Abstract
Substantial variability exists in the presentation of complex neurological disorders, and the study of single nucleotide polymorphisms (SNPs) has shed light on disease mechanisms and pathophysiological variability in some cases. However, the vast majority of disease-linked SNPs have unidentified pathophysiological relevance. Here, we tested the hypothesis that SNPs within the miRNA recognition element (MRE; the region of the target transcript to which the miRNA binds) can impart changes in the expression of those genes, either by enhancing or reducing transcript and protein levels. To test this, we cross-referenced 7,153 miRNA-MRE brain interactions with the SNP database (dbSNP) to identify candidates, and functionally assessed 24 SNPs located in the 3’UTR or the coding sequence (CDS) of targets. For over half of the candidates tested, SNPs either enhanced (4 genes) or disrupted (10 genes) miRNA binding and target regulation. Additionally, SNPs causing a shift from a common to rare codon within the CDS facilitated miRNA binding downstream of the SNP, dramatically repressing target gene expression. The biological activity of the SNPs on miRNA regulation was also confirmed in induced pluripotent stem cell (iPSC) lines. These studies strongly support the notion that SNPs in the 3’UTR or the coding sequence of disease-relevant genes may be important in disease pathogenesis and should be reconsidered as candidate modifiers.
Introduction
Genome wide association studies (GWAS) (1,2) and work defining quantitative trait loci (QTLs) (3,4) have shown that some single nucleotide polymorphisms (SNPs) associate with the onset or susceptibility of complex diseases (5). Although disease-associated SNPs commonly alter the amino acid sequence of the translated protein, SNPs located within a disease allele may not cause obvious protein misfolding or impair transcript stability, and may even present as conservative or synonymous mutations. While many SNPs are in linkage disequilibrium, there is limited information connecting the actual SNP to disease pathophysiology (6).
The majority of human genes are under post-transcriptional regulation by miRNAs (7) and single base changes can alter their effectiveness in repressing transcript levels. We postulate that an SNP-based effects on protein expression or stability could occur if they reside within a microRNA recognition element (MRE). In this manner, SNPs could create, destroy, or modify microRNA (miRNA) interactions with the mRNA. This phenomenon was demonstrated earlier when the SNP rs7646 was shown to alter miR-197 binding to the MTHFD1L 3′UTR, causing more efficient repression in the presence of the allele associated with increased risk of neural tube disorders (8). Similarly, SNPs alter miR-330-3p mediated regulation of MAP2K5, potentially contributing to mood and anxiety-related traits (9). Given the relevance of miRNAs in regulating neural patterning (10), neural stem cell differentiation (11), cell type specification (12), synaptic plasticity (13), and neuropsychiatric disorders (14–16) we looked at a role for SNPs in miRNA-based regulation on a broad scale.
Bioinformatics driven approaches predict a number of possible miRNA-MRE interactions in the human brain (17). A transcriptome-wide map of Argonaute (Ago2) CLIP-seq data has refined our understanding of those miRNAs engaged in gene silencing in mouse (18) and human brain (19). Of the 7,153 miRNA-MRE reported interactions constituting stringent Ago2 binding sites for the most abundant brain miRNAs, functional regulation of >3,000 genes are likely. The mapped interactions occur nearly equally in the coding sequence (CDS) and the 3’ untranslated region (3’ UTR) of mRNAs (18,19). While 3’ UTR binding is the classical mechanism that prediction algorithms focus on, CDS interactions can occur, as MREs in transcript coding regions have been shown to impart gene regulation (20,21).
In this study, we cross-referenced SNP locations from dbSNP (22) against the 7,153 miRNA-MRE reported interactions in the human brain (19). The pipeline for prioritization of SNPs was to i) require that the SNPs reside within the seed sequence of the MRE, and ii) exclude genes with additional distinct MRE targets to more abundant miRNAs (by Ago2-CLIP). We selected 24 candidate genes for validation. In 13 the MRE resides in the CDS and in 11 the MRE is in the 3’UTR of the target gene. We also tested the impact of synonymous SNPs in the CDS on gene expression. In all, we demonstrate that SNP-resident MREs can alter mRNA and protein levels of brain-expressed transcripts.
Results
SNPs in the 3’UTR MRE modify miRNA-mediated regulation of target genes. (A) Flowchart describing steps utilized to prioritize candidate miRNA-MRE interactions for further study. Of the 24 candidates selected, 11 SNP-resident MREs are in the 3’UTR and 13 SNP-resident MREs are in the CDS of target genes. (B) Schematic describing how a SNP in a MRE could enhance or disrupt miRNA-mediated target gene regulation. (C–H) HEK293 cells were transfected with constructs containing the reference SNP (Ref), the alternative ‘candidate’ SNP (Alt), or a scrambled MRE (Scr) for the following genes (C) RAPGEF1, (D) SOD1, (E) CARF, (F) TUBB4A, (G) FBXO45 and (H) RPRD1A and analysed for RNA and protein expression. RNA and protein was harvested 48 h post-transfection. Left panel: RT-qPCR for Renilla (normalized to Firefly mRNA), presented as fold change relative to WT (n = 9). Right panel: Relative luminescence units (RLU) normalized (Renilla/Firefly) and presented as fold change relative to WT (n = 12). All panels: data presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test, **P < 0.01, ***P < 0.001.
