Molecular Vision 2021; 27:1-16
Received 09 September 2020 | Accepted 04 January 2021 | Published 06 January 2021
Yihua Zou, Jiakai Li, Peiyan Hua, Tingyi Liang, Xunda Ji, Peiquan Zhao
Department of Ophthalmology, Xinhua Hospital, Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Correspondence to: Peiquan Zhao, Department of Ophthalmology, Xinhua Hospital, Affiliated to Shanghai Jiao Tong University School of Medicine, No. 1665, Kongjiang Road, Shanghai, China, 200092; Phone: +86-13311620396; email: email@example.com
Purpose: Retinoblastoma (RB) is a pediatric ocular malignancy due to biallelic inactivation of the RB1 gene. Genetic testing is critically important for treatment decisions for this disease. Targeted next-generation sequencing (NGS) has been demonstrated to be an effective strategy for discovering all types of mutations in the RB1 gene. The aim of this study is the application of targeted NGS in a cohort of Chinese patients with retinoblastoma to identify germline mutations in the RB1 gene.
Methods: Blood samples were collected from 149 unrelated probands with retinoblastoma (62 bilaterally and 87 unilaterally) and their parent(s). Genomic DNA was analyzed with custom panel-based targeted NGS, and the panel was designed to include exons 1–27 of the RB1 gene with flanking intronic sequences. Single nucleotide variations (SNVs) and small insertions/deletions (InDels) identified were confirmed with Sanger sequencing. If the Sanger sequencing of a low-frequency variant (LFV) detected with targeted NGS was negative, PCR-based deep NGS was conducted for added confirmation. Copy number variations (CNVs) detected with targeted NGS were confirmed with multiplex ligation-dependent probe amplification (MLPA).
Results: Overall, 74 germline mutations were detected in 48.3% of the probands (72/149, 56 bilateral and 16 unilateral cases). The total detection rate in the bilateral cases was 90.3% (56/62). These mutations included 64 SNVs and InDels (25 nonsense, 20 splicing, ten frameshift, eight missense, and one synonymous variants) and ten CNVs. All CNVs were confirmed with MLPA. Twenty-four (32.4%, 24/74) variants detected were novel, including nine splicing, six frameshift, five missense, and four nonsense variants. Eight LFVs (10.8%, 8/74) were found with targeted NGS; six of which were identified with Sanger sequencing, and two were identified with PCR-based deep NGS (13.16% and 3.000% mutant rates, respectively).
Conclusions: This study expanded the spectrum of germline mutations in RB1 using targeted NGS technology, which is a cost-saving and efficient method for genetic sequencing of retinoblastoma and may improve the molecular diagnosis of retinoblastoma.
Retinoblastoma (RB; OMIM 180200) is the most common primary ocular malignancy in childhood, originating from progenitor cells of retinal photoreceptors . The incidence of retinoblastoma is constant at one case per 15,000–20,000 live births worldwide, corresponding to about 9,000 new cases every year [1,2]. The majority of retinoblastomas is initiated by biallelic inactivation of the retinoblastoma susceptibility gene, RB1 (Gene ID:5925; OMIM 614041; NM_000321) [3,4]. The RB1 gene is a prototypic tumor suppressor gene, and its production is a protein (pRB, NP_000312.2) with 928 residues [5,6]. The structure of pRB is divided into three domains composed of an N-terminus, an R motif and A/B “pocket,” and a C-terminus . The mutations in the RB1 gene that inactivate the pRB can make the cell cycle progression uncontrolled, driving benign retinomas into malignant retinoblastomas . With the recent advances in genomic and epigenetic analysis methodologies, it has been reported that loss of function of the RB1 gene is insufficient for the development of retinoblastoma. Some other candidate oncogenes (MDM4 [Gene ID:4194; OMIM 602704], KIF14 [Gene ID:9928; OMIM 611279], MYCN [Gene ID:4613; OMIM 164840], CDH11 [Gene ID:1009; OMIM 600023], DEK [Gene ID:7913; OMIM 125264], E2F3 [Gene ID:1871; OMIM 600427]) have been found to drive retinoblastoma progression through the genomic gain or loss or overexpression in gene expression .
