“Integrative genomics study of microglial transcriptome reveals effect of DLG4 (PSD95) on white matter in preterm infants”

Preterm birth places newborn infants in an adverse environment that leads to brain injury linked to neuroinflammation. To characterise this pathology, we present a translational bioinformatics investigation, with integration of human and mouse molecular and neuroimaging datasets to provide a deeper understanding of the role of microglia in preterm white matter damage. We examined preterm neuroinflammation in a mouse model of encephalopathy of prematurity induced by IL1B exposure, carrying out a gene network analysis of the cell-specific transcriptomic response to injury, which we extended to analysis of protein-protein interactions, transcription factors, and human brain gene expression, including translation to preterm infants by means of imaging-genetics approaches in the brain. We identified the endogenous synthesis of DLG4 (PSD95) protein by microglia in mouse and human, modulated by inflammation and development. Systemic genetic variation in DLG4 was associated with structural features in the preterm infant brain, suggesting that genetic variation in DLG4 may also impact white matter development and inter-individual susceptibility to injury. Preterm birth accounts for 11% of all births 1, and is the leading global cause of deaths under 5 years of age 2. Over 30% of survivors experience motor and/or cognitive problems from birth 3, 4, which last into adulthood 5. These problems include a 3-8 fold increased risk of symptoms and disorders associated with anxiety, inattention and social and communication problems compared to term-born infants 6. Prematurity is associated with a 4-12 fold increase in the prevalence of Autism Spectrum Disorders (ASD) compared to the general population 7, as well as a risk ratio of 7.4 for bipolar affective disorder among infants born below 32 weeks of gestation 8. The characteristic brain injury observed in contemporary cohorts of preterm born infants includes changes to the grey and white matter tissues, that specifically include oligodendrocyte maturation arrest, hypomyelination and cortical changes visualised as decreases in fractional anisotropy 9–13. Exposure of the fetus and postnatal infant to systemic inflammation is an important contributing factor to brain injury in preterm born infants 12, 14, 15, and the persistence of inflammation is associated with poorer neurological outcome 16. Sources of systemic inflammation include maternal/fetal infections such as chorioamnionitis (which it is estimated affects a large number of women at a sub-clinical level), with the effect of systemic inflammation in the brain being mediated predominantly by the microglial response 17. Microglia are unique yolk-sac derived resident phagocytes of the brain 18, 19, found preferentially within the developing white matter as a matter of normal developmental migration 12. Microglial products associated with white matter injury include pro-inflammatory cytokines, such as interleukin-1β (IL1B) and tumour necrosis factor α (TNF-α)20, which can lead to a sub-clinical inflammatory situation associated with unfavourable outcomes 21. In addition to being key effector cells in brain inflammation, they are critical for normal brain development in processes such as axonal growth and synapse formation 22, 23. The role of microglia in neuroinflammation is dynamic and complex, reflected in their mutable phenotypes including both pro-inflammatory and restorative functions 24. Despite their important neurobiological role, the time course and nature of the microglial responses in preterm birth are currently largely unknown, and the interplay of inflammatory and developmental processes is also unclear. We, and others, believe that a better understanding of the molecular mechanisms underlying microglial function could harness their beneficial effects and mitigate the brain injury of prematurity and other states of brain inflammation25, 26 A clinically relevant experimental mouse model of IL1B-induced systemic inflammation has been developed to study the changes occurring in the preterm human brain 27, 28. This model recapitulates the hallmarks of encephalopathy of prematurity including oligodendrocyte maturation delay with consequent dysmyelination, associated magnetic resonance imaging (MRI) phenotypes and behavioural deficits. Here, we take advantage of this model system to characterise the molecular underpinnings of the microglial response to IL1B-driven systemic inflammation and investigate its role in concurrent development. In preterm infants MRI is used extensively to provide in-vivo correlates of white and grey matter pathology, allowing clinical assessment and prognostication. Diffusion MRI (d-MRI) measures the displacement of water molecules in the brain, and provides insight into the underlying tissue structure. Various d-MRI measures of white matter have been associated with developmental outcome in children born preterm 29–32, with up to 60% of inter-individual variability in structural and functional features attributable to genetic factors 33, 34. White matter abnormalities are linked to associated grey matter changes at both the imaging and cellular level 10, 35, 36, with functional and structural consequences lasting into adulthood 37, 38. Tract Based Statistics (TBSS) allows quantitative whole-brain white matter analysis of d-MRI data at the voxel level while avoiding problems due to contamination by signals arising from grey matter 39. This permits voxel-wise statistical testing and inferences to be made about group differences or associations with greater statistical power. TBSS has been shown to be an effective tool for studying white matter development and injury in the preterm brain 40, providing a macroscopic in vivo quantitative measure of white matter integrity that is associated with cognitive, fine motor, and gross motor outcome 11, 41, 42. In this work we take a translational systems biology approach to investigate the role of microglia in preterm neuroinflammation and brain injury. We integrate microglial cell-type specific data from a mouse model of perinatal neuroinflammatory brain injury with experimental ex vivo and in vitro validation, translation to the human brain across the lifespan including analysis of human microglia, and assessment of the impact of genetic variation on structure of the preterm brain. We add to the understanding of the neurobiology of prematurity by: a) revealing the endogenous expression of DLG4 (PSD95) by microglia in early development, which is modulated by developmental stage and inflammation; and b) finding an association between systemic genetic variability in DLG4 and white matter structure in the preterm neonatal brain.