Seventeen unique miRNAs were represented across the 24 candidates. We cloned each miRNA into a vector whereby expression was driven by Pol III U6. Their biological activity was confirmed by assessing their suppressive action on Renilla luciferase with each miRNA showing a dose dependent repression of their respective perfect targets (Supplementary Material, Fig. S1). All miRNAs under study were also expressed and interacted with target mRNAs by PAR-CLIP in HEK293 cells (25).
SNPs in 3’UTR MREs alter miRNA-mediated regulation
Next, we cloned the entire 3’UTR (containing the ancestral SNP, referred to as reference allele) of the 11 target genes with a 3’UTR-resident MRE (Supplementary Material, Table S2) downstream of a Renilla luciferase reporter (referred to as ‘Ref’ for reference). We generated the altSNPs by site-directed mutagenesis (referred to as ‘Alt’ for alternative) and scrambled the MRE at seed location 2-5 to abrogate putative miRNA binding (referred to as ‘Scr’). Fig. 1B depicts how a C->G SNP could disrupt, and how a G->C SNP could enhance miRNA-mediated repression of a target gene.
When HEK293 cells were transfected with RAPGEF1-Alt or SOD1-Alt, a significant decrease in Renilla mRNA and protein levels was observed by RT-qPCR and luciferase assay, relative to the Ref construct (Fig. 1C and D). In contrast, there was a significant increase in target mRNA and protein levels in cells transfected with the Scr plasmid, indicating regulation by endogenous miR-124-3p (targets RAPGEF1) and miR-139-5p (targets SOD1) is lost (Fig. 1C and D). The effect of the miRNA on regulation of the altSNP-resident MRE was also confirmed by varying miRNA doses using miRNA overexpression constructs. Increasing miR-124-3p expression plasmid levels resulted in increased knockdown of RAFGEF1-Alt (compared to Ref) at every ratio measured, while scrambling the seed prevented this effect (Supplementary Material, Fig. S2A). SOD1-Alt showed similar results on overexpressing miR-139-5p (Supplementary Material, Fig. S2B). Thus SNPs rs3739497 (RAPGEF1) and rs41391245 (SOD1) improve miRNA-mediated repression.
Introducing the altSNP in CARF (rs62183749, Fig. 1E), TUBB4A (rs8113500, Fig. 1F), FBXO45 (rs78195212, Fig. 1G) and RPRD1A (rs78544189, Fig. 1H) disrupted miRNA-mediated repression, resulting in increased Renilla mRNA and protein levels compared to WT, and often to the same extent as the Scr controls. Similarly, while increasing the dose of miR-181-5p (targets CARF, Supplementary Material, Fig. S2C), miR-1301-3p (targets TUBB4A, Supplementary Material, Fig. S2D), miR-381-3p (targets FBXO45, Supplementary Material, Fig. S2E) and miR-23a-3p (targets RPRD1A, Supplementary Material, Fig. S2F) boosted repression of the Ref constructs, it was less effective in reducing the levels of 3’UTRs harbouring the relevant altSNPs.
Other genes with altSNPs residing in 3’UTR MREs show a similar pattern of regulation, namely PDE8B (Supplementary Material, Fig. S2G) and RAB5A (Supplementary Material, Fig. S2I). In both cases, the Ref gene product showed significant repression by endogenous miRNAs, while the altSNP significantly diminished miRNA-mediated repression (PDE8B-rs1580, RAB5A-rs8682). Overexpressing the targeting miRNA boosted PDE8B-Ref (Supplementary Material, Fig. S2H) and RAB5A-Ref (Supplementary Material, Fig. S2J) repression, and introducing the altSNP dampened it, at least at low miRNA expression plasmid doses. We also found instances where the altSNP had no effect. For example, for PPIP5K2-Alt (rs8488, Supplementary Material, Fig. S2K-L) and UNC80-Alt (rs73080885, Supplementary Material, Fig. S2M and N) we found similar levels of miRNA-mediated repression to the Ref. SSB is probably not a real target (Supplementary Material, Fig. S2O and P), as significant repression occurred only with miR-7-5p overexpression. Thus, 8 of 11 candidate genes with 3’UTR MREs show strong altSNP-dependent effects on miRNA-mediated repression.
SNPs in CDS MREs alter miRNA-mediated regulation
We identified 13 genes where the SNP-resident MRE is located within the CDS of the target mRNA (Supplementary Material, Table S2). Their cDNAs, derived from human brain RNA to ensure that the appropriate isoform is studied, along with a Flag- and His-tag epitope, were cloned downstream of a CMV promoter (referred to as ‘Ref’) for expression. The relevant altSNP (referred to as ‘Alt’) or a scrambled MRE (seed nt 2-5; referred to as ‘Scr’) was introduced by site-directed mutagenesis. Of note, we made synonymous mutations with similar codon frequencies to the ancestral codon to generate the Scr constructs.
SNPs in the CDS MRE modify miRNA-mediated regulation of target genes. (A–F) HEK293 cells were transfected with Ref, Alt or Scr vectors for the following genes (A) SREBF1, (B) TUBA1A, (C) MAN2A2, (D) MARCH7, (E) SCARB1, and (F) WBSCR17. RNA and protein was harvested 48 h post-transfection. Left panel: RT-qPCR for respective genes (normalized to HPRT), presented as fold change relative to WT (n = 9). Right panel: ELISA (anti-Flag Ab) for the respective genes, presented as fold change relative to WT (n = 10). All panels: data presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test, **P < 0.01, ***P < 0.001.
