Assessment of Target Enrichment Platforms Using Massively Parallel Sequencing for the Mutation Detection for Congenital Muscular Dystrophy

Abstract

Sequencing individual genes by Sanger sequencing is a time-consuming and costly approach to resolve clinically heterogeneous genetic disorders. Panel testing offers the ability to efficiently and cost-effectively screen all of the genes for a particular genetic disorder. We assessed the analytical sensitivity and specificity of two different enrichment technologies, solution-based hybridization and microdroplet-based PCR target enrichment, in conjunction with next-generation sequencing (NGS), to identify mutations in 321 exons representing 12 different genes involved with congenital muscular dystrophies. Congenital muscular dystrophies present diagnostic challenges due to phenotypic variability, lack of standard access to and inherent difficulties with muscle immunohistochemical stains, and a general lack of clinician awareness. NGS results were analyzed across several parameters, including sequencing metrics and genotype concordance with Sanger sequencing. Genotyping data showed that both enrichment technologies produced suitable calls for use in clinical laboratories. However, microdroplet-based PCR target enrichment is more appropriate for a clinical laboratory, due to excellent sequence specificity and uniformity, reproducibility, high coverage of the target exons, and the ability to distinguish the active gene versus known pseudogenes. Regardless of the method, exons with highly repetitive and high GC regions are not well enriched and require Sanger sequencing for completeness. Our study demonstrates the successful application of targeted sequencing in conjunction with NGS to screen for mutations in hundreds of exons in a genetically heterogeneous human disorder.

Next-generation sequencing (NGS) applications can have a tremendous impact on molecular medicine.¹ Sequencing whole genomes for personalized medicine may soon become possible with next-generation sequencing applications, such as whole-genome de novo sequencing, transcriptome sequencing, microRNA profiling, and targeted sequencing.^2–5 Technological advances have significantly increased the speed and throughput while decreasing the cost for these applications.^6–9 At present, targeted analysis of candidate genes is most suitable for diagnostic applications facilitating functional interpretation of sequence variations and overcoming limitations in computational power.¹⁰ DNA diagnostic sequencing of selected genes consists of two steps: an enrichment step and massive parallel sequencing using one of the commercially available NGS platforms.

A variety of methods have been developed in the last few years to enrich a selected portion of the genome, including solid phase–based microarrays^11–15 and solution phase–based methods, namely, SureSelect (SS).¹⁶ These strategies have the ability to enrich for megabase intervals or a full complement of protein-coding exons (exome). An alternative approach has been developed for enrichment of a desired genomic region by a microdroplet-based PCR approach, RainDance Technologies (RDT).¹⁷ This technology uses emulsion chemistry to generate millions of microdroplet-based PCR reactions, each representing a single amplification of desired target loci. Each droplet supports an independent PCR and is made to contain a single primer pair, along with the genomic DNA template and other reagents necessary for the PCR reaction. The entire population of droplets represents hundreds to thousands of distinct primer pairs and is subjected to thermal cycling, after which the emulsion is broken, and the PCR products are recovered. The mixture of DNA amplicons can be subjected to shotgun library construction and sequenced by NGS technology. By combining these RDT and SS target enrichment technologies with NGS, they can become important sequencing tools with the potential to be implemented in a clinical diagnostics laboratory. These enrichments methods and NGS technologies are relatively new, and as such, quality control of the sequence data has not yet been well defined. Guidelines will need to be put in place before NGS technology can be used routinely in molecular diagnostics laboratories.

Some of the most common heterogeneous genetic disorders for which genetic diagnosis is sought include inherited breast cancer, intellectual disability, ataxia, congenital sensory disorders, and inherited muscle disorders.¹⁰ Over the past decade, molecular understanding of the congenital muscular dystrophies (CMDs) has greatly expanded.¹⁸ Congenital muscular dystrophy disorders can be classified into four major groups, based on the affected genes and the location of their expressed protein: i) abnormalities of extracellular matrix proteins (LAMA2, COL6A1, COL6A2, and COL6A3); ii) abnormalities of membrane receptors for the extracellular matrix (FKTN, POMGNT1, POMT1, POMT2, FKRP, LARGE, ITGA7, and DAG1); iii) abnormal endoplasmic reticulum protein (SEPN1); and iv) intranuclear envelope protein (LMNA).¹⁹ Inheritance patterns range from classic autosomal recessive to de novo dominantly acting mutations (COL6 and LMNA). A specific diagnosis can be challenging because muscle pathology may not yield a definitive diagnosis, and access to and expertise in using immunohistochemical stains is limited. Muscle biopsy and genetic test findings must be interpreted in a clinical context, yet the majority of diagnostic testing is not accompanied by a standard clinical data set. When clinical features are manifested in patients and recognized by clinicians, the proportion of people who have the disease who test positive (clinical sensitivity) for merosin-deficient, Ullrich/Bethlem, and Walker-Warburg syndrome CMDs and related CMDs is 100%, 60% to 65%, and 60% to 65%, respectively.²⁰ Mutation screening by conventional Sanger-based DNA sequencing is not a scalable technology in a clinical laboratory setting because of the time requirements to complete an analysis and quality assurance procedures that exponentially increase with genetic complexity.¹⁰

The goal of this study was to evaluate the RDT and SS sequence enrichment technologies in conjunction with NGS sequencing by comparing them to each other and to the current gold standard, Sanger sequencing. We selected 12 genes associated with CMDs, which span a 65-kb exonic region (321 exons ± 50 bp on either side of each intron/exon boundary), as the clinical target in this pilot study. Given the genetic complexity and the fact that a specific diagnosis can be challenging, the genes that make up our CMD panel are ideal for evaluating the effectiveness of these target enrichment methods for clinical applications. In this report, we describe the identification of sequence variants in 12 CMD genes using two target enrichment methods (RDT and SS) and an NGS sequencing platform (SOLiD 3; Life Technologies, Grand Island, NY) under optimized conditions, and we discuss our findings in light of the validation and clinical laboratory implementation of these enrichment approaches.

Materials and Methods

Patients

All development was performed at the Emory Genetics Laboratory, which is a Clinical Laboratory Improvement Amendments and College of American Pathologists–accredited high-complexity laboratory. To demonstrate that NGS data could be used to identify different types of mutations, five positive control samples (C11, C12, C13, C14, and C15) whose mutations had previously been identified by Sanger sequencing of CMD panel genes were included in the analysis. The CMD comprehensive panel that consisted of Sanger sequencing COL6A1, COL6A2, COL6A3, FKRP, FKTN, ITGA7, LAMA2, LARGE, POMGNT1, POMT1, POMT2, and SEPN1 was performed on sample C12. The Bethlem Myopathy/Ullrich CMD sequencing panel was performed (Sanger sequencing of COL6A1, COL6A2, and COL6A3) on samples C11 and C13. Moreover, the merosin-deficient CMD Type 1A (MDC1A) sequencing panel (Sanger sequencing of LAMA2) was performed on sample C14. Similarly, the muscle-eye-brain disease sequencing panel was performed (Sanger sequencing of POMGNT1) on sample C15. Furthermore, a normal control sample (wild-type, W16) known to lack any mutations was added to the sample list.

