Identification of Novel FMR1 Variants by Massively Parallel Sequencing in Developmentally Delayed Males

Abstract

Fragile X syndrome (FXS), the most common inherited form of developmental delay, is typically caused by CGG-repeat expansion in FMR1. However, little attention has been paid to sequence variants in FMR1. Through the use of pooled-template massively parallel sequencing, we identified 130 novel FMR1 sequence variants in a population of 963 developmentally delayed males without CGG-repeat expansion mutations. Among these, we identified a novel missense change, p.R138Q, which alters a conserved residue in the nuclear localization signal of FMRP. We have also identified three promoter mutations in this population, all of which significantly reduce in vitro levels of FMR1 transcription. Additionally, we identified 10 noncoding variants of possible functional significance in the introns and 3’-untranslated region of FMR1, including two predicted splice site mutations. These findings greatly expand the catalogue of known FMR1 sequence variants and suggest that FMR1 sequence variants may represent an important cause of developmental delay.

INTRODUCTION

Fragile X syndrome (FXS [MIM 300624]) is the most common inherited cause of intellectual disability. In addition to a variable degree of intellectual impairment, FXS patients often exhibit autism-like behaviors, such as gaze avoidance, hand-flapping, and tactile defensiveness. Other classic features of FXS include macroorcchidism and an elongated face with large everted ears. However, due to the subtlety and variable expressivity of the more distinguishing characteristics, the identification of a causal mutation is necessary for the diagnosis of FXS [Garber et al., 2008].

The most common causal mutation leading to FXS is the expansion of the CGG trinucleotide repeat located within the 5’-untranslated region of the FMR1 gene (MIM 309550) [Ashley et al., 1993; Verkerk et al., 1991]. This expansion, referred to as the full mutation, represents expansion beyond 200 repeats. The full mutation leads to hypermethylation of the FMR1 promoter, thereby preventing expression of FMR1 and its gene product, FMRP [Chiurazzi et al., 1998; Sutcliffe et al., 1992]. FMR1 deletions are the second most common known cause of FXS, although far less common than the repeat expansion mutation [Coffee et al., 2008]. While it seems plausible that sequence variants affecting the expression or function of FMRP could represent a third important cause of FXS, only three such mutations, a missense mutation (I304N) and two small deletion nonsense mutations, have been reported since FMR1 was identified in 1991 [De Boulle et al., 1993; Lugenbeel et al., 1995]. Therefore, FMR1 sequencing is rarely performed in the clinical setting, due to the expectation of a low diagnostic yield. Also, methodological constraints have previously prevented a thorough assessment of FMR1 sequence variation in a large number of patients, leaving the true significance of pathogenic sequence variants in FMR1 unknown [Castellvi-Bel et al., 1999; Chiurazzi et al., 1994; Collins et al., 2010; Gronskov et al., 1998; Reyniers et al., 1996; Shinahara et al., 2004; Vincent et al., 1996; Wang et al., 1997].

Massively-parallel sequencing (MPS) vastly improves upon the cost-effectiveness and throughput of traditional Sanger sequencing, enabling facile detection of sequence variation at a scale that was previously impractical [Shendure and Ji, 2008]. One of the many applications that have emerged for MPS is targeted resequencing to detect novel mutations in particular genomic regions, such as a collection of candidate genes [Dahl et al., 2007] or the entire exome [Ng et al., 2009]. Due to the Gigabase-scale capacity of MPS, targeting of a single candidate gene in a single patient is generally inefficient. However, through the use of a pooled-template design, a single gene can be sequenced in multiple individuals simultaneously to screen for the presence of rare or novel sequence variants, thus allowing for efficient, cost-effective large-scale targeted resequencing [Druley et al., 2009; Ingman and Gyllensten, 2009; Koboldt et al., 2009; Out et al., 2009].

To detect potentially pathogenic FMR1 sequence variants, we employed pooled-template MPS to assess the promoter, all 17 exons, and a substantial portion of the intronic sequence of FMR1 in 963 developmentally delayed males referred for FMR1 repeat testing but found not to have the full mutation. We identified one patient with the novel missense change p.R138Q, which alters a conserved residue within the nuclear localization signal of FMRP. Furthermore, we report three novel promoter variants, all of which reduce the in vitro expression of FMR1, and several novel sequence variants in conserved noncoding regions of FMR1, including two predicted splice site mutations. Together, these novel variants suggest that there may be clinical utility in diagnostic FMR1 sequencing for developmentally delayed males.

