Introduction

Mobile elements (MEs) are DNA segments that can propagate through the genome using an RNA intermediate. In humans, three groups of MEs are still active: Long Interspersed Nuclear Elements 1 (L1), Alu and SINE-VNTR-Alu (SVA). L1 are autonomous MEs because they code for the proteins essential to their mobility [1]. Alu and SVA are non-autonomous MEs that use L1 machinery [1,2,3]. Taken together, they represent more than 25% of human genome base pairs: 16.9% for L1, 10.6% for Alu and 0.2% for SVA [4].

During retrotransposition, MEs duplicate their targeted region while inserting, creating repeated sequences anywhere in the genome [5]. This mechanism can lead to the creation or deletion of genes, thus participating in the evolution of the human genome. The rate of retrotransposition differs between MEs. New mobile element insertions (MEIs) are estimated to occur in 1/40 births for Alu, 1/63 births for L1 and 1/63 births for SVA [6]. It has been estimated that about 0.3% of all disease variants in the human genome are caused by de novo MEIs [7]. Specifically, retrotransposition can affect gene structure and/or expression, leading to genetic diseases or cancers. It can potentially involve exonic, intronic, splicing or UTR regions, and can cause deletions. Retrotransposition can thus lead to gene loss of function or gene expression modifications [8]. The first case described in 1988 was a patient with hemophilia A resulting from an L1 insertion in the F8 gene [9]. Since then, additional new genetic diseases or forms of cancer involving MEs have been reported. In 2016, a literature review identified 119 cases in which MEs were responsible for human diseases, including 76 L1 insertions, 30 Alu insertions and 13 SVA insertions [10]. The existence of diseases caused by Alu, L1 or SVA retrotransposition highlights the importance of detecting MEs within the patient’s panel, exome or genome sequencing data, especially in case of unexplained rare genetic diseases.

When recently inserted, MEs are not present within the human genome reference. Before the development of next-generation sequencing (NGS), MEs were identified by targeted gene sequencing (Sanger sequencing). Restriction digest, Southern blot, fragment cloning and Sanger sequencing were used to detect and characterize MEIs. The current large-scale use of exome sequencing (ES) has led to a significant increase in the diagnostic rate for unexplained rare genetic diseases [11]. However, the huge amount of pangenomic data has not been completely explored. Indeed, genetic anomalies other than copy number variants or single nucleotide variants are rarely considered in ES data pipelines. MEs cannot be easily identified with classical exome or genome pipelines because they contain repeated sequences or produce split- or multimapped reads. Tools to identify retrotransposons [12] exist, but they are usually not included in ES or genome sequencing (GS) analysis pipelines for rare diseases [13]. The few studies on MEs using ES data suggest that between 0.04% and 0.1% of suspected genetic diseases are caused by MEs localized in exonic regions [13]. Indeed, based on the DDD cohort of 9738 trios for individuals with developmental disorders, Gardner et al. used MELT to identify MEIs on ES data, which resulted in a diagnostic rate of 0.04% [14]. One year later, Torene et al. analyzed ES data from a cohort of 89,874 samples, including 38,871 cases with neurodevelopmental delay in particular. They implemented a specific tool for targeted capture sequencing data and found a similar diagnostic rate of 0.03% involving MEIs. This tool was then used by Demidov et al. [15] on 6584 probands with rare diseases or cancers. Two cases were found to be caused by a de novo germline MEI, again leading to a diagnostic rate of 0.03%.

The categories of variants identified with ES are increasingly diverse, and there is a trend towards using ES a unique genetic test to increase the diagnostic rate in genetic diseases. With this in mind, we included the detection of MEIs in our routine ES bioinformatics pipeline. Using MELT [16], we retrospectively analyzed ES data from a large cohort with a wide range of congenital conditions, including developmental anomalies and/or neurological disorders, in order to estimate the diagnostic rate and compare it to previous studies on ES data.

Materials and methods

Individuals

MEI detection was performed retrospectively on ES data from 3232 individuals, including 2410 probands, 384 trios, and 1 family of 4 individuals. About 80% of the probands had developmental anomalies and 20% had a primary neurological disorder (45% females and 55% males). About 68% of the ES results were negative or inconclusive.

The capture and sequencing steps were performed with several different kits and sequencing technologies (Table S1). BAM files were obtained using previously described methods [17]. Informed written consent was obtained from individuals or parents for ES analysis.