The most common initial sign of retinoblastoma is leukocoria, followed by strabismus and proptosis . No racial or gender predisposition has been found for the development of retinoblastoma. Patients with hereditary retinoblastoma have an increased risk for developing secondary malignancies, such as osteosarcoma, soft tissue sarcomas, and melanomas [11,12]. Early detection and early treatment are important for successful preservation of life and vision of patients with retinoblastoma, and identification of causative mutations is essential to access the risk of tumor development in the relatives of patients with retinoblastoma [13,14].
Mutations in the RB1 gene have been reported in patients from East Asia (Table 1). Two studies from Japan published in 1994 reported a detection rate of mutations in RB1 in tumors of 58% with PCR-SSCP and direct genomic sequencing and 40% with Southern blotting and RT–PCR [15,16]. A study from South Korea reported a detection rate of mutations in the RB1 gene of 71.4% in peripheral blood with RB1 exome sequencing and multiplex ligation-dependent probe amplification (MLPA) in 2012 . In a Chinese population, Cheng et al. examined genomic DNA from peripheral blood samples and tumor tissue samples from 47 patients with retinoblastoma with a detection rate of 21% with PCR-SSCP-DNA sequencing . The He study reported germline mutations in peripheral blood samples from 85 patients with retinoblastoma with a total detection rate of 47.1% with Sanger sequencing and MLPA . Recently, Lan analyzed germline mutations in peripheral blood from a large-scale Chinese cohort with Sanger sequencing and methylation-specific MLPA and reported a detection rate of 41.7% (75/180) . Most of the studies above were based on conventional detection methods, such as Sanger sequencing and MLPA, especially in China. In addition, none of the recent studies above mentioned the detection of low-level mosaic mutations in the RB1 gene.
Low-level mosaic mutations, which mean mutations with a lower mutant frequency than 50%, always less than 20%, play a crucial role in pathogenesis of sporadic retinoblastoma. Allele-specific PCR (AS-PCR) and deep sequencing have been demonstrated to be efficient for detecting mutations with a mutant frequency of less than 20% as previously reported [21-23]. However, these methods are still difficult for frequent use as they are time-consuming. Sanger sequencing can cover the whole genomic regions of the RB1 gene but is not powerful enough to identify low-level mosaic mutations.
Recently, targeted next-generation sequencing (NGS) has been demonstrated to be a rapid and effective strategy for detecting all types of mutations, in which only protein-coding regions of one or multiple specific genes are sequenced [24,25]. Therefore, the aim of the present study was to use custom panel-based targeted NGS to identify germline mutations in the RB1 gene from 149 unrelated Chinese patients with retinoblastoma. At the same time, considering the limitation of confirmation of low-frequency variants (LFVs) with Sanger sequencing, we first used PCR-based deep NGS to confirm the LFVs as an addition. With targeted NGS, Sanger sequencing, PCR-based deep NGS, and MLPA, the detection rate of germline mutations was 48.3% (72/149), with 90.3% (56/62) in the bilateral cases and 18.4% (16/87) in the unilateral cases.
A total of 149 unrelated probands with retinoblastoma and their parents were selected for this study from Xinhua Hospital affiliated with Shanghai Jiao Tong University School of Medicine, Shanghai, China, between 2017 and 2020. The clinical diagnosis of retinoblastoma was made by clinical examination and radiological investigations (CT/MRI) along with Retcam imaging at the hospital. All clinical information and blood samples used in this study were obtained with informed written consent from the families and in accordance with the Declaration of Helsinki.
2 ml of peripheral venous blood from each proband and the proband’s families was drawn into an anticoagulant tube containing EDTA, stored at -20 ºC prior to use. Genomic DNA was extracted with the High Pure PCR template preparation kit (Roche Diagnostics, Pleasanton, CA) in accordance with the manufacturer’s protocol as previously described . Each DNA sample was accurately quantified with the Qubit Quantification Platform (Invitrogen, Paisley, UK) or the NanoDrop 2000 spectrophotometer (Thermo Scientific, Waltham, MA).
