affected genes are listed for each brain area, with downand upregulated genes being represented by blue and red color, respectively.
All data are based on N(CON) = 6, N(POL) = 6. poly(I:C) exposure affected the expression of 6 main genes involved in myelination, both in the mPFC and in the NAc: myelin and lymphocyte protein (MAL), myelin-associated glycoprotein (MAG), myelin-associated oligodendrocytic basic       (Cldn11), and myelin regulatory factor (Myrf). The first 5 of these, which were found to be downregulated by poly(I:C) in the gene array analysis, resulted decreased by the prenatal manipulation also when analysed with real-time qRT-PCR (Fig. 3A). Ingenuity Pathway analysis (Fig. 3B) and GO enrichment analyses (Supplementary Table 2) suggested that these genes are implicated in a variety of biological processes and functions implicated in myelination and ensheathment of axons. We further sought evidence for the possibility that changes in the expression of myelin-related genes would correlate with the behavioral and cognitive performance. In support of this notion, we found that the mRNA levels of MOG in the mPFC correlated positively with the animals' performance in the spatial recognition memory test (Supplementary Table 3). On the other hand, the gene expression levels of MOG, MAG, MAL and MOBP in the NAc showed a marked positive correlation with social approach behavior (Supplementary Table 6).
Whole-brain, voxel-wise macroscale MRI phenotyping of immune-challenged and control offspring At 12 weeks of age, total brain volume did not significantly differ between behaviorally naïve CON (572 ± 7mm 3 ) and POL offspring (584 ± 3mm 3 ). Voxel-wise TBM analysis revealed sparse neuroanatomical differences in relative volume between the 2 groups (q = 0.05 FDR corrected; Fig. 4). Biologically plausible clusters of voxels showing apparent volume increases were present in the right hemisphere in the primary motor, somatosensory cortex (barrel field) and visual cortex (Fig. 4). In contrast, biologically plausible clusters of voxels showing apparent volume decreases were present bilaterally in the piriform cortex, anterior commissure, interfasicular nucleus, third ventricle, left periaqueductal gray nucleus, left external capsule, extending into the left fimbria, right amygdala and the right ventral mesencephalon (Fig. 4). Interestingly, the cerebellum showed the most pronounced volumetric alterations in POL offspring. Apparent volume increases were present in the crus 1 of the ansiform lobule and simple lobule and the inferior cerebellar peduncles (Fig. 4). In contrast, apparent volume decreases were present in the fifth cerebellar lobule, right paraflocculus and paramedian lobule of the cerebellum (Fig. 4).
We next tested for the presence of group-level differences between CON and POL offspring in the tissue parameters derived from the mcDESPOT dataset. Brain-wide voxel-wise cluster analysis revealed little or no change in T 1 , although sparse, but biologically plausible clusters of increased T 1 (spinlattice) relaxation time (+100 ms) were present bilaterally in the anterior nucleus accumbens and the inferior cerebellar peduncles  (q = 0.05 FDR corrected; Fig. 5A). In contrast, T 2 (spin-spin) relaxation times were significantly decreased in POL offspring (−20 ms) in the prefrontal, anterior cingulate, insular, retrosplenial granular, motor, somatosensory, visual and auditory cortices, with stronger decreases (−40 ms) in the piriform cortex (q = 0.05 FDR corrected; Fig. 5B). Bilateral decreases in T 2 (−20 ms) were also present in the hypothalamus, ventral thalamus, dorsal, and ventral hippocampus, ventral mesencephalon and cerebellar gray and white matter, respectively (Fig. 5B). Myelin water fraction (MWF) increased significantly in POL as compared to CON offspring, particularly in the cortex, hippocampus, and cerebellar gray and white matter (q = 0.05 FDR corrected; Fig. 5C). These increases ranged from +2% in the prefrontal cortex to +7% in the cerebellum (Fig. 5C). Notably, increases in MWF largely overlapped with the topographical distribution of decreases in T 2 , but not T 1 .

Prenatal immune activation affects protein expression levels of MOBP
Based on the preceding findings, we sought to verify whether the myelination-related changes in gene expression and MRI translate into, and stem from, differential protein expression. Thus, we analysed the protein levels of MBP, a major constituent of the myelin sheath, and MOBP, one of the candidates that emerged from our genome-wide study, with immunohistochemistry and optical densitometry (OD) in mPFC. Consistent with our gene expression study, but not the apparent increase . Sparse alterations in neuroanatomy following late prenatal immune activation revealed by brain-wide tensor based morphometry analysis of 3D FSE T 2weighted MR images. Mice were subjected to prenatal poly(I:C) treatment on gestation day 17 (POL; N = 10), or they were exposed to prenatal control (CON, N = 10) treatment. Data shown are significant group-level voxel-wise differences in the scaled log Jacobian determinant (corrected for total brain volume) thresholded at q < 0.05 (FDR-corrected). Distances from bregma are indicated in mm for each slice shown. R, right; L, left; S, superior; I, inferior. in MWF in the MR images, there were no significant differences in the area fraction of MBP-immunoreactive pixels in the mPFC of POL relative to CON offspring (Fig. 6A). In contrast, the area fraction of MOBP-immunoreactive pixels was reduced in the mPFC of POL relative to CON offspring (Fig. 6B).