MATERIALS AND METHODS

Clinical Population

While loss of function mutations in FMR1 may cause a phenotype resembling FXS, we would anticipate that mutations that simply reduce FMRP function or expression would result in a more subtle phenotype. Therefore, we decided to sequence FMR1 in patients who had tested negative for FMR1 repeat expansion at the Emory Genetics Laboratory over a five year span. Because the current standard of care is for all children presenting with developmental delay to be tested for FMR1 repeat expansion, the patients in this clinical population do not necessarily exhibit the classic FXS phenotype, but rather represent the more general diagnosis of developmental delay. Indeed, only 2–3% of such referred samples test positive for the full mutation of FMR1. For ease of interpretation, we elected to only sequence males, in which a variant would be hemizygous and more likely to be penetrant. Patients older than age 18 at the time of testing were excluded, as the clinical indication for their FMR1 repeat test was more likely to have been triggered by a question concerning transmission risk or the appearance of later onset premutation-like tremor/ataxia phenotype, not an early onset developmental delay. Racial identification was available for only 241 of the 963 patients sequenced (25.0%). Among these, 164 (68.1%) were of European descent, 74 (30.7%) were of African descent, and 3 (1.2%) were of Asian descent.

Genomic DNA Samples

We obtained deidentified aliquots of genomic DNA from the Emory Genetics Laboratory, Department of Human Genetics, Emory University School of Medicine, for every male under age 18 who was referred for and tested negative for the FMR1 full mutation from April 2002 to August 2007. In total, 1392 aliquots were obtained. The genomic DNA samples had previously been extracted from whole blood by standard methods in a CLIA-certified environment. The Emory University Institutional Review Board approved this use of deidentified clinical samples.

Massively-Parallel Sequencing

Targeting FMR1

As seen in supporting information Figure 1, four long range PCR (LR-PCR) amplifications were designed to target FMR1 (supporting information Figure 1 can be found in the online version of this article). The LR-PCR primer pairs were as follows: FMR1A-F: 5’-CAGACTGCGCTACTTTGAACC-3’ and FMR1A-R: 5’- CTACATACCAACAAACGCACTACTGCTACAT-3’; FMR1B-F: 5’- AATTTCCAGTATACTTGTCTATTTTTCGAGATG-3’ and FMR1B-R: 5’- TTTTGGGAGATAGCTACCTACAGGGTATCTGATT-3’; FMR1C-F: 5’- GTTGAACATTAAATTGCAGTTCAGAATACATAG-3’ and FMR1C-R: 5’- GAGACATATCCAATCCACTTGCCGTTATAGT-3’; FMR1D-F: 5’- AATAATCTGATACGTTTAAAAGGTTGCTATTGA-3’ and FMR1D-R: 5’- TTAATATGGTTTAGTGGCACCCTATGTAATAAA-3’. Each LR-PCR-A reaction contained 50 ng of genomic DNA, 100 ng of each primer, 5 µl of dNTPs (Takara Bio Inc., Otsu, Shiga, Japan), 12.5 µl of 2x GC Buffer II (Takara), and 0.5 µl of Ex Taq (Takara), in a total of 25 µl. The following PCR conditions were used for LR-PCR-A: initialization at 95°C for 4 minutes; 37 cycles of denaturation at 95°C for 30 seconds and annealing/elongation at 60°C for 4 minutes; and a final elongation step of 72°C for 9 minutes. Each LR-PCR-B, -C, and -D reaction contained 50 ng of genomic DNA, 100 ng of each primer, 4 µl of dNTPs (Takara Bio Inc., Otsu, Shiga, Japan), 2.5 µl of Ex Taq Buffer (Takara), and 0.4 µl of Ex Taq (Takara), in a total of 25 µl. The following PCR conditions were used for LR-PCR-B: initialization at 94°C for 4 minutes; 30 cycles of denaturation at 94°C for 20 seconds and annealing/elongation at 64°C for 8 minutes; and a final elongation step of 68°C for 13 minutes. The same conditions were used for LR-PCR-C, but 35 cycles of denaturation and annealing/elongation were used instead of 30. The same conditions used for LR-PCR-C were used for LR-PCR-D, but the annealing/elongation at 64°C was continued for 9 minutes instead of 8 minutes. The expected sizes of the LR-PCR amplicons were visually confirmed with gel electrophoresis on the E-Gel system (Invitrogen, Carlsbad, CA). Failed LR-PCRs were repeated one time before the sample was excluded from the study. Since clinical testing for the full mutation would have revealed large deletions of FMR1, the most common cause of repeated failure of amplification was DNA degradation. After LR-PCR, 963 samples remained for sequencing.

