Background

Intellectual disability (ID) has a prevalence of 2.3%, making it a prime socio-economical problem [1]. ID is a very complex and heterogeneous disorder that can be caused by genetic factors, environmental factors or a combination of both. As a result, the etiology remains unknown in ~ 30% of cases. X chromosome-linked ID (XLID) has served as a model for the genetics underlying ID, in part because it is approximately 30% more prevalent in males than in females, suggesting that important causative genetic loci are located on the X-chromosome [2].

In the last 15 years, candidate gene mutation screening [3, 4], hybridization-based array screens [5, 6] including high resolution array-CGH [7,8,9,10], and massively-parallel sequencing (MPS) screens [11,12,13,14] led to the identification of many genes associated with ID. It became clear that genetic causes of ID are highly heterogeneous, as the reported mutations explain only a small number of ID families [2]. For example, a Sanger sequencing-based screen of all exons on the X-chromosome in 208 XLID families only revealed causal mutations in 25% of families [15]. Later, MPS allowed for much higher throughput identification of disease-associated mutations, deletions and duplications. However, this groundbreaking method could not resolve more than 20% of the remaining cases, as illustrated by an X chromosome-specific exome MPS screen in 405 XLID families [14]. Thus, despite the large number of studies and significant technological progress, the etiology of ID remains unsolved in at least 40% of XLID families.

These figures strongly suggest that the missing mutations should be searched for in the non-coding regions of the X chromosome, or in regions that currently escape analysis. An intergenic variant identified by targeted MPS on the complete linkage interval of a large XLID family has been associated with enhanced expression of HCFC1 in a family with nonsyndromic ID [16] demonstrating that indeed regulatory mutations contribute to ID.

Another group of regions that have been neglected are the tandem repeats. Tandem repeats largely escape mutation analysis because their larger sizes are not covered by short read sequencing technologies. In addition, the sequence reads often fail to be mapped back to the reference genome due to their repetitive nature.

Tandem repeats are DNA sequences consisting of multiple (almost) identical copies of a short (typically 1–50 nt) unit sequence that is repeated in a head-to-tail manner. Such repeats are arbitrarily divided into microsatellites and minisatellites, depending on repeat unit length. Tandem repeats are abundantly present in the human genome including the coding sequences and promoters [17], but thorough variation analysis is lacking due to technical challenges. The mutation rate of repeat regions is typically at least one order of magnitude higher than those in non-repetitive DNA, and as a consequence, variation in repeat length in coding or regulatory regions has a high probability to influence the function or expression of genes [18,19,20]. Despite all those features, repeats are often overlooked as prime targets for disease-related mutations. Moreover, the most commonly used MPS instruments from Illumina and Thermofisher provide average read lengths of 150–200 nt, which is too short to read through most repeats. Even paired-end sequencing does not increase read length in this case, because in order to obtain reliable sequences both reads of a pair should span a full tandem repeat with flanks. Therefore, long-read sequencing technologies are more suitable to study repeat variations. Recently, single molecule real-time sequencing has been introduced to study tandem repeats through long-range PCR amplicons spanning a single repeat of interest. Despite the significant error rate of this newest MPS technology, an accurate consensus tandem repeat can be reconstructed via a local de novo assembly [21,22,23]. In addition, this long read MPS platform is especially valuable to study expansions because of the circular nature of the reads. Multiple passes through the read sequence allow to generate a consensus sequence which facilitates discrimination between sequencing errors and PCR artefacts (“stutters”), that are commonly obscuring tandem repeat analyses.

To test the hypothesis that tandem repeat expansions are a hidden cause of XLID, we set out to selectively target repetitive sequences on the X chromosome and characterize them by single molecule sequencing using the PacBio platform. Specifically, we targeted more than 1800 tandem repeats on the X-chromosome in 3 families with idiopathic X-linked intellectual disability in whom previous methods did not detect any potential genetic cause. Our analysis identified one candidate causal repeat expansion in one family. Gene expression analysis showed down-regulation of the neighboring MIR222 and, indirectly, FMR1 and NEFH overexpression. This study suggests that tandem repeat mutations may be a hidden cause of XLID and potentially of other diseases as well.

Methods

Selection of tandem repeats and capture probe design

A list of tandem repeats on the X chromosome was obtained from the hg19 human reference genome (UCSC [24]) with the ETANDEM tool (EMBOSS package [25]) and was complemented by repeats from several other sources [19, 26,27,28,29] as described by Duitama et al. [30] bringing the total number of target repeats to 43,106. They included repetitive loci with a unit size 1–50 bp, copy numbers of 2–809, full length 22–4048 bp, and GC content including the extreme values of 0 and 100%.

All tandem repeats were annotated according to their position relative to a gene and divided into two groups: presumably functional (i.e. located in coding and regulatory regions) and likely non-functional as previously described [30]. The variability potential of these repeats was predicted by the SERV score [17] based on the following characteristics: unit length, copy number in the reference genome, and intralocus homology. SERV values 1–3 correspond to the highly variable tandem repeats that are usually used for genotyping.