development and inter-individual susceptibility to injury.
Taken together, these analyses suggest that the gene co-expression relationships observed in 280 microglial cells upon in vivo IL1B treatment are at least in part conserved at the protein level; 281 these relationships can be synthetized by two major protein-protein interaction modules (SPNs).

282
Therefore, these SPNs might represent functional modules in protein-protein interaction networks 283 that have been detectable in response to IL1B exposure and development. These SPNs are 284 functionally distinct and enriched for diverse disease gene annotations, with SPN1 specifically 285 enriched for genes involved in neuropsychiatric disorders linked to prematurity, such as DLG4 in 286 schizophrenia and autism [70][71][72] , SHANK1 in autism 73 and CAMK2A in several phenotypes 74 .

288
Transcriptional regulation of SPNs

289
Following the identification of gene co-expression networks and the resulting functional modules 290 at the protein level (SPN1 and SPN2), we set out to examine their potential regulation by 291 transcription factors (TFs) 75 and also searched for potential regulatory relationships mediated by 292 TFs involving the members of SPN1 and SPN2. TFs can determine coordinated expression of 293 several target genes (i.e., resulting in a co-expression network) by binding directly to DNA (e.g.,

294
at gene promoters), a process that can be mediated by other TFs or proteins that do not 295 themselves interact with DNA directly 76 .

297
To identify candidate TFs for the regulation of SPN1-2, we used the PASTAA algorithm 77 to 298 analyse the transcription factor binding-site (TFBS) motifs at the promoters of genes that encode 299 for the proteins defining both SPNs (Methods). This analysis across SNP1-2 indicated that a unsupervised genome-wide network analysis, we were interested to note that the combined PPI 305 and TF analysis revealed a possible functional relationship between two protein interaction sub-306 networks. The same relationship was not observed between the STATs TFs and SPN2 genes 307 (Supp. Table 17). The link between the STAT family of TFs and SPN1 genes was also supported 308 by an independent analysis using the Gene Set Enrichment Analysis (GSEA) tool in the 309 Molecular Signatures Database (MSigDB) database (Broad Institute) 78 to assess the overlap 310 between 615 gene sets (containing genes that share a common TFBS, defined in the TRANSFAC or primary microglia were stimulated by IL1B + IFNg and exposed to vehicle or a small molecule IFNg induced expression of several pro-inflammatory and immunomodulatory markers and

406
Overlay, Supplementary Video 2). This DLG4 staining disappears by P3, and the protein is still 407 absent from IBA1 positive cells at P45 (data not shown). Following exposure to IL1B however, 408 DLG4 can still be seen at the microglial membrane at P3 along with an apparent change in 409 morphology from a fully ramified to a slightly more amoeboid form ( Figure 6

