Abstract
INTRODUCTION
Duchenne Muscular Dystrophy (DMD) (MIM 310200), Becker Muscular Dystrophy (BMD; DMD milder allelic variant) (MIM 300376), Intermediate Muscular Dystrophy (IMD) forms and isolated cardiomyopathy without skeletal muscle involvement (XLDC; MIM 302045) are all caused by mutations in the dystrophin gene (DMD, MIM 300377) located on Xp21. On the basis of DMD prevalence in male newborns (1:3,500) and using Haldane’s rule (i.e., for a disease with a stable prevalence, the rate of appearance of mutated alleles must equal the removal of affected individuals due to fitness decrease), the disease-associated mutation rate for the DMD gene is estimated to be around 1:10,000 meioses [1–3]. Copy number variations (CNVs) are the most frequent mutations in this gene, particularly CNV deletions that account for approximately 65% of all mutations [4]. Although CNVs can occur along the entire DMD gene, deletion frequency is higher in a minor hot-spot at the 5’ end (exons 2 to 20) and particularly in the central portion of the gene (exons 45 to 55) [5–8]. Indeed, despite the large size of this gene (2300 kb in length and 79 exons), which could represent a huge target for chromosomal rearrangements [9], CNVs do not occur randomly but mainly in hot spot regions. Therefore, DMD size is not sufficient to explain the high CNV incidence.
CNVs are considered to be the consequence ofDNA damage repair mechanisms that were first des-cribed in model organisms, particularly bacteria. Three major mechanisms have been proposed for genomic rearrangements in the human genome: Non-allelic homologous recombination (NAHR), non-homologous end joining (NHEJ) and micro-homology mediated replication-dependent recombination (MMRDR). Non-allelic homologous recombination (NAHR) is mostly mediated by recombination between low copy repeats (LCRs). LCRs are large-sized (10–300 kilobases) and highly similar (usually >95% identity) sequences present in few copies in a haploid genome. NAHR was originally described in Charcot-Marie-Tooth disease type 1A (CMT1A) and hereditary neuropathy with liability to pressure palsies (HNPP) [10]. Since then, this mechanism has been shown to be involved in most diseases due to recurrent CNVs (CNVs with a common size and clustered breakpoint (BP)) [11, 12]. In rare cases, NAHR utilizes repetitive sequences (RE) (LINE, SINE, LTR and low-complexity AT rich elements), rather than LCRs, as homologous substrates. However, the length of homology between two RE sequences is much shorter than between two usual LCRs, which may explain the lower frequency of the RE-mediated recombination events [13]. The mechanisms of non-recurrent CNVs (CNVs with different sizes and distinct BPs) rarely implicate NAHR. They do not require LCRs, but may be promoted by these sequences or by other genomic architecture features, such as REs, TTTAAA sequences or palindroms [14]. The two main mechanism of non recurrent CNVs are NHEJ, which has a role in DNA double strand break (DSB) repair [15], and MMRDR. In NHEJ, DNA DSBs are repaired by ligation with another break and loss of a DNA segment of variable length. The product of NHEJ repair often contains additional nucleotides at the DNA end junction, leaving a “molecular scar” [13, 16]. In MMRDR, CNVs are formed when the replication polymerase is dislodged, upon encountering a breakpoint (BP), and re-associated with another close-by template, based on micro-homology (i.e., ≥2 bp at both ends), to continue replication [17]. MMRDR-based mechanisms include fork stalling and template switching (FoSTeS), break-induced serial replication slippage (BISRS) [18] and micro-homology-mediated break-induced replication (MMBIR) [18–21].
Sequencing analysis of BP junctions suggests that NHEJ and MMRDR are the main mechanisms of DMD gene rearrangements [4, 22–27]. We took advantage of the implementation in the clinical laboratory of a gene-specific, high-resolution, comparative genomic hybridization (CGH) array to fine-map the distribution of intronic BPs in a cohort of 57 patients with BMD/DMD who harbored deletions within the major hot-spot of deletions (intron 44 to 55) of the DMD gene. We then sequenced BP junctions in 21 of these patients to search for genomic elements at BP junctions that may suggest specific mechanisms. We also investigated the possible correlation between clinical severity and BP intronic localization.
