Molecular Diagnosis of Autosomal Dominant Polycystic Kidney Disease Using Next-Generation Sequencing

Abstract

Autosomal dominant polycystic kidney disease (ADPKD) is caused by mutations in PKD1 and PKD2. However, genetic analysis is complicated by six PKD1 pseudogenes, large gene sizes, and allelic heterogeneity. We developed a new clinical assay for PKD gene analysis using paired-end next-generation sequencing (NGS) by multiplexing individually bar-coded long-range PCR libraries and analyzing them in one Illumina MiSeq flow cell. The data analysis pipeline has been optimized and automated with Unix shell scripts to accommodate variant calls. This approach was validated using a cohort of 25 patients with ADPKD previously analyzed by Sanger sequencing. A total of 250 genetic variants were identified by NGS, spanning the entire exonic and adjacent intronic regions of PKD1 and PKD2, including all 16 pathogenic mutations. In addition, we identified three novel mutations in a mutation-negative cohort of 24 patients with ADPKD previously analyzed by Sanger sequencing. This NGS method achieved sensitivity of 99.2% (95% CI, 96.8%–99.9%) and specificity of 99.9% (95% CI, 99.7%–100.0%), with cost and turnaround time reduced by as much as 70%. Prospective NGS analysis of 25 patients with ADPKD demonstrated a detection rate comparable with Sanger standards. In conclusion, the NGS method was superior to Sanger sequencing for detecting PKD gene mutations, achieving high sensitivity and improved gene coverage. These characteristics suggest that NGS would be an appropriate new standard for clinical genetic testing of ADPKD.

CME Accreditation Statement: This activity (“JMD 2014 CME Program in Molecular Diagnostics”) has been planned and implemented in accordance with the Essential Areas and policies of the Accreditation Council for Continuing Medical Education (ACCME) through the joint sponsorship of the American Society for Clinical Pathology (ASCP) and the American Society for Investigative Pathology (ASIP). ASCP is accredited by the ACCME to provide continuing medical education for physicians.

The ASCP designates this journal-based CME activity (“JMD 2014 CME Program in Molecular Diagnostics”) for a maximum of 48 AMA PRA Category 1 Credit(s)™. Physicians should only claim credit commensurate with the extent of their participation in the activity.

CME Disclosures: The authors of this article and the planning committee members and staff have no relevant financial relationships with commercial interests to disclose.

Autosomal dominant polycystic kidney disease (ADPKD) affects 1 in 400 to 1 in 1000 live births worldwide.¹ It is the most common inherited kidney disease, accounting for approximately 5% of the end-stage renal disease population.² ADPKD is initiated by gene mutations in renal tubular epithelial cells, which seem to be more sensitive to haploinsufficiency, resulting in increased proliferation and cyst formation.³ The consequent increase in the number and size of kidney cysts causes progressive chronic kidney disease.⁴ ADPKD is mainly caused by mutations in two large genes, PKD1 and PKD2, accounting for 75% to 85% and 15% to 25% of cases, respectively, in clinically well-characterized populations. PKD1 spans 46 exons and encodes polycystin-1 with 4303 amino acids.⁵ PKD2 spans 15 exons, encoding polycystin-2, which consists of 968 amino acids.⁶ Chromosome 16 includes six homologous genes (ie, pseudogenes) that share 97.7% sequence identity with the PKD1 gene exons 1 to 33.^7,8

The clinical diagnosis of ADPKD is established by family history and renal imaging modalities, such as ultrasonography, computed tomography, and magnetic resonance imaging.⁹ However, these diagnostic test results are often ambiguous, particularly in young individuals. Consequently, genetic testing plays an increasingly important role in the diagnosis and management of patients with ADPKD.¹⁰ Moreover, with the development of potentially effective pharmacologic treatments for ADPKD,¹¹ the need for accurate diagnostic genetic tests has become more compelling.

The key step in ADPKD genotyping procedures is amplification of the PKD1 gene region while excluding the pseudogenes. This was traditionally achieved by using long-range PCR (LR-PCR) with primers located to the rare mismatch sites that distinguish PKD1 and the pseudogenes, followed by nested PCR of the individual exons, whereas the single-copy regions of PKD1 and PKD2 were directly amplified from genomic DNA. Amplicons were then directly analyzed by Sanger sequencing or by sequencing coupled with a mutation screening step to lower the testing cost.^12,13 However, the genetic analysis of ADPKD is challenging, especially owing to the large size, complex genomic structure, and allelic heterogeneity of PKD1 and PKD2.¹⁴ Next-generation sequencing (NGS) technology has revolutionized the field of human genetics and molecular diagnostics.^15,16 Recently, Rossetti et al¹⁷ reported a mutation screening strategy for analyzing PKD genes using NGS by pooling LR-PCR amplicons and multiplexing bar-coded libraries. This approach was designed to have a high throughput and has been successful for screening mutations in large cohorts. However, the method had low sensitivity and slow turnaround time, mainly because of the sample pooling strategy used.

