Evidence for genetic regulation of mRNA expression of the dosage-sensitive gene retinoic acid induced-1 (RAI1) in human brain

Abstract

RAI1 (retinoic acid induced-1) is a dosage-sensitive gene that causes Smith-Magenis syndrome (SMS) when mutated or deleted and Potocki-Lupski Syndrome (PTLS) when duplicated, with psychiatric features commonly observed in both syndromes. How common genetic variants regulate this gene, however, is unknown. In this study, we found that RAI1 mRNA expression in Chinese prefrontal and temporal cortex correlate with genotypes of common single nucleotide polymorphisms (SNPs) located in the RAI1 5'-upstream region. Using genotype imputation, "R(2)-Δ(2)" analysis, and data from the RegulomeDB database, we identified SNPs rs4925102 and rs9907986 as possible regulatory variants, accounting for approximately 30-40% of the variance in RAI1 mRNA expression in both brain regions. Specifically, rs4925102 and rs9907986 are predicted to disrupt the binding of retinoic acid RXR-RAR receptors and the transcription factor DEAF1 (Deformed epidermal autoregulatory factor-1), respectively. Consistent with these predictions, we observed binding of RXRα and RARα to the predicted RAI1 target in chromatin immunoprecipitation assays. Retinoic acid is crucial for early development of the central neural system, and DEAF1 is associated with intellectual disability. The observation that a significant portion of RAI1 mRNA expression is genetically controlled raises the possibility that common RAI1 5'-region regulatory variants contribute more generally to psychiatric disorders.

Figure 1. Identification of genotyped SNPs within the RAI1 locus that correlate with mRNA expression in prefrontal cortex and temporal cortex.

(top) Plots of coefficients of determination (R²) from linear regression analysis of RAI1 mRNA expression in prefrontal cortex (upper panel) and temporal cortex (lower panel) vs. genotypes for 37 SNPs located within the neighborhood (~273 kb) of the RAI1 gene. SNPs exceeding the threshold for nominal statistical significance (i.e., P < 0.05, uncorrected for multiple testing) are indicated by blue bars. (P = 0.05 thresholds denoted by horizontal dotted lines.) (middle) Model of the RAI1 gene showing its intron/exon structure, direction of transcription and locations of key SNPs. The red bar indicates the location of the PCR product amplified from the cDNA used to quantify RAI1 mRNA expression. (bottom) Plot of pairwise Δ² ( = r²) LD constants based on genotype data for Han Chinese individuals in our collection generated using Haploview. Dark gray boxes: SNP pairs with 0.9 < Δ² < 1; various shades of gray squares (from light to dark): 0.02 < Δ² < 0.9; white squares: 0 < Δ² < 0.02. Numbers at each end of the triangle indicate the locations of the chromosome segment analyzed (based on GRCh37hg 19 coordinates).

Figure 2. The RAI1 “core” regulatory region.

(top) Screen shot from the UCSC genome browser (http://genome.ucsc.edu) for the indicated region of chromosome 17, showing the locations of the 5′-end of the RAI1 gene, histone III lysine 27 acetylation (H3K27Ac) and DNase sensitive clusters. (bottom) Plots of linear regression R² values for RAI1 SNPs with measured or imputed genotypes (total: 96 SNPs within 90.2 kb). SNPs with nominally significant correlations between genotype and RAI1 mRNA expression in prefrontal cortex or temporal cortex are indicated by blue bars.

Figure 3. Visualization of SNP families that contribute to RAI1 mRNA expression in prefrontal cortex.

(a) R²-Δ² plot showing the contributions of three SNP families associated with the “index” SNPs rs12449964, rs10401011 and rs4636969, respectively, to measured R² values for 96 genotyped or imputed SNPs within the RAI1 core regulatory region. (b) R²-R² plot showing measured vs. predicted R² values for the same set of SNPs. (c) Regression equation derived from multivariable linear regression analysis of measured R² values vs. Δ²_ij values associated with the index SNPs rs12449964, rs10401011 and rs4636969. The variance inflation factors (VIFs) for the Δ²_ij independent variables were all < 5.0; range: 1.8–3.6 (see Methods for details). P-values for regression coefficients and the ANOVA-based P-value and adjusted R²_model for the overall fit of the model are listed in the panel. (d) Plot of measured vs. predicted R²-values for the model based on the index SNPs rs12449964, rs10401011 and rs4636969.

Figure 4. Visualization of SNP families that contribute to RAI1 mRNA expression in temporal cortex.

(a) R²-Δ² plot, (b) R²-R² plot (c) Regression equation and associated P-values and R²_model, (d) Plot showing linear correlation between measured and predicted R² values as described in the legend to Fig. 3, but based on the “index” SNPs rs12449964, rs4925102 and rs8071107. As above, the VIF associated with each Δ²_ij was < 5.0.

Figure 5. Candidate RAI1 regulatory variants.

(top) Screen shot from the UCSC Genome Browser for the indicated region of Chromosome 17 showing the positions of the RAI1 gene, histone III lysine 27 acetylation and DNase sensitive clusters. (bottom) Locations of the “index” SNPs rs12449964, rs4925102 and rs9907986, RXR-RAR and DEAF1 consensus binding sequences, and DNA sequences protected by transcription factor binding in DNase footprint assays as listed in RegulomeDB database (http://www.regulomedb.org/).

Figure 6. Chromatin immunoprecipitation assays demonstrate binding of RXRα and RARα to the RAI1 segment containing the putative RXR::RAR_DR5 binding site and rs4925102.

Neuronal SH-SY5Y cells were treated with 1 μM all-trans-retinoic acid (ATRA) for 135 min prior to cross-linking with formaldehyde and isolation of chromatin. Immunoprecipitation of the cross-linked and fragmented chromatin was performed using mouse anti-RARα (left), anti-RXRα (middle) antibodies, or normal mouse IgG (right). Following purification of immunoprecipitated chromatin and reversal of cross-linking, quantification of target DNA sequences was carried out by real-time PCR using primers designed to amplify a DNA segment containing the putative RXR::RAR_DR5 binding site and rs4925102 (black bars) or a DNA segment from RAI1 exon 3 (grey bars). Each measurement was performed in triplicate. Y-axis: amounts of precipitated target DNA sequences expressed as percent (%) of input target DNA sequences in 50 μg chromatin. Statistical significance was assessed by ANOVA followed by post hoc Tukey tests: ****P < 0.0001, **P = 0.0035 (adjusted Tukey test P-values).