MELT pipeline and filters

After positive control analysis (see Supplementary Materials and Methods), we chose to use MELT (v2.1.5), with default settings, to detect MEs. First, each ES depth was determined using SAMtools [18] (v.1.2) (Fig. 1). This value was then used by MELT. Three main VCF files were obtained, one each for Alu, L1 and SVA. No MELT-specific filtering was performed. After MELT analysis, the pipeline determined for each proband whether it had a solo or a trio (proband and parents) analysis. Data from each family member were extracted from the 3 main VCF files, generating 3 VCF files per proband and per ME. These files were concatenated per family. The resulting file was annotated with AnnotSV (v1.2) [19]. A final tabulated file report was generated using an in-house Python2 (v2.7.15) script. MEs were then filtered to retain only those located in non-intronic regions in genes classified as morbid by OMIM, and present in less than 5 individuals in the cohort. Individual ID and sequencing depth on the MEI site were added. Remaining MEs were manually analyzed in order to detect any concordance between patient phenotypes and OMIM descriptions.

Fig. 1: MEs detection pipeline with the MELT tool on ES data.
figure 1

“Preprocessing”, “IndivAnalysis”, “GroupAnalysis”, “Genotype” and “MakeVCF” steps were realized by the MELT tool. Python scripts were used to mean depth computation, familial data extraction and proband final report generation.

Full size image

Candidate validation with orthogonal methods

Other bioinformatics tools

Each ME candidate was also checked in Tangram (v0.3.1) [20], Mobster (v0.2.4.1) [21] and SCRAMble (v1.0.1) [13] results (see Supplementary Data), which are three other tools used to detect MEs in NGS data. Tangram used discordant read pairs (DP) and split-reads (SR) to identify class 1 transposable elements. Mobster only used DP to detect Alu, L1, SVA or HERV-K (Human Endogenous RetroViruses K). SCRAMble used clusters of soft-clipped reads. This made it possible to compare the use of DP + SR versus DP only.

Polymerase chain reaction (PCR)

PCR testing was used to validate candidate MEIs. Elongation time depended on DNA fragment size: 2 min for Alu insertion and 10 min for L1 insertion. Regions of interest, spanning breakpoints, were amplified with specific primer couples (Table S2) located on the human genome reference and using PrimeStar GXL kit (Takara Bio Inc.) as recommended by the provider. PCR amplification was then checked by 1.0% TBE agarose gel electrophoresis.

RNA analyses

Blood and fibroblast total RNA extraction, quantification and quality control conditions are available in the Supplementary Data.

Determination of the impact of a variation on RNA splicing (cDNA sequencing): cDNA was synthesized from RNA using QuantiTect Reverse Transcription kit (Qiagen GmbH) following the provider’s recommendations. All coding sequences of the region of interest were extracted from UCSC genome browser (http://genome-euro.ucsc.edu/index.html). Primers (Table S3) were located in exons on both sides of those containing the variation of interest, using Primer3 (ref. [22]) program. PCR amplification and MiSeq sequencing were done as described above and in the Supplementary Data. Sequencing data were aligned using STAR (v2.5.2b) [23].

Cell culture

Fibroblasts were obtained from healthy controls and the subject with a FERMT1 mutation following written consent for skin biopsy. Cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) containing 4.5 g/L glucose (Thermo Fisher Scientific Inc.) supplemented with 10% fetal calf serum (FCS) and 1% antibiotics (ZellShield, Minerva Biolabs GmbH), in an incubator at 37 °C with 5% CO2 in a humid atmosphere.

Results

Data filtering

Among the 3232 individuals analyzed with the MELT pipeline, 2394/2409 probands had at least one detected MEI (Fig. 2), totaling 496,312 suspected MEIs (404,236 (81.45%) Alu, 87,751 (17.68%) L1 and 4325 (0.87%) SVA). One proband was excluded due to abnormal results. Eighty-nine percent of the detected MEIs were removed because they were not located in an exon, a promoter, a transcription terminator region or a UTR region. Among the remaining candidates (54,305), 4484 (9%) were inserted in a gene known to be involved in a human disorder (OMIM list of morbid genes). The frequency of the MEIs was checked in the 1000 Genomes database, and the candidates were kept when their frequency was below 1% (86.77% of the 4884 remaining MEs). Only MEIs present in less than 5 individuals in the cohort were conserved, removing 74.75% of the previously filtered candidates. This filtering removed 99.78% (n = 495,242) of the initial number of suspected MEIs. The remaining 1070 candidates were detected in 516 probands, consisting of 50.47% (540/1070) Alu, 49.25% (527/1070) L1, and 0.28% (3/1070) SVA. The genes are listed in Table S6. The mean and the median were respectively 2.07 and 1 MEI per proband. The remaining candidates were then filtered by concordance between the proband’s phenotype and the clinical synopsis recorded in OMIM.