Relationships between MRI and postmortem histology
Correlative analyses revealed a significant negative correlation between MWF (%) and T 2 (ms) in the prefrontal cortex of CON offspring, such that longer T 2 relaxation times were associated with smaller MWF (Supplementary Figure 2). These data are in good agreement with relaxometry studies in dys-myelinated mice (Shiverer mutants), which show extended T 2 relaxation times relative to their WT littermates (Dyakin et al. 2010) and our prior work in the cuprizone mouse model (Wood et al. 2016a(Wood et al. , 2016b. T 2 relaxation time is also reported to correlate with MWF values in human datasets, although these relationships appear to be inconsistent (positive and negative) over different developmental periods (Deoni et al. 2012). A strong trend for the same relationship was also present in POL offspring (Supplementary Figure 2). There were no other significant correlations between MRI parameters (T 1 , T 2 , MWF, or volume) in either CON or POL offspring (data not shown).
We next investigated relationships between the values of T 1 , T 2 , and MWF against the area fraction of MBP and MOBP Figure 6. Immunohistochemical alterations following late prenatal immune activation. Mice were subjected to prenatal poly(I:C) treatment on gestation day 17 (POL), or they were exposed to prenatal control (CON) treatment. staining in the mPFC, where these data were available for the same animal. The results of this analysis are presented in Supplementary Table 9. We found no significant relationship between the measured MWF and MBP area fraction in either CON or POL offspring. In contrast, a significant negative correlation was found between T 2 and MOBP area fraction in the mPFC of POL offspring, such that shorter T 2 values were associated with a greater area fraction of MOBP (Supplementary  table 9). The T 1 values in the mPFC were also negatively related to MOBP area fraction in POL offspring, although this failed to reach statistical significance at the 2-tailed level (r = −0.739; P = 0.09; Supplementary Table 9) Prenatal immune activation alters the methylation profile of the MOBP promoter We were further interested in examining whether epigenetic modifications may be a plausible mechanism underlying the transcriptional effects of prenatal immune activation. To test this hypothesis, we analysed DNA methylation profiles of the MOBP promoter in the mPFC and NAc using the EpiTYPER technique, which detects and quantifies DNA methylation using MALDI-TOF mass spectrometry (Suchiman et al. 2015). We focused on MOBP because the initial microarray and subsequent RT-PCR analyses consistently revealed decreased expression of this myelination-related gene in offspring exposed to prenatal immune activation relative to control offspring (Figs  2A and 3A).
We found hypermethylation of multiple CpGs sites in 2 distinct MOBP promoter segments that are adjacent to (amplicon 1) or incorporate (amplicon 2) the transcriptional start site (TSS) (Fig. 7). In the mPFC, the infection-induced changes in CpG methylation levels differed as a function of genomic location: The percent CpG methylation encompassed in amplicon 1 was increased at CpG 507 and CpG 457 in POL compared to CON offspring, whereas it was decreased at CpG 343 in the former relative to the latter group (Fig. 7B). On the other hand, the CpG sites encompassed in amplicon 2 were generally hypermethylated in POL relative to CON offspring (Fig. 7B). Prenatal immune activation also led to genomic location-dependent changes in accumbal MOBP DNA methylation: the methylation levels at CpG 239 (amplicon 1) and at CpG 210 (amplicon 2) were increased in POL compared to CON offspring (Fig. 7C).