In addition to the six control samples, six patients with CMD phenotypes were selected from the cohort of families collected by Dr. Carsten Bonnemann (National Institutes of Health, Bethesda, MD) under an approved institutional review board from the Children's Hospital of Philadelphia, and written informed consent was obtained for all patients. In all families, the affected members suffered from an underlying muscular dystrophy. The underlying gene mutations had not been identified for three patients, whereas the mutations were known for three patients, and these represented our blinded sample set.

Microdroplet-Based PCR Primer Design

A list of the 12 CMD genes was provided to RainDance Technologies (Lexington, MA), and they designed a custom CMD panel using their custom primer design pipeline based on the Primer3 algorithm (http://frodo.wi.mit.edu/primer3). The custom panel was prepared, and primers were designed to target all 321 coding exons (383 amplicons) of these 12 genes, including 50 bp of intronic sequence flanking each exon in the design. The amplicons in the panel ranged in size from 200 to 600 bases, with a GC content of 25% to 87%, and represented a total coding sequence of 65 kb. All single nucleotide polymorphisms (SNPs) and repeat regions were filtered from the primer selection region. The RDT design was quality checked in our laboratory to ensure that none of the primers were designed over known SNPs (dbSNP build 130) using an in-house Perl script against the National Center for Biotechnology Information SNP database. Primers were also verified to avoid repetitive regions of the genome using the program RepeatMasker (http://www.repeatmasker.org). The primers for the 383 amplicons varied in annealing temperature from 57°C to 60°C, with a primer length range of 15 to 22 bases. Other rules for primer design included BLASTing the primers to the chromosome that had the gene of interest and in silico PCR, using the UCSC Genome Browser, to match the designed primers to the PCR product sequence and size for the gene of interest.

Microdroplet-Based PCR

The samples were fragmented to 3 to 4 kb by shearing the genomic DNA with the Covaris S2 instrument (Covaris, Woburn, MA) following the manufacturer's instructions. To prepare the input DNA template mixture for targeted amplification, 3 μg of the purified genomic DNA fragments were added to 4.7 μL of High-Fidelity buffer (Invitrogen, Carlsbad, CA), 1.26 μL of magnesium sulfate (Invitrogen), 1.6 μL of 10 mmol/L dNTP (Invitrogen), 3.6 μL of 4 mol/L betaine (Sigma-Aldrich, St. Louis, MO), 3.6 μL of Droplet Stabilizer (RainDance Technologies, Lexington, MA), 1.8 μL of dimethyl sulfoxide (Sigma-Aldrich), and 0.7 μL of 5 units/μL of Platinum High-Fidelity Taq (Invitrogen). The samples were brought to a final volume of 25 μL with nuclease-free water. PCR droplets were generated on the RDT1000 instrument (RainDance Technologies). The CMD panel consists of an emulsion that contains a collection of unique primer droplets in which each primer droplet contains a single matched forward and reverse primer for each amplicon in the panel. Each panel contains multiple replicates of each unique primer droplet. Careful control is achieved in the manufacture of each panel to ensure that the volume of each droplet is consistent. This ensures that the concentration of the forward and reverse primers are consistent across all PCR reactions. Furthermore, the manufacture of each panel allows the counting of each unique primer droplet so that the representation of each unique primer droplet is consistent within the panel. This ability to uniformly represent each primer droplet within each panel allows a uniform representation of each PCR reaction per sample, resulting in low bias between all of the amplicons in the panel. The RDT1000 generates a PCR droplet by pairing a single genomic DNA template droplet with a single primer droplet. The paired droplets flow past an electrode that is embedded in the chip and are instantly merged to create a single PCR droplet. All of the resulting PCR droplets are dispensed as an emulsion into a PCR tube and then transferred to a standard thermal cycler for PCR amplification. Each single sample generates more than 1,000,000 singleplex PCR droplets. Samples were cycled in an Applied Biosystems (Foster City, CA) GeneAmp 9700 thermocycler as follows: initial denaturation at 94°C for 2 minutes; 55 cycles at 94°C for 15 seconds, 54°C for 15 seconds, 68°C for 30 seconds; final extension at 68°C for 10 minutes, and a 4°C hold. After PCR amplification, the emulsion was broken to release each individual amplicon from the PCR droplets. For each sample, an equal volume of Droplet Destabilizer (RainDance Technologies) was added to the emulsion of PCR droplets, the sample was vortexed for 15 seconds, and spun in a microcentrifuge at 13,000 × g for 5 minutes. The oil below the aqueous phase was carefully removed from the sample. The remaining sample was purified using a MinElute column (Qiagen, Valencia, CA) following the manufacturer's recommended protocol. The purified amplicon DNA was then tested on an Agilent Bioanalyzer (Agilent Technologies, Santa Clara, CA) to confirm that the amplicon profile (mixture of all amplicons of sizes ranging from 200 to 929 bp) matched the expected amplicon profile.

Microdroplet-Based PCR Amplicon Concatenation and Shearing

The ends of the amplicons were blunt-end repaired by adding the reagents to the purified DNA (diluted to 68 μL): 10 μL of 10× blunting buffer (Epicenter, Madison, WI), 10 μL of 2.5 mmol/L dNTP Mix (Invitrogen), 10 μL of 10 mmol/L ATP, 2 μL of End-It enzyme mix (Epicenter, Madison, WI), and sterile water to a total reaction volume of 100 μL. The reaction was incubated at room temperature for 30 minutes, and the DNA was immediately purified using Ampure XP beads (Agencourt, Danvers, MA). The amplicons were subsequently concatenated using the NEB Quick Ligation kit according to the manufacturer's protocol. DNA was purified using Ampure XP beads and eluted in 105 μL of low TE buffer. An Agilent Technologies 7500 Bioanalyzer chip was run to confirm the concatenation of PCR products. The sample was fragmented as described in the standard SOLiD workflow.

SureSelect Probe Design and Synthesis

The biotinylated-cRNA probe solution was manufactured by Agilent Technologies and was provided as capture probes. The sequences corresponding to the 12 CMD genes (321 exons) were uploaded to the Web-based probe design tool, eArray.²¹ The coordinates of the sequence data in this study are based on NCBI Build 36.1 (UCSC hg18). The following parameters were chosen for the probe design: 120-bp capture-probe length, 20× capture-probe tiling frequency, a 20-bp allowed overlap, and avoidance of repetitive regions. In total, 54,420 probes were designed, synthesized on a wafer, subsequently released off the solid support by a selective chemical reaction, PCR-amplified through universal primers attached on the probes, and then amplified and biotin-conjugated by in vitro transcription.¹⁶

Genomic DNA Fragment Library for SureSelect

Genomic DNA fragment libraries were prepared according to the manufacturer's instructions (Agilent Technologies). Briefly, 3 μg of each genomic DNA was fragmented by Adaptive Focused Acoustics (Covaris S2; Covaris), resulting in fragmentation of the genomic DNA to a size range of 150 to 180 bp. After end repair, the SOLiD barcoding adaptors were ligated and the libraries cleaned up using Ampure XP beads. A high-sensitivity bioanalyzer chip was run to ensure that ligation was successful. Each fragment library was nick-translated and enriched by a PCR amplification step. The PCR-amplified fragment libraries were quantified by a NanoDrop (ND1000; NanoDrop Technologies, Wilmington, DE).

