Severe cerebellar malformations in mutant mice demonstrate a role for PDGF-C/PDGFRα signalling in cerebellar development

Platelet-derived growth factors (PDGFs) play a number of important roles during vertebrate development and organogenesis (Andrae et al., 2008). The PDGF family consist of four different ligands (PDGF-A, -B, -C and -D) that signal through two tyrosine kinase receptors (PDGFR-α and -β). PDGF-C binds primarily to PDGFR-α (Li et al., 2000). We have previously shown that expression of PDGF-C is needed for proper formation of the cerebral meninges. Genetically modified mice lacking PDGF-C (Pdgfc^−/−;Pdgfra^GFP/+) develop thin and malformed meninges in combination with neuronal over-migration from the cerebral cortex into the leptomeningeal space and spinal cord defects (spina bifida). Developmental defects are also seen outside of the central nervous system (Andrae et al., 2016).

The cerebellar development in mice starts around E8-9 with the formation of the isthmic organiser. The process continues for several weeks postnatally, with the final positioning of neurons in the different cell layers and the foliation that generates the typical pattern of the cerebellum. Each step in the process is dependent on specific genes, and a multitude of cerebellar malformations have been identified both in mice and in humans (Aldinger and Doherty, 2016; Haldipur and Millen, 2019). Cerebellar malformations are commonly accompanied by malformations elsewhere in the brain.

The human Dandy Walker malformation (DWM) affects the developing cerebellum and is characterised by cerebellar hypoplasia and upwards, dorsal rotation of the vermis (midline structure) in combination with enlargement of the fourth ventricle (Parisi and Dobyns, 2003). Dysplasia of the most posterior cerebellar lobes, referred to as vermian tail, can be observed in DWM using foetal MRI. The tail sign has been suggested to be specific for the DWM, but several questions remain unanswered regarding its clinical significance (Bernardo et al., 2015; Teresa and Kapur, 2018). Although much is still to learn about genetic alterations that cause DWM, Foxc1-mutant mice have been suggested as a model for DWM (Aldinger and Doherty, 2016; Aldinger et al., 2009; Haldipur et al., 2017) because they display similar cerebellar abnormalities, including defective cerebellar foliation, hypoplasia of the posterior vermis and defective layering of Purkinje cells (Haldipur et al., 2017). Foxc1 encodes a transcription factor (Foxc1) that regulates expression of SDF-1 (Cxcl12) in meningeal cells (Zarbalis et al., 2007), which in turn bind to CXCR4 on granule cells in the external granular layer. Loss of SDF-1/CXCR4 interaction causes cerebellar malformations due to defective migration of cells in the external granular layer (EGL) cells and Purkinje cells (Vilz et al., 2005; Aldinger et al., 2009; Haldipur et al., 2017).

Here we describe a cerebellar phenotype including hypoplasia and malformation of the posterior vermis in Pdgfc^−/−;Pdgfra^GFP/+-mice. The phenotype is similar to that reported for Foxc1-mutant mice, and we therefore suggest that also PDGF-C has a role in DWM. In addition, we show that denudation of the ependyma in the fourth ventricle is associated with the cerebellar phenotype in Pdgfc^−/−; Pdgfra^GFP/+ mice.

We found cerebellar hypoplasia combined with a specific loss of the posterior inferior vermis in Pdgfc^−/−;Pdgfra^GFP/+ mice. The phenotype was only observed in the combination of complete loss of Pdgfc and heterozygous loss of Pdgfra, suggesting that both PDGF-C and PDGF-A signaling via PDGFRα are important for cerebellar development. Abnormal cerebellar morphology was primarily seen in the caudal lobes closest to the rhombic lip at the sagittal midline (Fig. 1). All late embryos and newborn Pdgfc^−/−;Pdgfra^GFP/+ mice were affected (Fig. 2). The severity of the malformation varied slightly between individuals, which was in line with some mice surviving a few days longer than others. Pdgfc^−/−;Pdgfra^GFP/+ mice show several different developmental defects, but the exact cause of death is still not known (Andrae et al., 2016). Most mutant mice died at birth or had to be euthanised within a few days postnatally. As we have reported earlier (Andrae et al., 2016), Pdgfc^−/−;Pdgfra^GFP/+ individuals were readily spotted in newborn litters of mixed genotypes by the presence of a visible red stripe in the skin above the caudal spinal cord; this stripe is associated with an underlying spina bifida.