We further investigated whether the infection-induced changes in DNA methylation might correlate with the gene expression levels of MOBP and with the animals' behavioral performance. As summarized in Supplementary Table 4 and  Supplementary Table 5, the methylation levels of CpG 457 positively correlated with the expression levels of MOBP in the mPFC, whereas the methylation levels of CpG 343 negatively correlated with the animals' performance in the spatial recognition memory test. There were no significant correlations between MOBP DNA methylation and gene expression levels in the NAc, nor were there any significant correlations between accumbal DNA methylation and behavioral performances of primary interest (Supplementary Table 7

Discussion
The present study analysed the gene expression profiles of the mPFC and NAc following exposure to late prenatal immune activation, and implemented a multi-system approach based on MR imaging, immunohistochemistry and epigenetic analyses to follow-up the gene expression changes it uncovered. Late prenatal immune activation led to a dysregulation of a variety of different genes in both brain areas, and to common prefrontal and striatal changes in myelin markers of functionality and stability. These findings were confirmed and extended by MRI and immunohistochemical results, while epigenetic analyses pointed to altered DNA methylation as a putative molecular mechanism underlying some of the transcriptional effects. Our study thus adds to the characterization of the longlasting molecular signatures of late prenatal immune activation in adult corticostriatal regions.
Our transcriptomic and epigenetic analyses in immunechallenged and control offspring were conducted using behaviorally and cognitively characterized animals. Advantages of this approach include that the examination of possible molecular effects takes place against the background of overt behavioral and cognitive phenotypes, which appears particularly important for (immune-mediated) neurodevelopmental disruption models that may contain a certain degree of litter-to-litter variability (Meyer et al. 2009). Here, we confirmed the deleterious effects of prenatal poly(I:C) exposure on spatial short-term memory and social interaction (Bitanihirwe et al. 2010a(Bitanihirwe et al. , 2010bRichetto et al. 2013), suggesting that prenatal viral-like immune activation leads to robust deficits in these cognitive and behavioral domains. Against the background of these impairments, we further identified transcriptomic, epigenetic, and neuroanatomical abnormalities in offspring exposed to prenatal immune activation. This approach thus allowed us to assess possible molecular correlates of the behavioral and cognitive abnormalities induced by prenatal immune challenge. For example, we identified a positive correlation between accumbal (and, to a lesser extent, prefrontal) expression levels of myelinationrelated genes such as MAG, MOG, MOBP and MAL and the animals' behavioral performance, especially when considering the domains of social interaction. These correlations, however, should be interpreted with caution in view of the relatively modest group sizes.
Our genome-wide gene expression analyses revealed transcriptomic alterations in a number of genes, many of which have been associated with schizophrenia. In particular, when comparing our results with the currently recognized risk genes for schizophrenia, 5 of the affected genes in the mPFC (ADORA2A, CNP, DRD2, MAG, PPP1RB1), and 21 of the affected genes in the NAc (APOD, CACNA 1B, DRD3, FOXP2, GLS, GRIN2A, HIST1H2BC, HRT1A, HRT2A, HRT4, MAG, MGST1, MYO16, OXT, PBRM1, PDYN, ST3GAL1, SULT4A1, TSNAX, VAMP4) fall into this list (Butler et al. 2016). In agreement with previous reports by Connor et al. (2012) and Smith et al. (2007), prenatal immune activation leads to less extensive gene expression changes in the mPFC (n = 116 genes) as compared to the NAc (n = 251 genes) (Smith et al. 2007;Connor et al. 2012). Moreover, consistent with these earlier reports (Connor et al. 2012), most transcripts in prenatally infected offspring show less than 2-fold changes from the control group. Such effect sizes are not unprecedented given the early prenatal timing of the environmental insult, which typically leads to pathological changes in brain and behavior that are widespread but often relatively mild in terms of effect size (Meyer et al. 2007).
Despite the magnitude and number of changes, these effects may still be pathophysiologically relevant and may prime the organism to altered neuronal functions when challenged with other environmental stressors or behavioral and cognitive demands.
Notably, some of our data are consistent with previous findings in the prenatal influenza infection model developed by Fatemi and colleagues, suggesting that at least parts of the transcriptomic changes induced by prenatal influenza exposure may be mediated indirectly via activation of the maternal immune system (Fatemi et al. 2012). As in the prenatal influenza model, we observed changes in the expression of FOXP2, PPP1R1B, DEAD box polypeptide, ATP13, CRYGE, and various other genes implicated in the pathophysiology of schizophrenia and autism (Fatemi et al. 2005(Fatemi et al. , 2008(Fatemi et al. , 2009a(Fatemi et al. , 2009b. The similarities between the 2 models are even more evident when considering the region-overlapping transcriptomic effects identified here. Indeed, among the 14 genes commonly affected in the mPFC and NAc, 6 of these (MAG, MOG, MOBP, MAL, CLDN11 and MYRF) are similarly affected by prenatal exposure to influenza (Fatemi et al. 2005(Fatemi et al. , 2009a and are involved in myelin functionality and stability. The consistency between our findings and those reported by Fatemi and colleagues thus suggests that reduced expression of markers of myelin stability and functionality could be a long-lasting molecular signature of various prenatal immune challenges. Additional support for this hypothesis also stems from recent proteomic analyses demonstrating similar effects of prenatal viral-like immune activation on myelination-related proteins (Farrelly et al. 2015). Indeed, Farrelly et al. (2015) also uncovered changes in myelinrelated proteins, such as MBP1 and rhombex 29, suggesting that prenatal infection may contribute to neurodevelopmental abnormalities through mechanisms involving myelin formation and functionality (Farrelly et al. 2015).
