mTOR inhibition in primary neurons and the developing brain represses transcription of cholesterol biosynthesis genes and alters cholesterol levels

Dysregulated mammalian target of rapamycin (mTOR) activity is associated with various neurodevelopmental disorders ranging from idiopathic autism spectrum disorders to monogenic syndromes as for example Tuberous sclerosis complex. Thus, maintaining mTOR activity levels in a physiological range is essential for brain development and functioning. Upon activation, mTOR regulates a variety of cellular processes such as cell growth, autophagy and metabolism. On a molecular level, however, the consequences of mTOR activation are not well understood, especially in the brain. Thus, while it was shown that in cells outside the central nervous system mTORC1 activity is necessary for activating gene transcription of different metabolic pathways this mechanism is ill defined in the brain. By combining mTORC1 inhibition with RNA-sequencing we identified numerous genes of the sterol/cholesterol biosynthesis pathway to be downstream targets of mTORC1 in vitro in primary neurons and in vivo in the developing cerebral cortex of the mouse. Of note, reduced expression of these genes upon mTORC1 inhibition translated into reduced cholesterol levels. We further show that while mTORC1 does not regulate chromatin accessibility or RNA stability of these genes it drives transcription of their DNA. Using a bioinformatics approach, we identified binding sites for the transcription factors SREBP, SP1 and NF-Y to be enriched in the promoters of mTORC1 target genes and confirmed binding of NF-YA by ChIP-qPCR. Altogether, our results indicate that mTORC1 is an important regulator of the expression of sterol/cholesterol biosynthesis genes in the developing brain. Altered expression of these genes may be an important contributing factor in the pathogenesis of neurodevelopmental disorders associated with dysregulated mTOR signaling.


Introduction
Although changes in the activity of the mTOR pathway are associated with numerous neurodevelopmental disorders the molecular basis of disease development remains poorly understood. The mTOR pathway is a central signaling pathway in the cell that controls gene expression at multiple levels. While mTOR is most famous for its role in regulating translation of specific target mRNAs it in addition regulates RNA stability and gene transcription [1][2][3][4][5][6][7]. In proliferating cells such as fibroblasts and cancer cells, mTORC1 controls transcription of glycolytic, lipid and lysosome biogenesis and mitochondrial metabolism genes [1,3]. It activates several transcription factors involved in metabolic processes including sterol regulatory element binding proteins (SREBPs) and hypoxia inducible factor-1α (HIF1α) [1,[8][9][10].
Most likely, dysregulation of transcriptional processes contributes to disease development in neurodevelopmental disorders that are associated with changes in mTOR activity -yet mTOR mediated transcriptional regulation in the brain is heavily understudied.
In the brain, the mTOR signaling pathway controls various different processes such as neuronal differentiation, neuronal cell size, axon guidance, dendritogenesis and synaptic plasticity [11]. Dysregulation of the mTOR pathway is associated with neurodevelopmental disorders including epilepsy, autism spectrum disorder (ASD) and intellectual disability (ID). In this context, both hyper-and hypoactive mTOR signaling are connected to disease. While mTOR hyperactivity was observed in disorders such as fragile X syndrome (FXS), neurofibromatosis 1 (NF1) and tuberous sclerosis mTOR hypoactivity was characteristic of a mouse model of Rett syndrome and of human embryonic stem cell (hESC) derived neurons that carried a MECP2 loss-of-function allele [12][13][14][15][16]. In addition, hemizygous germline mutations in the X chromosomal MID1 gene lead to decreased mTOR signaling and Opitz BBB/G syndrome, a midline malformation disorder and ID syndrome [17].
The evolutionary conserved mTOR kinase comprises the core of two protein complexes, mTOR complex 1 (mTORC1) and 2 (mTORC2) [18]. mTOR activity is regulated intracellularly by nutrients, energy level and stress factors (e.g. hypoxia) and extracellularly by growth factors (e.g. brain derived neurotrophic factor (BDNF)), hormones (e.g. insulin), neurotransmitters and cytokines. Upon activation mTORC1 phosphorylates eukaryotic translation initiation factor 4E (eIF4E) binding proteins (4E-BP1 and 4E-BP2) and S6 kinases (S6K1 and S6K2). Both phosphorylation of the 4E-BPs and S6Ks comprise critical steps in the initiation of cap-dependent translation. For instance, phosphorylation of the 4E-BPs leads to their inactivation and dissociation from 5'cap bound EIF4E whereas phosphorylation of the S6Ks leads to their activation. S6Ks in turn phosphorylate proteins (e.g. ribosomal protein S6) that control different steps of translation [18]. Two studies showed that mTOR inhibition in fibroblasts and prostate cancer cells has a moderate effect on the translation of most mRNAs. It, however, strongly downregulated translation of a subgroup of mRNAs, many of which involved in protein synthesis, that contained specific sequence motifs, the 5′-terminal oligopyrimidine tract (5'TOP) or a pyrimidine-rich translational element (PRTE), in their 5'-untranslated regions (5'UTRs) [19,20].
While translation regulation by mTOR has been studied extensively in the brain, much less is known about mTOR mediated transcriptional regulation.
Here we used 3'mRNA sequencing (3'mRNA-Seq) to identify mTOR dependent genes in neurons treated with the mTOR inhibitor temsirolimus. We found that temsirolimus treatment downregulated expression of numerous genes of the sterol/cholesterol biosynthesis pathway resulting in decreased cholesterol levels.
Injection of rapamycin into pregnant dams confirmed our results in vivo in the developing cerebral cortex. While we did not find changes in RNA stability or chromatin accessibility, we identified binding sites for the transcription factors SP1, SREBP and NF-Y that were enriched in the promoter regions of the downregulated genes. ChIP-qPCR analyses confirmed binding of NF-YA to mTOR target gene promoters. Interestingly, NF-YA binding to these promoters increased in temsirolimus treated neurons. Our findings strongly suggest an important role for the mTOR pathway in regulating the expression of metabolic genes in neurons and in the developing brain.