MATERIALS AND METHODS
Patients and clinical assessment
Fifty seven patients carrying a DMD gene deletion within exons 45 to 55 were included in this study. They belonged to 51 different families (51 probands and 6 affected relatives) (Table 1).
The 57 patients were followed at different neuromuscular centers in France. The detailed records of the medical history, made available by their referring physicians, were reviewed to determine the long-term disease course and clinical severity. For each patient, age of onset (AOO) of the first symptoms, age at the last examination and age of ambulation loss (AAL), if applicable, were recorded. All types of muscular (delayed motor milestones, exercise intolerance, motor abnormalities), cognitive or cardiac symptoms were considered to determine the AOO. The results of cardiac function investigations were also reviewed, when available, to identify the AOO of cardiomyopathy. According to previous studies, cardiomyopathy was defined by a left ventricular ejection fraction below 55% or shortening fraction below 32% or both [28]. Cognitive assessment was based on clinical information (given by the referring physician and psychologists), daily life, communication and social skills, academic achievement, or neuropsychometric assessments with intelligence quotient (IQ) scores.
For 29 patients, the diagnosis was supported by western blot analysis of dystrophin expression level in muscle protein extracts prepared from frozen muscle biopsies, as previously described [29].
Molecular analyses
Blood samples, collected for diagnostic purposes, were obtained after written informed consent in accordance with the rules defined by the local ethics committees. Genomic DNA was extracted from leukocytes using standard procedures (phenol extraction or Wizard® Genomic DNA Isolation System, Promega). DNA concentration and quality were evaluated with a NanoDrop 2000 spectrophotometer (Nano-drop Technologies, Wilmington, DE) and agarose gel migration.
Deletion BP mapping by CGH array analysis
The DNA samples from the 57 patients were analyzed by using a custom oligonucleotide-based CGH array (Roche-Nimblegen) [30] and experimental conditions validated for diagnostic purposes [31]. Exons, introns, promoters and a region of 2000bp around the 5’ and 3’ terminal exons were covered by overlapping 60 mer probes with a tiling interval (i.e., spacing between the 5’ ends of probes) of 50 bp. Probes distribution was similar in exons and introns.
DNA sequencing of BPs
The DMD gene sequence was extracted from the UCSC human genome website (Ref Seq NM_004006).
Intronic BPs were characterized by direct sequencing in 21/57 patients (20 unrelated). PCR primers were first selected within a 500bp interval around the BPs mapped by CGH array and then within a 1kb interval, if necessary. Oligonucleotide primer pairs (sequences available on request) were designed using the Primer3Plus online tool (http://www.bioinformatics.nl/cgi-bin/primer3plus/primer3plus.cgi). Purified PCR products were sequenced bidirectionally using an Applied Biosystems 3130XT automated capillary sequencer (Applied Biosystems, Foster City, CA, USA) (protocol available on request).
Analysis of BP motifs
For each BP, 1000bp of reference sequence flanking both proximal and distal BP (500bp from the non-deleted side and 500bp from the deleted side) were downloaded. Searches for repetitive elements (RE), such as short interspersed nuclear elements (SINE), long interspersed nuclear elements (LINE), long terminal repeats (LTR) and DNA repeat elements, were performed by using the RepeatMasker program (Version 3.2.7). The presence of DNA palindromic sequences between the proximal and distal junction sequence was assessed with Mobyle.pasteur.fr. In the palindrome analysis, sliding window parameters were as follows: Minimum length of the palindrome = 4bp, maximum length = 100bp with 0 mismatches.
Analysis of BP distribution and statistical analyses
Each intron was divided in five equal blocks and the Chi2 test was used to determine whether the number of BPs located in each block differed significantly from the random expectation. The Chi2 test of homogeneity (or Yates corrected Chi2 test, when appropriate) was used for comparing the observed and expected frequencies in a random distribution. Significance levels are indicated as P levels.
RESULTS
This study included 57 patients with a dystrophinopathy who harbored a deletion within the DMD gene hot spot region going from intron 44 to 55. The cohort consisted of 48 sporadic probands and nine patients from three families: Family 1 (#1019, two brothers and a maternal cousin), family 2 (#360, two brothers) and family 3 (#737, two brothers, one maternal cousin and a more distant nephew) (Table 1).