AltSNPs disrupted miRNA binding in MAN2A2 (rs2677743, Fig. 2C, Supplementary Material, Fig. S3D), MARCH7 (rs17813964, Fig. 2D, Supplementary Material, Fig. S3E), SCARB1 (rs5889, Fig. 2E, Supplementary Material, Fig. S3F), and WBSCR17 (rs17143818, Fig. 2F, Supplementary Material, Fig. S3G). AltSNPs in the CDS of MARCH7 and WBSCR17 abrogated miRNA-mediated repression to levels similar to Scr. However, altSNPs in the CDS of MAN2A2 and SCARB1, while not as effective as the Scr, still caused significant de-repression of both genes. The significant de-repression in the setting of the Scr for these genes supports regulation by the endogenous corresponding miRNAs.
Overexpression of let-7a-5p (targets MAN2A2 and MARCH7, Supplementary Material, Fig. S4E and F) and miR-124-3p (targets SCARB1, Supplementary Material, Fig. S4G) caused the increased repression of the Ref construct, and in each case, the altSNP construct showed less repression compared to Ref. Overexpression of miR-30b-5p (targets WBSCR17, Supplementary Material, Fig. S4H) caused increased repression of WBSCR17-Alt, divergent to the regulation observed under endogenous miRNA levels (Fig. 2F). Of the remaining 6 genes in the CDS screen, PPP3R1, SYN1, and SYT showed regulation by endogenous miRNAs (Supplementary Materials, Figs S3H–J, S4I–N), but the respective altSNPs had no impact. The remaining 3 genes were not regulated by the corresponding candidate miRNAs (Supplementary Materials, Figs. S3K–M, S4O–T). Thus, 7 of 13 CDS-resident MRE candidates show strong SNP-dependent effects on miRNA-mediated repression.
Rare codon-generating altSNPs trigger miRNA-mediated repression
Synonymous codon changes generate active MREs. (A) Flowchart describes steps utilized to prioritize genes. (B) Schematic describing how a synonymous codon change via a SNP in the CDS could cause ribosomal pausing, resulting in miRNA-mediated silencing. HEK293 cells were transfected with the Ref or Alt plasmids for (C, D) ABCA7 or (E–F) PICALM. (C, E) Left panel: RT-qPCR for respective genes (normalized to HPRT), presented as fold change relative to WT (n = 9). Right panel: ELISA (anti-Flag Ab) for respective genes, presented as fold change relative to WT (n = 10). (D, F) Actinomycin-D was added 48 h post-transfection and chase performed (RNA harvest) at noted time points. RT-qPCR for the respective genes (normalized to HPRT), presented as fold change relative to time 0 (n = 3). (G, H) HEK293 cells were co-transfected with Ref or Alt plasmids along with a non-targeting oligonucleotide control (Nt) or (G) anti-miR-34a-5p or (H) anti-miR-34a-5p + anti-miR-103a-3p. RNA and protein was harvested 48 h post-transfection. Left panel: RT-qPCR for respective genes (normalized to HPRT), presented as fold change relative to Ref + Nt/anti-miR. n = 9. Right panel: ELISA (anti-Flag Ab) for respective genes, presented as fold change relative to Ref + Nt/anti-miR. n = 10. Panels D, F: data presented as mean with standard error; remaining panels: data presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test, *P < 0.05, **P < 0.01, ***P < 0.001. Two-way ANOVA was used to test for interaction, ###P < 0.001.
ABCA7-Alt (rs4147914) mRNA and protein levels were significantly reduced relative to the Ref construct (Fig. 3C, Supplementary Material, S5A and B). To test if the altSNP-resident mRNA was less stable, we performed an Actinomycin-D pulse chase. These data show that the half-life of the mRNA went from greater than 24 h to ∼2 h with the introduction of the altSNP (Fig. 3D). Similar results were obtained for PICALM (rs592297) (Fig. 3E and F, Supplementary Material, Fig. S5C and D) and HMGCS1 (rs56257144) (Supplementary Material, Fig. S5E–G), wherein the altSNP significantly decreased gene expression. SNPs in the CDS of CNTN6, GRIN2A and SORL1 (Supplementary Material, Fig. S5H–M) had no effect on mRNA or protein levels when tested in HEK293 cells.
To test which miRNAs mediated the observed reduction in mRNA levels, we searched for predicted MREs (of the top 50 brain expressed miRNAs) on the CDS of ABCA7, PICALM, and HMGCS1 downstream of the SNP site using the computational tool RNA22 V2 (24) (Supplementary Material, Table S4). There are 46 MREs on ABCA7, comprising binding sites for 15 of the top 50 brain expressed miRNAs. PICALM has 8 predicted MREs (6 miRNAs) and HMGCS1 has 1 predicted MRE. miR-34a-5p has 12 MREs on ABCA7. Although not the most abundant miRNA targeting ABCA7, based on the site type and folding energies for each MRE, miR-34a-5p is the top predicted regulatory miRNA. PICALM has two MREs each for miR-34a-5p and miR-103a-3p. To test their relevance, we co-transfected HEK293 cells with Ref or Alt plasmids for each gene along with the anti-miRNAs for miR-34a-5p or miR-103a-3p. We observed significant de-repression of both ABCA7 (Fig. 3G, Supplementary Material, Fig. S5B on transfecting anti-miR-34a-5p) and PICALM (Fig. 3H, Supplementary Material, Fig. S5D on transfecting anti-miR-34a-5p + anti-miR-103a-3p). These results suggest that the activity of miR-34a-5p on ABCA7 and the activity of miR-34a-5p + miR-103a-3p on PICALM are directly responsible for target repression when the altSNP is present.