Methods
Mice, cell culture and drug treatment NMRI (8-12 weeks old) mice were ordered from Janvier labs (Saint Berthevin, France) and sacrificed by cervical dislocation. For neuron culture primary cortical neurons were isolated from E14. 5

RNA-Seq data analysis
After the sequencing bcl2fastq v2.17.1.14 conversion software (Illumina, Inc.) was used to demultiplex sequence data and convert base call (BCL) files into Fastq files.
Sequencing adapters (AGATCGGAAGAG) were trimmed and reads shorter than 6 nucleotides were removed from further analysis using Cutadapt v1.11 [21]. Quality control checks were performed on the trimmed data with FastQC v0.11.4 [22]. Read mapping of the trimmed data against the mouse reference genome and transcriptome (mm9) were conducted using STAR aligner v2.5.3 [23]. To estimate the expression levels of each transcript the mapped reads were assigned to annotated features using the Subread tool featureCounts v1.5.2 [24]. The output of the raw read counts from featureCounts were used as an input for the differential expression analysis using the combination of DESeq2 v1.16.1 [25] and edgeR v3.26.8 [26] packages. Pairwise comparison analysis of two different conditions was carried out with edgeR to normalized the expression levels of known genes and only genes with CPM (counts per million) > 10 were further analyzed. The differentially expressed gene analysis was performed on the normalized genes expression using DESeq2 with |log2FoldChange| > 0.5 and padj (adjusted p-value) > 0.05.
Chromatin Immunoprecipitation (ChIP)-qPCR: DMSO or temsirolimus treated cells were pelleted and resuspended in cold PBS and later fixed with 1% formaldehyde in PBS for 10 min at room temperature, followed by quenching with 125 mM glycine for 5 min.

ATAC-seq data analysis:
ATAC-Seq data quality check was performed using reads FASTQC v0.11.8 [22]. Further, adaptors were removed using Trimmomatic v0.39 [27]. Paired-end ATAC-Seq reads were mapped to Mus musculus genome (mm10) UCSC annotations using Bowtie2 v2.3.5.1 [28] with default parameters. Properly paired end reads with high mapping quality (MAPQ ≥10) were retained in analysis with help of Samtools v1.7 [29]. Next, using Picard tools MarkDuplicates [30] utility duplicates were removed. ATAC-Seq peaks were called using MACS2 v2.1.1.20160309 [31]. ATAC peaks were visualized with USCS genome browser [32]. Gene ontology analysis and KEGG pathway analysis: For differentially up-and down-regulated expressed genes of each pairwise comparison as well as overlapping genes for SREBP, NFY and SP1 motifs, an over representation analysis (ORA) was carried out with clusterProfiler v3.4.4 [33]. All expressed genes within the pairwise comparison samples with CPM > 10 served as a background for the analysis. The Bioconductor org.Mm.eg.db v 3.8.2 mouse annotation package [34] and mouse KEGG.db [35] was used for the gene ontology analysis and KEGG analysis, respectively. The default parameters were used for all over-representation tests.