Non random distribution of intronic breakpoints
CGH array analysis of genomic DNA from the 57 patients (51 unrelated) allowed assessing the distribution of 102 independent BPs located in intron 44 (n = 39), 47 (n = 24), 48 (n = 17), 49 (n = 13), 50 (n = 2), 51 (n = 1), 53 (n = 5) and 55 (n = 1) (Table 1 and Fig. 1).
Identical BPs
Twelve BPs had at least another apparently identical BP as further confirmed by genomic DNA sequencing (Table 2). Identical deletions from intron 44 to 53 and from intron 47 to 49 were found in two patients (#5550 and #1580) and three patients (#10500, #7347 and #2974), respectively. Two other patients (#10507 and #2210) carried deletions with identical BPs in intron 48, while the 5’ BPs were more than 10Kb distant in intron 44 (Hg18:31,906,933 and 31,917,676) (Table 1).
Clustered BPs (located in a interval <700bp)
In intron 44, which is the largest DMD gene intron (250kb in length), 48.7% of BPs (n = 19/39) were found within the most distal 50kb genomic region of this intron (Fig. 2). This differed significantly from the random expectation of 7.8 BPs in each block of 50kb of this intron (P < 0.0001). Moreover, 13 BPs in intron 44 were clustered in six genomic regions of less than 500bp (Table 1 and Fig. 2). Similarly, 14/24 BPs in intron 47, 6/17 BPs in intron 48 and 4/13 BPs in intron 49 were located in clustered genomic regions of less than 700bp (Table 2).
In conclusion, among the 102 studied BPs, 12 (11.8%) were identical, 37 (36.2%) were located within intervals of less than 700bp (clustered BPs), and 53 (52%) were not in close proximity of another BP (scattered BPs) (Fig. 3a). This distribution is different from a random one, in agreement with the presence of genomic elements that might promote such rearrangements.
Identification of specific genomic features involved in DNA breaks and repair mechanisms
The presence of REs, known to promote CNVs, was investigated in 1kb genomic sequences surrounding the proximal and distal BPs in 82/102 BP regions (Table 2). Twenty one junctions (17 with different BPs) were then sequenced to identify other genomic sequences that are known to promote DNA breaks and elements that might suggest specific DNA repair mechanisms. The location of the BPs characterized by CGH array was used to choose oligonucleotides for sequencing junction fragments in the genomic DNA samples (Table 3). On average, the distances between the real BP and the proximal and distal breakpoints mapped by CGH array were 190bp (standard deviation, sd = 242bp) and 267bp (sd = 275bp), respectively. In most cases (n = 33 sequences), this distance was less than 500bp and sequencing was successful using the first set of primers chosen on the basis of the CGH array results
REs are more frequent in the vicinity of identical and clustered BPs
REs were more frequently found in the vicinity of identical BPs (12/12) and clustered BPs (32/37, 86.5%) compared with scattered BPs (38/53, 71.7%) (p < 0.05) (Fig. 3b), in agreement with a role for these genomic sequences in BP occurrence.
In all patients carrying deletions with identical BPs (n = 5), a RE from the same family was found on either side of the deletion junctions (Table 1). Nucleotide homology within LINE sequences located between proximal and distal BPs was detected in two identical BPs (patients #5550 and #1580, deletion from intron 44 to 53). However, such homologous sequences were of limited size (82% identity in 27bp) (Fig. 4a). In another patient (#1562), a 302bp sequence between two SINE sequences, located 8bp next to the distal BP and 479bp distant from the proximal BP, showed 87% similarity and 74% (224bp) identity (Fig. 4b). In the other cases, the RE sequences in the distal and proximal BPs were not similar.
Identification of other genomic sequences that might promote BP occurrence
In four patients (#6504, #1348, #275 and #7347), palindromic sequences of 4bp were detected at either side of the junction. These sequences were close to both BP in most cases (six sequences at <20bp and two sequences at 121bp and 274bp). The TTTAAA sequence was identified within thesurrounding 100bp of one side of the junction in two patients (#6504 and #1348), 44bp and 42bp away from the BP.
Identification of genomic features associated with DNA break repair mechanisms
Micro-homologies of 2 to 5bp between sequences at either side of deletions were identified in 13 patients (76%) (Fig. 5). In addition, insertions of small sequences (3 to 8bp) were found in three patients (#3431, #3054 and #3375), in the absence of micro-homology for two of them (Table 3). Alignment of the inserted sequences with the BP flanking sequences showed that, in patient #3054, the inserted AAC sequence aligned with the template proximal to the BP. This suggested a backward slippage of the polymerase on the template sequence upon encountering the BP during replication (Fig. 6a). A single-nucleotide deletion, located 3bp proximal to a BP, was identified in one patient (#7630). Three patients did not have any specific sequence signature.