SureSelect Solution-Based Hybridization and Target Enrichment

The target-enrichment step for each sample was performed according to the manufacturer's instructions (Agilent Technologies). Briefly, in a 96-well PCR plate, the capture probes were mixed with RNase block solution and kept on ice in a separate PCR plate; 500 ng of each genomic DNA-fragment library (“B” row) was mixed with SureSelect Block Mix and transferred into the B row, heated for 5 minutes at 95°C, and held at 65°C thereafter in the thermocycler GeneAmp PCR System 9700 thermocycler (Applied Biosystems). While maintaining the plate at 65°C, hybridization buffer was added into the “A” row of the PCR plate and incubated at this temperature for at least 5 minutes. The capture library mix was added to the “C” row in the PCR plate and incubated for 2 minutes at 65°C. The hybridization mixture was added to the capture probes, followed by the addition of the DNA fragment library. The solution hybridization was performed for 24 hours at 65°C.

After the hybridization, the captured targets were selected by pulling down the biotinylated probe/target hybrids by using streptavidin-coated magnetic beads (Dynal DynaMag-2; Invitrogen). The magnetic beads were prepared by washing 3 times and resuspending in binding buffer [1 mol/L NaCl, 1 mmol/L EDTA, and 10 mmol/L Tris-HCl (pH 7.5)]. The captured target solution was added to the beads and rotated for 30 minutes at room temperature. The beads/captured targets were pulled down by using a magnetic separator (DynaMag-Spin; Invitrogen), removing the supernatant, resuspending in wash buffer #1 (Agilent Technologies), and incubating for 15 minutes at room temperature. The beads/captured probes were pulled down with the magnetic separator and washed by resuspension and incubation for 10 minutes at 65°C in wash buffer #2. After three warm washes, elution buffer (0.1 mol/L NaOH) was added and incubated for 10 minutes at room temperature. The eluted captured targets were transferred to a tube containing neutralization buffer [1 mol/L Tris-HCl (pH 7.5)] and desalted with the MinElute PCR Purification Kit (Qiagen). Finally, the targets were enriched by 20 to 30 cycles of PCR amplification by using 5 μL per sample as a template, and the amplified targets were purified by Ampure XP beads. The samples were processed by the standard SOLiD workflow.

Variant Annotation and Identification of Causative Mutations

SNP and indel information was extracted from the alignment data using the NextGENe (Softgenetics, State College, PA) software. Analysis was limited to ±20 bp on either side of each exon. Additional custom filtering criteria were imposed to minimize the false-positive rates. Variants were filtered first for those that are novel (not present in dbSNP or the 1000 Genomes databases) and for those that are likely deleterious. We predicted that damaging SNPs would be novel silent, missense, nonsense, or splice-site SNPs, whereas damaging indels would be in coding regions. The variants that met these criteria were used for downstream analysis. Specifically, a large list of variants were identified by NGS; however, variants with <20× coverage (coverage is the average number of reads representing a given nucleotide in the reconstructed sequence) were removed from the list to be Sanger sequence confirmed, unless a variant was listed in the Human Gene Mutation Database as a frameshift or a nonsense change. In addition, variants with high frequency were also removed, but Human Gene Mutation Database, frameshift, and nonsense changes were kept. Similarly, synonymous coding variants previously found in the normal population were removed from the list. There were examples of variants in exons with <20× coverage that were selected for confirmation based on the likelihood of being real changes as indicated by the allele percentage (the percentage of a nucleotide at a specific location given by NextGENe visual sequence output; for a heterozygote and a homozygote, a 50%:50% and >90% allele percentage is expected, respectively) and Phred-like sequencing quality score (q), defined as q = −10log₁₀(p), where p = error probability for the base (ie, if q = 20 and P = 0.01, then the error rate is 1 in 100) at that site. Finally, all variants that met all of the specified criteria were Sanger sequence confirmed.

Validation of Mutations and Polymorphisms by Sanger Sequencing

Primers were designed to amplify each exon, including 50 bp of flanking intronic regions of LAMA2, COL6A1, COL6A2, COL6A3, FKTN, POMGNT1, POMT1, POMT2, FKRP, LARGE, ITGA7, and SEPN1. Samples were prepared by fluorescence sequencing on the ABI 3730XL DNA analyzer with BigDye Terminator chemistry and the BigDye XTerminator purification kit (Applied Biosystems).

Results

A total of 12 samples were evaluated in this comparative study. Five positive control samples were chosen to have different types of CMD-causing mutations to evaluate the ability of the target enrichment methods to identify different types of sequence variations. In addition, a wild-type sample was included to serve as a normal control reference. Moreover, six blinded samples were examined by NGS to identify the underlying mutation(s) responsible for the clinical features described for each patient and to evaluate each method for its diagnostic potential.

RDT and SS Enrichment Strategies Demonstrate Similar Sequencing Coverage and Genome Mapping

A single run on the SOLiD produced a read average of 385,459 and 419,749 for RDT and SS, respectively, with no statistically significant difference between the two enrichment methods (Table 1). RDT and SS samples had a similar average percentage mapped to the genome, namely, 54% and 55%, respectively. However, on average, a higher percentage of the RDT reads mapped to the target region relative to the SS reads. A range of 29% to 38% of RDT and 14% to 23% of SS unique reads mapped to the targeted regions, percentages that are consistent with other studies.^17,22 Approximately 96% of the region of interest had at least a 5× coverage in the SS samples compared to 88% of the RDT samples. Similarly, 95% of the region of interest had at least 20× coverage in the SS samples and lower in the RDT samples (87%).

Table 1.

