The BDNF Val66Met Variant Affects Gene Expression through miR-146b

Abstract

Variation in gene expression is an important mechanism underlying susceptibility to complex disease and traits. Single nucleotide polymorphisms (SNPs) account for a substantial portion of the total detected genetic variation in gene expression but how exactly variants acting in trans modulate gene expression and disease susceptibility remains largely unknown. The BDNF Val66Met SNP has been associated with a number of psychiatric disorders such as depression, anxiety disorders, schizophrenia and related traits. Using global microRNA expression profiling in hippocampus of humanized BDNF Val66Met knock-in mice we showed that this variant results in dysregulation of at least one microRNA, which in turn affects downstream target genes. Specifically, we show that reduced levels of miR-146b (mir146b), lead to increased Per1 and Npas4 mRNA levels and increased Irak1 protein levels in vitro and are associated with similar changes in the hippocampus of hBDNF^Met/Met mice. Our findings highlight trans effects of common variants on microRNA-mediated gene expression as an integral part of the genetic architecture of complex disorders and traits.

Introduction

Genetic and genomic variability is intimately linked to differential disease risk among individuals. Gene expression can be modified by various genetic variants ranging from large structural variants (such as CNVs) to single nucleotide polymorphisms (SNPs). It is estimated that SNPs and CNVs capture 84% and 18%, respectively, of the total detected genetic variation in gene expression (Stranger et al., 2007). Such genetic variants affecting gene expression are referred to as expression quantitative trait loci (eQTL). eQTL can affect transcription either in trans (by affecting expression at distally located genomic loci, including ones in different chromosomes) or in cis (by affecting expression of genomic loci at their immediate vicinity). While the mechanism by which cis-acting variants affect gene expression is well established, how exactly variants acting in trans modulate gene expression and contribute to disease susceptibility remains under investigation (Cookson et al., 2009). microRNAs (miRNAs) (Bartel, 2004) bind to their target mRNAs (Lewis et al., 2003) and play a fundamental role in regulating gene expression primarily through post-transcriptional gene silencing via mRNA degradation or translational repression. As such, miRNAs could play an important role in mediating trans-acting effects on gene expression. In that respect, the potential of miRNAs to regulate expression of multiple genes could be an important component of the genetic architecture of complex disorders and traits. We have previously shown that 22q11.2 microdeletions, rare but recurrent de novo CNVs that predispose to cognitive dysfunction and schizophrenia mediate their effects at least in part due to pervasive miRNA-dependent effects on gene expression (Ambros et al., 2003; Stark et al., 2008; Xu et al., 2013). This finding established that miRNA-mediated trans effects on gene expression as an integral part of the pathogenesis of psychiatric and cognitive disorders (Xu et al., 2012). However, whether the contribution of miRNAs to the genetic etiology of complex disorders extends beyond rare structural mutations to common risk variants remain unknown. Addressing this issue is a key step towards obtaining a comprehensive view of the role that miRNAs play on mediating the trans regulatory effects of various rare and common disease-predisposing genetic variants.

To investigate whether miRNAs contribute to the trans regulatory effects of common genetic variants, we took advantage of a common SNP at codon 66 in the pro-domain of the Brain-Derived Neurotrophic Factor (BDNF) gene, which results in substitution of a Valine by a Methionine (Val66Met), leading to disrupted transportation of both the mRNA and protein to neuronal terminals and reduced activity-dependent release of BDNF protein both in vitro and in vivo (Chen et al., 2004; Chiaruttini et al., 2009; Egan et al., 2003). Val66Met SNP is a human-specific common genetic variant (Tettamanti et al., 2010). The allelic frequency of BDNF^Met is ~20–30% in populations of European origin but ranges from 0% in some populations of African origin to more than 50% in Asian populations (Petryshen et al., 2010). Val66Met SNP has been associated with reduced volume of hippocampal and prefrontal cortical grey matter and altered performance in memory tasks (Egan et al., 2003; Hariri et al., 2003; Rybakowski et al., 2006). Moreover, there is suggestive evidence that Val66Met modulates risk of depression (Verhagen et al., 2010), anxiety disorders (Tocchetto et al., 2011) and schizophrenia (Neves-Pereira et al., 2005). Studies in in vitro neuronal cultures have shown a role of BDNF in regulating miRNA levels (Fiore et al., 2009; Huang et al., 2012; Vo et al., 2005), suggesting that the effects of natural variation at the BDNF locus may be mediated via changes in miRNA levels. However, whether BDNF affects miRNA expression in vivo and most importantly, whether BDNF Val66Met SNP modulates risk of associated diseases and traits via miRNAs remains unknown. In this study we begin to address this question by utilizing a mouse model of humanized BDNF Val66Met variant.

Results

We have previously generated two mouse strains where the mouse Bdnf coding sequence was substituted by the corresponding human BDNF sequence carrying either a Met (hBDNF^Met) or a Val (hBDNF^Val) allele (Cao et al., 2007). To determine the impact of the humanized BDNF variant upon miRNA gene expression in hBDNF^Met/Met and hBDNF^Val/Val mouse model, we conducted a miRNA microarray-based expression profiling and identified 23 miRNAs with significantly altered expression (FDR-corrected P-value < 0.05) (Figure 1A and Table 1, see Experimental methods). Thirteen of these miRNAs were significantly upregulated in hBDNF^Met/Met mice, while 10 of them were significantly downregulated in hBDNF^Met/Met mice. Expression changes (fold change, FC) ranged from 1.29 to 0.79, except for miR-700 (FC = 1.66).

Figure 1. miR-146b expression levels are decreased in the hippocampus of hBDNF^Met/Met mice.