We aimed to sequence around 2000 repeat loci on the X chromosome. We reasoned that a total of at least 500 flanking base pairs should be kept in a 1 kb consensus sequence for probe annealing sites and repeat variation, hence the maximal length threshold of 500 bp for tandem repeats. We also kept a number of intronic and intergenic tandem repeats on the X chromosome to narrow down the linkage intervals with inheritance patterns of more than 2 haplotype specific alleles per repetitive locus.

First, tandem repeats were pre-selected based on their characteristics, presumed functionality and/ or predicted degree of polymorphism (Table 1): 1) 353 tandem repeats in coding regions with the total size up to 500 bp; 2) 174 tandem repeats with SERV score ≥ 1 and total length ≤ 500 bp, which are located in regulatory sites (CpG islands, transcription factor binding sites, regions upstream or downstream from a gene, including micro-RNA genes); 3) 390 regulatory repeats of any size with SERV scores 0.4–1; 4) 68 regulatory tandem repeats within 1000 bp distance from the 112 genes that are known to be involved in XLID (Greenwood Genetic Center, [31]) and not yet included for probe design; 5) 1000 non-functional tandem repeats evenly distributed over the X chromosome with SERV score > 0.8, at least 15 copies in the reference genome, unit length > 1 bp, total length ≤ 500 bp, GC content 30–70%, and at least 3 different capture probes available.

Table 1 Initial selection of possible targets for subsequent probe design, and final selection of tandem repeats for capture and sequencing
Full size table

Then probe design was performed as previously described [30] and included left and right flanking probes, spanning probes centered on tandem repeats, and double probes containing both flanks of a target. In total, 9969 probes generated for 4503 tandem repeats (715 functional and 3788 non-functional repeats) matched uniqueness criterion, i.e. were not predicted to hybridize aspecifically (Table S1). Subsequently, for those 270 repeats where it was not possible to generate unique capture probes, another round of probe design was performed with modified settings: flanking probes and two parts of a double probe were allowed to shift outwards from a repeat by up to 500 bp, and the most proximal available unique probe was chosen for the application (see Additional file 1). This approach allowed us to generate 118 new probes and add 66 tandem repeats to our final repeat selection. To increase the total number of available probes from 1 to 3 or 4, the same strategy was applied to the 7 ‘XLID repeats’ and was successful for 6 of them.

Finally, we examined the distribution of 204 functional repeats which could not be targeted after the above described steps. These repeats were tested for homology to other genomic loci using the Bowtie alignment tool [32] with the following settings: -e 200 -n 3 -y -l 15 -k 10 (or -k 30, depending on the expected number of locus-specific alignments). When a repeat together with its capture probes revealed a local specificity, but no homology to unrelated regions (e.g. showed homology only with other sequences within a cluster of a locally duplicated region), this repeat was kept in the final selection.

Following this approach, two distinguishable clusters of untargeted repeats were found at both ends of the X chromosome. Eighteen tandem repeats on the p arm fall in the pseudoautosomal region 1 (PAR1) with a 100% homology with the corresponding region on the Y chromosome. For 14 out of 18 repeats in PAR1 there was no high homology detected other than with the Y chromosome, and 1 to 3 locus-specific probes were available per repeat. Similarly, we reanalyzed probes for 13 repeats in PAR2 on the q arm, but no locus-specific probes were found due to a high similarity of these sequences to other regions on the chromosomes X, Y or autosomes.

Two other clusters of so far untargeted tandem repeats were detected on the q arm of the X chromosome. A total of 23 presumably regulatory repeats, located in CpG islands, build up a cluster at Xq23. They belong to a 53 kb region (chrX:114,952,840-115,006,118) with a complex structure, which reveals a number of local duplications. For this reason, all the fragments of the original probes were realigned to the reference genome to scrutinize more top alignments. As a result, 3 probes became available for each repeat within the cluster, which cross-align to repeats within this region, but not to other positions in the genome. All 23 repeats were added to the final selection of tandem repeats.

A cluster with 15 untargeted coding repeats represents members of the cancer/testis gene family 47, also known as CT47. It is comprised of 13 nearly identical loci clustered in a 118 kb region at Xq24 (chrX:120,002,680-120,120,440). Following the same approach, we enriched our final selection of tandem repeats with 11 loci, each having 3 cluster-specific probes.

This approach resulted in 1837 tandem repeats. They contain 837 (83%) presumably functional repeats on the X chromosome (Table 1) including repeats implicated in spinal bulbar muscular atrophy, fragile X and Fragile X E syndromes.

All probes obtained for tandem repeat capture were replicated based on the available types of probes and their GC content, as described by Duitama et al. [30] (Additional file 3: Figure S1), except that each probe with a GC content below 40% or above 70% was multiplied by 4. The resulting probe design contains 21,386 probes with a total capture size of 0.49 Mb and is given in the Additional file 2. It includes all the necessary information for ordering the SureSelect probes (Sheet 1), and a full description of the available probes for each target including their genomic positions, replication numbers and efficiency of a particular probe combination (Sheet 2). Each tandem repeat in the final selection is targeted by 4 to 20 probes of 1 to 4 types (Additional file 3:Figure S1).