Fig. 2: Filters applied to MELT detected MEs in 2394 probands.
figure 2

The majority of the detected MEs were Alu, followed by L1 and SVA. The 5 applied filters and indicated on the right of the figure permitted to reduce the number of potential candidates by more than 99%. Approximately 78% of the patients were removed from the list.

Full size image

Candidate MEIs

Once the phenotypic concordance was established, 9 candidates were retained in ADGRG6, NPRL3, FERMT1, SLC26A2, KMT2D, SETD5, TTN, SYNE1 and GRIN2B genes (Table S5). Six were Alu elements (ADGRG6, NPRL3, FERMT1, SLC26A2, KMT2D and GRIN2B) and 3 were L1 elements (SETD5, TTN and SYNE1).

The candidates and segregation were checked by PCR (Figs. 3, S3). None of the expected insertions were confirmed for ADGRG6, SLC26A2, KMT2D, SETD5, TTN and SYNE1: they were considered to be false positive results. The profile of FERMT1, GRIN2B and NPRL3 showed abnormal PCR products consistent with MEIs. The previous identification of another candidate variant and the segregation analysis led us to consider the NPRL3 insertion as inconclusive.

Fig. 3: FERMT1, GRIN2B and NPRL3 ME candidates validation and segregation.
figure 3

a FERMT1 MEI segregation for the proband (blood + fibroblasts) and his children (blood). Expected size (bp) without/with insertion: ~548/~829. Pr proband. b GRIN2B MEI segregation for the proband (blood) and her parents (blood). Expected size (bp) without/with insertion: ~640/~915. c NPRL3 MEI segregation for the proband (blood) and his parents (blood). Expected size (bp) without/with insertion: ~417/~750. (+) WT fragment amplification control. (−) PCR negative control without DNA.

Full size image

MEI candidate in FERMT1

The proband was an 80-year-old man referred for severe childhood-onset poikilodermia. He had no family history of skin disorders, but was born from parents native to the same village. He had three unaffected older sisters (two of them had died) and three unaffected children (Fig. S4a). Poikilodermia was more pronounced on exposed areas such as the face, neck, and forearms. Clinical examination also revealed microstomia, labial, palatine and right jugal leucokeratosis, cheilitis, bilateral ectropion, bilateral Dupuytren’s contracture, palmar keratodermia and nail dystrophy with trachyonychia (Fig. S5). The clinician’s hypotheses included atypical Zinsser-Cole Engman syndrome, Kindler syndrome or atypical poikiloderma. Solo ES was inconclusive.

MELT detected a heterozygous Alu insertion (NC_000020.10:g.6078235_6078236insN[281]; NM_017671.5:c.892_893insALU, r.850_957del, p.?; ClinVar VCV001341677) in FERMT1 (OMIM *607900), the gene involved in Kindler syndrome, an autosomal recessive disease. The patient’s phenotype and molecular information were therefore highly concordant. SCRAMble detected this insertion at the same position. Neither Tangram nor Mobster identified the event (Table S5).

Analysis of PCR products by gel electrophoresis confirmed that the proband had an insertion in exon 7. This insertion of ~280 bp is similar in length to an Alu element. Only the abnormal PCR product was identified in the proband’s DNA (Fig. 3a). This was in favor of a homozygous MEI rather than a heterozygous insertion as suspected by MELT results. The homozygous state was clearly confirmed on the Integrative Genomics Viewer (IGV) profile. The DNA of the proband’s parents was not available, but his three children and his remaining sister were analyzed. The segregation showed that the children were all heterozygous for the insertion with one 548 bp band (band 3) and one ~829 bp band (band 2). Moreover, an additional band (band 1, Fig. 3a) corresponding to DNA heteroduplex was detected (see Supplementary Data). The unaffected sister did not present an abnormal PCR product.