The first signs of malformations in the cerebellum of Pdgfc^−/−;Pdgfra^GFP/+ mice were observed at E15-16, in the area close to the rhombic lip (Fig. S1A,B). In control mice, the cerebellum expanded into a rounded appearance and as a consequence, the choroid plexus and the rhombic lip rotated ventrally into the fourth ventricle (Fig. 2E). The Pdgfc^−/−;Pdgfra^GFP/+ cerebellum instead displayed a flat and elongated shape with the rhombic lip positioned dorsally (Fig. 2F). At birth, the four primary fissures and the five cardinal lobes had started to form in control mice and were clearly visible at the sagittal midline (Fig. 2G,I). In mutant mice, the two rostral cardinal lobes appeared normal and a central lobe was present but had an abnormal shape. The caudal lobes were, however, essentially missing, which changed the position of the rhombic lip to a more dorsal position, away from the fourth ventricle (asterisk in Fig. S1F). A similar phenotype has been reported in Foxc1-mutant mice, where it was suggested to model the DWM (Haldipur et al., 2017).

The majority of the Pdgfc^−/−;Pdgfra^GFP/+ mice die shortly after birth (Andrae et al., 2016), before the cerebellum is fully developed. A limited number (< 5%) of Pdgfc^−/−;Pdgfra^GFP/+ mice displayed a somewhat less severe phenotype and survived until P15-P19, which gave us the opportunity to analyse the cellular organisation in a few postnatal brains. The overall organisation of granular-, molecular- and Purkinje cell layer and the white matter were largely intact in the rostral lobes (Fig. 1). At more posterior locations in the midline vermis, the neuronal layers appeared progressively disorganised, involving all cell layers. For example, a reduced number of EGL cells (Fig. 1D), ectopic clusters of granular cells, irregular multicellular layers of Purkinje cells, and an uneven molecular layer were observed (Fig. 3). The position of Purkinje cells in Pdgfc^−/−;Pdgfra^GFP/+ mice was abnormal already at birth, where the multicellular layer of calbindin-positive Purkinje cells was uneven and did not follow the shape of the developing lobes (Fig. 3A,D). At P15, ectopic clusters of Purkinje cells were still observed in some areas, whereas other areas had a more normal appearance with Purkinje cells positioned one-by-one (Fig. 3C,F). In the caudal cerebellum, which displayed the highest degree of cellular disorganisation, the pattern of cell nuclei was completely disrupted (Fig. 3L). Bergmann glia lacked their typical perpendicular orientation in mutants and were regionally missing, and glial endfeet were absent (Fig. 3L′). The orientation of radial glia in the anterior lobes was not as irregular, even though fewer glial endfeet seemed to connect at the pial surface compared to the control mice.

As a consequence of the altered cerebellar phenotype, the ventricular lining from the colliculus to the choroid plexus in the fourth ventricle was elongated in Pdgfc^−/−;Pdgfra^GFP/+ mice compared to the control mice (yellow dotted line, Fig. 4A,B). In addition, the cells lining the ventricular zone (VZ) closest to the rhombic lip (RL) in the mutant mice did not organise into a continuous layer of polarised ependymal cells (Fig. 4C,D). This loss of ependymal cells was visible by Hematoxylin and Eosin staining already at embryonal day 17.5 (E17.5) (encircled area in Fig. 2F). The mutant ependymal cells in this area had lost their epithelial-like morphology and were completely lacking apical podocalyxin (Podxl) expression (Fig. 4C′,D′). Loss of podocalyxin expression was largely restricted to the sagittal midline and was not seen in lateral sections (data not shown). To confirm the absence of ependymal cells in the posterior region of the ventricular zone, we used ependymal markers including microtubular protein βIV-tubulin (expressed in the cilia of the ependymal cells) and the cytosolic calcium-binding protein S100B (Fig. 5). Neither βIV-tubulin nor S100B were observed at the fourth ventricle midline in mutants, i.e. the region also lacking Podxl expression (Fig. 5B,D), which further confirms the loss of ependymal cells in this region. With the observed disrupted ependyma in Pdgfc^−/−; Pdgfra^GFP/+ mice, the brain parenchyma close to the rhombic lip would suffer direct exposure to cerebrospinal fluid present in the fourth ventricle. Disruption of the ependymal cell layer is also referred to as denudation (Sival et al., 2011). Previous studies have observed how denudation activates a repair mechanism by accumulation of astroglial processes (Sarnat, 1995) and thereby upregulate expression of glial fibrillary acidic protein (GFAP). In Pdgfc^−/−;Pdgfra^GFP/+ mice, GFAP was indeed upregulated and covered the area with disrupted ependymal cells (Fig. 5F), suggesting activated repair involving GFAP-positive astrocyte processes. We also analysed the expression of additional ependymal markers and junction proteins in the mutant mice (Fig. S2). In intact ependymal cells, the neuroepithelial junction protein n-cadherin (Cdh2) and the gap-junction protein connexin 43 (Cx43) were observed on the lateral/apical pole of the cells. In cells lacking podocalyxin, this expression was partly or completely lost from E17.5 (Fig. S2D). Earlier, at E15.5, we did not observe abnormal ependymal denudation and marker expression in this area.