(A) miRNA expression profile in hippocampus of hBDNF^Met/Met animals versus hBDNF^Val/Val animals (n = 5 each genotype). Volcano plot showing the FDR-corrected P-value and corresponding relative expression of each miRNA. Green dots, miR-146b and miR-337-3p; blue dot, miR-700. (B) Expression levels of miR-146b and miR-337-3p in hippocampus of hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met mice (n = 6 each genotype), as measured by qRT-PCR. Expression levels in hBDNF^Val/Met and hBDNF^Met/Met mice were normalized to hBDNF^Val/Val animals. (C) Expression levels of miR-146b and miR-337-3p in acute hippocampal slices (n = 9, from 3 animals, each treatment) treated with BDNF for 2 hours, as measured by qRT-PCR. Expression levels following BDNF treatments were normalized to mock treatment control. Results are expressed as mean ± SEM. *P < 0.05, **P < 0.01, ANOVA with post-hoc Bonferroni's test (B); Student’s t-test (C).

Table 1.

Significantly dysregulated microRNAs in the HPC of BDNF^Met/Met mice, as compared to BDNF^Val/Val mice

mmu- miR	FDR P-value	regulation	Fold Change	Stem-loop Accession	Chr	Chr_start	Chr_end	Strand
146b	2.83E-03	down	0.83	MI0004665	19	46417252	46417360	+
700	2.83E-03	up	1.66	MI0004684	4	134972470	134972548	−
337–3p	3.54E-03	down	0.83	MI0000615	12	110823999	110824095	+
130b	3.54E-03	up	1.17	MI0000408	16	17124154	17124235	−
127–5pa	4.38E-03	up	1.12	MI0000154	12	110831056	110831125	+
328	7.71E-03	up	1.23	MI0000603	8	107832264	107832360	−
291a-5p	9.73E-03	down	0.93	MI0000389	7	3218920	3219001	+
10a	1.03E-02	up	1.15	MI0000685	11	96178479	96178588	+
20b	1.40E-02	down	0.88	MI0003536	X	50095290	50095369	−
337–5p	2.51E-02	up	1.15	MI0000615	12	110823999	110824095	+
532–5p	2.61E-02	up	1.23	MI0003206	X	6825528	6825623	−
674	3.05E-02	up	1.26	MI0004611	2	117010863	117010962	+
27b	3.30E-02	down	0.92	MI0000142	13	63402020	63402092	+
874	3.70E-02	down	0.88	MI0005479	13	58124486	58124561	−
804	4.29E-02	up	1.15	MI0005203	11	50171287	50171381	−
7a-1-3pb	4.29E-02	down	0.83	MI0000728	13	58494140	58494247	−
485–3pc	4.29E-02	up	1.14	MI0003492	12	110973112	110973184	+
491	4.29E-02	up	1.15	MI0004680	4	87767944	87768029	+
721	4.29E-02	down	0.79	MI0004708	5	136851586	136851673	−
688	4.29E-02	down	0.87	MI0004653	15	102502223	102502297	−
342–3p	4.40E-02	down	0.87	MI0000627	12	109896830	109896928	+
378	4.45E-02	up	1.29	MI0000795	18	61557489	61557554	−
185	4.51E-02	up	1.09	MI0000227	16	18327494	18327558	−

^apreviously mmu-miR-127*,

^bpreviously mmu-miR-7a*,

^cpreviously mmu-miR-485*

We followed up expression changes of the top three miRNAs (miR-146b, miR-700 and miR-337-3p) with absolute FC > 1.2 and FDR-corrected P-value < 0.005 (Figure 1A). We verified the downregulation of miR-146b and miR-337-3p by quantitative real-time PCR (qRT-PCR) in adult HPC. As compared with levels in hBDNF^Val/Val animals, miR-146b was decreased by 8% (P = 0.20) and 19% (P < 0.01) in hBDNF^Val/Met and hBDNF^Met/Met animals respectively, and miR-337-3p was decreased by 19% (P < 0.01) and 10% (P = 0.06) in hBDNF^Val/Met and hBDNF^Met/Met animals respectively (Figure 1B). qRT-PCR did not confirm the significant change in miR-700 levels in the microarray data (FC = 1.08, P = 0.48, hBDNF^Met/Met versus hBDNF^Val/Val) (Supplementary Figure 1). It is worth noting that both miR-146 and miR-337-3p are among the top 20 most enriched miRNAs in synaptic compartments of adult mouse forebrain, suggesting they control synaptic-related functions (Lugli et al., 2008). The expression pattern and level of miR-700 during brain development remains unknown. Since validation was attempted for only 3 of the microRNAs in Table 2, the full extent of microRNA dysregulation due to BDNF Val66Met SNP remains to be determined in future experiments.

Table 2.

Functional enrichment analysis of predicted miR-146b targets

Category	Term	Count	%	Fold Enrichment	P Value	Benjamini FDR
Annotation Cluster 1	Enrichment Score: 2.85
GOTERM_BP_FAT	GO:0030182~neuron differentiation	14	8.97	3.97	4.62E-05	0.0226
GOTERM_BP_FAT	GO:0030030~cell projection organization	12	7.69	4.26	1.11E-04	0.0271
GOTERM_BP_FAT	GO:0031175~neuron projection development	10	6.41	5.19	1.25E-04	0.0246
GOTERM_BP_FAT	GO:0007409~axonogenesis	8	5.13	5.56	5.72E-04	0.0502
Annotation Cluster 2	Enrichment Score: 2.42
GOTERM_MF_FAT	GO:0003723~RNA binding	16	10.26	2.93	3.07E-04	0.0391
Annotation Cluster 3	Enrichment Score: 2.37
GOTERM_MF_FAT	GO:0003677~DNA binding	32	20.51	2.21	1.76E-05	0.0046
GOTERM_BP_FAT	GO:0010557~positive regulation of macromolecule biosynthetic process	15	9.62	3.20	2.18E-04	0.0267
GOTERM_BP_FAT	GO:0031328~positive regulation of cellular biosynthetic process	15	9.62	3.08	3.30E-04	0.0357
GOTERM_BP_FAT	GO:0009891~positive regulation of biosynthetic process	15	9.62	3.05	3.62E-04	0.0352

We speculated that if reduced expression levels of miR-146b and miR-337-3p were due to reduction in regulated BDNF release in these animals, their expression would be induced by BDNF. To test this hypothesis, we applied physiological levels of BDNF to acute hippocampal slice preparations from wildtype animals and monitored expression levels of miR-146b and miR-337-3p following 2 hours of incubation. We found that acute BDNF stimulation increased the levels of miR-146b by 22% (P = 0.01). We also observed a non-significant increase in the levels of miR-337-3p (21%, P = 0.11) (Figure 1C). These experiments demonstrated that expression of at least miR-146b is likely to be under the direct control of BDNF and could be altered in response to reduction in regulated BDNF release.