In order to confirm that the inserted fragment was an ME, the PCR product from the proband was sequenced (MiSeq) after gel extraction. The sequences were aligned to the human genome reference sequence, FERMT1 gene reference sequence modified with an Alu insertion and Alu reference sequence (Fig. S6a–c). Sequencing results for the proband confirmed a homozygous Alu insertion in the FERMT1 gene. All PCR products for the 3 children were also sequenced and aligned (Fig. S6d–f). PCR and MiSeq profiles (Fig. S9b) were identical for the 3 children, in favor of heterozygosity for the paternally-inherited Alu insertion.

In order to confirm these results, RNA expression and splicing were analyzed in the proband’s fibroblasts. Sequencing of RT-PCR products was performed (exons 5 to 10) and reads were aligned to the human reference genome (Fig. 4a). Unlike control fibroblasts, the proband’s profile showed no reads aligned with exon 7, in which the Alu element insertion was detected. Sashimi plots, which allow visualization of splice junctions, confirmed the skipping of exon 7 during pre-mRNA splicing in the proband’s fibroblasts (Fig. 4b). cDNA sequencing data were then aligned with FERMT1 cDNA reference sequence, confirming the exon 7 deletion (Fig. S10).

Fig. 4: FERMT1 exon 7 skipping in proband’s cDNA fibroblasts.
figure 4

The first two profiles were obtained for two control fibroblasts cDNA sequencing data and the third for proband’s fibroblasts cDNA. Only exons 5 to 10 were studied. a FERMT1 cDNA alignment with human genome reference sequence. As expected, the control fibroblasts profiles showed reads aligned with the exons 5 to 10 (reverse strand) with homogeneous coverage. The proband’s profile did not present any reads aligned with the exon 7. The other exons alignments are similar to controls’. Thus, the proband’s cDNA did not contain the exon 7 sequence. b Exons 5–10 sashimi plots. As expected, the control fibroblasts cDNA did not have splice anomaly as each exon was present in mRNA. For the proband, this plot confirmed the exon 7 skipping during FERMT1 mRNA splicing: no peak could be identified for the exon 7.

Full size image

The homozygous insertion of an Alu element within FERMT1 exon 7 caused a splice defect leading to an in-frame exon 7 skipping (108 nucleotides). Familial segregation and abnormal splicing resulting from the Alu insertion in exon 7 combined with a strong clinical and biological correlation led us to consider this Alu insertion as pathogenic.

ME candidate in GRIN2B

The proband was a 5-year-old girl referred for developmental disability. She had axial hypotonia, developmental delay and stereotypies. No specific facial features were reported. Birth parameters and growth were normal. The parents were unaffected (Fig. S4b). Solo ES was negative.

MELT suggested a heterozygous Alu insertion (NC_000012.11:g.13716543_13716544insN[275]; NM_000834.3:c.3628_3629insALU, p.?; ClinVar VCV001341678) in GRIN2B (OMIM *138252) at the position chr12:g.13716543 (GRCh37). Mobster and SCRAMble detected this insertion at the position chr12:g.13716609 and chr12:g.13716543, respectively. This gene has been involved in intellectual developmental disorder with or without seizures (OMIM #613970), and developmental and epileptic encephalopathy (MIM #616139). Causal variations are mostly de novo [24].

Analysis of PCR products by gel electrophoresis confirmed an Alu insertion in exon 13 in the proband (Fig. 3b). This insertion of ~280 bp was similar in length to an Alu element. As expected, the proband was heterozygous for the insertion with one normal PCR product (640 bp, band 3) and one abnormal PCR product (~900 bp, band 2). Moreover, an additional PCR product corresponding to DNA heteroduplex (band 1) was evidenced. Parental profiles were similar to the control without any abnormal PCR products.

In order to confirm that the inserted fragment was an ME, the proband and parent’s PCR products were sequenced (MiSeq). All sequences were analyzed using GRIN2B modified reference as described for FERMT1 (Figs. S12, S13). The results confirmed a de novo Alu heterozygous insertion for the proband in exon 13 of GRIN2B. RNA expression and splicing analyses in the proband’s blood were inconclusive. No cDNA amplification was obtained, confirming that GRIN2B is not expressed in blood.