Selection of XLID cases

Initially, we selected familial cases of ID recorded by the University Hospital of Leuven where consecutive studies, such as karyotyping, individual screening of known ID genes [4], X-array-CGH [8], or X-exome sequencing [14] did not reveal any clear pathogenic variants. The protocol was approved by the appropriate Institutional Review Board of the University Hospital of Leuven, Belgium, and informed consent was obtained from the parents of the affected patients and their healthy family members. In these families, we first checked for autosomal linkage with Merlin and for X-chromosomal segregation with Minx with default parameters (MERLIN package) using available STR profiles of the family members. For several families, SNP-arrays were additionally performed to confirm or disprove X-linkage. The HumanCytoSNP-12v2.1 BeadChips (Illumina) were prepared according to manufacturer’s instructions, and their results were analyzed with Merlin. In two families, idiopathic ID was confirmed to be X-linked: L020 (or 5X, or MRX51) [33], L061 (or 37SX). For one more family, L084 (or 78X), a suggestive linkage interval on the X was obtained. In each family, an EBV-PBL cell line from the proband was available. Per family, DNA from an affected and an unaffected male were chosen for sequencing of the X-chromosomal tandem repeats: 5X20 and 5X15; 78X28 and 78X19; L061_Y and L061_S (Fig. 1). Genomic DNA samples were provided by the DNA biobank of the University Hospital of Leuven.

Fig. 1
figure 1

Pedigrees of the selected families with idiopathic XLID. Probands are marked with a black arrow head. Grey filling indicates ID phenotype. Blue arrows point out individuals selected for targeted capture and sequencing of tandem repeats. ‘DNA’ stands for available genetic material. Ideogram of the X-chromosome with a zoom into the linkage interval is depicted for each family. Red and blue boxes indicate initial and refined linkage intervals, correspondingly

Full size image

The affected individuals of the L020 family all present with non-syndromal ID ranging from mild to moderate as described by Claes et al. [33]. It is important to note that the 5X9 member of the L020 family was treated as affected, and 5X16 as unaffected. The affected individuals of the L061 family presented with non-syndromal moderate ID. Two individuals also presented with epilepsy. They were non-dysmorphic and had normal neurological examination except for individual L061_MJ, who experienced a cerebrovascular accident at the age of 47 years. Family L084 includes 3 affected males with mild to moderate non-syndromal ID. The youngest individual also presented with spastic paraplegia starting in young adulthood.

Library preparation and sequencing

DNA samples were sonicated in a Focused-ultrasonicator (Covaris) into fragments with an average size of 800–1000 bp. Library preparation was done following the manufacturer’s instructions (SureSelect Target Enrichment System for Roche 454 GS FLX and GS Junior Sequencing Platforms). SureSelect libraries were directed for SMRTbell library preparation (2 kb Template Preparation Procedure: DNA damage repair till first purification of SMRTbell templates using 0.6X AMPure PB beads) and sequenced on the Pacific Biosciences RS II machine with the P5-C3 or P6-C4 reagent kit. Each library was run on three SMRT cells. Sequencing was performed by the Genomics Core of the University Hospital of Leuven. Fastq files were obtained with a minimal requirement of 6 subreads per read of insert.

Analysis of the sequencing data

We developed a bioinformatical pipeline for retrieving information on the targeted tandem repeats from the sequencing data and their subsequent genotyping (Additional file 3: Figure S2). First, adaptor sequences were removed by trimming 32 nucleotides at both sides of a read (see Additional file 4). Alignment of the trimmed reads to the reference genome was performed using Burrows-Wheeler Aligner and Smith-Waterman alignment algorithm (BWA SW) [34]. PCR duplicates were removed with the MarkDuplicates tool (Picard tools package [35]). BEDtools [36] were used to obtain the aligned reads, which were mapped within 1000 bases from the target sites, intersected the targeted tandem repeats, or spanned full tandem repeats. A custom script (see Additional files 5 and 6) was used to estimate the number of reads spanning tandem repeats together with their 20 nt-long flanking regions, considering soft-clipped reads with several partial alignments to the flanks.

Reads of insert were further analyzed with the TSSV tool [37] with a -d option, but before that an additional step was introduced to increase specificity and speed up the analysis. For each tandem repeat a custom script (see Additional file 7) filtered those reads that were mapped within 300 bases distance from the targeted region, and created separate input files for the TSSV tool: a fasta file with filtered reads and a corresponding TSSV library. TSSV libraries included names of tandem repeats, left and right flanking sequences which were fetched from the reference genome on the Galaxy platform, and repetitive units with expected ranges of copy number (lowest and highest copy numbers set equal to the reference copy number in this case). For tandem repeats included in clusters at Xq23 and Xq24, unique cluster representatives were searched for in fasta files containing all reads mapped to the X chromosome.

TSSV output was processed with a custom script (see Additional file 8), which analyzes allele lengths, calculates copy numbers, implements genotyping principles described in Duitama et al. [30], and determines which copy numbers correspond to partial reads where only one of the flanking regions was found by the TSSV.