449
To describe in more detail the regional pattern of expression in the human brain including white 450 matter, we used human brain whole tissue gene expression data from the UK Brain Expression

500
between infants with or without the minor allele (A) for SNP rs17203281, which was consistent 501 and replicated in both cohorts ( Figure 9). There were no significant differences in other clinical 502 features (gestational age at birth, age at scan, days of ventilation) between infants with or without 503 the minor allele (Supp.   . Table 21), of which one SNP (rs3826408) is in high 540 linkage disequilibrium (i.e., is a proxy SNP) 115 with rs17203281 (r 2 = 0.667, D' = 1). Therefore regulation of DLG4 mRNA expression in the brain, in particular, including the SNP rs17203281 543 which we previously and independently associated with white matter features in preterm infants.
epilepsy in an independent gene network study using exome sequencing data in cohorts of 548 patients with neuropsychiatric disorders 70 . This identified DLG4 within a small module of 24 549 genes involved with synaptic function, in which de novo and more severe missense mutations 550 were more likely in individuals with significantly higher intellectual impairment. We 551 investigated the overlaps between the main gene modules reported in 70 and our SPN1 and SPN2 552 genes using a formal test for intersecting gene lists 116 (Supp.  553   15). We found that there were significant overlaps between SPN1/2 and the reported disease-554 associated modules. In particular, we found that DLG4 was the most frequently occurring gene in 555 these overlaps (11/51 occurrences) followed by another SPN1 gene, CAMK2A (4/51). Also of 556 note, our identification of two networks SPN1/2 was broadly captured by the autism Modules 1/2 557 (reported in 70 ), with DLG4 is being found in Module 2, considered by the authors to be the 558 synaptic module, in keeping with our interpretation of the broad function of SPN1.

560
In summary these observations suggest that DLG4 genetic variation at the level of the organism 561 affects white matter structure in preterm infant, with potential broader implications for associated 562 neuropsychiatric disorders. The latter is suggested by previously published genetic susceptibility 563 data in patients with neuropsychiatric disorders. Our study suggests that the observed differences 564 in white matter features in preterm infants might be mediated by the rs17203281 variant, 565 variation in which is in turn linked to altered gene expression of DLG4 in the human brain.

569
In this work, we present a translational systems biology study of the effects of IL1B and 570 development on the microglial transcriptome, in which we integrate mouse and human data from 571 complementary sources to gain a deeper understanding of the neurobiology of preterm brain 572 development and injury. The primary finding of this study is that DLG4 (PSD95) is expressed in microglia during early development and that this developmentally regulated expression pattern is 574 altered by neuroinflammation associated with brain damage in preterm born infants. Specifically, using a mouse model of preterm brain inflammation we analyse cell-type specific patterns of gene 576 co-expression, protein interactions and transcriptional regulation. We identify two networks 577 (SPNs) of interacting proteins, one of which (SPN2) is predicted to interact with the other 578 nervous system-oriented network (SPN1) via the transcriptional regulation by STAT3 of its 579 targets in SPN1. STAT3 TF signalling has previously been linked to microglial immune activity 580 in response to stimuli including lipopolysaccharide (LPS), and the Stat3 gene has been found to 581 be involved in neuroprotective pathways upregulated in microglia from old mice 117 . We 582 prioritised DLG4 (PSD95) as a key player given its role as hub of the SPN1 network, its 583 biological importance in neurodevelopment and previously unknown importance in microglia.