LR-PCR Amplicon Pooling

The concentrations of LR-PCR amplicons were measured by fluorometric quantification with PicoGreen dsDNA reagent (Molecular Probes, Eugene, OR). Equimolar pools were created by first combining across 19 patients within a given LR-PCR (i.e. A, B, C, or D as shown in Fig. S1). The following amplicon amounts were used per patient: 33.4 ng of LR-PCR-A; 145.8 ng of LR-PCR-B; 172.1 ng of LR-PCR-C; and 200 ng of LR-PCR-D. Included in each LR-PCR-C pool was 172.1 ng of LR-PCR-C from a patient with the single known missense mutation in FMR1 [De Boulle et al., 1993]. This I304N positive control was intended to serve as sentinel for singleton detection in each pool and as a direct measure of the false negative rate of our approach. Each amplicon pool was purified to remove excess primers with the PureLink PCR purification kit (Invitrogen, Carlsbad, CA). Purified amplicon pools were quantified with the Nanodrop 1000 spectrophotometer (Thermo Scientific, Wilmington, DE). For each set of 19 patients, purified amplicon pools A, B, C, and D were combined in equimolar fashion to a total of 4 µg.

Target Library Construction

Amplicon pools were fragmented by sonication (Sonicator S-4000, Misonix, Farmingdale, NY) with the following parameters: 8 pulses of 30 seconds each, with 2 minutes rest between pulses, at a power output of 20%. The fragmentation range, as visualized by gel electrophoresis, was from 100 – 400 bp. End-repair, adaptor ligation, and PCR amplification were performed as described previously [Okou et al., 2009].

Genome Analyzer Sequencing and Analysis

From each processed amplicon pool, a total of 6 pmol of the library were added to one lane of the Genome Analyzer (GA) flowcell (Illumina, San Diego, CA). Single-end sequencing for 26 cycles was performed on the GA using the Illumina protocols for cluster generation and sequencing-by-synthesis. GA image analysis and base-calling were performed, respectively, with the Firecrest and Bustard software packages from the Illumina GA Pipeline. Mapping was performed with the MAQ software package [Li et al., 2008], using the following mapping parameters: a maximum mismatch (−n) of 3, a mutation rate (−m) of 0.01, and a maximum sum of mismatching base qualities (−e) of 140. The short sequence reads were mapped to the FMR1 region of the human genome reference (NT_011681.15, g.3435700 to g.3475545). The reference sequence was modified at the following SNPs, where the minor allele is incorrectly included in the reference sequence: rs1270092, rs1270091, rs4824232, rs4824233, rs5904650, rs11342854, rs61419778, rs5904816, and rs68020458. The reference sequence was further altered by masking repeat elements with RepeatMasker.

Variant Detection

Only base-calls with a GA quality score greater than 45, corresponding to an error rate of 0.003%, were considered for variant detection. The frequency of each base was calculated at each nucleotide position in the pileup of mapped short reads. To account for context-specific errors in GA sequencing [Dohm et al., 2008; Harismendy et al., 2009], we measured the error rate at each nucleotide position by performing a similar GA sequencing run on a single male individual, in which every base position should have a theoretical 100% “major allele frequency.” Call this experiment the “null experiment.” The observed “minor allele frequency” at each nucleotide position in this “null experiment” was used as the expected error rate for each of the pooled sequencing runs. Since each pool contained 20 individual alleles, true variants would be expected to be observed in 5% or more of the reads. To call any given base, we compared two models. In the first model, the number of minor allele reads was assumed to be binomially distributed, with minor allele frequency equal to the error rate estimated in the “null experiment.” This model corresponds to the “null” model of no variants present. In the second model, we assume the minor allele count is also binominally distributed, but with frequency greater than or equal to 5%. We compare the two models with a likelihood ratio test, and call a variant if the variant model fits better than the null model with p < 1 × 10⁻⁴.