Blood samples of all probands were subjected to targeted NGS. DNA libraries were prepared using the Illumina standard protocol. The amplified DNA was captured with custom panel-based targeted region–related biotinylated oligoprobes provided by the GenCap Enrichment Kit (MyGenostics, Beijing, China) or Agilent’s SureSelect Library Prep Kit (Amplicon-gene, Shanghai, China). The panel was designed to contain the 27 exons and flanking intronic sequence of the RB1 gene. Briefly, 3 μg of genomic DNA from each sample was fragmented for 350- to 400-bp products with nebulization, and Illumina adapters were then ligated to the fragments, which were then amplified with PCR. Libraries were prepared, and then the captured regions were enriched using the methods above according to the manufacturer’s protocol. The enrichment process was validated with the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Quantitative PCR were used to check the quantity of the library. The enrichment libraries were finally sequenced on an Illumina NextSeq 500 sequencer (MyGenostics) for paired read 2 × 100 bp or Illumina MiSeq (Amplicon-gene) for paired read 2 × 150 bp.
The raw sequence reads were mapped to the UCSC GRCh37/hg19 human reference genome using the Burrows–Wheeler Aligner (BWA, V 0.7.11). The sequence of the RB1 gene (accession number L11910) was used as reference. Variants were locally realigned with the GenomeAnalysisTK (GATK, V 3.3) toolkit and identified using ANNOVAR based on the Human Gene Mutation Database (HGMD), ClinVar, ExAC, ESP, and the 1000 Genomes project. Multiple bioinformatics tools, including PROVEAN, SIFT, PolyPhen-2, MutationTaster, GERP++, and REVEL, were used to predict the pathogenicity of genetic variants. The pathogenicity of genetic variants were finally accessed according to the 2015 American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) guidelines . The gene variant nomenclature was in accordance with the Human Genome Variation Society (HGVS) recommendations.
All single nucleotide variations (SNVs) and small insertions/deletions (InDels) detected in probands and their parents were subjected to Sanger sequencing. Genomic DNA was extracted using the method above. Primers flanking each variant were designed. The corresponding exons of the variants were amplified with PCR and purified according to the preset program in the Beckman automated workstation (Beckman Coulter Inc., Melville, MA). Purified PCR products were analyzed on a 3130XL Genetic Analyzer (Life Technologies, Carlsbad, CA). If the result of the Sanger sequencing of an LFV detected with targeted NGS was negative, then PCR-based deep NGS was conducted for added confirmation. In the PCR-based deep NGS, the only difference with targeted NGS was the amplified DNA that was a 200-bp-long sequence flanking the specific variant detected with targeted NGS rather than all the DNA fragments. All potential copy number variations (CNVs) in probands were further confirmed with MLPA. MLPA was strictly performed according to the formal protocol of the manufacturer with the SALSA MLPA kit P047-RB1 (MRC-Holland, Amsterdam, the Netherlands). Data were analyzed with Coffalyser software (MRC-Holland) downloaded from the MLPA website.
Statistical software (SPSS v.22; IBM Corp., Armonk, NY) was used for data analysis. One-way analysis of variance (ANOVA) and the chi-square test were performed to compare the differences between continuous or categorical variables. A p value of less than 0.05 was considered statistically significant.
This study included 149 probands with retinoblastoma (85 male patients (57%) and 64 female patients (43%), 87 unilateral cases (58.4%) and 62 bilateral cases (41.6%)) from a Chinese cohort with targeted NGS (Appendix 1). The general clinical profiles of this cohort are summarized in Table 2. Germline mutations were found in 72 probands (16 unilateral and 56 bilateral). The total detection rate was 48.3% (72/149), with a detection rate of 90.3% (56/62) in the bilateral cases and 18.4% (16/87) in the unilateral cases. The mean age at diagnosis of the probands was 13.0±1.30 months (mean ± standard error of the mean [SEM]) in the positive group and 27.1±2.30 months in the negative group. The majority of probands were classified into Group D or E (the International Intraocular Retinoblastoma of Classification, IIRC) at diagnosis, accounting for 95.8% in the positive group and 100% in the negative group.