Mapping Data of Reads Obtained Following Sequence Enrichment

Patient no.	Targeted method	High-quality reads (n)	Mapped to human genome (%)	On target (%)	Mean read length, (bases)	Bases covered for ROI (5×) (%)	Bases covered for ROI (20×) (%)
C11	RDT	209,312	57	36	47	88	86
	SS	456,576	60	18	47	96	95
C12	RDT	338,596	57	36	47	88	87
	SS	589,792	57	19	47	96	95
C13	RDT	265,206	56	35	47	88	87
	SS	585,719	55	18	47	96	96
C14	RDT	162,715	53	29	47	87	84
	SS	364,716	48	14	47	96	95
C15	RDT	343,911	55	38	47	88	87
	SS	408,524	56	16	47	96	95
W16	RDT	482,435	57	31	47	87	86
	SS	358,043	60	17	47	96	95
1	RDT	461,154	53	38	46	88	87
	SS	219,550	51	18	47	95	94
3	RDT	526,152	55	35	47	89	88
	SS	533,999	54	23	47	96	95
7	RDT	328,037	51	37	46	88	87
	SS	536,653	53	17	47	96	94
8	RDT	558,078	55	36	47	88	87
	SS	433,943	58	20	47	96	95
9	RDT	237,236	43	29	47	88	86
	SS	223,520	46	14	47	96	94
10	RDT	712,674	54	36	47	89	88
	SS	325,952	56	16	47	96	95
Ave. ± SD	RDT	385,459 ± 164,064	54 ± 4	35 ± 3	47 ± 0	88 ± 1	87 ± 1
Ave. ± SD	SS	419,749 ± 127,594	55 ± 4	18 ± 3	47 ± 0	96 ± 0	95 ± 1

Reads were obtained from RainDance Technologies and SureSelect sequence enrichment of human genomic DNA samples for congenital muscular dystrophy target regions using a SOLiD sequencing platform.

Ave., average; C, positive control samples; RDT, RainDance Technologies; ROI, region of interest; SS, SureSelect; W, wild-type sample.

Sequence Coverage and GC Content of Target Region

The target sequence complexity has a strong effect on the efficiency of DNA amplification and capture for individual exons. The mean gene depth was similar between the RDT and SS samples (Figure 1). The mean gene depth of coverage across all RDT samples ranged from 0× for POMT1, to 108× for POMGNT1, with an average of 51 ± 28× across all genes. By contrast, the mean gene depth of coverage across all SS samples ranged from 2× for POMT1, to 87× for LARGE, with an average of 53 ± 22× across all genes. In some instances, RDT enriched certain genes, namely, POMGNT1 and POMT2, better than SS (Figure 1). The opposite is also true: SS enriched LARGE better than RDT.

Figure 1.

Average gene coverage among all of the congenital muscular dystrophy genes following RainDance and SureSelect target enrichment and next-generation sequencing.

Despite the high mean gene-read depth and target region coverage, several exons, including exon 1 of COL6A1, exon 1 of LAMA2, exon 1 of SEPN, and all exons of POMT1, had low average coverage in both RDT and SS runs (see Supplemental Figures S1–S4 at http://jmd.amjpathol.org and Table 2). In addition, a closer examination showed that the first coding exons were poorly covered (see Supplemental Figures S1–S12 at http://jmd.amjpathol.org). All of the exons of POMT1 failed due to the average percent GC content of 56%, and several exons have a percent GC content >60%. Interestingly, 26 exons of the 321 targeted exons (8%) had low coverage in RDT compared to 2 exons (0.8%) in SS samples (Table 2).

Table 2.

Common and Unique Exons with Consistently Low Coverage across All RDT and SS Samples

RDT			SS			RDT and SS
Gene	Exon #	%GC	Gene	Exon #	%GC	Gene	Exon #	%GC
COL6A1	5	65	LAMA2	53	33	COL6A1	1	73
COL6A1	12	67	SEPN1	3	58	LAMA2	1	71
COL6A1	24	68				LAMA2	44	17
COL6A1	30	59				SEPN1	1	87
COL6A1	34	50				POMT1	2	60
COL6A1	35	67				POMT1	3	45
COL6A2	3	65				POMT1	4	39
COL6A2	6	65				POMT1	5	59
COL6A2	7	64				POMT1	6	40
COL6A2	14	68				POMT1	7	56
COL6A2	16	71				POMT1	8	57
COL6A2	22	73				POMT1	9	55
COL6A2	24	63				POMT1	10	59
COL6A2	26	62				POMT1	11	51
COL6A2	27	59				POMT1	12	67
COL6A3	13	48				POMT1	13	57
COL6A3	15	63				POMT1	14	49
FKTN	6	41				POMT1	15	67
FKTN	7	45				POMT1	16	64
ITGA7	15	62				POMT1	17	59
ITGA7	25	66				POMT1	18	57
LAMA2	27	36				POMT1	19	61
LAMA2	47	43				POMT1	20	55
POMT2	9	62
POMT2	10	40
SEPN1	6	87

Coverage was <20× average.

%GC, percent GC content.

Variant and Mutation Identification in Control Samples

The Sanger method is considered by the clinical laboratory community as the gold standard for sequencing. To validate the enrichment methods, we compared the variant calls obtained by Sanger sequencing to those obtained by NGS using five control samples with known mutations and one wild-type normal control as follows:

• 1Sample C11: COL6A1, COL6A2, and COL6A3 sequenced for Bethlem myopathy/Ullrich CMD panel;
• 2Sample C12: LAMA2, COL6A1, COL6A2, COL6A3, FKTN, POMGNT1, POMT1, POMT2, FKRP, LARGE, and ITGA were sequenced for CMD comprehensive panel;
• 3Sample C13: COL6A1, COL6A2, and COL6A3 were sequenced for Bethlem myopathy/Ullrich CMD panel;
• 4Sample C14: LAMA2 was sequenced for CMDC1A panel;
• 5Sample C15: POMGNT1 was sequenced for the muscle-eye-brain panel;
• 6Sample W16: wild-type control.

Different sets of genes were sequenced for the control samples because, traditionally, the clinical features of the patients were used to determine which genes should be pursued. Specifically, we sequenced the five positive controls, which encompass a variety of variants, such as deletions (sample C14), duplications (samples C12 and C13), missense (sample C11), and splicing changes (sample C15), post-RDT or SS enrichment using the SOLiD platform and the Sanger method (Table 3).

Table 3.

Number of Sequence Variants Identified by RainDance Technologies and SureSelect Using Filtered Data to Minimize the False-Positive Rates