Validation of sequencing derived genotypes and their inheritance in a family

The genotypes obtained by massively parallel sequencing and data analysis were validated by fragment analysis in family members and if necessary in up to 100 controls. Control sampling comprised unaffected members of other families and patients admitted to the hospital with other, non-neurological diseases. PCR was performed in two rounds consisting of 15 and 20 cycles respectively. The first round was performed on 50 ng genomic DNA in a 25 μl mixture using Taq DNA Polymerase (Invitrogen) and 0.2 μM unlabeled specific primers designed with Primer3 [38]. All forward primers contained a 21 bp extension of the M13 sequence at the 5′-end. Of the first PCR product, 2 μl were used as a template for the second reaction containing a FAM-labelled M13 primer and a locus-specific reverse primer. Final products were run on an ABI3500xL Genetic Analyzer (Thermofisher) with the GeneScan 500 ROX Size Standard (Thermofisher). Fragment lengths were analyzed with the GeneMapper v4.1 software (Thermofisher). For the loci possibly expanded and located within a linkage interval, we performed Sanger sequencing on the first PCR products with the respective unlabeled primers and the BigDye v3.1 cycle sequencing kit. Products were analyzed on the ABI3500xL, and resulting sequences were aligned with the BioEdit v7.1 software (Ibis Biosciences) to count the exact number of units in a tandem repeat. PCR primers used in this study are given in Additional file 9.

Quantitative PCR

For MIR222 expression analysis RNA was extracted from EBV-PBL cell lines using mirVana miRNA Isolation Kit (Thermofisher) following small RNA enrichment procedure. RT-PCR for small RNA was performed using the TaqMan MicroRNA Reverse Transcription Kit (Thermofisher). Expression level was measured by qPCR using miRNA-specific TaqMan Small RNA Assays (Thermofisher) with 2 endogenous control miRNAs: hsa-let-7f-5p, MIR98. This was done in two independent RT-qPCR experiments. For expression analysis of other genes, total RNA was extracted from the non-confluent cell cultures using RNeasy Mini Kit (Qiagen), and cDNA was synthesized with Superscript Reverse Transcriptase and random primers (Thermofisher). Expression levels were measured 2–3 times by qPCR using SYBR Green on the LC480 apparatus (Roche) with 3 endogenous control genes: GUSB, HPRT1, PORCN. Primers used for qPCR in this study are given in Table S2, Additional file 3.

Total RNA sequencing

Total RNA was extracted from the non-confluent EBV-PBL cell cultures using RNeasy Mini Kit (Qiagen), and cDNA was synthesized with Superscript Reverse Transcriptase and random primers (Thermofisher). Total RNA single-end Illumina sequencing generated 50 bp reads, which were mapped to the hg19 human reference genome using Tophat version 2.0.6 [39]. BAM files were handled with SAMtools version 0.1.18 [40]. Quantification of reads per gene and differential expression analysis was performed with Cufflinks version 2.0.2 [41]. Differentially expressed genes were first pre-selected with the false discovery rate of 5%. To filter the most deregulated genes, they were ranked according to the ratio (R) of the difference between the patient expression value (P) and the closest control value (C) to expression range within controls: R = |P-C|/(Cmax-Cmin). Loci with statistically significant difference of expression (p < 0.001) in patient comparing to three controls were subjected to pathway enrichment analysis using IPA (Qiagen).

Results

Tandem repeat capture and sequencing in XLID families

We analyzed three families with idiopathic XLID for X-chromosomal tandem repeat variation. In the past, in families L020, L061 and L084 neither full coverage X chromosomal microarrays, nor X-exome sequencing [4, 14] revealed any pathogenic variants. In those families, linkage analysis results in LOD scores of respectively 2.406, 2.23 and 0.932, suggesting X-linkage (see Materials and Methods). The family trees, affected and unaffected family members selected for targeted resequencing, as well as the linkage intervals are shown in Fig. 1 and Table 2. Taken that the average prevalence of intellectual disability in Western countries is 2%, and the majority of cases are sporadic, we could use this frequency to estimate the probability of at least two or three causal factors (de novo mutations and/or environmental factors) co-occurring within one family: 4 × 10− 4–8 × 10− 6. Therefore, we consider a combination of several different etiologies within a single family to be unlikely, especially when the detected linkage interval on the X chromosome is significant (LOD > 2).

Table 2 Linkage analysis confirmed X-linkage in 3 familial cases of intellectual disability
Full size table

For each XLID family an affected and an unaffected male was selected for targeted capture and long-read single-molecule sequencing of the tandem repeats on the X-chromosome (Fig. 1). On average, more than 135,000 reads are obtained per sample, of which 28% map within 1 kb from our targets, and almost 20% of the reads are useful for genotyping, as they span a target together with both flanks (Table 3). For 8.68% of the targets we could not obtain any reads, and for 1.61% of the loci we only obtained sequences that do not span the full repeat length. We obtained full sequences for 88 to 93% of the targets in the sequenced individuals with an average coverage of 10 to 23 consensus reads per locus. All obtained genotypes are given in Additional file 10.