Phenotype-genotype correlations studies
At the last assessment, the patients’ age ranged between 2.6 to 75.8 years (36.6±15.3). Patients with DMD (n = 2) had an out of frame deletion of exons 48 to 50. The other patients carried in frame deletions. Specifically, patients with BMD (total n = 50) had in frame deletions from exon 45 to 47 (n = 13), from exon 45 to 48 (n = 13), from exon 45 to 49 (n = 9), from exon 45 to 53 (n = 5), from exon 45 to 55 (n = 1), from exon 48 to 49 (n = 4), from exon 48 to 51 (n = 1), or an isolated deletion of exon 48 (n = 3), or exon 49 (n = 1). Patients with IMD (n = 4) included two relatives from family 1 with a deletion from exon 45 to 47 (#1019–01 and #1019–41) and two relatives from family 3 with a deletion from exon 45 to 49 (#737–01 and #737–11). These patients’relatives (#1019–71, #737–21 and #737–31) had a milder BMD phenotype. The patient with XLDC (#9782) was 20-year-old at the last examination and had an in frame deletion from exon 48 to 49. Only two patients (#1562 and #8663, with deletions from exon 45 to 47 and from exon 45 to 48, respectively) presented cognitive abnormalities (Supplementary Table 1).
To determine whether, for each deletion type (for example for deletions from intron 44 to 48), the length of the intronic deletions was correlated with the clinical severity of BMD patients, three major clinical features were analyzed (AOO of symptoms, AOO of cardiomyopathy and AAL) (Supplementary Table 1). No statistically significant correlation was found, probably due to the low number of patients with each type of deletion. Then, the phenotype of patients carrying deletions with both identical BPs from the same families (n = 9) and from different families (n = 5) was analyzed. In the nine patients from three families, the clinical evolution appeared variable within two families. Two disease course patterns were observed: A severe one leading to rapid loss of ambulation before the age of 20 years, and a more progressive course in which loss of ambulation occurred in the 5th decade or later. Among the five patients from different families with identical BPs, clinical heterogeneity was also quite important concerning the AOO of skeletal muscle symptoms and disease progression.
DISCUSSION
In the present study, we took advantage of the implementation in the clinical laboratory of a custom oligonucleotide CGH array [30, 31] to study BP distribution and perform target sequencing on junctions in a large cohort of patients with deletions within the intron 44–55 interval, to improve our understanding of the mechanisms favoring CNV occurrence in the DMD gene. The accuracy of the custom oligonucleotide CGH array was sufficient to allow targeted BP sequencing because the average distance between BPs indicated by the CGH analysis and determined by sequencing was <300bp [31]. Therefore this technique constitutes an undeniable advance for BP identification.
Analysis of BP distribution in the hot spot region of deletions between introns 44 and 55 identified 20 regions of BP clustering in introns 44, 47, 48, 49 and 53. Miyazaki D et al. already reported (study on three patients) that BPs in intron 44 clustered in the 3’-half of that intron [32]. We confirmed this finding by showing that 48.7% of BPs in intron 44 were located within the 50kb more distal part of the intron. Similarly, Toffolatti and collaborators detected a relative clustering of DNA breakages in a region of 5kb in intron 47 (5 of 18 patients); however, this was not significantly different from the random distribution [26]. We confirmed this clustering within intron 47 by identifying six genomic regions of less than 700bp that included 71% of intron 47 BPs. Sironi et al. [25] also identified clustered BPs in introns 47 and 48. Specifically, they found that 3/12 BPs in intron 47 occurred within a 1.25kb interval and all sequenced deletion ends in intron 48 were located within a 23kb region. In our series, 35% of BPs in intron 48 were located in clustered regions <700 bp and two BPs were identical. To our knowledge, clustered BPs in introns 49 and 53 of the DMD gene have never been reported. Previous study on BP distribution did not highlight such clustered distribution possibly because they usedinaccurate tools for BP identification [9], or because the number of tested patients was too small to detect such clustering [23, 33]. The observation of BP clustering is in agreement with the involvement of REs and of other genomic elements in DNA breaks. In our study, the presence of REs in the vicinity of most BPs in identical or clustered BPs (90%), and less frequently near to scattered BPs (72%) (P < 0.005), illustrates the important role of such elements in the occurrence of DMD deletions. RE often contain palindroms that could form hairpin loops predisposing to DNA DSBs [26]. In rare cases, RE are involved as homologous sequences for NAHR [34]. In two patients, we found homologous sequences between proximal and distal REs. However, such homologous sequences were of limited size, not supporting NAHR as the underlying mechanism. We also identified sequences known to promote DNA instability and DSB formation [14, 32] near to junction fragments. Specifically, in four cases (23%) we found palindromic sequences that promote DNA hairpin loop structures, which predispose to DSB formation. Moreover, in two BPs we detected TTTAAA sequences that can induce a curvature in the DNA molecule, thus favoring the occurrence of breakage [14]. This is higher than expected by chance, on the basis of the frequency of TTTAAA sequences (1/1420 pb) in the human genome [35].