The effects of prenatal immune activation on myelinrelated dysfunctions are particularly interesting in light of the potential role of myelination and white matter abnormalities present in schizophrenia and other neurodevelopmental disorders (Haroutunian et al. 2014;Mighdoll et al. 2015;Chavarria-Siles et al. 2016). Indeed, myelin provides the basis for rapid impulse conduction in the central nervous system and acts as electrical insulation for the unsheathed axon, which both helps to preserve the amplitude and increase the conduction velocity of the propagating axonal potential (Nave and Werner 2014;Normand and Rasband 2015). Given these essential functions, it is not surprising that damage to the myelin structure has been implicated in a variety of neurodevelopmental disorders. The expression of MAG, MOG, MOBP, MAL and CLDN11 is physiologically enriched in myelin-forming oligodendrocytes and is downregulated in schizophrenic subjects (Hakak et al. 2001;Aston et al. 2004;Katsel et al. 2005;Le-Niculescu et al. 2009). Moreover, alterations in white matter, such as volume reductions in prefrontal areas and increased density in subcortical areas, morphologic abnormalities in oligodendroglia and myelin-related gene abnormalities have all been related to schizophrenia (Sanfilipo 2000a(Sanfilipo , 2000bConnor et al. 2011). Our findings here, together with those reported by Fatemi et al. and Farrelly et al. (Fatemi et al. 2005, 2009a, 2009bFarrelly et al. 2015), highlight that the pathological relationship between prenatal infection and neurodevelopmental psychiatric disorders involves the disruption of myelination-related processes.
We performed additional MRI and immunohistochemical investigations to examine whether abnormal expression of myelination-related genes may have an impact on gray and white matter structure. In general, our brain-wide voxel-wise analysis revealed sparse differences in volume and T 1 , but widespread decreases in T 2 and increases in the MWF between CON and POL offspring. In the NAc, we observed focal increases in T 1, decreases in T 2 , but no change in the MWF, whereas the mPFC of POL offspring showed significant decreases in T 2 and an apparent increase in the MWF. Our data represent the first application of mcDEPOST to a rodent model of an epidemiologically informed risk factor for neurodevelopmental disorders. A single prior clinical study has utilized mcDESPOT in a small number of adult individuals with ASD (N = 14) (Deoni et al. 2015). This revealed increases in T 1 bilaterally within the cerebellum, thalamus, and internal capsule; and in right temporal and occipital WM, accompanied by decreases in the MWF, but no change in T 2 . Our T 1 data, especially in the cerebellum, thus partially overlap with these data, but the majority of our findings go in the opposite direction. This may not be surprising since we only model a single disease factor, whereas most neurodevelopmental disorders such as ASD are clearly multifactorial in origin.
A previous longitudinal in vivo MRI study in rats has demonstrated that exposure to poly(I:C)-induced immune activation on GD 15 leads to an abnormal developmental trajectory of brain maturation (Piontkewitz et al. 2011). This raises the question of whether our cross-sectional MRI data are representative of a similar phenomenon in the mouse brain following immune activation on GD 17. Longitudinal in vivo MRI studies of typically maturing C57/Bl6 mice suggest 2 important findings in this context. First, most gray matter regions reach their final volume within the first 8 postnatal weeks (Hammelrath et al. 2015). Second, there is an initial decrease in T 2 from 3 to 8 weeks postnatal, after which T 2 steadily increases up to 24 weeks of age (Hammelrath et al. 2015). Taken together, our findings of widespread decreases in T 2 in POL offspring at 12 weeks postnatal suggest a delay in the normal process of mouse brain maturation following maternal immune activation. Although longitudinal in vivo MRI studies will be helpful to clarify this, we have recently shown that prenatal immune activation on GD 17 in mice leads to an "immature" cortical GABAergic network in early adulthood, supporting a "delayed maturation" hypothesis (Richetto et al., 2014). Importantly, rats and mice show differential timescales of brain maturation, thus these findings may not translate precisely across species (Mengler et al. 2014;Hammelrath et al. 2015) How do these macroscale findings relate to those at the transcriptional level, particularly with respect to myelination? Our postmortem immunohistochemistry analysis of the mPFC is in line with the gene expression data, with no change in MBP, but a decrease in MOBP. What then, underlines the changes in T 1 and T 2 , and in particular the seemingly paradoxical increase in the MWF in the PFC of POL offspring? Prior studies have shown that the MWF derived from a conventional multiple spin-echo MCR approach are strongly correlated with histological estimates of myelin content (Webb et al. 2003;Stanisz et al. 2004;Laule