We further conducted a functional enrichment analysis using the DAVID functional annotation tool upon the potential targets of these top altered miRNAs (FDR p < 0.005) as predicted by TargetScan. Among the six miRNAs analyzed, the number of predicted miR-146b targets is relatively modest but they show a highly significant enrichment of the GO term “neuron differentiation”, which is consistent with the well-established neurotrophic effects of BDNF (Table 2). In contrast, all other miRNA targets were either enriched in broad GO terms (such as regulation of transcription) or did not show any significant enrichment (Supplementary Data Sheet 1). Notably, miR-146b was initially isolated from the HPC (He et al., 2007) and was found to represent one of the most abundant miRNA in mouse HPC. In addition, miR-146b is also downregulated in a Mecp2-null mouse model of the Rett syndrome, which presents BDNF signaling abnormalities (Urdinguio et al., 2010). Collectively, the evidence outlined above points to miR-146b as a good candidate BDNF-modulated miRNA that deserves further analysis.

To determine whether the observed miRNA dysregulation is sufficient to affect expression of downstream targets, we sought miR-146b targets whose expression levels are in turn altered by Val66Met SNP. Luciferase reporter clones of longest 3’UTRs of putative miR-146b targets (Table 3) were generated and the repressive effects of pre-mir-146b mimic on these clones were measured in luciferase reporter assays using N18 cells, a mouse neuroblastoma cell line. Luciferase expression from 3’UTR constructs of Irak1, Traf6, Per1, Stx3, Syt1, Kctd15, Sort1, Dlgap1, Npas4 and Lin28A was significant downregulated (P < 0.05) (Figure 2A) by pre-mir-146b, as compared to a pre-scramble control. The repression of luciferase expression by pre-miR-146b was especially pronounced on the 3’UTR of Irak1 (59%, P < 0.01), Traf6 (81%, P < 0.01), Per1 (54%, P < 0.01), Lin28A (55%, P < 0.01), Npas4 (44%, P < 0.01), while constructs with 3’UTR of other candidate targets were modestly repressed (Figure 2A). The repressive effects of pre-mir-146b on Per1 3’UTR reporters were antagonized by anti-miR-146b LNA oligonucleotides, demonstrating the specificity of the luciferase assay screen for miR-146b targets (Figure 3). Overall, miR-146b was able to repress in vitro expression of all predicted targets through their 3’UTR sequences.

Table 3.

Predicted miR-146b targets selected for validation

	TargetScanMouse									microT	PicTar	miRanda	mirDB
	version 6.2, 6/2012									version 5	03/2007	11/2010	04/2012
Target	total context+ score	Conserved sites				Poorly conserved sites
		total	8mer	7mer- m8	7mer- 1A	total	8mer	7mer- m8	7mer- 1A	miTG score	PicTar score	mirSVR score	target score
Irak1	−0.77	2	2	0	0	1	0	0	1	0.98	8.79	−1.65	94
Traf6	−0.53	3	3	0	0	0	0	0	0	>0.99	6.28	−0.72	95
Per1	−0.31	0	0	0	0	2	1	0	1	0.43	2.52	−0.66
Stx3	−0.25	1	1	0	0	0	0	0	0	0.82	1.06	−0.46
Syt1	−0.22	1	1	0	0	1	0	0	1	0.92	4.44	−1.15
Cask	−0.21	1	0	1	0	1	0	0	1	0.80	1.02	−0.48	65
Robo1	−0.19	1	0	0	1	0	0	0	0	0.83	1.30	−0.52
Kctd15	−0.17	1	1	0	0	0	0	0	0	0.57	5.00	−0.20
Sort1	−0.13	1	1	0	0	0	0	0	0			−0.22	50
Dlgap1	−0.09	0	0	0	0	1	0	1	0	0.76	0.75	−0.18
Srrd	−0.08	0	0	0	0	1	0	1	0	0.44	1.56	−0.03
Bsn	−0.07	0	0	0	0	1	0	0	1	0.46		>−0.01
Gria3	−0.03	0	0	0	0	1	1	0	0	0.78	0.83	>−0.01
Npas4	−0.02	1	0	1	0	0	0	0	0	0.56	2.02	−0.02
Lin28A	−0.01	0	0	0	0	1	0	0	1	0.65	2.53	−0.01
Akt3	−0.01	0	0	0	0	1	0	0	1	0.54		−0.31

Figure 2. miR-146b targets, Per1, Npas4 and Irak1 are upregulated in hBDNF^Met/Met mice.

(A) Repressive effects of pre-mir-146b on 3’UTRs of a group of putative miR-146b targets (see Supplementary Data Sheet 1) were examined by a dual-luciferase reporter assay performed in N18 neuroblastoma cell line (n = 3 for each reporter). Expression value of each target reporter was normalized to no 3’UTR control. A pre-scramble oligo was used as control for each target tested. (B) Repressive effects of pre-mir-146b on Per1 and Npas4 3’UTRs (left and right, respectively) were abrogated by mutations in miR-146b binding sites, as analyzed by luciferase reporter assay. Expression value of each mutant reporter was normalized to Wt 3’UTR reporter. (C) Expression levels of Per1 and Npas4 in hippocampus of hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met mice (n = 6 each genotype), as measured by qRT-PCR. Expression levels in hBDNF^Val/Met and hBDNF^Met/Met mice were normalized to hBDNF^Val/Val animals. (D) Western blots analysis of Irak1 (left) and Traf6 (right) protein levels in in hippocampal lysates of hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met mice (n = 7 each genotype). Upper: Representative western blot assays of Irak1 and Tarf6. β-actin was used as loading control. Lower: Quantification of Irak1 and Traf6 protein levels. Expression levels in hBDNF^Val/Met and hBDNF^Met/Met mice were normalized to hBDNF^Val/Val littermates. Results are expressed as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ANOVA with post-hoc Dunnett's test for multiple comparisons relative to control (A, B); ANOVA with post-hoc Bonferroni's test (C, left, D); Student’s t-test (C, right).