Motif identification
Motif discovery analysis was performed using HOMER (Hypergeometric Optimization of Motif EnRichment) Software v4.9 [36]. ATAC-Seq sequencing data from 186 genes in FASTA format served as input for the findMotif.pl function. Scrambled of input sequences (randomized) was created automatically by HOMER and used as a background for the motif analysis.

Transcriptional targets downstream of mTOR in neurons
To identify mTOR dependent transcriptional changes we treated primary cortical neurons with the mTOR inhibitor temsirolimus for 5 h or 24 h. Temsirolimus treatment caused strong downregulation in mTOR activity as measured by the ratio of the mTOR downstream effector pS6/S6 ( Figure 1A Because mTOR is best known for its function in gene activation, we decided to focus on the set of genes downregulated upon mTOR inhibition in subsequent analyses. Several studies have reported mTOR regulated transcriptional changes in non-neuronal cells.
For instance Duvel et al. used Tsc1-/-and Tsc2-/-fibroblasts, which exhibit growth factor independent activation of mTORC1, in combination with rapamycin treatment.
The authors found that mTORC1 activates expression of numerous genes involved in glycolysis, pentose phosphate pathway and lipin/sterol biosynthesis. We used the dataset published by Duvel et al. to identify mTOR-dependent transcriptional changes shared between fibroblasts and neurons. Surprisingly, when we compared our set of downregulated genes with the 130 genes found to be upregulated by mTORC1 in fibroblasts only 13 genes overlapped between both sets after 5 h of temsirolimus treatment and only 20 genes after 24 h (Figure 1 D, Supplementary table S2).

mTOR regulates expression of genes of the cholesterol pathway in primary neurons
Gene ontology analysis revealed that genes downregulated after 5 or 24 h of temsirolimus treatment were enriched for metabolic terms (Figure 2A). These terms comprised sterol/lipid biosynthesis/metabolism processes including the cholesterol biosynthesis pathway (Figure 2A). Other metabolic pathways previously described to be dependent on mTOR signaling like glycolysis and pentose phosphate pathway were, however, not enriched. By directly analyzing gene expression changes from our RNA-seq data we could confirm a general downregulation of sterol/cholesterol pathway genes.
Genes of the glycolysis pathway, however, remained unaltered in their expression ( Figure 2B). Also, by RT-qPCR several genes of the cholesterol biosynthesis process showed a 40 -60% downregulation after 5 h of temsirolimus treatment (Figure 2 C).
Westernblot experiments confirmed a significant downregulation of two of the three tested genes at protein level after 24 h of temsirolimus treatment (Figure 2 D). In contrast to genes of the cholesterol biosynthesis pathway, genes of the glycolysis pathway (Pfkp, Pgm2 and Pdk1) remained unaltered in their expression ( Figure 2C).
Temsirolimus is a derivative of rapamycin, a binder of FK506-binding protein-12 (FKBP-12) and allosteric partial inhibitor of the mTORC1 kinase. In contrast to rapamycin (and temsirolimus) ATP-competitive inhibitors like Torin 1 inhibit the phosphorylation of all mTORC1 substrates. The observed effect that temsirolimus treatment caused inhibition of the genes of some but not all metabolic pathways previously described to be dependent on mTOR activity could theoretically be due to temsirolimus being a partial mTORC1 inhibitor. In a next step we therefore treated neurons with Torin 1 for 5 h.
Similar to temsirolimus treatment the cholesterol biosynthesis pathway genes Ldlr and Dhcr7 were significantly downregulated after 5 h of Torin 1 treatment (Supplementary figure S1). In addition and in contrast to temsirolimus treatment, however, the glycolysis pathway genes Pfkp, Pgm2 and Pdk1 were also significantly downregulated (Supplementary figure S1).