Variant Confirmation

Standard Sanger sequencing was used to confirm all variant calls. Fresh LR-PCRs were generated from each individual sample in a pool positive for a given putative variant. The amplicons were purified and Sanger sequenced using primers targeted to confirm the variant. Each chromatogram was visually inspected for the presence of the putative variant.

Control Genotyping

The control samples used for the genotyping of novel variants were obtained from the NIMH Human Genetics Initiative. All controls were adult males of European descent who had been screened to rule out psychiatric disorders. Genotyping was performed by the iPlex Gold method (Sequenom, San Diego, CA) as per the manufacturer’s instructions, using primers (Supporting information Table I, available in the online version of this article) designed with the Sequenom Assay Design 3.1 software. The single-base primer extension method failed for three variants, c.−332G>C, c.−254A>G, and c.−67G>C, described below. These three variants obliterate restriction sites for SacII, EcoNI, and BseYI, respectively. Thus, restriction digestion was used to genotype for these three variants. For both iPlex and restriction digestion genotyping, a positive control was included in every plate to confirm the sensitivity of the assay. After genotyping, a fresh PCR was produced for all control samples in which a minor allele was detected, and traditional Sanger sequencing was used to confirm the presence of the minor allele.

In Silico Analysis

Assessments of the cross-species conservation of the FMRP amino acid sequence and the FMR1 promoter sequence were performed with the ClustalW2 sequence alignment program. Predictions of the effects of amino acid substitution were performed with the programs PMut [Ferrer-Costa et al., 2005], PANTHER PSEC [Thomas and Kejariwal, 2004], SIFT [Ng and Henikoff, 2003], and PolyPhen [Ramensky et al., 2002]. For each variant position, the regional conservation across placental mammals was assessed by phastCons and the basewise conservation across placental mammals was assessed by phyloP; these values were obtained from the UCSC Genome Browser [Rhead et al., 2009; Siepel et al., 2005]. The program NNSplice was used to predict splice sites that may be created or obliterated by novel sequence variants [Reese et al., 1997]. Variants predicted to alter miRNA binding to the FMR1 3’UTR were identified with the program miRanda [Betel et al., 2008; John et al., 2004].

Luciferase Assays

Plasmid Construction

The pFMR1-luc plasmid has been previously reported [Smith et al., 2004]. A multistep process was used to introduce the novel promoter variants into pFMR1-luc. First, the LR-PCR-A amplicon from each of the patients identified with novel promoter variants was cloned using the TOPO TA Cloning kit (Invitrogen, Carlsbad, CA). The TA clones were cut with NarI and NruI and ligated into pBluescript, and then cut with KpnI and HindIII and ligated into the pFMR1-luc plasmid. Sanger sequencing was used to confirm that the three variant plasmids contained the novel promoter variants and that all four plasmids contained an equivalent number of CGG repeats, which was determined to be 8. The pGL3-Basic and pRL-TK plasmids were purchased from Promega (Madison, WI).

Cell Culture and Transfections

HeLa cells were cultured at 37°C with 5% CO₂ in DMEM with 10% fetal bovine serum. Twenty-four hours before transfection, 1×10⁶ cells were plated in 2 ml of media in each well of 6-well cell culture dishes. Transfections were carried out in Opti-MEM (Invitrogen, Carlsbad, CA), using 10 µl of Lipofectamine 2000 (Invitrogen, Carlsbad, CA) and 1 µg of total plasmid in a 10:1 ratio (firefly plasmid: control Renilla plasmid). Each plasmid was transfected into six separate wells. Four hours after transfection, the media containing the transfection reagent and plasmids was replaced with DMEM with 10% fetal bovine serum. Forty-eight hours after transfection, cells were harvested with 500 µl 1x Passive Lysis Buffer (Promega, Madison, WI) by rocking at room temperature for 15 minutes. Lysates were cleared of cell debris by centrifugation at 14000 rpm for 5 minutes at 4°C.