Figure 3. The repressive effect of pre-mir-146b on Per1 3’UTR is specific.

Pre-mir-146b significantly repressed Per1 3’UTR over a concentration range of 1 nM to 0.1 nM, as analyzed by luciferase reporter assay. The repressive effects of pre-miR-146b can be completely neutralized by anti-miR-146b (Exiqon) transfection (n = 3 for each condition) at a concentration ratio of ~1:10. Results are expressed as mean ± SEM. **P < 0.01, ***P < 0.001, ANOVA with post-hoc Dunnett's test for multiple comparisons relative to control.

As mammalian miRNAs inhibit target expression predominantly through decreasing mRNA levels rather than repressing translation (Guo et al., 2010), we hypothesized that most of the physiologically relevant target genes in vivo would also be upregulated at the transcript levels in hBDNF^Met/Met animals, which have lower levels of mir-146b in HPC. Therefore, we measured and compared the expression levels of the predicted candidates in hBDNF^Val and hBDNF^Met knock-in animals by qRT-PCR. Expression levels of only Per1 and Npas4 (out of 13 predicted targets with good amplification signals in qRT-PCR assays) were significant influenced by genotype (ANOVA, P < 0.01 for Per1; P < 0.05 for Npas4) and were upregulated in hBDNF^Val/Met (24% for Per1 and 33% for Npas4) and hBDNF^Met/Met mice (28% for Per1 and 65% for Npas4) as compared to hBDNF^Val/Val mice (Table 4, left panel and Figure 2C). As a control, we tested an equal number of genes without a miR-146b seed sequence but none showed significantly elevated expression in hBDNF^Val/Met and hBDNF^Met/Met mice (Table 4, right panel). Therefore, Per1 and Npas4 represent particularly sensitive targets whose modulation by the Val66Met SNP can be readily detected at both mRNA levels in vivo and protein levels in vitro.

Table 4.

Expression levels of candidate miR-146b targets and non-targets in the HPC of hBDNF^Val/Met and hBDNF^Met/Met mice, as compared to BDNF^Val/Val mice

Genes with miR-146b Binding Site(s)				Genes without miR-146b Binding Site
	hBDNFVal/Met	hBDNFMet/Met	ANOVA		hBDNFVal/Met	hBDNFMet/Met	ANOVA
gene	fold change	fold change	P-value	gene	fold change	fold change	P-value
Per1	1.24	1.28	1.1E-3	B3gat1	0.91	0.97	0.09
Npas4	1.33	1.65	0.03	Kif5c	1.13	1.06	0.10
Robo1	1.10	1.04	0.06	Cdk5r1	1.07	1.09	0.12
Akt3	1.10	1.10	0.14	Ldb2	1.02	0.95	0.18
Traf6	1.03	0.94	0.19	Cpeb2	1.07	1.01	0.31
Syt1	1.13	1.05	0.29	GSK3b	1.03	0.97	0.37
Irak1	1.02	0.96	0.30	Cdk5r2	1.05	1.15	0.39
Sort1	1.10	1.03	0.33	vGlut1	0.99	0.95	0.49
Cask	1.01	0.88	0.36	Agxt2l1	1.10	1.24	0.64
Gria3	0.99	1.11	0.40	Nptx2	0.97	1.05	0.67
Stx3	0.91	1.01	0.41	Hspa8	0.98	0.94	0.71
Srrd	1.09	1.00	0.53	Clec16a	1.01	1.01	0.97
Bsn	1.03	1.05	0.55	Mef2C	0.99	0.99	0.98

Note: While expression levels of two of the selected miR-146b candidate targets (Per1 and Npas4) were significantly altered (upregulated) in HPC due to Val66Met SNP (P < 0.05, ANOVA), none of the genes without miR-146b binding site shows change in expression levels, as assayed by qRT-PCR.

To investigate if miR-146-mediated repression on Per1 and Npas4 expression is specific and operates directly via the target sites as predicted by TargetScan (Table 3), we engineered Per1 and Npas4 3’UTR luciferase reporters carrying mutated miR-146b binding sites. Per1 3’UTR contains two cognate miR-146b binding sites at position 190–197 (an 8-mer site) and position 505–511 (a 7mer-1A site). 3 different Per1 3’UTR mutants (Mut1: Site 1 mutant; Mut2: Site 2 mutant; Mut1&2: Site 1 and 2 mutants), and a Npas4 3’UTR mutant (Mut) with the 7mer-m8 site abrogated were generated. Compared with Wt reporter, Mut1, Mut2 and Mut1&2 in Per1 3’UTR increased luciferase activity by 49% (P < 0.01), 26% (P < 0.01), and 58% (P < 0.01) respectively, while Mut in Npas4 3’UTR increased the luciferase activity by 29% (P < 0.01) (Figure 2B). Thus both miR-146b binding sites in Per1 3’UTR control miR-146b–mediated regulation on Per1 expression, although the 8mer site seems to have a larger impact. The miR-146b binding site in Npas4 3’UTR similarly controls miR-146b–mediated repression of Npas4 expression. Overall, we have identified Per1 and Npas4 as targets of mir-146b in vivo. Reduced miR-146b levels in animals carrying one or two hBDNF^Met alleles result in the corresponding rise in Per1 and Npas4 transcript levels.