SNPs in MREs alter miRNA binding
SNPs in MREs alter miRNA binding. HEK293 cells were co-transfected with Ref + Alt plasmids of the noted genes. Ago2 immunoprecipitation (anti-Ago2 Ab) and associated RNA was harvested 24 h later. RT-qPCR using allele-specific primers (Ref vs Alt) for the respective genes (normalized to HPRT and total gene expression Ref + Alt) was performed. Data are presented as the ratio of Alt/Ref (n = 12). Data are presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test (SNP vs WT), *P < 0.05, **P < 0.01, ***P < 0.001.
SNPs modify miRNA-mediated regulation in induced pluripotent stem cells (iPSCs)
To test the effects of the altSNPs on miRNA-mediated regulation in situ, we screened a library of 14 iPSC lines generated from population control donors for heterozygosity. iPSC lines were classified as altSNP-/- (homozygous for ancestral/reference allele), altSNP+/+ (homozygous for the alternative SNP), and altSNP+/- (alternative SNP/reference allele). As expected from genetic variation between individuals (30,31), the donors show a range of gene expression levels (Supplementary Material, Fig. S6).
SNPs in MREs alter miRNA-mediated regulation in iPSCs. Three iPSC lines representing the following genotypes: altSNP-/- (homozygous for reference allele), altSNP+/+ (homozygous for the alternative candidate SNP), and altSNP+/- (alternative SNP/reference allele) were selected for each gene. iPSC lines were left untreated (Ut) or transfected with the non-targeting (Nt) control or respective miRNA reagents: (A) RAPGEF1: anti-miR-124-3p; (B) SOD1: anti-miR-139-5p; (C) CARF: mimic of miR-181-5p; (D) SREBF1: anti-miR-29b-3p; (E) ABCA7: anti-miR-34a-5p; (F) PICALM: anti-miR-34a-5p + anti-miR-103a-3p. RNA and protein was harvested 48 h post-transfection. Top panel: RT-qPCR for the respective genes (normalized to HPRT), presented as fold change relative to WT (n = 4). Bottom panel: Densitometry of immunoblots performed for each gene (shown in Supplementary Material, Fig. S7) presented as fold change relative to Ut (n = 3). iPSC lines used (reference Supplementary Material, Fig. S6): RAPGEF1 (1,4,7,2,13,14); CARF (1,6,12,13,2,9); SOD1 (1,3,6,4,10,12); SREBF1 (1,8,10,4,5,6); ABCA7 (4,11,9,1,7,10); PICALM (3,7,11,1,5,10). Data are presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test, *P < 0.05, **P < 0.01, ***P < 0.001.
SNPs in the MRE alter miRNA binding in iPSCs. Three altSNP+/- iPSC lines for each of the 6 genes were left untreated (Ut) or transfected with the non-targeting (Nt) control or respective miRNA reagents (RAPGEF1:anti-miR-124-3p, SOD1:anti-miR-139-5p, CARF: mimic of miR-181-5p, SREBF1:anti-miR-29b-3p, ABCA7:anti-miR-34a-5p, PICALM: anti-miR-34a-5p + anti-miR-103a-3p). Ago2 immunoprecipitation (anti-Ago2 Ab) and associated RNA was harvested 24 h later. RT-qPCR using allele-specific primers (Ref vs Alt) for respective genes (normalized to HPRT and total gene expression Ref + Alt) was performed. iPSC lines used (reference Figure S6): RAPGEF1 (2,13,14); CARF (1,6,12); SOD1 (4,10,12); SREBF1 (4,5,6); ABCA7 (1,7,10); PICALM (3,7,11). Data are presented as the ratio of SNP/WT (n = 12). Data are presented as box & whiskers plot (min to max) and analysed by one-way ANOVA with Sidak’s multiple comparison test, *P < 0.05, **P < 0.01, ***P < 0.001.
Discussion
While disease-associated SNPs have been identified and mapped, the pathophysiological relevance for many remains largely anonymous (32). A functional consequence of SNPs within the MRE of a miRNA is that they can impart changes in the expression of host genes, either by enhancing or reducing transcript and protein levels. Many groups have described strategies to reveal SNPs within MREs that cause altered gene expression and in some cases, pathophysiological effects. Currently available computational tools include MirSNP (33), a computational approach to identifying SNPs in MREs by combining TargetScan (17) predictions, miRNA seed information, and dbSNP data to compute and predict the occurrences of ‘seed-altering’ SNPs. However, because TargetScan is based upon MREs in the 3’UTR, and the model was trained with data from a single cell type, not all SNP-MRE interactions will be captured. The PolymiRTS (34) database identifies and annotates SNPs in MREs which have been biologically tested, but again focuses on miRNA binding regions in the 3’UTR. Finally, mrSNP (35) provides a web service for researchers working with RNA-seq data to predict the impact of an SNP in the 3’UTR on miRNA binding. While all these approaches are useful to some degree, we developed an approach to capture SNP-resident MREs in the brain and brain diseases.