584
We document novel dynamics of Dlg4 mRNA and of DLG4 protein that are endogenous to 585 microglia, and modulated by inflammation and development. In sum, there appears to be a 586 biologically driven discrepancy between Dlg4 gene expression and DLG4 protein synthesis in 587 response to IL1B, which we hypothesise might be due to post-translational mechanisms 588 modulated by inflammation that have been reported to play an important role in DLG4 protein 589 biology (review in 102 ). We translate these findings to humans by showing evidence of 590 developmentally regulated DLG4 gene expression throughout human brain tissue, and similarly 591 localise DLG4 protein expression to the membrane of developing human brain microglia. We 592 investigate possible effects of the DLG4 gene in-vivo in preterm infants, and show a significant 593 impact of common genetic variation in DLG4 (rs17203281) on preterm human brain imaging 594 features in two independent cohorts. This DLG4 effect on white matter could be due to measured 595 differences in expression of DLG4 mediated through a possible cis-eQTL action of rs17203281,

596
suggesting that inter-individual genetic variability in DLG4 gene could affect the response to 597 perinatal inflammation linked to IL1B.

599
The importance of DLG4 within this biological context could be due to various factors. From a 600 network topology perspective, SPN1 has a star configuration secondary to its central hub DLG4

601
(where a star consists of a single central node, or network-hub, that is connected to several 602 peripheral nodes). It has been suggested that the star network topology may have specific

663
although we cannot in this study specifically attribute this effect to the action of microglia.

665
Regarding a potential role for DLG4 protein in microglia and using our in vitro model of 666 microglial activation (exposure to IL1B+ IFNg), we investigated the effects on activation state of 667 blocking the archetypal interaction between the DLG4 protein and the N-methyl-D-aspartate 668 receptor (NMDAR) using a TAT-N-dimer that specifically disrupts the molecular interaction 669 between DLG4 and the NMDAR N2B subunit 130 . We have previously shown that microglia 670 express functional NMDAR and that these have a moderate but significant effect on microglial 671 activation state and response of the brain to perinatal injury 131 . We did not see effects of between DLG4 protein and the NMDAR and its known functions in neurons, we conjecture that glutamatergic and GABAergic signalling, which has been investigated in inflammation and injury In this work we use cell-type specific microglia data from a relevant mouse model, as well as 705 cell-type specific data from microglia in developing human brain tissue, plus human white matter genetic variation and gene expression to seek complementary insight into cellular-level genomics approach, since genotype data from infants reflect their systemic profiles while current relationship observed in the imaging-genomics analysis between DLG4 variability and brain 712 structure in preterm infants is general, and not attributable specifically to microglia. However,

713
we believe that we have provided a valuable attempt to bridge these conceptual gaps with 714 supporting biological validation.

716
In focusing our imaging analysis on brain features that best co-localise anatomically with white 717 matter, we hope to capture brain regions that are likely to exhibit the effects of microglial activity 718 since it has been observed that the developing white matter is highly populated with microglia in 719 the perinatal period in both humans and mouse 106 , and even in adults microglia are typically 720 much more numerous in white matter than in the corresponding regions of neocortex 108, 145 . It 721 has previously been discussed that TBSS is a reductionist approach to image analysis, and that the anisotropy. In addition we note that our analysis is oriented towards identifying an association 729 between genetic variability and general diffusion properties of white matter rather than attempting 730 a precise spatial localization of this effect within white matter, which would constitute a 731 subsequent specific enquiry making use of higher quality imaging than was available for the 732 current study. We are also currently collecting data to extend a similar approach to the study of     instructions. In brief, 50,000 primary microglial prepared as described above were plated in 48 995 well plates and after 12 hours of incubation with IL1B at 50 ng/mL + IFN-at 20ng/ml in the 996 presence or absence of TAT-N-dimer at 30 nM, medium was changed to serum free media 997 containing the recommended suspension of bioparticles. Cells were incubated for five hours, Forty-eight hours after plating, human CD11B+ microglia were treated for 4 hours with DMEM we had too few cells available to perform a dose response. We had confidence that the high 1073 potency of LPS would lead to a pro-inflammatory response and this was confirmed by the 1074 increased expression of TNFa (Supplementary Figure 13).

1076
Immunohistochemistry of isolated ex-vivo human microglia and brain sections   image and corrected for differences in spatial distortion due to eddy currents. Non-brain tissue 1214 was removed using the brain extraction tool (BET) 185           the human forebrain and possible involvement in periventricular white-matter injury of preterm