Intriguingly two other target genes (Traf6, Irak1) which were strongly affected by pre-mir-146b in luciferase assays (Figure 2A) but whose mRNA levels were unaffected in hBDNF^Met/Met mice (Table 4) have been previously identified as targets of miR-146 in human monocytes (Taganov et al., 2006). It is likely that these targets are repressed by miR-146b translationally without any changes in their transcript levels (Table 4). Indeed, Western blot assays of protein extracts from the HPC of hBDNF^Val and hBDNF^Met knock-in mice (6 pairs of 8 week-old male mice, group-housed under standard rearing condition) showed increases of Irak1 in hBDNF^Val/Met (25%, P > 0.05) and hBDNF^Met/Met animals (59%, P < 0.05) as compared to hBDNF^Val/Val animals (Figure 2D). In contrast, Traf6 levels did not vary significantly with the genotype of the mice (ANOVA, P > 0.05) (Figure 2D). Thus, although we did not detect altered mRNA levels, Irak1 protein levels were significantly elevated in hBDNF^Met/Met mice likely due to reduced miR-146b–mediated translational repression.

It was recently reported that BDNF application can rapidly elevate in vitro expression of several miRNA biogenesis proteins, including Dicer1 and Lin28A, resulting in a general increase in mature miRNA levels but a decrease in levels of let-7 family miRNAs whose degradation is promoted by Lin28A–induced pre-miRNA uridylation (Huang et al., 2012). Derepression of let-7 family miRNA targets leads to increased expression of proteins such as GluA1 and Homer2 in response to BDNF. Altered expression of miR-146b and downstream targets could therefore be secondary to changes in levels of Dicer1, Lin28A or let-7 miRNA in hBDNF^Met/Met mice. To test whether Val66Met regulates protein levels of microRNA biogenesis components, we used Western blot analysis to monitor levels of Dicer1, Dgcr8, Lin28A and Ago2 in six pairs of 8 week-old male mice (grouped-housed under standard rearing condition) but did not observe any genotypic differences (ANOVA, P > 0.05) (Figure 4). In addition, expression levels of previously reported BDNF-induced targets, GluA1 and Limk1 (Huang et al., 2012; Schratt et al., 2006) as well as let-7 family miRNAs, let-7b and miR-98 were not altered in hBDNF^Met/Met mice (Supplementary Figure 2 and 3, Table 5). These results argue against a generic effect of Val66Met on miRNA biogenesis and possibly suggest a more direct mode of action on the expression control of a subset of miRNAs including miR-146b (see Discussion).

Figure 4. Baseline expression levels of miRNA biogenesis enzymes are not altered in hBDNF^Met/Met mice.

(A) Expression levels of Dicer1, Lin28A, Ago2 and Dgcr8 in hippocampal lysates of hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met mice (n = 6 for each genotype), as assayed by Western blots. Expression levels in hBDNF^Val/Met and hBDNF^Met/Met mice were normalized to hBDNF^Val/Val littermates. Results are expressed as mean ± SEM. P > 0.05, ANOVA with post-hoc Bonferroni's test. (B) Representative western blot assays of Dicer1, Lin28A, Ago2 and Dgcr8 in hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met lysates. α-tubulin was used as loading control.

Table 5.

Expression of let-7 family microRNAs in the HPC of hBDNF^Met/Met Mice, as compared to hBDNF^Val/Val mice

miRNA ID	FDR P-value	Rank	FC Absolute	Regulation
mmu-let-7a	0.6020	337	1.09	down
mmu-let-7a*	0.1698	66	1.20	up
mmu-let-7b	0.9133	524	1.02	up
mmu-let-7b*	0.6653	389	1.07	up
mmu-let-7c	0.9444	541	1.01	down
mmu-let-7c*	0.8927	506	1.01	up
mmu-let-7d	0.9113	515	1.02	down
mmu-let-7d*	0.2736	116	1.24	up
mmu-let-7e	0.7696	422	1.06	up
mmu-let-7f	0.9067	510	1.02	down
mmu-let-7f*	0.9067	511	1.03	up
mmu-let-7g	0.7345	413	1.02	down
mmu-let-7g*	0.1972	78	1.27	up
mmu-let-7i	0.3366	182	1.04	up
mmu-let-7i*	0.0691	31	1.20	up
mmu-miR-98	0.8246	450	1.11	up
mmu-miR-107	0.5905	326	1.05	down
mmu-miR-143	0.5916	329	1.04	up

Note: None of the let-7 family miRNA showed significantly altered expression (FDR-corrected P > 0.05) in HPC in miRNA microarray. See Figure 1 for details of the microarray by LC Sciences.

Discussion

Our results clearly demonstrated that BDNF Val66Met is associated with altered expression of a specific subset of miRNAs and most importantly of their downstream targets. Val66Met affects miRNAs levels not through miRNA biogenesis (Huang et al., 2012) but more likely through modulating BDNF-dependent transcription factor binding in regulatory regions. Among the affected miRNAs, miR-146b and its modulated downstream targets, which are particularly enriched in neuronal genes, are likely to have substantial contribution to the biological effects of the Val66Met variant. Indeed, it is well established that BDNF induces the expression of neuronal genes through activation of MAPK pathway and Akt1/2 downstream of TrkB receptor (Lyons and West, 2011), which in turn leads to activation of CREB, MEF2 and NF-κB transcription factors (Huang and Reichardt, 2003; Minichiello, 2009; Shalizi and Bonni, 2005; Yoshii and Constantine-Paton, 2010). Recent in vitro studies in neuronal cultures have shown that exogenous BDNF transcriptionally activates the expression of miRNAs via CREB and MEF2 activation (Fiore et al., 2009; Vo et al., 2005). Interestingly, it was shown that miR-146 transcription is in part controlled by NF-κB (Perry et al., 2009; Taganov et al., 2006).