In this study, 64 SNVs and small InDels, which consisted of 25 nonsense, 20 splicing, ten frameshift, eight missense, and one synonymous variants, were identified in 41.6% (62/149) of the samples (Table 3). The most common type of variant was nonsense, followed by splicing and frameshift variants, representing 39.1% (25/64), 31.3% (20/64), and 15.6% (10/64), respectively. In addition, ten CNVs, which included nine large deletions and one large duplication, were identified in the remaining ten probands. Mean age at diagnosis of probands with splicing variants was less than 12 months, while probands with missense variants had the oldest mean age at diagnosis (17.3±4.10 months) in this study. Early stages at diagnosis of retinoblastoma were rare; two were splicing variants, and one was CNV. The mean age at diagnosis and the laterality were compared among different types of mutations, which were not statistically significantly different (one-way ANOVA and the chi-square test).
The SNVs and InDels were comprised of 40 known and 24 novel variants, identified in 48 bilateral and 14 unilateral cases (Table 4). Of them, 55 were pathogenic, four were likely pathogenic, and five were of uncertain significance. The pathogenic variants were comprised of 45 substitutions, eight small deletions, and two small insertions, including 24 nonsense, 19 splicing, nine frameshift, and three missense variants. Of the pathogenic variants, 46 variants were de novo, seven were inherited from the father, one was inherited from the mother, and one was unavailable. Specifically, 83.3% (20/24) of the pathogenic nonsense variants were due to C to T transitions, and 85% (17/20) of them were located at the CGAarg codon of the RB1 gene. The likely pathogenic variants included one missense, one nonsense, one frameshift, and one splicing variants. Of them, one was de novo, and three were unavailable. The missense variant c.137G>A affected codon 46 of the RB1 mRNA, changing the amino acid arginine to lysine. It fell at the last nucleotide of exon 1, which may affect the splice site of this exon. The other three likely pathogenic variants were classified as likely pathogenic due to the lack of samples from the probands’ mothers and inconsistent effects predicted by multiple algorithms. The variants of uncertain significance included four missense and one synonymous (splicing) variants, of which two missense variants were inherited from the father, one missense variant was inherited from the mother, and two were unavailable. The synonymous variant c.861G>A (p.Glu287=) fell at the last nucleotide of exon 8, which may cause aberrant splicing in published studies [28,29]. Because these variants were predicted to be neutral, tolerated, or benign by multiple algorithms, the variants were classified as variants of uncertain significance.
Of the 40 known variants, six variants (c.751C>T, c.763C>T, c.1399C>T, c.1654C>T, c.1666C>T, c.265–2A>G) were concurrent in two unrelated probands, one variant (c.1215+1G>A) was concurrent in three unrelated probands, and one variant (c.958C>T) was concurrent in four unrelated probands (Table 4). All eight probands were classified into advanced retinoblastoma at diagnosis except proband CA1258B (c.265–2A>G). The unilateral retinoblastoma cases were diagnosed at a relatively older age than the bilateral cases with the same variant, whereas the difference in clinical stages at diagnosis between patients with the same variant was unclear.
The 24 novel variants were identified in 19 bilateral and four unilateral cases, with an average age at diagnosis of 10.9 months. The types of mutations included nine splicing, six frameshift, five missense, and four nonsense variants. Of them, 20 variants were pathogenic, one was likely pathogenic, and three were of uncertain significance (Table 5). Most novel variants were de novo and were predicted to be deleterious by PROVEAN, disease-causing by MutationTaster, and conserved by GERP++ tools. The frameshift variant c.1432dupA was classified as likely pathogenic due to the lack of samples from the probands’ parents. The three missense variants were of uncertain significance as they were predicted to be benign by multiple bioinformatic tools. In addition, proband 17C024236, a 22-month-old male patient with a unilateral eye affected at diagnosis, was found to carry two different novel variants.
There were 21 variants (20 bilateral and six unilateral) located at the regions encoding two conserved domains A (residue 393–572) and B (residue 646–772) of pRb, of which eight were nonsense, six splicing, five frameshift, and two missense variants (Figure 1). The pathogenicity analysis indicated that 85.7% (18/21) of these variants were pathogenic, and the other three were likely pathogenic (c.1421+3A>T, c.1432dupA, c.2268T>A). The three variants were classified as likely pathogenic due to the absence of samples from the probands’ parents, although c.2268T>A was predicted to be disease-related by multiple algorithms.