The novelty in our approach is three fold. First, we captured >95% of all predicted miRNA-mediated gene regulation events in the brain by focusing exclusively on the top 50 expressed miRNAs. Second, not being constrained by a 3’UTR bias in identifying candidate miRNA-MRE interactions allowed us to investigate physiologically relevant CDS events. Third, we excluded genes with additional distinct MREs to more abundant brain expressed miRNAs, prioritizing the most significant SNP-resident MREs. This was done to maximize the possibility that the altSNP altered activity of the most abundant, and possibly the most relevant miRNA, regulating that gene. Overall, 24 genes were investigated; 14 genes showed significant differences in gene expression levels when the altSNP was introduced in the MRE. These changes corresponded to predicted enhanced or reduced repression. It is possible that those MRE-resident SNPs which did not affect gene expression (3 of 11 in the 3’UTR; 6 of 13 in the CDS) may represent false negatives caused by binding of multiple lower-abundant miRNAs involving MREs distinct from the one queried.
We also asked if synonymous altSNPs causing rare codons could result in significant mRNA repression. We identified three genes whose altSNP-resident mRNAs exhibited increased association with Ago2, higher mRNA decay rates, and significantly lower protein levels. Finally, while the use of exogenous expression systems in HEK293 cells is widely used to test effectiveness of miRNA-mediated target repression, using a non-native promoter to drive gene expression of the target could affect the impact of the altSNP. To validate our findings, we screened 14 iPSC lines for the altSNPs that altered miRNA binding, and show a similar impact.
Cis-acting regulatory SNPs can induce allelic imbalance in gene expression (36–40). While dramatic shifts in gene expression cause disease, as in the case of disease-associated mutations, allelic imbalances cause minor shifts in gene expression, whose consequences can accumulate with age. This can have a dramatic impact, especially on the pathogenesis and pathophysiology of neurologic diseases (41–44). This consequence was recently demonstrated with regard to the pathogenesis of Parkinson’s disease (PD), where cis-acting SNPs either increased or decreased α-synuclein expression in human neurons, thereby increasing or decreasing PD risk respectively (45).
AltSNP-resident MREs could therefore result, over time, in the accumulation of allelic imbalances and alter disease pathophysiology for the better or worse. Calcium response factor (CARF) is a transcription factor that regulates BDNF expression in a calcium- and neuron-selective manner (46,47). Strong evidence point to deficits in BDNF signalling as a contributing factor to the pathogenesis of several major diseases and disorders such as Huntington’s disease, Alzheimer’s disease (AD), amyotrophic lateral sclerosis (ALS), and depression (48,49). The SNP rs62183749 increased CARF expression, possibly rendering a protective influence on disease pathogenesis. Superoxide dismutase 1 (SOD1) is important in apoptotic signalling and oxidative stress (50). Over 150 different toxic gain-of-function mutations in SOD1 link it to familial ALS (51). Of interest however is that wild type SOD1, under conditions of cellular stress, is implicated in ∼90% of sporadic ALS cases (52). While the exact mechanism of how aberrant wild type SOD1 levels contribute to ALS is unknown, it lends an interesting perspective on how the SNP rs41391245 could influence ALS pathogenesis by reducing SOD1 expression. SNPs in the ATP-binding cassette transporter A7 (ABCA7) were recently identified to increase AD risk (53). Decreased expression or loss of ABCA7 levels doubled insoluble Aβ levels and amyloid plaques in the brain (54). Furthermore, recent studies suggest that AD risk associated SNPs in ABCA7 may also contribute to increased PD risk (55). Similarly, the phosphatidylinositol-binding clathrin assembly protein (PICALM) gene has also been identified as a susceptibility locus for late-onset AD (LOAD) incidence (56). PICALM plays a key role in mediating the clearance of Aβ at the blood-brain barrier, as well as mitigating Aβ toxicity in neurons (57). Early evidence suggests that increased PICALM expression may reduce AD risk (58). The SNPs rs4147914 and rs592297 dramatically reduce ABCA7 and PICALM expression respectively, possibly conferring a disease risk.
It is noteworthy that GWAS-SNPs were not prioritized in our search to identify functional SNP-resident MREs. However, of the 17 genes that showed altSNP-induced changes in miRNA-mediated regulation, 5 had GWAS associated hits. These include rs3739497 (RAPGEF1, associated with glycemic traits and epilepsy prognosis, GWAS Central Identifier: HGVM2257467), rs8113500 (TUBB4A, associated with Parkinson’s disease and age-related macular degeneration, HGVM8374170), rs8682 (RAB5A, associated with Parkinson’s disease, stroke and age-related macular degeneration, HGVM15436), rs17813964 (MARCH7, associated with psoriasis and movement-related adverse antipsychotic effects, HGVM11362713), and rs592297 (PICALM, associated with pulmonary function and epilepsy prognosis, HGVM379360).
Allele-specific differences in gene expression owing to cis-regulatory SNPs highlight the potential impact of SNPs on complex neurological disorders. We describe an experimental approach to identify functionally relevant SNP-resident MREs. Creating better predictors of SNPs in miRNA-MRE interactions will allow us to enhance our understanding of various disease mechanisms. Importantly, our experimental paradigm is applicable to any tissue of interest and can guide additional mechanistic studies to determine molecular consequences of risk alleles.