Table 3 Sequencing yield demonstrated high recovery rate for tandem repeats in the assay
Full size table

For tandem repeats with moderate GC content (including non-functional repeats) capture and sequencing success reaches 99.0–99.6% ( Additional file 3: Figure S3; Table 3), although for functional targets with high GC content it is ~ 70% lower. We obtained full sequences for almost all GC-poor loci (< 40% GC), while for GC-rich regions (> 70% GC) we observe a decrease in recovery rate to ~ 30% ( Additional file 3: Figure S3) despite the equal probe quadruplication for both groups. Efficiency of a corresponding probe combination is given for each targeted tandem repeat in the Additional file 2 (Sheet 2) together with the influence of the GC content, number of available probe types, total number of the used probes, and full length of a repetitive locus.

Expansions in large tandem repeats are even detected with partial reads

Because highly expanded alleles could exceed the size of fragments enriched in the libraries, we searched for loci that exclusively yielded reads that did not span the complete repeat. An example of such a large repeat expansion detected by this assay is an intronic repeat in the CLCN5 gene, represented by 15 copies of 26 bp in the reference genome. We estimated copy numbers for all partial alleles in the sequenced individuals, and the longest ones correspond to at least 24–35 copies (Additional file 10). Thus, the full repeat length of the longest allele is estimated to be more than 900 bp, which is considerably longer than the reference repeat length of 390 bp. This expanded repeat is present in both affected and unaffected individuals in all three families, and since it is located outside the linkage interval in 2 families, the repeat is likely not a causal variant for XLID.

XLID25 expansion in L020 is potentially linked to the phenotype

Apart from individual repeats, we also included clustered tandem repeats in our analysis. The following strategy was used in each family to narrow down the list of candidate variants. Loci within the linkage intervals, which provided a different unit copy number in the affected versus the unaffected male were then genotyped in other family members by fragment analysis. Moreover, linkage analysis was repeated using these polymorphic repeats as additional segregation markers, allowing to refine the linkage intervals for L020 and L084 families (Table 2; Additional files 11, 12 and 13). Since we expected that mutations in repeat copy number occurred independently in these three XLID families, we checked for unique copy numbers in the patients, which segregated with the phenotype in the family and were absent in the other families. The existence of such variants in the general population was further screened for in a control sampling of up to 100 males. Only copy number variants that were not found in controls were then considered as ID candidate loci.

For the L084 family, 18 tandem repeats were targeted by our assay in the linkage interval, of which 17 are successfully sequenced. Of those, only one intergenic tandem repeat at chrX:145,340,826-145,341,025 (XLID32) exhibits a copy number difference between the affected (78X19) and unaffected (78X28) individuals. However, this allele is also found in control samples and thus considered to be a benign variant.

For the L061 family, we obtained 315 presumably functional loci within the linkage interval with a coverage of at least 5X. Forty five of these repeats were found to be polymorphic within the normal variation range obtained in other XLID families; fragment analysis demonstrated that 3 variants (XLID75, XLID77, XLID79) were false positives; 1 other variant (XLID76) was in the normal size range when compared to additional controls (Table 4); and for 4 repetitive loci (XLID73, XLID74, XLID78, XLID80) the apparent unique copy number detected only in the affected males of the L061 family were also found in the unaffected control population (Table 4). Finally, the tandem repeat XLID72, which displays a shorter polyglutamine tract in the proband compared to his unaffected relative and other families, is located in the first exon of the well characterized AR gene. Since the array length of the tandem repeat chrX:66,765,149-66,765,262 is within the normal range, it is considered to be not related to ID.

Table 4 Polymorphic tandem repeats in family L061 that were also detected in a control sampling
Full size table

For the L020 family, 44 targeted tandem repeats are present in the linkage region, of which 43 are successfully sequenced. Of these, 5 repeats located in regulatory regions exhibit a copy number difference between the affected (5X20) and unaffected (5X15) family members. However, following fragment analysis, in 4 loci (XLID2, XLID20, XLID22, XLID27) alleles detected in 5X20 and his affected relatives were also detected in 20 control samples (Table 5).

Table 5 Repeats in L020 family confirmed potential phenotypical relevance for XLID25 allele upon a control screening
Full size table

The tandem repeat at chrX:45,606,270-45,606,355 (XLID25), which reveals a unique allele in the affected family members segregating with the phenotype, is located 65 bp downstream of the microRNA gene MIR222. It is a complex repeat consisting of two consecutive (CT)n and (GT)n sub-repeats separated by three thymidines (Fig. 2). After screening the entire control population of 100 healthy males by fragment analysis only two exhibited the same total length as in the L020 proband. We performed an additional screening by fragment analysis to differentiate (CT)n and (GT)n sub-repeats in those controls (Table S3). For that, we used a nested PCR primer annealing to the boundary that separates the sub-repeats, which gives an approximate estimation of CT copy number. To define the exact length of each individual repeat, amplicons of the 5X17 proband along with that of 23 control males were Sanger sequenced (Fig. 2). The sequenced controls included the samples with the estimated CT copy number equal to or one copy shorter than that of the proband, and the samples with the allele of the same total length as the proband. The latter control samples demonstrated different copy numbers for both (CT)n and (GT)n sub-repeats (17 CT and 37 GT copies or 20 CT and 34 GT copies), compared to the proband who exhibited an allele with the highest CT copy number (21), while the GT unit number (33) was in the normal range of 24–37 copies, observed in the controls. Notably, none of the sequenced samples revealed the (CT)n sub-repeat longer than 20 copies.