Patient no.	Targeted method	Total NGS variants⁎	Variants missed by NGS†	Sanger variants‡	False positive (%)	No. of exons with <20× coverage	INT	EX	SNP	N-SNP	N-SNP INT	N-SNP EX
C11	RDT	8 (8)	7 (6)	7 (5)	(37)	94	0 (3)	8 (5)	7 (6)	1 (2)	0 (0)	1 (2)
	SS	9 (16)	7 (6)	7 (12)	(25)	30	0 (3)	9 (13)	7 (12)	2 (4)	0 (0)	2 (4)
C12	RDT	(53)	(5)	(39)	(26)	93	(17)	(36)	(44)	(9)	(3)	(6)
	SS	(43)	(5)	(37)	(30)	27	(24)	(38)	(41)	(21)	(8)	(13)
C13	RDT	6 (14)	5 (7)	4 (12)	(14)	87	2 (5)	4 (9)	4 (12)	2 (2)	1 (0)	1 (2)
	SS	12 (19)	4 (2)	5 (18)	(5)	22	1 (7)	11 (12)	5 (15)	7 (4)	0 (1)	7 (3)
C14	RDT	7 (4)	15 (1)	5 (4)	(0)	186	1 (3)	6 (1)	5 (4)	2 (0)	0 (0)	2 (0)
	SS	16 (7)	9 (0)	9 (5)	(13)	38	2 (3)	14 (4)	8 (4)	6 (3)	0 (0)	6 (3)
C15	RDT	20 (6)	11 (0)	13 (4)	(33)	92	5 (1)	15 (5)	12 (2)	8 (4)	2 (1)	6 (3)
	SS	23 (4)	11 (0)	12 (4)	(0)	45	3 (1)	20 (3)	23 (2)	7 (2)	0 (1)	10 (1)
W16	RDT	21	12	8	NA	106	3	18	6	15	2	13
	SS	17	11	10	NA	55	5	12	5	12	4	8
1	RDT	18	10	8	NA	76	7	11	8	10	4	6
	SS	16	9	7	NA	61	8	8	8	8	6	2
3	RDT	15	6	8	NA	79	5	10	12	3	2	1
	SS	21	5	9	NA	25	5	16	13	8	1	7
7	RDT	17	9	11	NA	94	5	12	8	9	1	8
	SS	22	6	10	NA	82	5	18	10	12	0	12
8	RDT	24	6	14	NA	74	8	16	11	13	4	9
	SS	11	7	12	NA	29	9	18	12	15	27	4
9	RDT	15	18	8	NA	146	3	12	8	7	1	6
	SS	20	15	10	NA	107	3	17	11	9	0	9
10	RDT	20	8	12	NA	65	7	13	13	7	1	6
	SS	18	8	13	NA	45	7	11	12	6	1	5
Ave. ± SD	RDT	16 ± 6 (17 ± 20)	10 ± 4 (4 ± 3)	9 ± 3 (13 ± 15)	(22 ± 15)	100 ± 36	4 ± 3(6 ± 6)	11 ± 4(11 ± 14)	9 ± 3(14 ± 17)	7 ± 5 (4 ± 3)	2 ± 1 (1 ± 1)	5 ± 4 (3 ± 2)
Ave. ± SD	SS	17 ± 5 (18 ± 15)	8 ± 3 (3 ± 3)	9 ± 2 (20 ± 24)	(15 ± 13)	49 ± 26	4 ± 3(8 ± 9)	14 ± 4(14 ± 14)	10 ± 515 ± 16	8 ± 4 (7 ± 8)	4 ± 8 (2 ± 3)	7 ± 3 (5 ± 5)

The bold numbers indicate the sequence variants that were identified by RainDance Technologies and SureSelect next-generation sequencing using filtered data (bold numbers) to minimize the false-positive rates. The filtering criteria were applied to RDT and SS data on the blinded and positive control samples. Square brackets indicate the column numbers. The numbers in parentheses are the number of variants identified using unfiltered RDT and SS data to confirm previously identified Sanger method variants in positive control samples only.

Ave., average; C, positive control samples; EX, exonic variants; False pos. (%), false-positive percentage; INT, intronic variants; N-SNP, non-SNP variants; N-SNP EX, non-SNP exonic variants; N-SNP INT, non-SNP intronic variants; W, wild-type normal sample.

^⁎Total number of variants identified following each enrichment method and next-generation sequencing.

^†Total number of variants identified by Sanger sequencing and missed by NGS because such data was filtered by the criteria described in Material and Methods.

^‡Total number of variants identified by next-generation sequencing that were confirmed by Sanger sequencing.

NGS identified a large number of sequence variants within our targeted regions of the controls with <100% concordance to Sanger sequencing results (Table 3). The NGS-identified variants of the positive control samples were generated by using the unfiltered RDT and SS data, so that it would be a true variant number comparison between the enrichment methods and Sanger sequencing. After a close examination of parameters such as coverage, allele percentage, and Phred-like score, the false-positive rate (Table 3) decreased significantly by eliminating variant calls with low probability of being true positives. On average, the false-positive rate of the two methods, RDT (22% ± 15%) and SS (15% ± 13%), was not statistically different. However, in several cases, the RDT false-positive rate was lower than the SS rate, such as in sample C14. By contrast, false-negative variants not detected by the RDT NGS were often in exons with poor sequence coverage (<20×) and high GC content. For example, exon 22: IVS22+4G>A, change in COL6A2 was not observed using the RDT data due to 0× coverage and 73% GC content (Tables 2 and 4). However, this change was confirmed by Sanger sequencing.

Table 4.

Detection of Variants and Mutations Following Target Enrichment

Patient No.	Gene	Exon/intron	Mutation	Amino acid change/consequence (frameshift/del)	Status (Homo/Het)	RDT detected	RDT mutation coverage	SS detected	SS mutation coverage	Sanger confirmed	Notes
C11	COL6A1	30	c.1931G>A	p.R644Q	Het	−	11	+	66	+	VUS
	COL6A2	23	c.1770G>A	p.T590	Het	+	8	−	20	+	VUS⁎
	COL6A2	28	c.2994C>T	p.H998	Het	+	20	+	42	+	VUS⁎
C12	COL6A1	29	IVS29-8G>A		Het	+	7	+	34	+	VUS
	FKTN	9	IVS9-40C>A		Het	−	−	−	−	+	VUS†
	LAMA2	14	c.2084C>T	p.D695V	Het	+	82	+	63	+	VUS
	LAMA2	39	c.5614G>T	p.D1872Y	Het	+	29	+	88	+	VUS
	SEPN1	5	IVS5+39C>T		Het	−	−	−	−	+	VUS†
	SEPN1	11	IVS11-31C>T	p.V549 mol/L	Het	−	−	−	−	+	VUS†
	SEPN1	13	c.1645G>A		Het	+	78	+	159	+	VUS
	SEPN1	3′UTR	c.1773+44G>T		Het	−	−	−	−	+	VUS†
C13	COL6A1	26	IVS26+50C>T		Het	−	−	−	−	+	VUS†
	COL6A1	33	c.2424G>T	p.Q808H	Het	+	8	+	17	+	VUS
	COL6A2	24	IVS24-3dupC		Het	−	8	+	162	+	VUS
	COL6A3	38	IVS38-34C>T		Homo	−	−	−	−	+	VUS†
C14	LAMA2	22	c.3154A>G	p.S1052G	Het	+	17	+	40	+	VUS
	LAMA2	47	c.6617delT	frameshift	Het	−	1	+	75	+	Del mut
C15	POMGNT1	7	c.636C>T	p.F212	Het	+	34	+	30	+	Splicing mut‡
	POMGNT1	17	IVS17+1G>A		Het	+	105	+	73	+	Splicing mut§
W16
1	COL6A3	14	IVS14-8del29	Del	Het	+	87	−	−	+	Del mut
3	COL6A2	22	IVS22+4G>A	p.H1337R	Het	+	44	+	137	+
	LAMA2	27	c.4010A>G		Het	+	10	+	32	+	Possibly damaging
7	COL6A3	1	c.53C>A	p.A18X	Homo	+	111	+	35	+	Nonsense mut
8	COL6A1	6	IVS6-18CC>T		Het	+	66	+	122	+
	LAMA2	24	c.3412G>A	p.V1138 mol/L	Het	+	66	+	132	+	Mut-HGMD
	LAMA2	43	IVS43+5G>C		Het	+	66	+	15	+
9	COL6A1	21	IVS21-2A>G		Het	−	−	−	−	+
	COL6A2	26	c.2039G>A	p.R680H	Het	+	8	+	22	+
	COL6A2	22	IVS22+4G>A		Het	−	−	+	26	+
	COL6A3	41	c.9206C>T	T3069I	Het	+	26	+	15	+
	LARGE	15	c.1949G>A	p.R650Q	Het	+	10	+	49	+
	SEPN1	12	c.1506C>A	p.N502K	Homo	+	26	+	17	+
10	COL6A1	14	IVS14+1G>A		Het	+	19	+	14	+	Splicing mut