We have identified Per1 and Npas4 as genuine targets of mir-146b in vitro and our in vivo studies strongly suggest that both of them are mir-146b targets in vivo, as well. Interestingly, Per1 is a transcriptional repressor controlling circadian rhythm (Takahashi et al., 2008). Circadian and sleep disturbances are often associated with psychiatric disorder (Lamont et al., 2007). PER1 expression is altered in postmortem brains of schizophrenia patients (Aston et al., 2004), while its homolog PER3 is considered a candidate gene for bipolar disorder and schizophrenia (Mansour et al., 2006; Nievergelt et al., 2006). Npas4 is homologous to Npas2, which is another core component of circadian regulation. Npas4 is required for inhibitory synapse formation in HPC as well as for activity-dependent expression of neuronal genes (Lin et al., 2008; Ramamoorthi et al., 2011). It should be noted that Npas4 directly promotes activity-dependent transcription of Bdnf (Pruunsild et al., 2011) and therefore part of the observed increase in Npas4 levels in animals carrying one or two hBDNF^Met alleles may be compensatory in nature. We also identified Irak1 as an additional target, modulated only at the protein level. Irak1 is an essential component downstream of Toll-like and IL-1 receptor signaling (Gottipati et al., 2008). Activation of Irak1 enhances NF-κB-mediated transcription and leads to elevation of innate and pro-inflammatory responses in neuron and glial cells (Li et al., 2011). Whether the Val66Met variant increases the risk of inflammatory insult in the central nervous system remains to be determined. Independent of underlying mechanisms, our results suggest that BDNF Val66Met variant may mediate its effects on psychiatric and cognitive phenotypes in part due to miRNA-dependent effects on gene expression (Stark et al., 2008; Xu et al., 2013). Because the Bdnf gene and the Val66Met variant are located on mouse chromosome 2, while Per1, Npas4 and Irak1 are on chromosomes 11, 19 and X, respectively, this represents a typical trans eQTL effect mediated via microRNAs. Integrated analysis of genetic and expression data indicates that common variants at a large number of eQTLs are associated with differential expression of target genes either in cis or in trans. The impact of eQTLs on miRNA expression has been investigated recently in various cell types including fibroblasts, glioblastoma, liver cells and hippocampal tissues (Borel et al., 2011; Dong et al., 2010; Parsons et al., 2012; Su et al., 2011). Approximately 5–20% of miRNAs investigated are under the influence of eQTLs with half of them under the control of cis- and half under the control of trans-eQTLs. Because each miRNA can potentially modulate multiple downstream targets, eQTL control over miRNA expression could underlie coordinated expression of networks of genes. Moreover, because miRNAs often target functionally connected genes (Tsang et al., 2010; Xu et al., 2013; Zhang et al., 2009) the cumulative impact of miRNA alterations due to disease-associated eQTL variants could result in considerable cellular dysfunction, which in turn may contribute substantially towards disease risk and clinical phenotype. In that context, our results also reveal opportunities for interaction between common eQTL variants and rare mutations (such as CNVs and point mutations) that affect either individual miRNA genes or biosynthesis and action of miRNAs (Stark et al., 2008; Xu et al., 2013; Xu et al., 2008).

It is interesting to note that although miR-146b was able to repress in vitro expression of all predicted targets through their 3’UTR sequences (as indicated by our luciferase assays), only three predicted targets (30%) showed expected alterations in vivo. There are several potential explanations for the discrepancies observed among predicted targets (such as Lin28A and Traf6), and results from in vitro and in vivo assays. Our in vitro assay indicated that while these and other predicted targets represent genuine targets of miR-146b that could be regulated by overexpression of miR-146b, the interaction between miR-146b and target genes under more physiological conditions likely involves additional levels of complexity. For example, the accessibility of the miR-146b binding site may be under the control of sequestering mechanisms involving mRNA binding proteins (Kedde and Agami, 2008). In addition, it is possible that the interaction between a specific miRNA target and the cognate miRNA is very sensitive to miRNA levels and altered by competition among different miRNA targets over a limited miRNA pool (Salmena et al., 2011; Seitz, 2009). Finally, the predicted targets of miR-146b may not be spatially and temporally registered with miR-146 (i.e co-expressed in the same cell type and developmental window). Our findings strongly suggest that considerable caution should be exercised when extrapolating results from predicted targets and in vitro assays to more physiological settings.

Overall, taken together with previous results on rare structural variants (Stark et al., 2008; Xu et al., 2013), our findings highlight miRNA-mediated trans effects on gene expression as an integral part of the genetic architecture of complex disorders and traits caused by common disease-predisposing genetic variants.

Experimental methods

Human BDNF knock-in mice

Generation of hBDNF^Met and hBDNF^Val knock-in mice has been described in detail previously (Cao et al., 2007). In short, we replaced a segment in mouse Bdnf coding region that carries either Val allele or Met allele and encodes all the 11 amino acids that differ between mouse and human protein. This genetic manipulation generated knock-in alleles that express human BDNF genes controlled by endogenous mouse Bdnf regulatory elements. hBDNF^Val/Val, hBDNF^Val/Met and hBDNF^Met/Met littermates were from hBDNF^Val/Met X hBDNF^Val/Met cross and were genotyped as described (Cao et al., 2007). All animal procedures were conducted according to protocols approved by the IACUC established by Columbia University under federal and state regulations.

miRNA microarray expression analysis

Hippocampi were acutely dissected from 5 pairs of 8 week-old male mice, which were grouped-housed and reared under standard rearing conditions. Total RNA was isolated using the mirVana miRNA isolation kit (Ambion). RNA quality was assessed with Bioanalyzer (Agilent Technologies, Palo Alto, CA) and all RNAs had a RIN > 7.0. Small RNAs (<300 nt) were then isolated and processed for microarray analysis (LC Sciences). Briefly, purified small RNAs were labeled with Cy5 (hBDNF^Val/Val) or Cy3 (hBDNF^Met/Met) fluorescent dyes and hybridized to dual-channel microarray µParaFlo microfluidics chips (LC Sciences) containing 569 miRNA probes to mouse mature miRNAs. Each of the spotted detection probes consisted of a nucleotide sequence complementary to a specific miRNA sequence and a long non-nucleotide spacer that extended the specific sequence away from the chip surface. The miRNA probe sequences used were from the miRBase Sequence database version 10.1. We collected hybridization images using a GenePix 4000B laser scanner (Molecular Devices) and digitized them using Array-Pro image analysis software (Media Cybernetics). Raw data were imported in ArrayAssist 5.0 (Stratagene). The microarray data were corrected by removing spots with intensity equal to or below median background and then normalized with the LOWESS (locally weighted regression) method implemented in ArrayAssist 5.5 software (Stratagene). Differentiation analysis was conducted to determine the FDR P-value of each miRNA gene as described previously (Stark et al., 2008).