Luciferase Assays

Protein concentrations of the lysates were measured by the Bradford assay. The Dual-Luciferase Reporter Assay System (Promega, Madison, WI) was used to measure luciferase activity. From each lysate, 5 µg of protein was added in 20 µl total volume to a luminometer tube. To each luminometer tube, 100 µl of LAR II was added. A manual-load luminometer was used to measure the luminescence over a 10 second period, following a 2 second premeasurement delay. The luminometer measurement was repeated after the addition of 100 µl of Stop & Glo reagent. For each lysate, the firefly luciferase values were divided by the Renilla luciferase values. The results of six independent transfections were averaged and the standard deviation was calculated for each plasmid.

RESULTS

Sequence Variants in FMR1

Through the use of pooled-template MPS, we sequenced FMR1 in 963 developmentally delayed males, each to an average sequence depth of 130-fold coverage per sample. As shown in Table I, we identified 59 known polymorphisms in FMR1 in this population, 57 of which were previously catalogued in dbSNP (build 130). This provides evidence of the sensitivity of pooled-template MPS. The only FMR1 SNPs included in dbSNP that we did not detect were those with low or unknown population frequencies; these were likely not present in our patient population.

Table I.

FMR1 Sequence Variants Detected in 963 Developmentally Delayed Males

	Novel variants		Known Polymorphisms
	Unique	Recurrent	Unique	Recurrent
5'US	6	1	0	1
Promoter	3	0	0	0
5'UTR	0	1	0	0
Exon: Nonsynonymous	1	0	0	1
Exon: Synonymous	1	1	0	3
Intron	65	38	3	48
3'UTR	5	7	1	2
3'DS	1	0	0	0

Unique variants were detected in only one sample pool, while recurrent variants were detected in more than one sample pool. US = upstream sequence; UTR = untranslated region; DS = downstream sequence

Two other variants, c.18G>T and c.−418_−417insGGC, although not previously identified in controls, were previously identified in intellectually disabled patients [Gronskov et al., 1998; Mila et al., 2000]. We observed the c.18G>T variant in 13 samples (1.4%) but also in 19 of 1401 controls (1.4%), indicating this is a benign polymorphism of FMR1. Intriguingly, the previously identified c.−418_−417insGGC variant was detected in the current study in 8/963 patients (0.83%), but in 0/700 Caucasian control males (p < 0.05, Fisher's exact test). Consistent with this finding, Mila and colleagues identified the variant in 1/25 fragile X-like males, but in none of 250 control males. It is not obvious how this insertion might influence FMRP expression as the site of insertion in not well conserved and since the site of insertion is itself GGC; thus this variant is a 3 bp duplication. Moreover, Mila et al. saw no difference in steady-state FMRP levels by western analysis nor any differences in gel shift or gene reporter analysis. Since we do not know the ethnicities in our patient samples, it is possible that this significant association results from population substructure.

Additionally, we detected 130 variants in FMR1 which, to our knowledge, have not previously been reported. None were previously catalogued in dbSNP, nor were they detected in the first nine publicly available personal genomes [Ahn et al., 2009; Bentley et al., 2008; Levy et al., 2007; Siva, 2008; Wang et al., 2008; Wheeler et al., 2008]. Among these variants, 63.1% were only detected in one individual, while 36.9% were detected in multiple individuals. The majority of the novel variants were found in the introns of FMR1. However, novel sequence variants were detected in all regions of the gene. A full summary of the sequence variants detected is provided in supporting information Table II, available in the online version of this article.

Although pooled-template MPS is a highly sensitive method [Druley et al., 2009; Ingman and Gyllensten, 2009; Out et al., 2009], subtle imbalances in pool construction can result in the underrepresentation of a given template in the sequence output, which leads to false negatives. To assess our false negative rate, we included a LR-PCR amplicon containing the rare mutation p.I304N in every pool. Because the positive control amplicon was pooled in the same quantity and fashion as all of the patient amplicons, the frequency of successful detection of p.I304N likely reflects the frequency of successful detection of other singleton variants in a pool. At our detection threshold, p.I304N was not detected in 12/51 pools. Based upon this false negative rate of 23.5%, it is possible that we missed up to 40 previously unknown true variants in the population of 963 developmentally delayed males. [Note that we successfully detected 130 previously unknown variants, and assuming we missed 23.5% of the variants that were present, then we missed 0.235*130/(1−0.235) = 39.9.] However, it would be expected that we missed no more than one true promoter variant [0.235*3/0.765 = 0.92] and no more than one true missense variant [0.235*1/0.765 = 0.31], thus suggesting that we detected the majority of FMR1 variants that are most likely to have a functional effect in this population.