Materials and Methods
Expression constructs
Primary miRNAs were ordered as gBlocks® gene fragments from Integrated DNA Technologies (IDT), and cloned downstream of a U6 promoter by restriction digestion. Perfect targets and non-targeting controls for each miRNA were designed based on the mature miRNA sequence. Sequences were ordered as gBlocks® gene fragments from Integrated DNA Technologies (IDT), and cloned into a psiCHECK-2 vector by Gibson assembly. Complete 3’UTR and CDS of candidate genes were amplified by PCR from cDNA, generated from total RNA harvested from donor human brain samples. This was done to ensure brain expressed isoforms were being studied. The 3’UTR gene products were cloned into psiCHECK-2 vectors by Gibson assembly. The CDS genes were cloned downstream of a CMV promotor in a mammalian expression construct by Gibson assembly, along with a Flag and 6XHis tag on the C-terminal end. We used site-directed mutagenesis by overlap-extension PCR to generate altSNPs and Scr variants of each gene.
Transfection into HEK293 and induced pluripotent stem cells (iPSCs)
HEK293 cells (24 well plates) and iPSCs (6 well plates) were transfected with lipofectamine® 2000 (Thermo Fisher Scientific) using the manufacturers recommended protocols. Plasmid co-transfections were performed using molar concentrations of plasmids. Anti-miRs and miR-mimics were transfected RNAiMAXTM Reagent (Thermo Fisher) to a final concentration of 10nM using the manufacturers recommended protocols. Stock anti-miRs against miR-124-3p, miR-139-5p, miR-29b-3p, miR-34a-5p, and miR-103a-3p were procured from IDT. Stock non-targeting single stranded and double stranded control oligonucleotides were ordered from IDT. A modified mimic of miR-181-5p was ordered from IDT as a duplex:
Mature Mimic Strand:/5PHOS/rArArUrArUrUr CrArArCrGr CrU rGrUr CrGrGrUrGmAmGrU
Complementary Mimic Strand: mUm CrAmCrCmGr AmCrAm GrCmGrUmUrGmArAmUrAmUT
Luciferase assay
Protein was harvested from HEK293 cells 48 h post-transfection. We used the Dual-Luciferase Reporter Assay (Promega) and followed the manufacture’s recommended protocol.
RNA isolation
Total RNA from HEK293 cells and Embryoid bodies (generated from iPSCs) was isolated using the mirVana™ miRNA isolation kit (ThermoFisher Scientific) or the TRIzol® Reagent (Life Technologies), according to the manufacturer’s protocol. Plasmid and genomic DNA was degraded using DNase I treatment (New England Biolabs). Total RNA was tested for quality on an Agilent Model 2100 Bioanalyzer (Agilent Technologies).
Quantitative RT-PCR (RT-qPCR)
First-strand cDNA was synthesized using MultiScribe™ Reverse Transcriptase (Thermo Fisher Scientific) with oligo-dT or random-hexamer priming. Sequence and cDNA specific primers were designed for SYBR® green assays and were procured for the following human genes from IDT: RAPGEF1, SOD1, CARF, TUBB4A, FBXO45, RPRD1A, PDE8B, RAB5A, PPIP5K2, UNC80, SSB, SREBF1, TUBA1A, HSPD1, MAN2A2, MARCH7, SCRAB1, WBSCR17, PPP3R1, SYN1, SYT1, FKBP4, GTDC1, NDUFV2, ABCA7, PICALM, HMGCS1, CNTN6, GRIN2A, SORL1, GAPDH, HPRT, RENILLA, and FIREFLY. All reactions were setup using Power SYBR® Universal PCR Master Mix (Thermo Fisher Scientific) and run on the BioRad CFX96™ Real-Time PCR detection system. All experiments were performed in quadruplicate.
Allele-specific RT-qPCR
RNA was harvested, DNase treated, first-strand cDNA synthesis performed as described above. Primers were designed for each gene to capture total gene expression levels (Ref + altSNP). Allele-specific forward primers were made according to Gaudet et al. (59). A common reverse primer and the allele-specific forward primer were used for allele-specific quantitation (normalized to total). All reactions were in Power SYBR® Universal PCR Master Mix (Thermo Fisher Scientific) and run on the BioRad CFX96™ Real-Time PCR detection system. All experiments were performed in quadruplicate.