mTOR activity is essential for proper expression of cholesterol pathway genes in the embryonic cerebral cortex
Having shown that mTOR promotes expression of cholesterol pathway genes in primary neurons in vitro we next wanted to verify this type of regulation in the brain in vivo.
While neurons of the adult brain rely mainly on astrocytes for cholesterol providing a critical time window during prenatal development exists where neuronal cholesterol synthesis is essential for neurons to differentiate normally [37]. To analyze if mTOR activity is required for the expression of cholesterol pathway genes we therefore chose to inhibit mTOR prenatally, starting at E16.5 by injecting rapamycin which has the same mechanism of action as temsirolimus intraperitoneally into pregnant mice. Twenty-four hours after a single dose of rapamycin injection we observed a strong inhibition of the mTOR pathway as shown by westernblot analysis of the mTOR downstream effector pS6 ( Figure 3A). Already at this time point all four tested cholesterol pathway genes (Ldlr, Osc, Mvd, Dhcr7) were reduced in their mRNA expression by about 50% (Figure 3B). In contrast, only Ldlr was significantly reduced by about 20% at protein level at this time point ( Figure 3C). Because the low effect on protein level could possibly be due to enhanced protein stability we, in a next step injected rapamycin on three consecutive days. Surprisingly, after 3 days of injection mRNA levels of the four genes tested had returned back to their normal levels which might be due to compensatory effects ( Figure 3D). Protein expression of Ldlr, Mvd and Nsdhl was, however, significantly reduced by 45, 30 and 25%, respectively ( Figure 3E).

Fig 3: mTORC1 inhibition in the embryonic cerebral cortex downregulates cholesterol biosynthesis genes in vivo.
Pregnant mice received either a single injection at E16.5 or were injected on three following days (E14.5-E16.5) once per day with DMSO or rapamycin at a dose of 1 mg/kg. Subsequently, total RNA and protein was extracted from the cerebral cortex and RT-qPCR and western blot experiments were performed. (A) Inhibition of mTORC1 activity was confirmed by western blot anaylsis of the mTORC1 downstream effector pS6 (in comparison to S6) 24 hours after a single injection of rapamycin. (B,D) RT-qPCR to detect RNA expression of the four genes Ldlr, Osc, Mvd and Dhcr7 after a one-day (B) or three-days (D) i. p. injection of rapamycin compared to DMSO; mRNA expression was normalized to Gapdh. (C,E) Immunoblot analysis to detect LDLR, OSC, MVD and DHCR7 protein expression after a one-day (B) or three-days (D) i. p. injection of rapamycin compared to DMSO and normalized to Gapdh. Data represent the average of six biological replicates. For statistical analyses, student's t test was used. *P < 0.05; **P < 0.01; ***P < 0.001.

Reduced cholesterol production in mTOR inhibited neurons
After 5 and 24 h of temsirolimus treatment eight and fifteen genes of the cholesterol biosynthesis pathway, respectively, were significantly downregulated in their expression ( Figure 4A, Supplementary figure S2  Düvel and colleagues showed that, in MEFs, the binding site for SREBPs is overrepresented in the promoters of mTOR regulated targets. When using the software findM [38] to screen the promoter regions of the genes downregulated in neurons treated with temsirolimus for 5 h we found that 44.9% of them contained at least one such binding site (Fig. 5A).
In a next step we applied the software Homer [36] on 5h Ts vs DMSO downregulated genes to identify additional transcription factor binding sites in an unbiased manner. In non-neuronal cells, mTORC1 regulates nuclear abundance of SREBP1; the SREBP transcription factors are required for mTOR induced expression of metabolic genes [1,8]. Much less is, however, known about the connection between mTOR activity and NF-Y functioning. The NF-Y complex is a trimer and consists of NF-YA, NF-YB and NF-YC.
Having histone-like structures, NF-YB and NF-YC, upon hetero-dimerization build a platform for NF-YA to associate. NF-YA then provides sequence specific motif recognition. Although most studies report a function for NF-YA as transcriptional activator a repressor function has been described as well [40][41][42]. Inspection of our RNA-seq data revealed no change in gene expression of either NF-YA, NF-YB or NF-YC 5h or 24h after temsirolimus treatment. Also, NF-YA predominantly localized to the nucleus under all experimental conditions and thus its subcellular localization did not change in temsirolimus treated neurons as determined by immunofluorescence analysis (Supplementary Figure S4). To test for changes in NF-Y binding after mTOR inhibition and because NF-YA is responsible for directing the NF-Y complex to its target sequences, we treated primary neurons with DMSO or temsirolimus and performed ChIP-qPCR with NF-YA specific antibodies. Since NF-Y mainly acts as a transcriptional activator we expected to see a decrease in NF-YA binding to its target sites within the promoters of mTOR regulated genes upon mTOR inhibition. Surprisingly, however, we observed the opposite. NF-Y showed an increased binding to the promoter of the known target Aurka as well as to the promoters of the mTOR target genes Mvd and Nsdhl upon mTOR inhibition. ChIP was performed with NF-YA specific antibodies in neurons treated for 5 hours with DMSO or temsirolimus followed by qPCR of positive control Aurka and negative control MyoD (F) or the downregulated targets Mvd, Nsdhl and Dhcr7 (G). Data represent the average of three biological replicates. For statistical analyses, student's t test was used. *P < 0.05; **P < 0.01; ***P < 0.001.