Identification of the Novel Missense Variant p.R138Q

In the 963 developmentally delayed males sequenced, we detected only one novel missense variant in FMR1. This variant, c.413G>A (Figure 1A), was identified in a sample from a patient of European descent, but was not detected in 1385 control males of European descent. The c.413G>A variant encodes an arginine-to-glutamine substitution at codon 138. Arginine-138 is believed to be one of the basic residues comprising the nuclear localization signal (NLS) of FMRP [Bardoni et al., 1997; Eberhart et al., 1996] and is highly conserved through Drosophila (Figure 1B). The sole exceptions to this conservation are the pufferfish species Takifugu rubripes and Tetraodon nigroviridis, in which glutamine is used at this position. The R-to-Q missense substitution is predicted to be pathological by PMut (NN output: 0.84) [Ferrer-Costa et al., 2005] and PANTHER PSEC (subPSEC: −4.3) [Thomas and Kejariwal, 2004], but tolerated by SIFT (score: 0.22) [Ng and Henikoff, 2003] and PolyPhen (PSIC Score Δ: 0.11) [Ramensky et al., 2002].

Figure 1. The novel missense variant p.R138Q.

(A) DNA chromatograms of the wild-type and variant alleles. (B) Diagram of FMRP, depicting the functional domains (NLS = nuclear localization signal; KH = K homology domain; NES = nuclear export signal) and R138Q missense substitution. The amino acid sequence alignment corresponds to the NLS. Functional basic residues are shown in red. Asterisks denote identical residues, colons denote conserved substitutions, and periods denote semi-conserved substitutions.

Novel Variants in the FMR1 Promoter

Identification and Preliminary Functional Evaluation

We detected three novel sequence variants in the minimal promoter of FMR1 (Figures 2A, B), each occurring in only one of the 963 sequenced developmentally delayed males. The c.−332G>C variant is located within overlapping binding sites for the Sp1 and AP-2α transcription factors [Lim et al., 2005; Smith et al., 2004]. The c.−293T>C variant is located near transcription start site II, within both an Initiator-like (Inr-like) and a TATA-like sequence [Beilina et al., 2004; Hwu et al., 1993]. The third variant, c.−254A>G, is located within an Inr-like sequence near the primary transcription start site [Beilina et al., 2004; Kumari and Usdin, 2001]. All three of these variant bases are conserved in mammals (Figure 2C).

Figure 2. The novel promoter variants c.−332G>C, c.−293T>C, and c.−254A>G.

(A) Diagram of the minimal promoter and 5’UTR of FMR1. Roman numerals I–III represent the three transcription start sites of FMR1. The GC boxes bind the transcription factor Sp1. (B) DNA chromatograms of the three novel promoter variants. (C) Mammalian conservation of the overlapping AP-2 binding site and GC box, the overlapping Inr-like and TATA-like sequences at transcription start site II, and the Inr-like sequence at transcription start site I.

The c.−332G>C, c.−293T>C, and c.−254A>G variants were not detected in 1308, 1266, and 1304 control males of European descent, respectively. However, the ethnicity of the patients in whom these variants were detected is unknown. Thus, it remains possible that the suggested association of these promoter variants with developmental delay results from population substructure.

Effects on Promoter Activity

To assess the functional significance of the three promoter variants, we introduced each variant individually into the previously described pFMR1-luc plasmid, which uses the human FMR1 promoter to drive expression of firefly luciferase [Smith et al., 2004]. Each resulting plasmid was co-transfected with the control plasmid pRL-TK into HeLa cells. Firefly luciferase activity, measured 48 hours post-transfection, was normalized to Renilla luciferase activity and expressed relative to the wild-type FMR1 promoter activity. As seen in Figure 3, the c.−332G>C variant reduced FMR1 promoter activity to 5.9% of wild-type levels, the c.−293T>C variant reduced FMR1 promoter activity to 29.2% of wild-type levels, and the c.−254A>G variant reduced FMR1 promoter activity to 36.2% of wild-type levels. All of these reductions from wild-type promoter activity were statistically significant (p < 1×10⁻⁷).