As shown above, PDGF-C is necessary for proper development of the cerebellum. To better understand the process, we investigated the time and anatomical location of the onset of Pdgfc expression in the cerebellum. We took advantage of the Pdgfc knockout mice, in which introduction of a lacZ gene was used to generate the Pdgfc-null allele (Ding et al., 2004). X-gal staining in the hindbrain of mutant mice (Pdgfc^−/−;Pdgfra^GFP/+) was compared to heterozygous control mice (Pdgfc^+/−;Pdgfra^GFP/+) from E9.5 until birth. Pdgfc-lacZ was expressed in the notochord and floor plate close to the hindbrain as early as E9.5 (data not shown). Within the cerebellar tissue, the first expression was visualised in the caudal part of the cerebellar primordium at E12.5 (Fig. 6A-C). Interestingly, at this stage the only putative target cells of PDGF-C (i.e. cells expressing Pdgfra) were positioned in the overlying mesenchyme (Fig. 6G,H). In mutant embryos, X-gal staining was stronger than in control embryos. This was most likely due to the presence of double copies of lacZ in the mutant mice (Fig. 6D-F). At E16.5, when cerebella in Pdgfc^−/−;Pdgfra^GFP/+ embryos could be morphologically distinguishable from control embryos, strong lacZ expression was seen in granule cells of the transient EGL. At this stage, EGL cells are known to migrate anteriorly from the rhombic lip to cover the newly fused cerebellar anlagen (Fig. S1). Mutant embryos displayed a more widespread expression pattern suggesting ectopic migration of Pdgfc-expressing cells or, alternatively, the compensatory upregulation of Pdgfc by other cell types (Fig. S1B,D,F) .

We wanted to identify changes in gene expression that were associated to the loss of PDGF-C in the developing cerebellum. Knowing that the Pdgfc expression in the cerebellar area was initiated around E12.5 (Fig. 6C) and morphological abnormalities were observed around E15.5 (Fig. S1A), we hypothesised that secondary gene expression changes resulting from the loss of PDGF-C signalling might occur between E12.5 and E15.5. Microarray gene expression analyses were therefore performed on RNA from cerebellar tissue (liberated from meninges) at E13.5 and E14.5. At E13.5, gene expression was almost identical in Pdgfc^−/−;Pdgfra^GFP/+ mutants and control pups. The limited number of differently regulated genes observed could be explained by the sex distribution in the litter. However, 1 day later (at E14.5), a significantly higher number of differentially expressed genes were identified in the Pdgfc^−/−;Pdgfra^GFP/+ cerebellum compared to the controls (fold change >2, P<0.05) (Fig. 6I). Taken together, this fit well into the timeframe, with loss of Pdgfc expression at E12.5, transcriptional changes and altered gene expression around E14.5 and morphological changes around E15.5.

The large number of genes downregulated at E14.5 indicated that PDGF-C is an important player in the developing cerebellum (Table S1). Most of the GO terms were related to development and cellular processes, e.g. G-protein coupled receptor signalling pathway, sensory perception and nervous-system processes (Table S2). The top downregulated genes at E14.5 were manually analysed and compared to the published literature. A few relevant genes were identified; e.g. the ependyma-specific genes Mak and Bbox1 were totally missing in mutant mice, the Purkinje cell marker Calb1 was downregulated >5 fold and meningeal markers as Cxcl12 and Crym (DeSisto et al., 2020) were downregulated −3.24 fold and −2.41 fold, respectively. Among the downregulated genes, there were also a large number of olfactory and vomeronasal receptors, which we could not connect to specific cells.