Target enrichment achieved by RainDance Technologies and SureSelect in combination with next-generation sequencing. Out-of-frame deletion mutation predicted to result in a premature translation stop. This change is of the type predicted to cause disease.

+, present; −, not present; C, positive control samples; Del, deletion; Het, heterozygous change; Homo, homozygous change; MEB, muscle-eye-brain; Mut, mutation; UTR, untranslated region; VUS, variant of unknown clinical significance, pending functional analysis for reclassification; W, wild-type sample.

^⁎It is possible for silent changes to disrupt RNA splicing.

^†Mutation and/or variant of unknown clinical significance not detected because the bioinformative algorithm set to detect ±20 bases from exon/intron boundaries.

^‡This mutation has been reported in individuals with MEB disease.

^§This mutation results in a G to A change in the consensus donor site of the exon 17/intron 17 boundary and is predicted to result in aberrant splicing of POMGNT1 RNA. This mutation has been reported in individuals with MEB disease.

High coverage along with high uniformity and specificity render target enrichment methods suitable for the reliable detection of different types of sequence variants in positive control samples. RDT and SS in combination with NGS correctly identified most variants of unknown clinical significance and mutations in the positive control samples (C11 to C15), except for changes whose coverage was too low (Table 4 and Figure 2, A and B). Deep intronic variants were not detected because we limited the custom set of sequence analysis to ±20 bp on either side of each exon. Examples of mutations detected by all three methods include a POMGNT1splice site change in sample C15, one copy of IVS17+1G>A in intron 17 (Figure 2A and Table 4), and a missense change detected in sample C12 LAMA2, one copy of c.2084C>T (p.D695V) in exon 14 (Figure 2B). By contrast, changes that were found in one data set but not the other include: c.1931G>A (p.R644Q) in IVS24-3dupC in COL6A1 and c.6617delT in LAMA2 not detected in RDT samples C11 and C14, respectively, but that were found in the corresponding SS samples. Conversely, one change, c.1770G>A (p.T590) in exon 23 of COL6A2, was identified in the RDT data of sample C11, but not in the SS data of the same sample. The missed changes in the RDT data may be explained by the coverage being too low at these sites, which is a likely consequence of high GC content.

Figure 2.

Types of variants detected in RainDance and SureSelect samples that were Sanger sequence confirmed. A: Example of splice site mutation of sample C15. One copy of an IVS17+1G>A splice site mutation in intron 17 of POMGNT1 was detected by Sanger sequencing, RainDance, and SureSelect data. Arrows, splice site mutation. B: Example of missense mutation detected in sample C12. One copy of c.2084C>T (p.D695V) missense mutation in LAMA2 was observed in Sanger sequencing, RainDance, and SureSelect data. Arrows, missense mutation.

Variant Types Detected by NGS in Blinded Samples

RDT and SS in combination with NGS permitted the detection of different types of variants in the six blinded control samples, and a significant proportion of such variants were confirmed by Sanger sequencing (Table 3; see Materials and Methods for filtering data criteria). The types of identified variants were exonic and intronic variants and SNP and non-SNP variants. Examples of missense, deletion, nonsense, splice site alteration, and duplication changes that we found in our data are c.4010A>G (p.H1337R) in LAMA2 (sample 3), IVS14-8del29) in COL6A3 (sample 1), c.53C>A (p.A18X) in COL6A3 (sample 7), IVS14+1G>A in COL6A1 (sample 10), and IVS24-3dupC in COL6A2 (sample C14). The total number of RDT variant calls in the blinded control samples varied from 6 to 24, and approximately 60% of the calls were Sanger sequence confirmed. Similarly, SS variants were in the range of 9 to 23, and approximately 60% were also Sanger sequence confirmed. On average, 10 and 8 variants were not detected in the RDT and SS data, respectively. Many of these variants were in problematic regions that had low exon coverage due to the high GC content, as indicated by the number of exons with <20× coverage (Tables 2 and 3). In general, NGS successfully identified silent, missense, nonsense, splice site, deletion, and duplication changes in positive and blinded samples using RDT and SS data. Deletions of up to 30 bp have been detected by NGS.

Comparison of Enrichment Methods

Comparison of the RDT and SS enrichment methods for SOLiD sequencing, and Sanger sequencing demonstrated essential similarities and differences (Table 5). RDT offers the lowest cost per amplicon, which is significantly less than SS and Sanger sequencing. Though RDT requires specialized equipment for the enrichment, its advantage is the ability to offer automation for the enrichment process. In addition, the DNA requirement for a 65-kb target interval is 3 μg for RDT and SS, but it is much higher for Sanger sequencing (∼19 μg). RDT is much better than Sanger sequencing in terms of being able to process more samples per day (eight vs. one). The RDT throughput is similar to that of SS; however, the enrichment step of the RDT process is automated. Recently, the automated Agilent NGS Sample Preparation Workstation was developed to streamline the sample preparation of the targeted next-generation sequencing workflow, which greatly improves the throughput and reduces the hands-on time. RDT also has the advantage over SS in its ability to distinguish between a gene and its pseudogene targets during the enrichment process.

Table 5.