Discussion

After analyzing the results obtained from ES data, PCR validation revealed 6 false positives among the 9 candidate MEIs (Fig. S3). A second PCR with a primer designed in the Alu element confirmed that the three Alu false positives did not show any MEI (data not shown). The 6 false positives were detected by MELT and passed MELT and interpretation filters. But they were missing within the BAM files generated by MELT containing the reads used for MEI detection (Fig. S14). However, LP, RP and SR scores, which indicate the number of 5′, 3′ and split-reads for each candidate ME, did not present different profiles compared with the 3 PCR-validated candidates. We did not succeed in identifying parameters that could be used to discriminate false positives, probably due to an insufficient number of validated results. It would be interesting to test a large number of candidates by PCR to identify all false positives before repeating these analyses. Therefore, PCR validation remains an important step before individual candidate analysis. To improve the results, working with the latest human genome reference sequence GRCh38 should be favored since detected structural variants with this reference had less false positive results than with GRCh37 (ref. [25]). In future, we hope to decrease our false positive rate by using the most recent human genome reference sequence.

After PCR validation, we retained 3 MEI candidates in the NPRL3, FERMT1 and GRIN2B genes. The last two were characterized by MiSeq sequencing. While the first insertion was excluded after familial segregation, the impact of the two other MEIs on patient phenotypes required additional investigations including the study of the consequences on RNA to confirm their pathogenicity.

FERMT1 gene has been described in the autosomal recessive Kindler syndrome, a condition highly compatible with the patient’s phenotype. An Alu element inserted in exon 7 of this gene was found to be responsible for an in-frame exon 7 skipping (108 nucleotides, 36 amino acids in FERM domain). The homozygosity of this insertion was compatible with the suspected consanguinity of the patient. Segregation analysis also revealed that all children were heterozygous for this insertion (Fig. S9). The impact of this MEI was confirmed in the RNA of the proband’s affected skin tissue. In addition to truncating variations, it has been reported that large deletions and genomic rearrangements that cannot be identified with simple PCR-based screening experiments constitute a significant part of the causes of Kindler Syndrome [26]. In this study we identified an Alu insertion in FERMT1 exon 7 which causes in-frame exon skipping. ME-induced loss of exons have already been described in the literature and were generally considered as pathogenic [26,27,28,29]. However, they were caused by a genomic, often Alu-mediated deletion, and not by a splice anomaly. Nevertheless, Alu insertions in genic regions may cause mRNA splicing modification and can lead to the introduction or the deletion of splicing sites [7]. Exons losses in FERMT1 reported in the literature cause frameshifts, leading to premature stop codons. However, the deletion of the last two exons 14 and 15 (ref. [26]), although not involving any described protein domain or premature stop codon, was considered to be responsible for the phenotype in a patient with Kindler syndrome. The exon skipping identified in our patient (Fig. 4) could lead to an abnormal protein with the deletion of a part of the first FERM domain. This domain is involved in membrane association by direct binding to the tail or the cytoplasmic domain of integrin membrane proteins, especially in focal adhesions [30]. Kindler syndrome is caused by the destruction of focal adhesions [31], reinforcing the hypothesis of a pathogenic role of the loss of exon 7 in our patient’s phenotype. A Western Blot analysis would be required to confirm the production of a modified FERMT1 protein lacking the 36 amino acids encoded by exon 7. Furthermore, immunolabeling experiments could determine whether there is a protein mislocalization, a decrease or an absence of FERMT1 protein in the patient’s fibroblast membrane, confirming the pathogenic role of the homozygous Alu insertion in FERMT1 exon 7.

The GRIN2B gene has been described in autosomal dominant infantile epileptic encephalopathy and intellectual disability. The proband carrying a heterozygous Alu element inserted in the exon 13 of GRIN2B was referred for developmental delay and axial hypotonia. Segregation analysis indicated that this insertion occurred de novo. No RNA study could be performed for this gene. We considered this ME as a variant of uncertain significance.

The tests performed on data from the 1000 Genomes project demonstrated the feasibility of MELT analysis on our ES data (see Supplementary Data), and our data were consistent with previous control results obtained with this tool [16]. The minimal differences observed were attributed to the change in version between the first analysis (MELT v1) and our analysis (v2). Nevertheless, it is important to highlight the bias in this approach considering the absence of a detailed comparison between the results from Tangram and Mobster and the MELT results. The results obtained for the 3 candidates confirmed MELT as the tool of choice for the detection of MEIs in our study (Table S4) since the 3 candidates detected by MELT were not all identified by Tangram or Mobster. However, it would be interesting to conduct a comparison with the same approach by analyzing Tangram and Mobster results, and comparing them with the results from the two other tools. The new SCRAMble tool, which was published during our study, may also be interesting since it is specifically designed for targeted sequencing data and it is easier and faster to use than the three other tools.