During NHEJ, broken DNA ends are modified by addition or deletion of bases to make them compatible and ready for the final ligation and end repair, thus leaving a “molecular signature” [16, 36]. Wang et al. (2015) analyzed 26 deletions in several genes and found a high frequency (5/26) of micro- deletions or duplications within BP-flanking regions due to DSB repair [37]. In our series, three patients had micro-insertions that could be explained by such a mechanism. However, one patient also had a micro-homology (#3375) and another one short tandem repeats (#3054) at the BP junctions, suggesting replication repair mechanisms and aberrant firing of replication origins (see below). In four patients, the absence of any sequence feature at BP junctions could suggest a NHEJ mechanism for these deletions.
In MMRDR, if during replication the lagging strand in the replication fork encounters a DNA lesion, it might invade a second fork downstream or upstream in the same gene, resulting in a deletion or a duplication, respectively [4, 16]. Micro-homologies promote MMRDR mechanisms and are identified by aligning distal and proximal BP sequences. In our study, the high frequency of micro-homologies at BP junctions (76%), comparable to the findings of previous studies [27, 38], supports the idea that replication repair mechanisms play an important role in the pathogenesis of DMD gene deletions [4, 39]. Moreover, other studies reported high frequency of small deletions and insertions flanking BPs, apparently originating from polymerase slippage events (serial replication slippage, SRS), in addition to frameshifts and point mutations, due to the implication of a low-fidelity, error-prone DNA polymerase associated with replicative mechanisms for CNV formation [37, 40]. This mechanism could be the cause of the deletion observed in patient #3375 who carried a 8bp insertion and a 3bp micro-homology, and in patient #7630 who had a 1bp deletion located 3bp away from the BP.
However, Ankala et al. proposed that replication repair mechanism are plausible for short rearrangements involving few hundreds to a thousand of bases, but not for large genomic rearrangements [17]. Some studies have shown that replication origins are associated with chromosome fragility [41, 42]. At least six replication origins have been identified in the DMD gene [43], including replication origins in introns 43, 46 and 65. One of the replication termination sites was mapped in intron 44 [43]. Studies in prokaryotes have shown that such termination sites may also serve as deletion hot-spots [44]. An attractive hypothesis is that the deletion sensitivity of this region could be related to DNA replication; a scarcity of replication origins in this region could lead to incomplete DNA replication when cell division is initiated, yielding a deletion of the non-replicated sequences [45]. Based on these studies and on the identification of insertions of short-tandem repeats at deletion BPs, Ankala et al. hypothesized that aberrant firing of replication origins, coupled with incomplete rescue of replication on the other strand, leads to DNA deletions [17]. The observed micro-insertions at BPs would reflect serial slippage of the progressing replisome when encountering DSBs and re-replication events of these short segments of template, but subsequent failure to continue replication after the BP region. In our study we identified 3–8bp insertions in 3/17 patients (Fig. 6a), which is lower than what reported by Anakala et al. (13 insertions/50 DMD CNVs). In one case (patient #3054, exon 45–49 deletion) the inserted 3bp sequence aligned perfectly with the template proximal to the BP in intron 44, suggesting a backward slippage of the replisome on the template sequence during replication upon reaching the BP. This patient did not have any micro-homology around the BP junction, and therefore the mechanism could be an attempt to rescue the non-replicated template due to failure of the replication origin in intron 46 (Fig. 6b).