A third microarray analysis was performed at P0, this time on both cerebellar tissue and the overlying meninges from mutant and control mice. Very few significant gene expression changes were observed in the cerebellar tissue, but in the mutant meninges, there were clear gene expression changes (Table S3). The differentially downregulated genes were highly associated to Vasculature development and similar GO terms (Table S4). Among the differentially downregulated genes in the meninges of Pdgfc^−/−;Pdgfra^GFP/+ mice were e.g. Col15a1 (−3.2 fold), Fmod (−2.75 fold), Vnn1 (−4.18 fold), Slc22a2 (−3.19 fold) and Cxcl12 (−1.7 fold). These genes reflect different meningeal fibroblasts (Del Gaudio, manuscript in preparation).

In this study, we demonstrate that PDGF-C plays an important role during the development of the mouse cerebellum. Loss of the Pdgfc gene, and thereby reduced activation of the PDGFRα, affected the cerebellar formation already during the embryonal stage. Formation of the cerebellum is complex and in the Pdgfc^−/−;Pdgfra^GFP/+ mice several distinct defects were identified; e.g. hypoplasia of the posterior vermis associated to a relative change in the position of the rhombic lip, loss of ependymal cells in the fourth ventricle, ectopic migration of EGL cells, and uneven and multicellular layers of Purkinje cells. As a result, postnatal mice presented cerebella with defective foliation and a complete lack of the most caudal lobes of the cerebellum.

Many processes occur from the start of Pdgfc expression (E12.5) in the developing embryonic cerebellum to the fully mature cerebellum postnatally. One interesting aspect is that whereas Pdgfc was expressed within the cerebellar tissue, we could only detect possible responding Pdgfra positive cells in the meninges and in the surrounding mesenchyme. The most distinct Pdgfc expression was in the EGL-cells. This pattern coheres with scRNAseq data from the developing human cerebellum, where PDGFC is expressed in EGL cells and PDGFRA in oligodendrocyte progenitor cells, meninges and choroid plexus/ependyma (Aldinger et al., 2021).

Interactions between meningeal cells and cells in the developing brain are necessary for proper formation of the CNS. The importance of signalling between EGL cells and meningeal cells is well known for proper foliation of the cerebellum (Sievers et al., 1981). Foxc1 is only expressed in the meninges and still has such an important impact on formation both on the cerebellum (Aldinger et al., 2009) and on the cerebral cortex (Siegenthaler et al., 2009).

The cerebellar phenotype of the Pdgfc^−/−;Pdgfra^GFP/+ mice showed great similarity to Foxc1 mutant mice (Haldipur et al., 2014,, 2017). FOXC1 loss in mice contributes to the formation of typical anomalies used to diagnose the human cerebellar DWM; hypoplasia of the cerebellar vermis, upward rotation of the vermis away from the brainstem and an enlargement of the fourth ventricle (Haldipur and Millen, 2019). Foxc1 is expressed in the meninges, where reduced levels of FOXC1 leads to reduced levels of SDF-1 (Cxcl12) (Zarbalis et al., 2007). The levels of SDF-1 in the meninges were reduced also in newborn Pdgfc^−/−;Pdgfra^GFP/+ mice. This further strengthen the hypothesis of a common signalling pathway responsible for the cerebellar phenotypes in Pdgfc^−/−;Pdgfra^GFP/+ mice and in Foxc1 mutant mice. In addition, a genetic interaction between Foxc1 and Pdgfra has been indicated in a previous study on zebrafish, where Pdgfra was suggested to act downstream of Foxc1 (French et al., 2014).

The cellular interface between the fourth ventricle and brain parenchyma in the cerebellum was disrupted in Pdgfc^−/−;Pdgfra^GFP/+ mice. The lack of intact ependymal cells was confirmed as the polarised Podxl expression was replaced by upregulated Gfap expression in the ventricular zone close to the rhombic lip. Ependymal disruptions (also called ependymal denudation) can be explained by, for example, mechanical forces and by increased pressure in the ventricular system. Ependymal cells are generated from radial glial cells (Spassky et al., 2005). The morphology and position of Bergman glia were disrupted in other areas of the Pdgfc^−/−;Pdgfra^GFP/+ mice, hence migration defects could be one possible explanation. Further, it has been suggested that alterations in Cdh2 (N-cadherin) expression proceeds ependymal denudation (Oliver et al., 2013). N-cadherin is needed to keep the ependymal layer intact and to maintain communication between the ependymal cells, and N-cadherin expression was altered in the Pdgfc^−/−;Pdgfra^GFP/+ mice. Denudation of the ependyma in Pdgfc^−/−;Pdgfra^GFP/+ mice was consistently and exclusively observed in the normally acute angle of the ventricular zone, shortly after the first signs of vermis hypoplasia and mis-location of the rhombic lip. Hence, another possible explanation is that mechanical stretching and tearing, enforced by the rotation of the vermis, contributed to the disrupted ependyma.