A comparison of Two Enrichment Methodologies, RainDance Technologies and SureSelect, Relative to the Sequencing Gold Standard, the Sanger Method

Parameters	RainDance Technologies	SureSelect	Sanger
Clinical implementation	Yes	Yes	Yes
Cost per PCR amplicon	$1.56	$2.34	$5.48
Equipment	Use of specialized equipment, automation	Use of specialized equipment, automation or manual execution	None required
Ease of use	Easier, small number of steps	Complicated, many manipulations	Labor intensive
DNA requirement	3 μg for 1 Mb target	3 μg for up to 30 Mb target	19 μg for 65 Kb region
Length of region for enrichment	Up to 1 Mb	Up to exome	Most limited
Scalability (samples/day)	One patient per time (8 samples/day)	Eight patients per time (8 samples/day)	One exon per time (1 sample/day)
Variant calls	>20× coverage needed	>20× coverage needed	2×
Limitation	May be used for genes with pseudogenes	Cannot be used for pseudogenes	May be used for genes with pseudogenes
Analytical sensitivity	>85% (95% CI: 72–88%)	>85% (95% CI: 77%–91%)	>99.5% (95% CI: 95.6%–100%)
Analytical specificity	>99.5% (95% CI: 99.9%–100%)	85–99.5%⁎ (95% CI: 79.9%–99.9%)	>99.5% (95% CI: 99.9%–100%)
Uniformity	>85% uniform amplification	>85% uniformity capture	>95% uniform amplification
Reproducibility	>99% reproducibility at 10-fold between two samples	>85% reproducibility at 10-fold between two samples†	100% reproducibility at 2-fold between two samples

^⁎The specificity variation depends on which exons are captured in a given experiment. Exon capture variability has been observed in SureSelect. Therefore, RainDance Technologies amplifications are more reproducible than SureSelect captures.

^†The average exon coverage profile of each gene varies between SureSelect capture experiments, whereas the RainDance Technologies average exon coverage profile is much more consistent.

The analytical sensitivities of RDT and SS (>85%) are similar to each other, but lower than Sanger sequencing (>99.5%) (Table 5), whereas the analytical specificities of RDT and Sanger (>99.5%) sequencing is similar, but lower for SS (85% to 95.5%). RDT and Sanger displayed significantly higher reproducibility than SS. When two samples are compared to each other, the RDT data demonstrated >99% variant call identity, and SS was lower at >85%.

Discussion

In recent years, studies have shown that target enrichment in combination with NGS holds out the promise of becoming a useful diagnostic tool for the detection of mutations in families with complex monogenic disorders^10,23–26; however, the performance of this technology has to be addressed before implementing it in a clinical and diagnostic laboratory setting. Here, we evaluate strengths and limitations, including the ability to identify different types of sequence variations and the diagnostic potential, of two enrichment technologies, RDT and SS, in a cohort of patients presenting with congenital muscular dystrophies, a clinical classification with heterogeneous genetic causes (Tables 1 and 3, numbers in parentheses; Table 3, bold numbers). To this end, a wild-type control, which had all 12 CMD genes Sanger sequenced, was included to serve as a normal control reference, along with five positive control samples blinded to laboratory staff and six blinded samples with clinical features of the CMDs.

The percentage of reads successfully mapped to the human genome was similar between RDT and SS, but a higher percentage of RDT reads mapped to the target region in comparison to the SS reads (Table 1). The target region–mapped percentages are consistent with other reports.^1,17,22,27 Thus, this demonstrates that specificity is lower in SS samples. The off-target reads in RDT samples may be explained by nonspecific sequence amplification and contaminating traces of genomic DNA. High specificity and uniformity has been shown for RDT between different amplicons by a study that was investigating new disease-causing mutations in X-linked intellectual disability genes.¹ Compared to other hybridization methods, RDT is better suited for enriching for short neighboring exons.¹ PCR can be optimized to amplify target regions, whereas hybridization approaches will have to carry adjacent sequences together with real ones, thereby reducing specificity and lowering target percentages (Table 1). In addition to its specificity, RDT has less allelic bias, since most alleles can be equally amplified. For a simple two-allele example, the ability to see both alleles is attributed to the ability of RDT to represent less than a haploid copy of the genome within each PCR reaction. This results in RDT having the ability to achieve single-molecule PCR, which is not possible with any other target enrichment strategy, including other PCR-based approaches. The result of having limiting amounts of genomic DNA is that when an allele is rare, RDT will have many singleplex PCR reactions per amplicon, allowing the rare allele to be present within one PCR reaction in which it is represented as the only target, letting us mitigate allelic competition. In contrast, selection by hybridization will capture less of an allele that is significantly different from the reference, and this allelic bias can increase false-negative rates.¹ Moreover, the target mapped reads are lower in the SS samples compared with the RDT samples, because there may be other 100% identical matches to off-target regions on other chromosomes, which resulted in the reads to these exons not being mapped by the read alignment programs.²²

The target sequence complexity has a strong effect on the efficiency of DNA amplification and capture for individual exons. Even though the mean gene coverage/read depth was similar within and between RDT and SS samples, all exons of POMT1 were not amplified by RDT or captured by SS, and thus a low coverage was observed (Figure 1). Generally, a high mean gene read depth was observed for most exons of most genes, with the exception of POMT1. However, in many samples, the first exon was problematic and typically had a low average coverage in both RDT and SS runs (see Supplemental Figures S1–S12 at http://jmd.amjpathol.org and Table 2). High GC content may explain the low coverage in the first exons of genes, given that the mean GC content of the first coding exon of all CMD genes is 64%. Also, the exons of POMT1 have an average percent GC content of 56%, and several failed exons have a percent GC content >60% (Table 2). For example, exon 1 of SEPN1, COL6A1, and LAMA2 had the highest GC contents, with percentages of 87, 73, and 71, respectively (Table 2). In addition, most exons that dropped out in the RDT samples also had a high GC content, which may hinder PCR amplification and could explain the absence of sequence data and lack of variant calls in these regions. Moreover, the negative effect of high GC content, as seen in our study for the exons with very low or null coverage, is also consistent with previous reports.^1,22,28 In contrast to RDT, SS is also sensitive to sample base composition, and sequences at the extremes of high GC/AT content can be lost through poor annealing and secondary structure, respectively.²⁸ Another consideration is that it is seldom possible to capture all of a desired target region in a hybrid capture experiment; targets are generally subjected to repeat masking (see Materials and Methods) before probe design to avoid capture of homologous repetitive elements. For exonic targets, <5% to 15% of the primary target region can be lost in this way, leaving a region to which probes could be designed after repeat masking, or target capture region, that constitutes >85% to 95% of the primary target region. For contiguous regions, the percentage of primary target region that is represented in the capture target region is generally lower (∼50% to 65%), but this is highly variable between regions (Table 1).

Relative to RDT, SS capture–based enrichment had a smaller number of missed exons, which may be explained by our use of 120-bp probes due to their length and the 20× tiling redundancy that ensured capturing a target even if another probe missed it. The tiling redundancy would not be an advantage if larger genomic regions were to be targeted, since the most probes allowed is 55,000 per design. This means that RDT and SS enrichment technologies both have the potential for clinical use; however, Sanger sequencing will still be needed to confirm complex exons that are inefficiently amplified or captured. At this time, diagnostic applications of enrichment methods with NGS are complementary to Sanger sequencing efforts until they achieve high enough coverage to account for all possible variants at a given target region.