Nevertheless, none of these hypotheses has been confirmed by experimental assays, and it is highly probable that these mechanisms are not working alone but in combination to explain deletions in the DMD gene. Ankala et al. also believe that, as suggested by Mitsui, micro-homology at CNV breakpoints may be attributed to a repair mechanism, such as microhomology-mediated end joining (MMEJ), rather than to a recombination mechanism [38]. Chen et al. 2015 proposed another model that combines the MMRBR and NHEJ pathways [46]. According to their hypothesis, during replication and SRS, the single-stranded template finally breaks, leading to DSB formation, replication fork collapse and activation of an alternative NHEJ pathway [18].
Concerning the phenotype-genotype correlations, it remains unclear why identical gene mutations lead to different phenotypes even among siblings, suggesting the contribution of other genetic modifiers, such as myogenic factor [47] and osteopontin [48]. In-frame deletions located in the central part of dystrophin (spetrin-like repeats) usually allow a residual production of partially functional dystrophin and are generally associated with BMD or IMD phenotypes [5, 49]. However, the reading frame rule does not explain the phenotypic variability (disease progression and motor, respiratory and cardiac involvement) observed within the DMD and BMD phenotypic classes [50, 51]. In a study performed on 33 patients, Van den Bergen et al. [52] concluded that dystrophin levels (if higher than 10%) are not a major determinant of disease severity in BMD. This is in agreement with our study. All patients with tested dystrophin levels (n = 29) had dystrophin levels above 10% (except one), including seven with dystrophin levels above 50%. We did not find any evident correlation between dystrophin level and clinical severity in these patients with BMD. The patient with dystrophin level below 10% showed early symptom onset (at the age of 4.1 years), but was still relatively young (17 years of age) at the last evaluation to allow a full assessment of the disease course (Supplementary Table 1). Previous studies on different in-frame deletions in patients with BMD suggest that besides the amount of residual dystrophin, other factors also play a role in the phenotypic variability and long-term clinical outcome: The structure of mutant dystrophin at the deletion junction site due to the nature of the deleted exons [53], the alteration of neuronal-type nitric oxide synthase (nNOS) binding due to deletions of repeats 16 and 17 of the dystrophin rod-domain [54], genetic modifiers from distinct genomic loci, such as LTB4 variants [55], and regulating non-coding RNAs, such as microRNAs [56]. Gentil et al. 2012 observed variable amounts of nNOSμ in 12 patients with BMD and deletions of exon 45 to 55 [54]. They hypothesized that this expression variability could be linked to the different sequences surrounding the intronic BPs, via epigenetic mechanisms [54]. In our study we observed phenotypical variability in patients carrying identical BPs, in agreement with the involvement of other factors. We did not have enough patients to test whether the length of intronic deletions at boundaries of deletions involving the same exons was associated with phenotypic variability. Miyazaki et al. also suggested that, in two patients with similar phenotypes and BPs in close proximity within the intron 44 to 55 region, some trans- or cis-acting elements involved in the sequence within or around the deleted regions could have contributed to the pathogenesis of the similar phenotypes [32]. They concluded that detailed sequence analysis of introns 44 and 55 in more patients with dystrophinopathies and deletions of exons 45 to 55 are needed to confirm this hypothesis [32].
In conclusion, our study illustrates, in a large cohort of patients, the involvement of different mechanisms in DMD gene deletions: Mainly MMRBR, but also NHEJ and potentially aberrant firing of replication origins. A combination of these mechanisms may also be possible. Although we did not find any correlation between clinical severity and variability of deleted intronic sequences, our report on a large series of patients combining clinical data and intronic BP location may be useful for future studies on the potential effects of other elements such as micro-RNAs and epigenetic factors.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
Footnotes
ACKNOWLEDGMENTS
We are grateful to the patients for their contribution in this study. We also thank the physicians for referring patients to our laboratory. This work was supported by the Assistance Publique –Hôpitaux de Paris. RBY is supported by AFM-Telethon. We wish to thanks Prof. Michel Koenig, Dr. Sylvie Tuffery-Giraud and Dr. France Pietri-Roussel for reviewing the manuscript and fruitful suggestions.