Fig. 2
figure 2

ChrX:45,606,270-45,606,355 tandem repeat (XLID25) is located downstream of MIR222. Sanger sequencing revealed the longest (CT)n sub-repeat in L020 XLID patient (5X17, bottom line), compared to 23 unaffected males. Co18 is the unaffected sibling (5X19) from the same L020 family.

Full size image

Upstream MIR222 gene reveals decreased expression, and its targets FMR1, NEFH are up-regulated in the proband

To investigate the possible effect of the XLID25 CT-sub-repeat expansion on MIR222 expression, we performed a miRNA specific TaqMan assay on the enriched small-RNA samples extracted from EBV-PBL cell lines of the proband and three male controls. As no cell lines were available for the other family members, their MIR222 expression has not been tested. The results show that MIR222 expression is decreased at least 5-fold in the patient (Fig. 3a) compared to the control samples that correspond to ‘Co14’ (genotype 17 CT, 32GT), ‘Co15’ (17 CT, 32GT) and ‘Co17’ (17 CT, 35GT) males in the sequence alignment in Fig. 2. Additionally, we tested the expression of one of the MIR222 downstream targets, FMR1, known to be involved in intellectual disability. Interestingly, the FMR1 mRNA levels are elevated by 30% in the XLID patient compared to the 3 controls, who show highly similar levels (Fig. 3b).

Fig. 3
figure 3

The expression level of MIR222 (a) is decreased while one of its targets, FMR1 (b), is up-regulated in the patient (‘Pat’, red) opposed to 3 controls (‘Co1’, ‘Co2’, ‘Co3’, shades of green). Error bars show standard deviation of the normalized expression in 2 and 3 experiments, respectively. ‘Co1’, ‘Co2’, ‘Co3’ samples correspond to ‘Co14’ (17 CT, 32GT), ‘Co15’ (17 CT, 32GT), ‘Co17’ (17 CT, 35GT) males in the sequence alignment in Fig. 2

Full size image

To reveal additional deregulated genes that might be affected by the altered MIR222 expression, we performed RNA sequencing again on RNA extracted from EBV-PBLs of the patient and 3 controls, same as in the previous experiment. We obtained 46 loci that are significantly and consistently up- or down-regulated in the patient, however none of them is within the linkage interval of the family. Based on Ingenuity Pathway Analysis most of these genes are involved in anatomical structure morphogenesis, cellular component movement, locomotion and localization of the cell. R ratio ≥ 0.9 for expression deregulation in patient is observed in 35 genes, of which 31 are known to be expressed in brain. Of these, 21 genes (68%) are predicted targets of MIR222, while it is expected to regulate half as many (30%) in an entire pool of the brain-expressed genes. Nine genes were selected to confirm the RNAseq data by qPCR. Upon increasing the number of controls to 7, the altered expression in the patient remained apparent in 2 of the 9 selected genes: ARMCX2 and NEFH (Fig. 4), which are predicted targets of MIR222 (microRNA.org). Notably, the NEFH gene encoding the heavy neurofilament protein reveals a 52-fold increased expression level in the XLID patient in relation to the 7 controls. As for the genes within the linkage interval other than MIR222, for 7 of them (MAOA, DUSP21, PPP1R2P9, MAOB, NDP, EFHC2 and MIR221) we did not obtain any RNA sequencing data while other 8 (KDM6A, ZNF673, FUNDC1, CXorf36, ZNF674, KRBOX4, ZNF674-AS1 and CHST7) reveal expression differences that are not statistically significant (p-values 0.29–0.99).

Fig. 4
figure 4

Up-regulated expression of NEFH (a) and down-regulation of ARMCX2 mRNA (b) in the affected proband (‘Pat’, red) and 7 controls (‘C1–7’, shades of green). Error bars show standard deviation of the normalized expression in 3 experiments

Full size image

Discussion

Despite multiple large-scale array-CGH and exome and genome sequencing analyses, the genetic variation underlying X-linked intellectual disabilities remains elusive for a large number of families. We hypothesize that tandem repeat expansions have escaped detection mainly due to short-read sequencing technologies. Here, we developed an extensive screen for X linked tandem repeat expansions using a long read MPS approach. We captured and sequenced more than 1800 tandem repeats in three families with idiopathic XLID and demonstrate the feasibility of single molecule sequencing to accurately detect and size tandem repeats and tandem repeat variability. Moreover, in one family a tandem repeat expansion segregating with ID seems to affect the expression of the nearby MIR222 gene, previously implicated in ID [42] and affecting expression levels of genes associated with ID [43,44,45,46].

