Copy number variations in candidate genomic regions confirm genetic heterogeneity and parental bias in Hirschsprung disease

A total of 55 Italian sporadic HSCR patients fully genotyped at the RET locus [6] were retrospectively included in the study. To investigate the possible presence of interstitial deletions of the RET region, we selected 52 cases homozygous for the risk haplotype [5‐8] as well as 3 patients carrying very uncommon haplotypes, consistent with possible hemizygosity of the same region. Four additional HSCR patients were included as positive controls: two with a de novo deletion at the centromeric region of chromosome 10 [12, 13], one with an inverted duplication at chromosome 22, and another one with a trisomy 21 (in addition to two HSCR patients already included in the sample set and presenting with Down syndrome, OMIM# 190685). The whole sample analyzed is thus constituted by 59 HSCR patients.

Besides the major RET gene, other candidate genes and loci were selected for the analysis based on i) linkage with HSCR, ii) association with the disease, iii) mutation in syndromic and isolated HSCR patients, iv) involvement in the transcriptional regulation of RET, and v) preliminary evidence, not confirmed later. Finally, other loci were included because altered in disorders presenting HSCR with a higher prevalence than the general population (Table 1).

The selected HSCR patients and positive controls underwent custom aCGH. As reported in Additional file 1, a total of 75 calls were estimated from the raw data using the Agilent Aberration Detection method as described under Methods. Two of these calls corresponded to already known trisomies of chromosome 21, and four calls identified the alterations included as controls.

In addition, we evaluated the profiles of all the samples by visual inspection. This allowed us to add 6 calls to the list of aberrations, undetected by the software. Finally, although we did not expect any new variant in replicates, in a triplicate sample we found two aberrations that had not been detected in the array firstly investigated, but were present in both the two replicates and thus assumed as reliable, for a total of 83 aberrations detected in 64 different chromosomal locations in 44 samples (Additional file 1, Figure S1). Twenty-six of these aberrations had previously been reported on DGV; all of them were common CNVs (frequency > 5%), with the exception of the deletion at SEMA3A/3D, that is compatible with two CNVs detected by sequencing [26, 27] with an overall frequency of less than 1% (Table 2, Additional file 1: Figure S1). Four of the CNVs common in DGV were recurrent in the samples. The variants at 15q11 and 5q13, and the gain at 9p11, common on DGV, showed frequencies roughly similar to those reported for controls in the high resolution gnomAD (https://​gnomad.​broadinstitute.​org/) and deciphering developmental disorders (DDD) (https://​decipher.​sanger.​ac.​uk/​) databases [28, 29]. Losses at 9p11 and 9q31 had frequencies in between gnomAD and DDD. Interestingly, variants on 22q11.2 were all but one more frequent in our sample (Additional file 2). Moreover, three regions, found to carry anomalies in 5 samples, are compatible with CNVs reported on the Decipher database. The region 22:25672585–25,892,401 was found duplicated in two patients and deleted in a third patient, with anomalies also reported on DGV, with frequencies similar to controls in gnomAD and DDD and regarded as likely benign common CNVs. Also the deletion at 9:113025039–113,029,430 is common on DGV and likely benign, but interestingly it is reported in a patients affected by aganglionic megacolon, intellectual disability and short stature. Finally, the deletion at 1:146638075–149,224,043 is compatible with several deletions reported on decipher including the 1q21.1 recurrent microdeletion (OMIM# 612474).

In conclusion, excluding the control regions and chromosome 21 for the two Down syndrome HSCR patients, we have detected 51 novel aberrations, plus the one reported on DGV with very low frequency (Additional file 1, Figure S1). As two of these variants were recurrent in three patients each, we had a total of 48 distinct variants detected in 25 patients. Several of them looked unlikely at the visual inspection and, as a matter of fact, mainly not confirmed on the replicate, when available. Conversely, most of the variants classified as likely or possible at the visual inspection were also replicated [30] (Additional file 1).

Most of the regions reported as duplicated or deleted in other studies [16, 17] were not included in our selected regions, neither we could detect any aberration comparable with those already reported when covered by probes in our design.