We were able to visualise early gene expression changes in the developing cerebellum of E14.5 embryos. A drawback with microarray analysis of a whole tissue is that gene expression changes in individual cell types are easily masked and hidden, and it was difficult to pinpoint relevant downregulated genes in the bulk-RNA study. A large number of the differently regulated genes were olfactory and vomeronasal receptors. Expression of olfactory receptors outside of the olfactory tissues (e.g. in the developing mouse cortex) have been detected in various gene expression analyses (Feldmesser et al., 2006; Otaki et al., 2004).

In summary, Pdgfc^−/−;Pdgfra^GFP/+ and Foxc1 mutant mice share several phenotypes in the developing brain, as disruption of the cortical basement membrane, cerebrovascular defects and hemorrhage (Andrae et al., 2016; Mishra et al., 2016; Zarbalis et al., 2007) and the cerebellar defect presented here. Both models present cerebellar vermis hypoplasia with e.g. ectopic granule cells, multilayers of Purkinje cells, dysmorphic Bergman glia and severe malformation of the most posterior lobes adjacent to the rhombic lip. Whether migration of granule cell precursors in the rhombic lip of Pdgfc^−/−;Pdgfra^GFP/+ mice is affected and if the organisation of radial glial cells is defective throughout the cerebellum remain to be analysed.

In summary, we show that PDGF-C has a role in the complex network required to build a proper cerebellum and we suggest a potential role in the formation of the DWM.

All experiments were conducted in accordance with Swedish animal welfare legislation. Ethical permits were approved by the regional research animal ethics committee in Stockholm (N24/06, N72/07, N33/10, N15/12) and in Uppsala (C225-12, C115/15, 5.8.18-03029-2020). Mouse strains used were: Pdgfc knockout mice Pdgfc^tm1Nagy (Ding et al., 2004), where the lacZ gene replaces Pdgfc; and Pdgfra^GFP/+ (Pdgfra^{tm11(EGFP)Sor}, Hamilton et al., 2003), where expression of nuclear green fluorescent protein (GFP) replaces Pdgfra. The two strains were crossed to generate double transgenic mice Pdgfc^−/−;Pdgfra^GFP/+. Littermate Pdgfc^+/+;Pdgfra^GFP/+ mice were used as controls, if not indicated differently. The majority of the mice were on a pure genetic background (C57BL/6J). Mice on a mixed genetic background (C57BL/6J;129S1) were also analysed and displayed an identical phenotype. Mutant Pdgfc^−/−;Pdgfra^GFP/+ mice were easily recognised with a hemorrhagic stripe over the lower spine (Andrae et al., 2016). Genotyping for Pdgfc was carried out with PCR and expression of Pdgfra-GFP was confirmed using a fluorescence lamp. Both female and male mice were included in the study.

Embryos and newborn mice were euthanised by decapitation, brains were dissected out, washed in PBS and fixed overnight in 4% paraformaldehyde (PFA). Mice older than 1 week were perfused through the heart with Hanks’ balanced salt solution (HBSS) and 4% PFA before dissection, and postfixation overnight. Fixed brains were either dehydrated, embedded in paraffin and sectioned at 7 µm or vibratome sectioned at 75 µm.

Primary antibodies at the following concentrations were used: goat-anti-mouse Podxl (1:200, R&D Systems, AF1556); rabbit-anti-GFAP (1:100, Invitrogen, 13-0300); rabbit-anti-GFAP (1:200, Dako, Z0334); rabbit-anti-S100B (1:100, Dako, Z0311); rabbit-anti-βIV tubulin (1:500, Abcam, ab179509); rat-anti-Cdh2 (1:50, Developmental Studies Hybridoma Bank, MNCD2-C); mouse-anti-Cx43 (1:50, Santa Cruz Biotechnology, sc-271837), rabbit-anti-Calbindin (1:200, Sigma-Aldrich, D2724). Secondary antibodies were produced in donkey and conjugated to Alexa488, Cy3, Alexa 568 or A647 (ThermoFisher Scientific). Additional antibody information in Table S5. All stainings were performed together with negative controls, where the primary antibodies were excluded. All stainings were performed on tissue from at least three different animals.