The mutant frequency was defined as the ratio of the total reads of mutant alleles to the total reads of wild-type alleles for the SNVs and InDels in the leukocyte DNA. Less than 20% of mutant frequency with targeted NGS was defined as LFVs. A total of eight variants (three bilateral and five unilateral) were identified as LFVs ranging from 3.39% to 19.0%, including four nonsense, three frameshift, and one splicing variants (Table 6). Six of the eight LFVs were identified with Sanger sequencing and two with PCR-based deep NGS (Figure 2), with a mutated frequency of 13.16% and 3.000%, respectively. Five of the LFVs were known, and three (c.2285_2286delAG, c.227T>G, c.1432dupA) were novel. Seven de novo LFVs were classified as pathogenic, while the variant c.1432dupA was likely pathogenic due to the lack of samples from the mother.
CNVs were identified in eight probands with bilateral retinoblastoma and two probands with unilateral retinoblastoma (Table 3); none of these probands’ parents were subjected to genetic testing. All CNVs detected were further confirmed with MLPA (Table 7). The ratios of detection ranged from 0.32 to 0.69 for the large deletions and from 1.42 to 1.59 for a large duplication. The proband with deletion of exons 1 and 2 was a girl with early stages of bilateral retinoblastoma at the age of 1 month at diagnosis. Her mother was bilaterally affected by retinoblastoma. The proband with deletion of exons 1–23 was a girl who was 34 months old at diagnosis of unilateral retinoblastoma, whose mother was unilaterally affected. Four probands with bilateral retinoblastoma and one proband with unilateral retinoblastoma were found to have deletions of the whole RB1 gene. Deletion of exons 1–17 were identified in a 2-month-old girl with bilateral retinoblastoma, and duplications of exons 18–23 were carried by a 13-month-old boy with bilateral retinoblastoma.
Three probands with bilateral retinoblastoma with different splicing variants were familial in this study (Figure 3). The disease–eye ratios (DERs) of these probands were 1.67, 2.00, and 1.50, all indicating a high disease penetrance (≥1.5). Variant c.265–2A>G was known and damaged the splice site between exon 2 and exon 3. However, another proband (RB99) with the same variant was de novo and had no similar family history. The other two splicing variants were novel, and both affected the probands bilaterally, which broke the splice sites between exons coding domain A. The variant of proband RB54 (c.1332+3A>C) was inherited from his mother who was unilaterally affected, and his sister was also affected. The variant of proband RB145 (c.1421+1G>A) was inherited from his father who was unilaterally affected.
NGS is increasingly used in gene mutation analysis of retinoblastoma. The utilization of targeted NGS makes the sequencing cost lower and the sequencing more efficient and convenient. Screening of mutations in the RB1 gene with targeted NGS substantially benefits the prepotency, early diagnosis, and treatment of retinoblastoma. If the mutations in RB1 are hereditary, the parents will be advised to have a second baby through in vitro fertilization techniques or early prenatal diagnosis for high-risk offspring. If the mutations in RB1 are not found in patients or are nonhereditary, the parents will not be worried about having a second baby, and the follow-up period for patients could be extended. In general, targeted NGS is popular in genetic sequencing of retinoblastoma for ophthalmologists and geneticists.
In this study, we reported 64 distinct germline SNVs and InDels and ten distinct CNVs in 72 probands from a cohort of 149 unrelated Chinese patients with retinoblastoma using targeted NGS. The total detection rate in this study was 48.3%, which is within the reported range of rates of 42.0–67.0% with Sanger sequencing and MLPA [20,30]. The detection rate of 90.3% in the probands with bilateral retinoblastoma is also consistent with previous reports (88.8–100%) [20,31,32]. The detection rate was relatively not that high, which may be due to the limitation of the panel designed in this study to report mutations in the promotor region or deep intronic variants which were actually uncovered. In addition, in this study, two variants were confirmed using PCR-based deep NGS rather than Sanger sequencing, which improved the detection rate from 47.0% (70/149) to 48.3% (72/149). To the best of our knowledge, this is the first report to combine the three methods of targeted NGS, Sanger sequencing, and PCR-based deep NGS for identification of low-frequency variants in the RB1 gene.