We elected to focus on gains/losses that seemed more promising at the visual inspection (classified as likely) and not reported on DGV, in addition to one deletion of the SEMA3A/SEMA3D region, compatible with a CNV reported on DGV but with a low frequency. We included also three aberrations unlikely at the visual inspection but particularly interesting as located in the SEMA3A/SEMA3D region and the RET locus, two master loci in HSCR development [1, 10, 11]. We have thus selected for validation 18 aberrations found in 14 patients (Table 3, Fig. 1, Fig. 2, Additional file 1).

Eleven of these 18 aberrations were confirmed in 11 patients (Table 3), including one variant that gave no conclusive results by validation with a different method but was confirmed on a replicate. Seven were instead not confirmed, mostly of which on RET and/or already considered as unlikely after visual inspection. In addition, one variant initially not selected for validation, because considered unlikely at the visual inspection, was confirmed on a replicate and resulted recurrent in two other samples that have not been further analyzed. Interestingly, some of these 12 true novel aberrations involved the NRG1, SEMA3A/SEMA3D and PHOX2B loci, three of the strongest candidates among our target regions. In particular, the short deletion detected between exon 6 and 7 of NRG1 affected a male isolated patient with L-form HSCR, the ~ 9 kb deletion in SEMA3A/3D was intergenic (between SEMA3A and SEMA3D), and was inherited from the mother by a male S-form HSCR isolated patient. We could also detect a deletion that involved almost the entire PHOX2B gene in a female isolated S-form HSCR patient, shown to be inherited by the mother. Other CNVs were found at known HSCR candidate loci 9q31 [3] (in four patients, although not overlapping between each other and covering a gene in only one patient), 16q23.3 (a short inter-exonic gain) [23], and 19q12 (not involving any known gene) [4]. In addition, a CNV recurrent in three samples has been detected at 22q11.2, a locus affected in DiGeorge and VeloCardioFacial syndromes, and in der (22) and Cat-Eye syndrome, disorders presenting HSCR with a higher prevalence than the general population [31]. Finally two CNVs, at 1q21 and 15q21, were found outside any of the HSCR candidate loci represented at high-density probes in the array (Table 3). Among these 12 losses and gains found in 11 patients, the M/F rate was 7/4, with an enrichment in females with respect to the whole sample analyzed (from 28.8 to 36.4%), seven patients had an S-form (70.0% with respect to 58.8% for the whole sample analyzed), while three were L and one unknown (Table 4). Nine were isolated cases, while one patient had Down syndrome and another one presented with cardiac and facial malformations. Finally, 5 patients carried either RET variants or RET locus deletions, with increased frequency of RET anomalies with respect to the original sample (45.5% vs 23.7%). Interestingly, the four patients with aberrations at 9q31 were all defective for RET.

No difference in the distribution of patients’ characteristics was detected compared to the whole sample analyzed, when considering both the newly detected 77 aberrations and the “true” 37 CNVs detected. Indeed, in both cases the median size of CNVs was larger among syndromic patients than among isolated cases: 1,5 Mb vs 16.5 Kb in 9 and 68 newly detected CNVs in syndromic and isolated patients respectively, and 2.8 Mb vs 514 Kb in 5 and 32 “true” CNVs, with a borderline p-value (p = 0,0866), as already reported by others [17]. No difference was detected instead with regards to patients gender, while RET negative patients carried CNVs larger on average than RET mutated patients.

A total of 55 Italian sporadic HSCR patients were retrospectively included in the study, as described under the “Results” section.

We performed high-resolution oligonucleotide array-CGH analyses on 20 candidate genes/loci known to be involved in HSCR, using a customized 8x15K array (Agilent Technologies, Santa Clara, CA, USA), in accordance with manufacturers’ instructions, whose details are reported in Table 1. Additional probes were also distributed along the whole genome, including those useful for sample replication and normalization. Genomic positions are based on the Human Genome GRCh37 (hg19) assembly (http://​genome-euro.​ucsc.​edu).