Multiple studies have been reported on large-scale analysis of tandem repeat polymorphism in several species, with various enrichment strategies, and utilizing different massively-parallel sequencing technologies. Information on thousands of repeats was previously retrieved from whole-genome or targeted sequencing data for 8 individual genomes [47], 550 and later 1009 individuals from the 1000 Genomes Project [48, 49], or 34 human gastric cancer cell lines and tissue samples [50]. An extensive characterization of > 390 thousand loci in more than 150 Drosophila strains by Fondon et al. [51] provided an important framework for further association studies in this species. However, Illumina read length limits target loci to microsatellites shorter than ~ 90 bp or less. On the contrary, the use of the 454 platform, generating longer reads, allowed to discover hundreds to thousands of new microsatellites in 14 species among animal, plant and fungi taxa [52], or selectively target human tandem repeats with a broader range of characteristics [30] with a total length up to ~ 300 bp. More recently, long read single molecule sequencing has been introduced to study tandem repeats, though limited to only a few loci so far [21,22,23]. Here, we combined the benefits of broad-range targeted sequencing with the large read size, which permitted analysis of virtually all coding and regulatory repeats on the X chromosome. Overall, 83% of all likely functional repeats could be targeted with unique probes, and on average more than 90% of targets were captured and sequenced.

Capture-based enrichment has previously been applied in studies of tandem repeat variation. The first approach [52] used 8 spanning probes designed to bind all microsatellites in the genome with probe-matching motifs. Later studies focused on a more selective and specific capture approach using probes complementary to the unique flanking regions of such repeats, which boosted the capture efficiency. Guilmatre et al. [47] utilized solely flanking probes and increased the on-target sequences to 39%. The design of longer flanking probes, as well as the addition of spanning and special double-flank probes that we previously described even raised the capture efficiency to 62% [30]. Our current probe design, based on the latter approach, however, resulted in a lower on-target rate of 28%. This apparent capture efficiency drop is likely due to a more than 5-time decrease in the number of targets compared to Duitama et al. [30]. According to the manufacturer (personal communication), this is a frequently observed effect of excess baits that are forced to bind aspecifically when there are not enough target molecules for a proper hybridization. Hence, it was suggested to test half of the recommended bait amount for the future hybridizations. Another difference in the probe design that probably has a minor effect is the ratio of the probes with extreme GC content. A higher percentage of probes with low GC content is likely to hybridize non-specifically to AT-rich Alu elements that are abundantly present in the genome, thereby increasing the off-target count. However, despite the large off-target rate our assay demonstrated high recovery rate of the targeted regions. It is clear though that optimizing the baits stoichiometry and further improvements to the probe design have the potential to boost the capture efficiency even more.

Interestingly, tandem repeat expansions can be detected even if the expansion exceeds the targeted fragment length. As an example, we detected expansion of more than 900 bp for an intronic tandem repeat in CLCN5 that is only 390 bp long in the reference genome. This expansion would not be detected with short-read technologies. We found such elongated alleles in all six sequenced individuals suggesting either a significant length variation in the population, or a local misassembly and a hidden gap in the reference genome. If so, this would not be surprising considering that multiple studies find repeat arrays in remaining gaps of genome assemblies [53,54,55].

A series of tandem repeats were shown to be variable in length between affected and unaffected relatives. However, all but one did not seem to be associated with ID as they were also found in healthy controls or were in disagreement with the clinical description. In one of the families (L020), a single repeat (‘XLID25’) consisting of two adjacent tandem repeat stretches revealed a unique expanded allele of 21 copies in the CT sub-repeat in the proband, whereas all other genotyped healthy individuals of the family as well as controls had unit numbers in the range of 11–20 copies.

CT-repeats (or GA-repeats in complement), also known as GAGA-elements, act as chromatin remodelling mediators by disrupting and displacing pre-assembled nucleosomes [56,57,58]. Emamalizadeh and colleagues [59] suggested that the (GA)11 repeat length in the promoter of RIT2 is crucial for obtaining the correct dosage of RIT2, important in regulating the neuronal function. A shorter allele (GA)5 in homozygous state has been detected only in a proband with schizophrenia thus linking it to the disease state. Similarly, length variation of (GA)n tandem repeats influences embryonic development and facilitates evolutionary adaptation by regulating MECOM and GABRA3 expression [60]. It has been reported that GAGA-binding protein in humans specifically binds to the elements of 8 GA-units [61], which explains why precise copy number ranges are extremely important in regulatory sites. GA-dinucleotides may also affect gene expression when located downstream of that gene [62], which is in agreement with our data.

Only 65 bp upstream of XLID25 a highly conserved microRNA gene, MIR222, is located. MIR222 is mainly expressed in telencephalon with a conserved pattern of expression in larval and adult brain in zebrafish [63]. The MIR221/222 cluster is known to play an important role in coordination of cell proliferation [64]. They were shown to regulate terminal differentiation of neurons in porcine cortex and cerebellum [65]. Moreover, MIR222 was demonstrated to regulate timing of neural development by blocking preliminary generation of bipolar neurons in Xenopus [66]. It has been suggested that MIR222, with or without MIR221, is a plausible candidate to cause intellectual disability associated with X-linked retinal dystrophy in the Xp11.3 deletion syndrome [42]. Interestingly, in a study by Chen et al. [67], screening of 13 brain-expressed miRNAs in 464 patients with X-linked intellectual disability revealed only 4 mutations, of which 2 segregated with the phenotype, and both were found in MIR222. One of these mutations was located near the Drosha ribonuclease cleavage site and therefore, could potentially affect mature miRNA formation. On the other hand, the high conservation of the brain-expressed MIR222 suggests an important function in this tissue. We demonstrated that in the proband of the L020 family expression levels of MIR222 were decreased. In consonance with our study, MIR222 and MIR221 levels were previously found to be down-regulated in hippocampal tissue of patients with a neurological disease – mesial temporal lobe epilepsy and hippocampal sclerosis [68]. All these studies point to a crucial role of MIR222 for proper brain functioning. Decreased levels seem to disturb yet enigmatic neuronal processes.