The general clinical findings of the probands in this study were similar to those for the distribution of Chinese patients with retinoblastoma reported previously [12,19]. Moreover, the number of probands with advanced retinoblastoma (IIRC D or E stage) at diagnosis was higher than 95% in this cohort, suggesting the severe condition of retinoblastoma at diagnosis in China. This may also be relevant to the fact that our hospital is a tertiary hospital in Shanghai, and diagnosis of patients with retinoblastoma who come to our hospital is often delayed. Additionally, we identified 40 known and 24 novel variants for SNVs and InDels, 53 of which were classified as pathogenic and six as likely pathogenic. The most common type of variant was nonsense, followed by splicing and frameshift variants. We found no statistically significant differences in the mean age at diagnosis and laterality among different types of mutations, which was as the same result as in Sagi’s study in Israel . However, the probands with missense variants tended to be older at diagnosis, and those with null mutations tended to have a younger average age at diagnosis. This perhaps could be explained by that missense mutations usually lead to reduction of function instead of loss of function of pRB. The familial cases, which were all splicing mutations, also revealed the high penetrance of splicing mutations as null mutations leading to the disruption of important splicing sites . In Mehyer’s study, splicing and frameshift mutations were associated with more advanced tumor stage , which was not found due to the rarity of patients diagnosed with an early stage of retinoblastoma in this cohort. As for the CNVs, all were identified with MLPA in the present study. Some patients may also have deletions or duplications of genes near the RB1 gene, which needs more studies to explore their relationships with the phenotypes of patients with retinoblastoma .
Importantly, low-level mutational mosaicism is still a challenge for molecular diagnostics of retinoblastoma. Conventional methods are difficult for routinely detecting low-frequency mutant alleles in the RB1 gene. It has been reported that Sanger sequencing is able to disclose mosaicism only for rates above 20% . However, several previous studies reported more sensitive methods. Rushlow et al. identified low-level mosaicism in 5.5% of bilateral retinoblastoma cases with AS-PCR of 11 mutational “hot spots” . Chen et al. reported that for retinoblastoma cases without mutations with Sanger sequencing, deep or next-generation sequencing technology of the RB1 gene can detect 30% and 6% low-level RB1 mosaic mutations in bilateral and unilateral cases, respectively . Amitrano et al. identified with deep sequencing low-level mosaic mutations in sporadic retinoblastoma cases at a frequency between 8% and 24% in blood DNA . Currently, targeted NGS, which has been reported in several recent studies, is considered a highly sensitive and efficient approach for the detection of genetic mutations [24,35]. Therefore, in this study, we noticed that the targeted NGS approach detected eight low-frequency variants with mutant frequency of less than 20%; six of them were identified with Sanger sequencing, but two were not. For the other two low-frequency variants not validated with Sanger sequencing, we performed targeted NGS and PCR-based deep NGS. We believe that the occurrence rate of the same NGS error at the same variant site in the two NGSs was very low. In addition, the coverage of PCR-based deep NGS was much higher than the targeted NGS ([11,280/1485 and 48,738/1628] and [173/33 and 598/21]), supporting the results of the NGS screening more validly.
However, although targeted NGS has good sensitivity and efficiency for the detection of mutations in RB1 and even low-level mosaic mutations, this method still needs additional related studies to explore its accuracy for a specific detectable range of the mutant frequency of mutations. Sanger sequencing and MLPA are needed to confirm the NGS results. In addition, functional studies on specific mutations in RB1 are needed to help understand the mechanism of pathogenesis of retinoblastoma.
In summary, targeted NGS is a cost-saving and efficient method for genetic sequencing of retinoblastoma. We identified 24 novel variants and eight low-frequency variants in the RB1 gene with targeted NGS, which expanded the spectrum of mutations in RB1 and may help improve the molecular diagnosis of retinoblastoma.
Appendix 1. General profiles of all retinoblastoma probands in this cohort.
We thank the patients and families who consented to participate in this study. The study was supported by National Natural Science Foundation of China (81470642, 81770964). Dr. Xunda Ji (firstname.lastname@example.org) and Dr. Peiquan Zhao (email@example.com) are co-corresponding authors for this paper.