Discussion
By the use of several non-neuronal cellular systems such as fibroblasts, regulatory T cells and cancer cell lines it was shown previously that the mTOR kinase is an important transcriptional activator of metabolic genes [1,3,5]. It has not yet been investigated whether mTOR also controls the transcription of metabolic genes in neuronal cells in the brain -although mTOR signaling is an essential pathway for brain development and mTOR dysregulation causes a variety of neurodevelopmental disorders. Therefore, dysregulation of metabolic gene expression may very likely contribute to neurological symptoms in mTOR associated disorders. Here we show that mTOR drives the expression of metabolic genes of the sterol/cholesterol pathway in primary cortical neurons in vitro and in the developing cerebral cortex in vivo.
Although in neurons like in non-neuronal cells mTOR promotes expression of metabolic genes we observed some major differences between the different cell types. In neurons treated with temsirolimus we observed downregulation of genes of the cholesterol biosynthesis pathway but not of the glycolysis and pentose phosphate pathway which were both affected in mTOR hyperactive fibroblasts treated with rapamycin [1]. One explanation for this discrepancy might be the different experimental conditions. Indeed, when we treated cortical neurons with the more potent mTOR inhibitor, Torin 1, which inhibits the phosphorylation of all mTORC1 substrates we not only observed gene downregulation in the cholesterol biosynthesis pathway but also in the glycolysis pathway. In addition, prior studies suggest that cell type-specific effects exist which are likely to contribute to the differences observed between non-neuronal cell types and neurons. Thus depending on the cell type, mTOR activates specific transcription factors such as HIF1α, SREBPs and TFEB in fibroblasts and IRF4 and GATA3 in regulatory T cells [1,3,43]. Even nuclear localization and DNA binding of mTOR itself was observed [5,9,44]. Which mechanisms contribute to regulating cholesterol gene expression by mTOR in neurons is still unclear. We found a significant enrichment of binding sites for the transcription factors SP1, SREBPs and NF-Y in the promoters of mTOR regulated genes in neurons. ChIP analyses confirmed the binding of NF-YA to all investigated promoters.
Of note NF-YA binding was increased upon mTOR inhibition. As a member of the NF-Y complex, which also contains NF-YB and NF-YC, NF-YA mediates DNA binding specificity to the CCAAT motif in the proximal promoter region [45]. NF-Y regulates an increasing number of genes and is ubiquitously expressed, so it's likely to have different roles depending on the context. The NF-YA knockout in mice causes early embryonic lethality [46]. In proliferating cells such as embryonic fibroblasts and hematopoietic stem cells, NF-Y is involved in cell cycle regulation [47][48][49][50]. While it is often downregulated during differentiation [51,52] NF-Y is active in mature neurons of the adult mouse brain where its deletion causes neurodegeneration [53,54]. NF-Y co-localizes with other transcription factors such as FOS at genomic sites [55]. In differentiating neurons and HEK293 cells it was found that NF-YA and JNK bind to the same genomic sites and that NF-YA recruits JNK to these sites [56]. In addition, it was shown that NF-Y interacts at target gene promoters in cooperation with SREBP1 or SREBP2 and SP1 and activates genes of metabolic pathways [39, 57,58]. The fact that NF-YA binding at mTOR responsive promoters increased after mTOR inhibition suggests a mechanism where NF-YA may be important to recruit other transcription factors, such as SREBPs to mTOR responsive promoters. In the case of mTOR downregulation a hypothetic feedback mechanism might act to increase binding of NF-YA to these promoters.