Quantitative RT-PCR analysis of mature miRNA and coding gene expression

Total RNA samples were extracted from the hippocampi that acutely dissected from 8 week-old male mice, which were grouped-housed and reared under standard rearing conditions, using mirVana miRNA Isolation Kit (Ambion) according to manufacturer’s protocol. We treated 3 µg of total RNA from each sample with DNA-free kit (Applied Biosystems). For RT of each mature miRNA, 100 ng of treatment RNA each sample was reverse transcribed using TaqMan Reverse Transcription Kit (Applied Biosystems) and RT primer provided in the individual TaqMan MicroRNA Assay (see below for catalog numbers). A glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene-specific RT primer (reverse primer from ABI # 4352339) was also included RT reaction. For RT of coding genes, the remaining DNase-treated RNA each sample was reverse transcribed using Random Primers (Invitrogen) and SuperScript II Reverse Transcriptase (Invirtogen). qRT-PCR was performed using ABI TaqMan method in a 7900 Sequence Detection System (Applied Biosystems). For each gene/miRNA, a duplex qPCR was performed using TaqMan Universal PCR Master Mix, no AmpErase UNG (Applied Biosystems) and custom designed primer and probe set, as well as primer and probe set for GAPDH (custom designed primer and probe set used in qPCR of coding genes, see below; ABI # 4352339 used in qPCR of miRNA). Quantification procedures were described previously (Stark et al., 2008). All statistical analyses were conducted in MS-Excel using Student t-test on data from at least 8 biological repeats and 5 technical repeats.

TaqMan MicroRNA Assays were purchased from Applied Biosystems. MicroRNA assay name and ID: hsa-miR-146b: 001097; mmu-miR-337: 002532; mmu-miR-700: 001634; hsa-miR-98: 000577; has-let-7b: 002619. All PCR primers and probes for coding genes were designed at Primer3 web site (http://frodo.wi.mit.edu/) and purchased from Sigma Genosys (Sigma-Aldrich) and the sequences can be found in Supplementary Data Sheet 2, except for Nptx2 (Mm00479438_m1, Applied Biosystems) and Clec16a (Mm00624340_m1, Applied Biosystems). All target gene probes were 5’ FAM and 3’ BHQ™-1 Dual labeled. Mouse GAPDH mRNA was used as the endogenous control. The custom made GAPDH gene probe was 5’ JOE™ and 3’ BHQ™-1 dual labeled. All statistical analyses were conducted in MS-Excel using Student t-test on data from at least 8 biological repeats and 5 technical repeats.

BDNF-treatment of hippocampal slices

Hippocampal slices of 250-µm-thick were prepared as described previously (Drew et al., 2011). Briefly, 250-µm-thick horizontal brain sections were prepared on a vibratome (Leica VT1200S) in dissection solution (in mM: sucrose 195, NaCl 10, KCl 2.5, NaH₂PO₄ 1, NaHCO₃ 25, glucose 10, MgCl₂ 5, MgSO₄ 1, CaCl₂ 0.5) from brain of 8 week-old Wt C57Bl/6J mice. Hippocampal regions were promptly cut out from ~10 horizontal brain sections and then immediately transferred to an interface chamber and allowed to recover for 1 h at 31–32 °C. Slices were then transferred to another chamber and incubated for 2 hours with pre-oxygenated artificial cerebrospinal fluid (aCSF) (bubbled with 5% CO2/95% O2) that had the following composition (in mM): NaCl 124, KCl 2.5, NaH₂PO₄ 1, NaHCO₃ 25, Glucose 10, MgSO₄ 1, CaCl₂ 2. In the BDNF treatment group, the aCSF also contained 50 ng/ml of BDNF (#B3795, Sigma-Aldrich). Slices were removed immediately after 2 hour incubation and total RNAs from the BDNF or sham treated slices were extracted using TRIzol (Invitrogen), following manufacturer’s instruction.

miRNA target prediction

39 genes were initially predicted by both TargetScan Mammal v.4.2 and PicTar (updated March 2007) miRNA target site prediction programs. Among them, 12 target candidates (Irak1, Traf6, Per1, Stx3, Syt1, Cask, Robo1, Kctd15, Dlgap1, Gria3, Npas4, Lin28A) were initially selected for further analysis due to their roles in neural development and/or plasticity (Table 1). We included in this list also Sort1, which encodes sortilin, a protein required for intracellular BDNF trafficking and activity-regulated release and has a miR-146b seed sequence in the 3’UTR of its mRNA. 3 additional targets predicted by miRanda only (Betel et al., 2008) (Akt3, Srrd, Bsn) were also tested because they are involved in important functions in neurotransmitter production (Srrd), postsynaptic density (Bsn) and TrkB signaling (Akt3), which are highly related to BDNF function. 16 predicted targets of miR-146b in total were included in this analysis. It is worth noting that most of these targets are also predicted by a more updated version of TargetScan v6.2 and miRanda (released: August 2010, updated: Nov 2010) and by additional programs – mirDB (Wang, 2008) (updated: April 2012) and microT v5.0 (Maragkakis et al., 2011). In fact, almost all candidates originally predicted by both TargetScan Mammal v.4.2 and PicTar are also predicted by 3 or more of these 5 programs (TargetScanMouse v6.0, microT v5.0, PicTar, miRanda 11/2010, mirDB 04/2012). Overall, we have tested the expression of a group of highly probable miR-146b targets in hBDNF^Val and hBDNF^Met knock-in mice.