We also tested the expression of one of the MIR222 targets namely FMR1. FMRP is an mRNA binding protein that has multiple functions in post-transcription gene regulation including mRNA stability, mRNA transport and localization, translation control, and pre-mRNA alternative splicing [69] with the latter being more prevalent in brain compared to other tissues [70]. FMR1 is usually the primary gene to test in the newly diagnosed ID patients [4]. Large tandem repeat expansions cause silencing of this gene that leads to intellectual disability with more severe forms in males. Contrary to this most common mechanism, we demonstrated a significantly increased expression of FMR1 in the affected male of L020. This controversy may suggest that precise concentration of FMRP in the brain is required for its proper function, and any dysregulation disrupting the balance can cause a disease. Our data is in line with the fact that the gene was shown to cause abnormal behavior when overexpressed in mice, a specific high-anxiety phenotype that is different from FMR1 knock-out mice [71]. Elevated FMR1 expression levels were also suggested to be causal in 5 ID patients carrying a duplication that harbors the FMR1 gene, amongst several others [43,44,45,46]. However, only one study [44] looked into mRNA expression in blood of a proband, which was within the normal range. Either the presence of two copies in males is not sufficient to cause overexpression or the expression levels in brain are different from those in blood cells leading to tissue-specific consequences.

In order to detect additional genes that could be regulated by MIR222, we performed total RNA sequencing and detected a second deregulated target of MIR222 namely NEFH. Its product is a component of neural cytoskeleton important for neuronal maintenance and plasticity, neurite outgrowth, axonal caliber and transport [72]. Interestingly, in the L020 proband this gene was 50-fold up-regulated. As shown by Collard et al. [73] overexpression of human NEFH in mice causes defects in axonal transport, which eventually leads to neuron degeneration. Notably, NEFH protein was described 1.5-fold up-regulated in children with cortical dysplasia with epilepsy [74].

All these findings provide indirect evidence that the unique tandem repeat variant of 21 copies of the CT/GA repeat is a strong candidate for the ID phenotype in the L020 family. The expanded repeat might cause decreased expression of the nearby MIR222 microRNA resulting in a decreased breakdown of neuronal target genes including FMR1 and NEFH.

To confirm a causal link between the detected tandem repeat variant and the XLID phenotype, further studies are required. Future screening of affected cohorts and healthy population for XLID25 expansions, microRNA MIR222 mutations, expression profiles of MIR222 might reveal more cases and statistical significant associations. In addition, cellular experiments will assess the impact of the repeat size on the MIR222 expression in the same patient cell line. Potentially, CRISPR/Cas9-induced double-strand break within the tandem repeat will lead to the reparation-induced repeat instability [75, 76]. This should allow to generate cell lines with a common genetic background and only different by the repeat copy number which would provide a reliable functional model. Finally, a knock-out of the microRNA in a mouse model would demonstrate its role in the XLID development.

To our knowledge, this is the first study capturing and single molecule sequencing targeted genomic loci. This strategy can be applied to other targets as well as to repeats elsewhere in the genome. Though this particular probe design is only applicable to the X-linked disorders, our tandem repeats screening approach is expandable to other chromosomes. With time, improved genome annotation might require an update of the list of tandem repeats that are potentially relevant for a disease development.

Whereas our method allows the detection of a majority of repeat expansions, it might fail on longer repeats. In this study, we targeted repeats up to 500 bp. This enabled sequencing over the repeat multiple times, generating proper consensus lengths. To accurately measure repeat lengths, the polymerase reads must be at least six times the size of the insert. With a mean polymerase read length of 15 kb, the repeat sizes should have a maximum length of 2–2.5 kb. Nevertheless, with ever expanding longer reads, the repeats sizes that can be measured, will also be expanded. However, such long repeats constitute only a small portion of all repeats in the genome. A second limitation is that the capture method has a reduced performance on GC-rich tandem repeats or fragments. Although the single molecule sequencing has no problem passing GC-rich repeats, those sequences show reduced capture efficiency.

Conclusions

Our findings provide indirect evidence that the unique tandem repeat variant of 21 copies of the CT/GA repeat is a strong candidate for the ID phenotype in one of the studied families with X-linked ID. The expanded repeat might cause decreased expression of the nearby MIR222 microRNA resulting in a decreased breakdown of neuronal target genes including FMR1 and NEFH. Present work is the first large-scale study of targeted sequencing of tandem repeats as a means to improve diagnosis of an inherited disease. Future application of the described assay in a large number of cases will allow to evaluate the general contribution of tandem repeat instability to XLID. Next to XLID, our design may be used to study other X chromosome related diseases too. Moreover, this approach is not restricted to the X chromosome, but is applicable to screen for tandem repeats on other chromosomes as well.