Allelic dropout due to SNPs in the PCR primer binding sites is a limitation inherent to all PCR-based assays, including Sanger sequencing.²⁹ RDT uses a library of primers to amplify the target regions and is therefore also susceptible to allele dropout if specific SNPs are in the primer binding sites. To minimize the absence of amplification of specific exons due to allele dropout, primers were designed in regions where SNPs have not been reported by using the Single Nucleotide Polymorphism and the 1000 Genomes databases. In the event that allele dropout occurs in RDT-targeted exons, sequence coverage cannot overcome allele dropout because there will be an absence of amplification, and such products consequently would generate zero to low sequence reads. By contrast, SS is a hybridization-based technology that relies on 120-bp probes to capture the region of interest. In this case, SS is likely to be less susceptible to allele dropout, meaning that a particular exon of interest will not be captured, because one SNP out of 120 bp may not be enough to affect the binding of the SS probe to its target. Also, to further decrease the chance for allele dropout, a 20× capture-probe tiling frequency and a 20-bp allowed overlap was used in the probe design. Even though the likelihood of allele dropout happening in an SS experiment is low, it is possible that the probes will not capture the targets of interest. Sequence coverage cannot overcome allele dropout, because no targets would be available to sequence. Although both technologies may be susceptible to allele dropout, the aforementioned considerations were taken into account to minimize such events. Furthermore, the importance of noting the clinical phenotype in the diagnosis of patients should be kept in mind to decrease the false-negative cases even further.

In assessing the validity of variants identified by NGS following RDT and SS enrichment, we saw that there was a deviation of NGS-identified variants in the targeted regions relative to Sanger sequencing results (Table 3). The analytical sensitivity of RDT and SS was lower than that of Sanger sequencing: >85%, >85%, and >99.5%, respectively (Tables 4 and 5). However, in positive control samples, RDT in combination with SS data was sufficient to correctly identify most variants of unknown clinical significance and mutations, with the exception of deep intronic variants and variants whose coverage was too low (Figure 2, A and B, and Table 4). The false-positive rates, possibly caused by inefficiency of amplification and artifacts of sequencing, decreased after investigators learned that the coverage levels, allele percentages, and Phred-like scores in combination were necessary to reach a threshold of <15% (Table 3).

NGS permitted different types of variants to be detected following RDT and SS enrichment in the six blinded control samples, including silent, missense, nonsense, small deletion, and small duplication changes (confirmed by Sanger sequencing; Table 4). The range of the total number of RDT variants was between 6 and 24, compared with SS variant calls of 9 to 23. However, on average, approximately 60% of such variants were confirmed by Sanger sequencing because of low exon coverage in some regions, as indicated by the number of exons with <20× coverage (Tables 2 and 3).

Several parameters are worth comparing between the two enrichment technologies and contrasting them to Sanger sequencing; namely, clinical implementation, cost, requirement for specialized equipment, ease of use, analytical sensitivity and specificity, and scalability (Table 5). RDT offers the lowest enrichment cost per amplicon when compared to SS and Sanger sequencing. An advantage of RDT requiring specialized equipment for the enrichment is that it allows full enrichment automation. By contrast, the manual SS procedure is more complicated and Sanger sequencing is labor intensive, since one must amplify and sequence each amplicon. However, an automated platform is now available for the SS enrichment method to fully automate the library construction steps. The RDT and SS DNA requirements are similar, but the Sanger method requires a significantly higher amount of DNA. RDT is more appropriate for situations requiring multiple genes with long exons because it can enrich a target interval up to 1 Mb. Up to eight samples can be processed per day. Similarly, SS can be used to process eight patients per day, but it can enrich a much larger target interval, up to the entire exome. By contrast, only one patient sample may be processed by Sanger sequencing in a day because it is such a large panel (383 amplicons). One important diagnostic issue is the ability to distinguish between gene and pseudogene targets; RDT and Sanger can readily address this issue by correctly choosing locations where the primers are to hybridize on the genomic DNA template. Since SS is a hybridization-based method, its limitation is not being able to distinguish between the gene and pseudogene targets. Therefore, RDT is most appropriate for the CMD panel because of the lower cost, ease of use, the length of the target interval that is appropriate (65 Kbp), the ability to distinguish between genes and pseudogenes, and the ability to process eight samples per day.

From an assay validation perspective, the analytical sensitivities of RDT and SS are similar, but lower than Sanger sequencing (>95%). The lower analytical sensitivity may be explained by the lower coverage of specific exons. However, the analytical specificities of RDT and Sanger were higher than SS. Furthermore, RDT and Sanger displayed a significantly higher reproducibility than SS, as shown by the comparison of two samples whose variants were tallied. For example, RDT data showed a >99% variant call identity between the two samples, whereas the SS was >85%. The analytical specificity of all three methods taken together approached 100%. As part of a diagnostic plan, exons with low coverage get reflexed to Sanger sequencing to ensure that no variants are missed. Therefore, RDT with Sanger sequencing will continue to have the analytical sensitivity of >99% by virtue of using Sanger sequencing for low-coverage exons.

Conclusions

Current single-gene approaches to identify mutations in patient samples do not offer the throughput and ease of use to screen the multiple genes often associated with many medical genetic disorders, such as CMDs. Targeted sequencing paired with NGS offers the first opportunity to effectively screen the complete coding regions for a panel of genes in a single experiment. The two targeted sequencing approaches evaluated in this study (RDT and SS) demonstrated the ability to quickly and accurately allow clinicians to simultaneously test a panel of 12 genes associated with CMDs. Both RDT and SS enrichment technologies proved suitable for use in a clinical laboratory setting (Table 5). On the basis of our findings, the RDT microdroplet-based PCR approach to targeted sequencing stands out as the appropriate solution for a clinical laboratory. Irrespective of the enrichment method used, some exons in highly repetitive and GC-rich regions are difficult to target with both of these approaches and will still require traditional Sanger sequencing. Our results support the notion that targeted molecular diagnostics of heterogeneous genetic disorders is now a reality. The adoption of a targeted sequencing approach in a clinical genetics laboratory will pave the way for a significant improvement in the diagnosis of heterogeneous genetic disorders and improve our understanding of disease genes.

Supplementary data

Supplemental Figure S1

RainDance and SureSelect average exon coverage for COL6A1.

Supplemental Figure S2

RainDance and SureSelect average exon coverage for LAMA2.

Supplemental Figure S3

RainDance and SureSelect average exon coverage for POMT1.

Supplemental Figure S4

RainDance and SureSelect average exon coverage for SEPN1.

Supplemental Figure S5

RainDance and SureSelect average exon coverage for COL6A2.

Supplemental Figure S6

RainDance and SureSelect average exon coverage for COL6A3.

Supplemental Figure S7

RainDance and SureSelect average exon coverage for FKTN.

Supplemental Figure S8

RainDance and SureSelect average exon coverage for FKRP.

Supplemental Figure S9

RainDance and SureSelect average exon coverage for ITGA7.

Supplemental Figure S10

RainDance and SureSelect average exon coverage for LARGE.

Supplemental Figure S11

RainDance and SureSelect average exon coverage for POMGNT1.

Supplemental Figure S12

RainDance and SureSelect average exon coverage for POMT2.