SDS-PAGE and Immunoblotting
HEK293 cells or Embryoid bodies (generated from iPSCs) were washed with PBS and lysed in freshly prepared lysis buffer (1% Triton, 25mM Tris pH 7.4, 150mM NaCl, protease inhibitors (cOmplete™, mini, EDTA-free, Roche)) for 30 min at 4°C. The lysates were centrifuged at 14,000 rpm for 20 min at 4°C, and the supernatant quantified by BCA Protein Assay kit (Pierce). Protein was denatured in 6X-Sample SDS buffer (375 mM Tris-HCl pH 6.8, 6% SDS, 48% glycerol, 9% 2-Mercaptoethanol, and 0.03% bromophenol blue). 20 μg of protein per lane was separated on a 4-15% SDS-PAGE gel for western blot analysis. Protein abundance was quantified by densitometry using an ChemiDoc MP Imaging System (Bio-Rad). Western blots were probed, stripped and re-probed as follows. PVDF membranes were first probed with the antibody against the gene of interest. After imaging, the PVDF membrane was stripped with Restore Western Blot Stripping Buffer (Thermo Scientific) for 15 min, washed in Tris Buffered Saline-Tween (TBS-T) and blocked in 5% Bovine Serum Albumin (BSA, Pierce) for 1 h. The membrane was washed in TBS-T and incubated with the goat anti-mouse or goat-anti-rabbit secondary antibody (1:10000, Sigma) for 1 h and imaged. If a signal was detected, the stripping procedure was repeated until no signal was observed. The membrane was washed in TBS-T, blocked for 1 h in 5% BSA and re-probed with the antibody against tubulin or GAPDH. Antibodies used in this study were procured from the following sources: Flag-M2 (F3165, Sigma); GAPDH (sc32233, SantaCruz); b-tubulin (sc5274, SantaCruz); RAPGEF1 (A301-965A, Bethyl Labs); SOD1 (A303-811A, Bethyl Labs); SREBF1 (ab3259, Abcam); ABCA7 (A304-427A, Bethyl Labs); CARF (A303-861A, Bethyl Labs); PICALM (ab172962, Abcam); His (ab9108, Abcam); Argonaute 2 (ab57113, Abcam).
Immunoprecipitation
Immunoprecipitation (IP) experiments were performed in HEK293 cells and SNP+/- iPS cells. To IP Argonaute 2, cells were lysed (as described above), and the supernatant (20–50 µg of protein) was incubated with anti-Ago2 Ab overnight at 4 °C, followed by incubation with protein G-agarose (Invitrogen) for 1 h at 4 °C. Beads were pelleted at 2,500 rpm for 30 sec, supernatant removed, and beads were resuspended in 500 mL RIP buffer. Three washes were repeated with RIP buffer, followed by one wash in PBS. The pellet was resuspended in TRIzol® Reagent (Life Technologies) (1 mL) and RNA harvested according to manufacturer’s instructions. RNA was eluted with nuclease-free water (20 μL). RIP buffer: 150 mM KCl, 25 mM Tris pH 7.4, 5 mM EDTA, 0.5 mM DTT, 0.5% NP40, 100 U/ml RNAase inhibitor (added fresh each time), protease inhibitors (added fresh each time).
6XHis-trap anti-FLAG ELISA
Cells (grown in 24 well plates) were lysed in lysis buffer (50 mM Tris-Hcl, 1M NaCl, 10 mM TCEP, 25 mM Imidazole, protease inhibitors) and processed as described above. Total protein (50 ug) was added to a single well on the PierceTM Nickel Coated Plates (Thermo Scientific) and incubated for 4–6 h at 4C. Each well was then gently washed 3 times with wash buffer (50 mM Tris-Hcl, 0.1M NaCl, 10 mM TCEP, 50 mM Imidazole). Each well was incubated with the anti-FLAG-M2-peroxidase antibody for 1 h at 4C. After washed each well 6 times with ice-cold PBS, the Amplex® UltraRed Reagent (Molecular Probes) assay solution was added (200 ul/well). Plates were incubated for 10-20 min at room temperature (in dark). 200 ul of the reaction was transferred to a 96well BLACK plate and read on the SpectraMax Minimax 300 Imaging Cytometer (Molecular Devices).
iPSC maintenance and orbital embryoid body generation
IPSC lines are stock control lines maintained by the Human Pluripotent Stem Cell Core at the Raymond G. Perelman Center for Cellular and Molecular Therapeutics, The Children’s Hospital of Philadelphia. Protocols to maintain and expand iPSCs are described by Sullivan et al. (60). After transfection, orbital embryoid bodies were generated from iPSCs to prevent contaminating DNA/RNA from mouse embryonic fibroblasts. Protocols to generate the embryoid bodies from iPSCs are described by Chen et al. (61).
| iPSC Line | Cell origin | Sex/Age (yrs) | iPSC Line | Cell origin | Sex/Age (yrs) |
|---|---|---|---|---|---|
| Line 1 | Fibroblast | Female, 2 | Line 8 | PBMC | Male, Adult |
| Line 2 | Fibroblast | Male, 3 | Line 9 | PBMC | Female, Adult |
| Line 3 | Fibroblast | Female, 5 | Line 10 | PBMC | Male, Adult |
| Line 4 | Fibroblast | Male, 2 | Line 11 | LCL | Male, 26 |
| Line 5 | Cord Blood | Female | Line 12 | LCL | Male, Adult |
| Line 6 | Bone Marrow | Male, Adult | Line 13 | Fibroblast | Male, newborn |
| Line 7 | PBMC | Male, Adult | Line 14 | Fibroblast | Male, newborn |
| iPSC Line | Cell origin | Sex/Age (yrs) | iPSC Line | Cell origin | Sex/Age (yrs) |
|---|---|---|---|---|---|
| Line 1 | Fibroblast | Female, 2 | Line 8 | PBMC | Male, Adult |
| Line 2 | Fibroblast | Male, 3 | Line 9 | PBMC | Female, Adult |
| Line 3 | Fibroblast | Female, 5 | Line 10 | PBMC | Male, Adult |
| Line 4 | Fibroblast | Male, 2 | Line 11 | LCL | Male, 26 |
| Line 5 | Cord Blood | Female | Line 12 | LCL | Male, Adult |
| Line 6 | Bone Marrow | Male, Adult | Line 13 | Fibroblast | Male, newborn |
| Line 7 | PBMC | Male, Adult | Line 14 | Fibroblast | Male, newborn |
LCL: Lymphoblastoid cell line, PBMC: Primary peripheral blood mononuclear cells.