Luciferase assays

Dual-Luciferase® Reporter Assay System and psiCHECK2 luciferase reporter construct which contains a Renilla gene as the reporter and a firefly genes as the internal control were purchased from Promega. The longest 3’UTRs of predicted miR-146b targets (Irak1, Tarf6, Per1, Stx3, Syt1, Kctd15, Sort1, Dlgap1, Npas4, Lin28A) were cloned into XhoI and NotI sites of psiCHECK2. mir-146b binding site mutant clones of Per1 and Npas4 were generated by PCR-based mutagenesis and subcloning. Per1 3’UTR mutant clone: Site Mut1 sequence (starting from position 186 in 3’UTR): TccAAGTTCagA (lower case letters denote altered nucleotide). Site Mut2 sequence (starting from position 495): GctCCCAGGTGTTacaA. Npas4 3’UTR mutant clone: Site Mut sequence (starting from position 309 in 3’UTR): TccGCCAGTTaca. All the clones were verified by Sanger sequencing. Mutations are predicted by RNAhybrid (Rehmsmeier et al., 2004) to disrupt the binding of miR-146b at the seeds and secondary binding sites. Luciferase assays were performed as described previously (Xu et al., 2013). Briefly, N18, a neuroblastoma cell line, was cultured in 24 well plates to 70% confluency and then transfected with various psiCHECK2 reporter constructs (100 ng per well) together with pre-mir-146 mimic or pre-scramble control oligo (1 nM = 0.5 pmol). Luciferase assays were performed 24 hr posttransfection using the Dual-Luciferase Reporter Assay System according to the manufacturer’s instructions. The luciferase signals were measured with a Turner BioSystems 20/20n Single-Tube Luminometer. The intensity of Renilla signal was normalized to the intensity of firefly signal for each assay. Each condition involved at least 3 biological repeats and 2 technical repeats. To control for the non-specific effects of introducing plasmids with different 3’UTRs and oligos, we normalized the luciferase level of each gene to the baseline (plasmid without 3’UTR) to control for the effect of introducing plasmids with different 3’UTRs and then compared the effect of pre-146b mimic for each gene to its corresponding pre-scramble control to determine the specific effect of pre-146b on each gene. All statistical analyses of Luciferase assays were conducted in MS-Excel with a student t-test.

Functional enrichment analysis of conserved targets of miRNAs

Conserved miRNA targets were predicted and extracted using TargetScan (http://www.targetscan.org/) with mouse homolog of human species. All targets were imported into The Database for Annotation, Visualization and Integrated Discovery (DAVID, v6.7, http://david.abcc.ncifcrf.gov/) for a functional annotation using its default settings (Huang da et al., 2009).

Western blot assays

HPC from 8-wk old mice were isolated and homogenized in ice-old modified RIPA buffer containing 1% Triton X-100, 0.2 mM EDTA, 100 mM KCl and 20 mM Tris pH 8.0 and Proteinase inhibitor cocktail (Roche). Homogenates were centrifuged at 12,000 × g at 4°C for 30 min. The supernatant was saved and the protein concentration in each sample was determined by DC Protein Assay (Bio-Rad). An aliquot of the supernatant equivalent to 50 µg (Dgcr8 and Dicer1 blots) or 20 µg (blots of all other protein) proteins were resolved on 4–12% polyacrylamide gel (Bio-Rad) and then transferred onto an ECF plus membrane (Amersham Biosciences) or Immobilon-FL membrane (Millipore). Antibody hybridization was performed as described previously.(Stark et al., 2008) Primary anitbodies used were Traf6 (#597, MBL), 1:1000 dilution; Irak1 (D51G7, Cell Signaling), 1:1000 dilution; Dicer1 (N1767/7, NeuroMab), 1:500 dilution; Dgcr8 (10996-1-AP, proteintech), 1:1000 dilution; Lin28A (#3978, Cell Signaling), 1:1000 dilution; Ago2/Eif2C2 (10686-1-AP, proteintech), 1:500 dilution; GluA1/GluR1 (AB1504, Millipore), 1:400 dilution; Limk1 (#3842, Cell Signaling), 1:500 dilution; β-actin antibody (A5441, Sigma-Aldrich), 1:10000 dilution; α-tubulin antibody (T5168, Sigma-Aldrich), 1:100000 dilution. Horseradish peroxidase conjugated secondary antibodies (1:5000 dilution) were subsequently used for probing the primary antibodies. The washed membrane was incubated with HRP substrate (Western Lightning Chemiluminescence Reagent, PerkinElmer Life Sciences) for 1 min, and chemiluminescence images were obtained using Alpha imaging system. Protein bands were subjected to densitometric analysis with ImageQuant (Molecular Dynamics) (Traf6 and Irak1) or NIH Image J (blots of all other proteins). Statistical analyses were conducted in MS-Excel with a student t-test upon the data from 3 biological repeats.

Database linking

The URL for data presented herein is as follows:

• microRNA Database (miRBase), http://www.mirbase.org/

• TargetScan Mammal v.4.2, http://www.targetscan.org/vert_42/

• TargetScanMouse v6.2, http://www.targetscan.org/mmu_61/

• PicTar prediction in vertebrates, http://pictar.mdc-berlin.de/cgi-bin/PicTar_vertebrate.cgi

• microT v.5, http://diana.cslab.ece.ntua.gr/micro-CDS/?r=search

• miRanda, http://www.microrna.org/microrna/home.do

• mirDB, http://mirdb.org/miRDB/

• Primer3, http://frodo.wi.mit.edu/

• TFSEARCH v.1.3, http://www.cbrc.jp/research/db/ TFSEARCH.html

• ConSite, http://asp.ii.uib.no:8090/cgi-bin/CONSITE/consite

• UCSC Genome Browser, assembly Dec. 2011 (GRCm38/mm10), http://genome.ucsc.edu/cgi-bin/hgGateway

• UCSC Genome Browser, assembly Feb. 2006 (NCBI36/mm8), http://genome.ucsc.edu/cgi-bin/hgGateway?hgsid=327728483&clade=mammal&org=Mouse&db=mm8

• Mouse miRNA promoter (Marson et al., 2008)http://www.cell.com/supplemental/S0092–8674%2808%2900938-0

Highlights.

• •miR-146b is dysregulated in the hippocampus of humanized BDNFVal66Met knock-in mice
• □Reduced miR-146b leads to increase of Per1, Npas4 and Irak1 in vitro and in vivo
• □miRNA-mediated trans effect of common variants could contribute to complex disorders

Abbreviations

BDNF

brain derived neurotrophic factor

SNPs

single nucleotide polymorphisms

miRNA

microRNA

eQTL

expression quantitative trait loci