et al. 2008Laule et al. , 2006, but to the best of our knowledge this has not been done for the mcDESPOT MCR technique. We have recently shown that the mcDESPOTderived MWF is sensitive to demyelination of white matter tracts in the mouse brain induced by chronic cuprizone exposure (Wood et al. 2016a(Wood et al. , 2016b. Specifically, demyelinated regions showed higher values of T 1 and T 2 , with a corresponding decrease in the MWF (Wood et al. 2016a(Wood et al. , 2016b. In the current study, we observed increases in T 1 and decreases in T 2 in the NAc and mPFC, respectively (as well as in other brain regions). Although preliminary, due to low statistical power, our correlation analysis suggests that the increase in prefrontal MWF signals is not related to the area fraction of MBP, but is negatively correlated to T 2 in both CON and POL offspring. Furthermore, there is a negative correlation between MOBP area fraction and T 2 in the PFC in POL offspring only. These data are in good agreement with observations that T 1 and T 2 are influenced by the entry of precursory macromolecules for myelin and myelin-associated proteins during human brain maturation (Paus et al. 2001). Furthermore, this is consistent with the robust topographical overlap between decreases in T 2 and apparent increases in the MWF in our MRI dataset.
Taken together, this suggests that the increase in MWF is not caused by an increase in the quantity of myelin itself, but changes to its 'quality'. Speculatively, decreased levels of myelin stability proteins would be predicted to affect the microstructure of the myelin sheath, altering the g-ratio. For example, if the myelin sheath is less compact, the fraction of water trapped between lipid bilayers could potentially increase, thereby driving an apparent MWF increase. Importantly, The function of MOBP has been associated with the compaction and stabilization of myelin membranes (Gould et al. 2000;Montague et al. 2006). This hypothesis could be tested by exploiting recent advances in the combination of myelin content information (obtained with mcDESPOT MCR) with information about axonal microstructure obtained through multi-shell DTI (Melbourne et al. 2014;Dean et al. 2016). Changes in the relaxation properties of brain tissues may however also reflect other biological processes, including iron accumulation and changes in tissue water compartmentalization through changes in axon fiber size, density or coherence (Paus et al. 2001;MacKay et al. 2006). Whilst further investigations are therefore warranted, our current data provide initial evidence suggesting that abnormal expression of myelination-related genes can be plausibly linked to changes in tissue relaxation time. Nevertheless, they also show that further validation of the mcDESPOT model at the postmortem level in rodent models would be helpful to the field.
Lastly, our study provides the first report concerning hypermethylation of the MOBP promoter following prenatal immune activation. Hence, we identify an epigenetic mechanism that could possibly underlie the effects of prenatal infection on long-lasting transcriptomic changes in myelination-related genes. The identified correlations between MOBP DNA methylation and mRNA levels readily support this notion. It should be pointed out, however, that the correlative analyses between DNA methylation and gene expression levels were based on small sample sizes (N = 5-6 animals per group), and therefore, they might have been underpowered. Nevertheless, our findings are in agreement with previous studies showing prenatal immune activation in rats or mice can induce that various epigenetic changes such as DNA hyper-or hypomethylation, histone modifications, and altered micro-RNA expression, some of which likely affect the expression of corresponding genes (Connor et al. 2012;Tang et al. 2013;Basil et al. 2014;Labouesse et al. 2015;Richetto et al., 2016).
In conclusion, our study further characterizes the molecular signature of prenatal viral-like immune activation in the offspring's corticostriatal regions. In particular, prenatal infectioninduced transcriptomic changes in myelination-related genes seems to be a common pathological feature in multiple brain areas. The current study also provides the first voxel-wise assessment of brain volume and application of MCR methodology, which together suggest a putative delay in brain maturation following prenatal viral-like infection. Furthermore, the study provides preliminary evidence that these macrostructural abnormalities may be plausibly linked to abnormal expression of myelination-related genes in the absence of demyelination per se. Future investigations are therefore needed to confirm the mechanistic links between these molecular modifications and MR-detectable brain alterations occurring in prenatally infected offspring. Nevertheless, our data highlight the power of combining system level assessments (MRI) with invasive cellular and molecular phenotyping of the same animal postmortem in the elucidation of disease-relevant mechanisms.