| iPSC Line | Cell origin | Sex/Age (yrs) | iPSC Line | Cell origin | Sex/Age (yrs) |
|---|---|---|---|---|---|
| Line 1 | Fibroblast | Female, 2 | Line 8 | PBMC | Male, Adult |
| Line 2 | Fibroblast | Male, 3 | Line 9 | PBMC | Female, Adult |
| Line 3 | Fibroblast | Female, 5 | Line 10 | PBMC | Male, Adult |
| Line 4 | Fibroblast | Male, 2 | Line 11 | LCL | Male, 26 |
| Line 5 | Cord Blood | Female | Line 12 | LCL | Male, Adult |
| Line 6 | Bone Marrow | Male, Adult | Line 13 | Fibroblast | Male, newborn |
| Line 7 | PBMC | Male, Adult | Line 14 | Fibroblast | Male, newborn |
| iPSC Line | Cell origin | Sex/Age (yrs) | iPSC Line | Cell origin | Sex/Age (yrs) |
|---|---|---|---|---|---|
| Line 1 | Fibroblast | Female, 2 | Line 8 | PBMC | Male, Adult |
| Line 2 | Fibroblast | Male, 3 | Line 9 | PBMC | Female, Adult |
| Line 3 | Fibroblast | Female, 5 | Line 10 | PBMC | Male, Adult |
| Line 4 | Fibroblast | Male, 2 | Line 11 | LCL | Male, 26 |
| Line 5 | Cord Blood | Female | Line 12 | LCL | Male, Adult |
| Line 6 | Bone Marrow | Male, Adult | Line 13 | Fibroblast | Male, newborn |
| Line 7 | PBMC | Male, Adult | Line 14 | Fibroblast | Male, newborn |
LCL: Lymphoblastoid cell line, PBMC: Primary peripheral blood mononuclear cells.
Bioinformatics Methods
Experimental and analytical details for the human brain AGO2 HITS-CLIP data were reported previously (19). In brief, human brain motor cortex (BA4) and striatum tissue samples were obtained from 11 individuals and lysed. Lysates were UV-irradiated to crosslink RNA-protein complexes. AGO2 RNP complexes were enriched by IP, and Illumina libraries were prepared from the purified AGO2-associated RNAs (both miRNAs and bound target sites). Libraries were sequenced on the Illumina HiSeq 2500 platform. FASTQ files for each library were trimmed of adapter sequence and then mapped to the human genome (hg19) using Bowtie1, requiring unique mapping. To avoid false-positive peak calls resulting from PCR amplification error, alignments were collapsed to unique chromosome:start:end:strand coordinates. P-values were calculated based on a gene-wise zero truncated negative binomial (ZTNB) model. Fisher’s method was used to combine the resulting positional p-values across all samples, and adjacent p-values < 0.05 were clustered. We manually examined each cluster to ensure the SNP resides within the MRE seed. In some cases, however, the interacting miRNA exhibited strong 3’ complementarity with the MRE. In such an instance, we chose to include these interactions as the SNPs still held the potential to disrupt miRNA binding and miRNA mediated regulation.
SNP annotations were obtained from dbSNP138 filtered to contain brain expressed genes (9,310 genes)(62), and downloaded from the UCSC Genome Browser (hg19), using the Table Browser tool. Additional information for coding SNPs, including codon and amino acid changes and identities of altered proteins, was also obtained from the UCSC Table Browser, using the snp138CodingDbSnp linked table. Coding SNP overlap with AGO2 clusters was based on matching Gene IDs. We used the gene ontology tool DAVID (23) for functional annotation based clustering of genes, while the computational tool RNA22 V2 (24) was used to identify novel miRNA binding sites on transcripts.
Statistical Analysis
All statistical analysis was done using GraphPad PRISM V6.0h. Normality of data was tested using the D'Agostino & Pearson omnibus normality test. One-way ANOVA with Sidak’s multiple comparison test was used to determine significance. Two-way ANOVA with Sidak’s multiple comparison test was used to test for significant interaction.
The statistics used in individual experiments along with method of presenting data and error bars are described in the figure legends.
Supplementary Material
Supplementary Material is available at HMG online.
Acknowledgements
We thank Carolyn Yrigollen, Pavitra Ramachandran and Deborah French for critical comments on the manuscript. We would also like to thank Jean Ann Maguire, Liang ‘Grace’ Ge, Alejandro Mas Monteys, Yong Hong Chen, Luis Tecedor, Gregory Cajka and Kasey L Brida for providing essential reagents and for technical assistance.
Conflict of Interest statement. B.L.D. is a founder of Spark Therapeutics, Inc., a gene therapy company.
Funding
The authors also acknowledge the support of the Human Pluripotent Stem Cell Core, The Children’s Hospital of Philadelphia. This work was supported by the National Institutes of Health [R01 NS076631 to B.L.D.]. Funding to pay the Open Access publication charges for this article was provided by Dr Davidson's grant funding.
References
Author notes
†Present address: The Sidney Kimmel Medical College, Thomas Jefferson University, Philadelphia, PA 19107, USA.