Downregulation of mTOR activity is thought to play an important role in the pathogenesis of neurodevelopmental disorders such as Rett syndrome and CDKL5 deficiency disorder [13,59,60]. How reduced mTOR activity contributes to disease development in these disorders is, however, not entirely clear. Defects in the cholesterol pathway may be one contributing mechanism. Even small perturbations in cholesterol metabolism can largely affect neuronal development [61][62][63][64]. Low levels of cholesterol have been associated with a variety of neurodevelopmental disorders. Cholesterol biosynthesis is a multistep process which is divided into a pre-and post-squalene pathway (Fig. 3A). Several genes of the cholesterol biosynthesis pathway are mutated in neurodevelopmental syndromes. The most common genetic syndrome associated with defects in cholesterol biosynthesis is the autosomal recessive Smith Lemli Opitz syndrome (SLOS; OMIM# 270400) which is caused by mutations in DHCR7 encoding the enzyme 7-dehydrocholesterol D7reductase. The neurological symptoms of SLOS include epilepsy, intellectual disability and behavioural problems, among others. Mutations in NSDHL are associated with the X-linked dominant disorder CHILD syndrome (OMIM #308050) and the X-linked recessive disorder CK syndrome (OMIM #300831), mutations in MVK with the autosomal recessive disorder Mevalonic Aciduria (OMIM #610377). All three syndromes are characterized by structural brain abnormalities and/or neurological symptoms including intellectual disability. The gene product of NSDHL, 3β-hydroxysteroid dehydrogenase is involved in one of the later steps in cholesterol biosynthesis, the gene product of MVK, mevalonate kinase, is a peroxisomal enzyme involved in cholesterol biosynthesis in the pre-squalene pathway. Of note expression of all three genes, Dhcr7, Nsdhl and Mvk was downregulated in temsirolimus treated neurons (Fig. 2B). Although these results suggest that perturbations in cholesterol biosynthesis in disorders associated with mTOR downregulation may contribute to disease development several things have to be addressed in future studies.
Thus, it is still unclear whether mTOR downregulation leads to long lasting defects in cholesterol biosynthesis and reduced cholesterol levels. In the in vivo experiments we observed that 24h after injecting a single dose of rapamycin the mRNA expression of all tested genes was decreased. After three days of rapamycin injection expression, however, had returned to normal levels (Fig. 4A, B). This observation hints to the existence of feedback or compensatory mechanisms. Indeed, cholesterol biosynthesis involves several feedback mechanisms [65]. Buchovecky and colleagues found that in the brains of a mouse model for Rett syndrome (Mecp2 null mice) total cholesterol was increased at P56 when mutant males have severe symptoms. At a later age (P70), however, brain cholesterol levels were comparable to wildtype levels which reflected reduced cholesterol synthesis. Likely, the over production of cholesterol had fed back to later decrease cholesterol synthesis. Of note genetic or pharmacologic inhibition of cholesterol synthesis ameliorated some of the symptoms in these mice.
Our studies were limited to mTOR mediated regulation of cholesterol biosynthesis in cortical neurons. While the brain needs to synthesize its own cholesterol even during development when the blood brain barrier has not fully formed yet, neuronal cholesterol synthesis is most important during a critical developmental time window [37]. Neurons in the adult brain rely mainly on astrocytes for providing cholesterol. In addition, during myelination oligodendrocytes synthesize huge amounts of cholesterol.
It was already shown in zebrafish that cholesterol is needed for mTOR activity in oligodendrocyte precursor cells and that mTOR regulates cholesterol-dependent myelin gene expression [66]. An open question therefore is whether mTOR signaling in astrocytes and oligodendrocytes is equally important for cholesterol biosynthesis as it is in neurons.