To assess genomic imbalances, we applied the Aberration Detection Methods ADM-2 with a threshold of 6, as recommended by Agilent. We also applied the centralization and the GC correction algorithms and considered as aberrant only those regions with a minimum of 2 consecutive probes and exceeding a Mean Absolute Log2 Ratio of 0.30 (referred as MALR> 0.30). In addition, sample profiles were evaluated at the whole genome level by visual inspection. Such a manual search was performed paying attention to log₂ ratios values above 0.5 and below − 1.0 and taking into consideration not only the overall profile of the single individual sample (and its quality) but also the profiles of the entire cohort. The above parameters were kept, with the exception of the MALR> 0.30 filter, and observations were confirmed by a second operator. Based on such inspection, we classified the CNVs detected as “likely”, “possible”, or “unlikely”. Loci with nearby gain or loss intervals and an intervening region of more than 2 probes were considered as separate CNVs, as well as those differing for 2 probes with opposite log₂ ratios (log₂ratio < − 0.3 for gain and > 0.3 for deletions). The quality of the experiments was evaluated on the basis of the QC metrics generated by the Genomic Workbench 5.0.14 software (Agilent Technologies), such as the DLRSpread (derivative log ratio spread), a measure of the log ratio noise for each sample. DLRSs and the other sample metrics are detailed elsewhere [30].

Ten samples with bad profiles (DRLS ≥0.3) together with other 16 arbitrarily selected samples, were replicated at least once on another array, for a total of 26 samples with at least one replicate. The successive variants search was performed in the replicates of overall better quality.

Aberrations were compared with CNVs observed in the normal population, as reported in the Database of Genomic Variants (DGV, http://​dgv.​tcag.​ca/​dgv/​app/​home), and with the CNVs reported in the DECIPHER database of phenotypes, v8.7 release (https://​decipher.​sanger.​ac.​uk/​). Data comparing is challenging as exact boundaries of the aberrations are not known, but only assumed to be between the last “normal” probe (outer) and the first “aberrant” probe (inner), depending on the mean coverage. However, we considered aberrations as consistent with those already reported if they showed an overlap ≥80%, did not differ for more than two probes with compatible log₂ ratios (that is ≥ |0.3|), and were of the same kind (gain or loss).

To get further insight into the aberrations thus identified, we also compared their frequencies in our sample with frequencies of corresponding CNVs publically available at a resolution similar to that of our design, considering comparable those CNVs with identical boundaries (Additional file 2). To this end we used the control data from the European population in the GnomAD website (https://​gnomad.​broadinstitute.​org/​) and from the DDD database, browsed through the UCSC genome browser (https://​genome-euro.​ucsc.​edu/​index.​html), and assessed statistical differences through the Fisher’s test or the Chisquare test with Yate’s correction for continuity when more feasible.

Results obtained with the custom aCGH, together with the concordance degree among the replicates on the same design array, showed that the replication rate was not very high, and that the visual inspection outperformed the mere software call [30]. However, a high false positive rate is not surprising as a few studies have shown a not infrequent presence of false positive and false negative results from aCGH [51‐54]. As a matter of the fact, Conrad et al. (2010), using quantitative PCR (qPCR) for initial validation of aCGH experiments on 450 HapMap samples, suggested a preliminary false-discovery rate of ~ 20%, then assessed to 15% when comparing the results with another CNV-typing array [55].

For this reason, the most promising regions, based on rare presence/absence on DGV and visual inspection, were selected to undergo validation with different approaches. Parents were also investigated, when available, to verify whether the aberrations were inherited or de novo. When the deleted region was sufficiently small, the DNA was amplified with PCR and checked for anomalous bands by electrophoresis. For the majority of the aberrations, validation was carried out applying quantitative PCR. Primer pairs were designed with Primer-blast (http://​www.​ncbi.​nlm.​nih.​gov/​tools/​primer-blast/​) according to stringent parameters to ensure successful assay. qPCR analysis was performed with the LightCycler 480 Instrument and relative software using the SYBR Green I Master following manufacturer’s recommendations (Roche, Manheim, Germany). In case of inconclusive results, we spotted the sample on a CGH 8x60K array (Agilent).

The probability that 7 out of 7 inherited CNVs were transmitted by mothers was checked by the binomial test.

Finally, we defined variant as “true” when either reported on DGV, confirmed on a second array replicate, or validated with a different approach as described above.