Paraffin sections were deparaffinised in xylene and graded series of EtOH. Antigen retrieval in Target Retrieval Solution pH 9 (S2367, Dako, Glostrup, Denmark) was used for Podxl, Cdh2 and Cx43. Sections were washed in PBS, blocked in 1% bovine serum albumin (BSA), 0.5% Triton X-100 (T8787, Sigma-Aldrich), 2.5% normal donkey serum in PBS for a minimum of 30 min at room temperature (RT). Incubation with primary antibody was in 0.5% BSA, 0.25% Triton X-100 (T8787, Sigma-Aldrich) in PBS overnight at +4°C. Sections were washed in PBS, incubated with secondary antibodies for a minimum of 1 h at RT, washed in PBS and stained with Hoechst (1:10,000 in PBS). Sections were mounted with ProLongGold antifade mountant.

Free-floating vibratome sections were stained as above, but slowly shaking and with longer incubation times to allow for proper penetration into the tissue. Blocking was overnight, primary antibodies were incubated for 48 h and secondary antibodies were incubated overnight. Sections were washed in PBS with 0.1% Tween 20 (P1379, Sigma-Aldrich).

X-gal staining was performed on Pdgfc^−/−;Pdgfra^GFP/+ and Pdgfc^+/−;Pdgfra^GFP/+ control mice. The method for x-gal-staining has been described before (Andrae et al., 2014). Up to E12.5, embryos were stained as whole mounts. From E13.5, brains were freely dissected before staining. X-gal-stained tissues were embedded in paraffin, sectioned at 7 µm and counterstained with Nuclear Fast Red (N3020, Sigma-Aldrich).

Hematoxylin and Eosin staining were performed on paraffin sections. Briefly, sections were deparaffinised, rehydrated and immersed in Mayer's Hematoxylin (HTX) (01820, HistoLab, Sweden) for 2 min. Blueing was performed in running tap water for 5 min before sections were immersed in Eosin (01650, HistoLab, Sweden) for 30 s. Sections were dehydrated and mounted in PERTEX (00811, HistoLab, Sweden).

Cerebellar tissue, cleared from meninges, from mutant embryos (Pdgfc^−/−;Pdgfra^GFP/+) were compared to all other embryos in the same litter (Pdgfc^+/−;Pdgfra^GFP/+, Pdgfc^+/+;Pdgfra^GFP/+, Pdgfc^−/−, Pdgfc^+/− and Pdgfc^+/+). Three litters were included; at E13.5 (three mutants and five controls), at E14.5 (two mutants and seven controls, and at P0 (four mutants and nine controls). From the P0-litter, the meninges were isolated and analysed separately. Dissected tissue was stored in RNAlater®(Ambion) before RNA isolation using the RNeasy microkit (Qiagen). RNA quality was checked in a 2100 BioAnalyzer (AgilentTechnologies, Santa Clara, CA, USA). Transcription profiling was performed with the Gene Chip Mouse Gene 1.0ST array, at the Bioinformatics and Expression analysis core facility, Karolinska Institute, Sweden (http://apt.bea.ki.se). Affymetrix raw data was normalised using PLIER algorithms (AffymetrixTechnical Note, Guide to Probe Logarithmic Intensity Error Estimation, http://affymetrix.com/support/technical/technotesmain.affx). To compare the gene expression difference between the mutant and control, t-test was applied and significance cut-off values were >2 in fold change with a P-value <0.05. Multiple-test correction was performed using the False Discovery Rate (FDR) method in R (Benjamini and Hochberg, 1995). To identify the functional enrichment among the differentially expressed gene list, GO analysis was performed using R package GOstat packages (version 2.48.0), GO.db (version 3.6.0) and org.Mm.eg.db (version 3.6.0). The P value <0.001 was used to select the significant enriched GO categories.

The entire microarray data set has been deposited in the NCBI GeneExpression Omnibus database (www.ncbi.nlm.nih.gov/geo/, accession numbers GSE77049 and GSE202835).

We thank Jana Chmielniakova, Pia Petersson and Cecilia Olsson for technical assistance with animals.