Figure 3. Functional effects of the novel promoter variants c.−332G>C, c.−293T>C, and c.−254A>G.

Luciferase activity is depicted as the ratio of firefly luciferase to Renilla luciferase and normalized to the activity of the wild-type FMR1 promoter. The results shown represent six independent transfections. Data are represented as mean ± 1 S.D.

Noncoding Variants in FMR1

To determine if any of the 127 novel noncoding variants in FMR1 are associated with developmental delay, we genotyped large numbers of control males of European descent for all variants occurring at highly conserved bases. We assessed the sequence conservation of the genomic region and nucleotide position for each variant by phastCons and phyloP scores, respectively [Siepel et al., 2005]. Variant positions with phastCons score > 0.8 and phyloP score > 1.5 were defined to be highly conserved. Similarly, we genotyped controls for variants predicted to alter splicing or miRNA binding. The splice prediction program NNSplice [Reese et al., 1997] was used to identify any variants that obliterate known splice sites or introduce novel splice sites. Two variants, c.880+885A>G and c.1472-521C>G, were predicted to introduce novel splice donor sites with high likelihood (> 0.85). The miRNA target prediction software miRanda [Betel et al., 2008; John et al., 2004] suggested that the novel variant c.*746T>C may reduce the binding of miR-548p, miR-891a, and miR-454 to the 3’UTR of FMR1. As seen in Table II, six novel intronic variants and four novel 3’UTR variants of possible functional impact were not identified in a large sample of control Caucasian males. Furthermore, one intronic variant and two 3’UTR variants exhibited a significantly different minor allele frequency in developmentally delayed subjects as compared to controls (Table II). However, because the ethnicity of most of the patients in whom these variants were detected is unknown, it is possible that these associations result from population substructure.

Table II.

Novel FMR1 Variants Only Found in Developmentally Delayed Males

Variant	Location	PhastCons	PhyloP	Patient frequency	Control frequency	P (Fisher's Exact Test)
Missense variant
c.413G>A	Exon 5 (p.R138Q)	0.99	2.94	1/963 (0.1%)	0/1385 (<0.07%)	N.S.
Promoter variants
c.−332G>C	GC Box/AP-2	0.72	0.65	1/963 (0.1%)	0/1308 (<0.08%)	N.S.
c.−293T>C	Inr-like/Tx. Start II/TATA-like	0.94	1.95	1/963 (0.1%)	0/1266 (<0.08%)	N.S.
c.−254A>G	Inr-like/Tx. Start I	1	2.10	1/963 (0.1%)	0/1304 (<0.08%)	N.S.
Other noncoding variants
c.105−8A>C	Intron 2	1	2.48	6/963 (0.6%)	0/1262 (<0.08%)	0.007
c.630+438A>C	Intron 7	1	2.55	1/963 (0.1%)	0/1263 (<0.08%)	N.S.
c.631−840G>A	Intron 7	1	1.76	1/963 (0.1%)	0/1239 (<0.08%)	N.S.
c.880+885A>G	Intron 9a	0	−0.73	1/963 (0.1%)	0/1084 (<0.09%)	N.S.
c.990+4T>C	Intron 10	1	2.46	1/963 (0.1%)	0/1248 (<0.08%)	N.S.
c.1472−521C>G	Intron 14a	0	0.40	1/963 (0.1%)	0/1254 (<0.08%)	N.S.
c.*23T>C	3’UTR	0.99	2.40	1/963 (0.1%)	0/900 (<0.11%)	N.S.
c.*746T>C	3’UTRb	1	2.22	6/963 (0.6%)	0/1260 (<0.08%)	0.007
c.*1867G>A	3’UTR	1	1.54	12/963 (1.2%)	0/951 (<0.11%)	0.0005
c.*2035C>T	3’UTR	1	2.47	3/963 (0.3%)	0/1270 (<0.08%)	N.S.