About 0.3% of all disease-causing variants in the human genome are caused by de novo MEIs [7], but lower values were expected for ES data because it covers only 2% of the genome. Gardner et al. studied a cohort of 9738 trios from the DDD project with MELT and identified 9 de novo MEs, 4 of which were classified as likely pathogenic, i.e., 0.04% [14] (Table 1). Another study published while our study was ongoing found a similar percentage in a cohort of 38,371 probands, mostly with neurodevelopmental delay [13]. Eight (0.02%) de novo and 5 inherited (0.01%) MEIs classified as pathogenic or likely pathogenic were identified from ES data and confirmed by Sanger. A third study identified 2 de novo MEIs out of a cohort of 6584 probands with rare disease or cancer, i.e., 0.03% [15]. Our study, which used ES data, detected one strong ME candidates in a cohort of 2410 probands, which is similar to the rates obtained by these three previous studies.

Table 1 The four studies on the identification of causal MEs in ES data from rare diseases and cancers cohorts.
Full size table

We had to overcome diverse bioinformatic and biological challenges during this study. First, MEs could not be detected with the classical pipeline used for SNV identification. No SNV nor SV classical callers were used, but specific read pairs analysis was performed. One of the major differences was the use of two different types of reference sequences. The identification of the discordant read pairs and the split-reads implied the simultaneous use of the human genome reference and MEs references. MELT used consensus sequences for the 3 MEs. Other tools, like Tangram and Mobster, used a MEs database containing several MEs subfamilies.

Another point raised during MEI identification was the determination of the breakpoint position. Using only ES data, breakpoints located in targeted regions could be precisely detected. The accuracy depended on the number of reads at the position and especially on split-reads, which are the best detectors. However, the accuracy can vary between tools. For example, the Alu element detected in GRIN2B by Mobster and MELT had two different positions: 13,716,609 (Mobster) and 13,716,543 (MELT) (Table S5). Nevertheless, they were close enough to be included in the Mobster confidence interval of 90% [13,716,523–13,716,668]. The characterization of the ME was thus less efficient than the detection, but this was not an obstacle for PCR confirmation. Breakpoint localization was precisely determined by MiSeq sequencing. It would be interesting to study this accuracy with long read GS data, which cover all genic regions and which would have better coverage, particularly for intron-exon boundaries.

MELT also determined the insertion genotype. However, there are inaccuracies as demonstrated by the FERMT1 insertion, which was classified as heterozygous by MELT and confirmed as homozygous by MiSeq sequencing (Figs. 3a, S7). Only split-reads were present when the breakpoint position in the BAM file was shown by IGV, confirming the homozygous status. Thus, the ME genotypes determined by MELT were not completely reliable. It is important to consider this parameter when analyzing the results for segregation study (trio) or comparing genotype and disease inheritance.

The annotation of results was necessary to analyze potential MEIs identified by MELT. MELT provided some information about each detected ME, but the information was not sufficient to analyze and filter the results. We therefore performed annotation with the AnnotSV tool. This additional information made it possible to filter MEs according to patient phenotype. Other annotations were also added for frequency (Fig. 2). The combination of these filters was used to remove irrelevant MEIs, resulting in a list containing less than 1% of the initial number of detected elements.

The second filter was applied to keep only OMIM morbid genes, which reduced the number of potential candidate MEIs by 91%. Although this filter was efficient, it had some limitations as it did not allow to discover new candidate genes [32]. Moreover, our study focused on non-intronic regions including exons, 3′ and 5′ UTR, promoters and transcription terminators. But it would be interesting to complete this analysis with intronic regions. Indeed, a ME inserted between 2 exons can have an impact on transcription, for instance by generating a new splice site or creating a premature polyadenylation signal. These changes do not only involve intronic regions. Thus, RNA analysis makes it possible to determine the influence of MEIs on splicing as well as on its expression level. Our results were then filtered based on the frequencies reported in the 1000 Genomes project [33]. Considering that our cohort was composed of patients presenting rare genetic diseases, only events with a frequency less than 1% were retained in order to remove polymorphisms [34]. We also chose to retain MEs present in less than 5 individuals in our cohort. A homozygous insertion inherited from both parents would therefore not be removed by this criterion. Similarly, a potential insertion transmitted by both parents to their 2 children (considering the family of 4 individuals in the cohort) would not be filtered out.

In conclusion, our work demonstrates that including MEI detection in diagnosis and research can improve the diagnostic rate in the challenging field of rare diseases.