^adenotes predicted splice site

^bdenotes predicted miRNA binding site

N.S. indicates p value is not significant.

DISCUSSION

Through the use of pooled-template MPS, we have identified 130 novel sequence variants in FMR1 in a deidentified population of 963 developmentally delayed males. While many of these are likely novel polymorphisms, several show evidence of functional effects and association with developmental delay. These findings have important implications for the diagnosis of developmental delay, the structure and function of FMR1 and FMRP, and the utility of pooled-template MPS for novel variant detection in a disease gene.

One of the most noteworthy findings in the sequenced clinical population was the identification of the novel missense change p.R138Q. If proven to be pathogenic, this variant would be only the second missense mutation to be identified in FMR1. However, the functional significance of p.R138Q remains unclear. The arginine-138 residue is largely conserved through Drosophila, but the pufferfish species Takifugu rubripes and Tetraodon nigroviridis both use a glutamine at this position. Furthermore, algorithms for the prediction of the effects of amino acid substitution give conflicting results for this change, as PMut and PANTHER PSEC predict that it is pathological, while SIFT and PolyPhen predict that it is benign. Finally, our initial studies of the effects of the p.R138Q substitution on NLS function have not revealed any change in the intracellular localization of FMRP (data not shown). However, the function of the FMRP NLS has historically been difficult to assess in its endogenous context [Bardoni et al., 1997; Eberhart et al., 1996; Fridell et al., 1996; Kim et al., 2009; Sittler et al., 1996]. Moreover, arginine-138 also lies within the N-terminus of FMRP, which has been implicated in RNA binding, domain dimerization and protein-protein interactions [Adinolfi et al., 2003]. Therefore, more thorough analyses of FMRP function will need to be performed to determine if the p.R138Q missense change has a deleterious effect.

Another important finding in this clinical population was the identification of three novel variants in the FMR1 promoter. To our knowledge, these are the first three variants to be identified in defined elements of the FMR1 promoter. Furthermore, all three independently reduced the activity of the FMR1 promoter in a luciferase reporter assay. While the ultimate measure of the effects of FMR1 promoter variants is the associated level of FMRP expression in patient tissues, this reporter assay data suggests that the three identified variants are likely to have a functional effect.

While the functional effects of the novel FMR1 variants identified in this study remain unclear, it is likely that several do indeed associate with developmental delay, such as the missense variant, the three promoter variants and the two splice site mutations. If these variants are all verified to be pathogenic, the frequency of FMR1 sequence variants causing developmental delay in our study population would be 0.8%, when accounting for our false negative rate. In comparison, the frequency of FMR1 repeat expansions, the most common inherited cause of developmental delay, is 2 – 3% in the same population [Biancalana et al., 2004; Hecimovic et al., 2002; Major et al., 2003; Pandey et al., 2002; Patsalis et al., 1999; Rauch et al., 2006]. Therefore, if the novel FMR1 sequence variants are indeed functional, such variants should be viewed as a significant contributor to the heterogeneous diagnosis of developmental delay, and FMR1 sequencing should be considered for diagnostic testing.

To our knowledge, this study represents the first application of pooled-template MPS for the identification of novel sequence variants in a clinical population. While several proof-of-principle papers had demonstrated that this approach was adequately sensitive and unbiased for the detection of rare novel variants [Druley et al., 2009; Ingman and Gyllensten, 2009; Out et al., 2009], the current study validates pooled-template MPS as a useful application of next-generation sequencing technologies for targeted studies of a single gene. With the continual market-driven increase in MPS capacity, pooled-template approaches will become even more important for the efficient use of MPS on single genes and other small genomic regions.

In summary, we have identified 130 novel sequence variants in FMR1 in a population of 963 developmentally delayed males. Among these variants are the novel missense change p.R138Q, which alters a conserved residue in the FMRP NLS, the first three sequence variants to be identified in the FMR1 promoter, all of which reduce transcriptional activity, and several noncoding variants of possible functional effect, including two splice site mutations. Taken together, these results provide avenues for structure-function studies of FMRP and FMR1, and suggest that pathogenic sequence variants in FMR1 may represent a significant cause of developmental delay.