Herein, we present a new NGS-based genotyping approach for patients with ADPKD that is better tailored to the standard clinical diagnostic setting, where rapid turnaround time and high sensitivity could be achieved by individually bar coding each patient in the run. The diagnostic performance of the new assay was evaluated using a panel of DNA samples previously analyzed by Sanger sequencing.¹⁸ The testing strategy, workflow, data analysis pipeline, costs, and other related issues are also discussed.

Materials and Methods

Study Patients

Study patients were participants in The Rogosin Institute Polycystic Kidney Disease Data Repository (http://www.clinicaltrials.gov, Identifier NCT00792155). This is a single-center, prospective, longitudinal study of genotype and phenotype characteristics of individuals with ADPKD. Study samples were randomly selected for analysis. All the participants underwent PKD genotyping by the Weill Cornell Medical College Molecular Pathology Research Laboratory (New York, NY) using direct sequencing or SURVEYOR nuclease–WAVE screening (Transgenomic Inc., Omaha, NE). In addition, we prospectively analyzed a new cohort of patients with ADPKD that has not been previously genotyped. The study was approved by the Institutional Review Board Committees at Weill Cornell Medical College and The Rockefeller University (New York, NY). All the participants provided written informed consent.

Long-Range PCR

Genomic DNA was extracted from peripheral blood lymphocytes using a Gentra Puregene blood kit (Qiagen Inc., Valencia, CA). The entire coding region, the exon-intron boundaries, and most of the 5′ and 3′ untranslated regions of PKD1 and PKD2 were amplified in a total of 10 (five reactions per gene) distinct PCR reactions using primers anchored either in the rare mismatched region with the human homologs or in the single-copy region of PKD1. The LR-PCR primers were designed using Primer3 software version 4.0.0 (http://bioinfo.ut.ee/primer3, last accessed October 25, 2013) (Table 1).¹⁹ The LR-PCR primers were modified at the 5′ end with NH2 to prevent overrepresentation of sequences at the amplicon ends in the ligation step and to increase sequence coverage uniformity.²⁰ LR-PCR was performed using either the GeneAmp high fidelity PCR system (Applied Biosystems, Foster City, CA) or the PrimeSTAR GXL DNA polymerase (Takara Bio Inc., Shiga, Japan). PCR amplification conditions for the various LR-PCR fragments are described in Supplemental Table S1. After purification with Agencourt AMPure XP beads (Beckman Coulter, Beverly, MA), the LR-PCR fragments from each patient were quantified using PicoGreen (Quant-iT; Invitrogen, Carlsbad, CA) and were pooled together at equal molar ratio.

Table 1.

LR-PCR Primers for NGS Sequencing

Fragment	Primers	Sequence	Size (kb)	Genomic location
PKD1_Ex1	PKD1_NGS_1F	5′-CGCAGCCTTACCATCCACCT-3′	2.3	chr16:2185030-2187307
	PKD1_NGS_1R	5′-TCATCGCCCCTTCCTAAGCA-3′
PKD1_Ex2-12	PKD1_NGS_2-12F	5′-CCAGCTCTCTGTCTACTCACCTCCGCATC-3′	8.7	chr16:2163080-2171636
	PKD1_NGS_2-12R	5′-CCACGGTTACGTTGTAGTTCACGGTGACG-3′
PKD1_Ex13-21	PKD1_NGS_13-21F	5′-TGGAGGGAGGGACGCCAATC-3′	7.9	chr16:2155145-2163036
	PKD1_NGS_13-21R	5′-ACACAGGACAGAACGGCTGAGGCTA-3′
PKD1_Ex22-34	PKD1_NGS_22-34F	5′-ATGCTTAGTGAGGAGGCTGTGGGGGTC-3′	7.8	chr16:2146980-2154794
	PKD1_NGS_22-34R	5′-ATGAGGCTCTTTCCACAGACAACAGAGGTT-3′
PKD1_Ex35-46	PKD1_NGS_35-46F	5′-CTGTGGGCGATGGGTTTATCAGCAG-3′	5.2	chr16:2139301-2144473
	PKD1_NGS_35-46R	5′-GAGACGGTGCAGGGAGTACGGTAGGA-3′
PKD2_Ex1	PKD2_NGS_ExP-1_F	5′-GTGGAGACAGAAGCCAACCAAAGAG-3′	1.4	chr4:88928226-88929584
	PKD2_NGS_ExP-1_R	5′-GGATGCGAGATGGAGCCCG-3′
PKD2_Ex2	PKD2_NGS_Exon2_F	5′-TTTCTTTCCATTTGCAATGTTTCATTC-3′	2.5	chr4:88938491-88940897
	PKD2_NGS_Exon2_R	5′-GGAAGATAGTCAATAAACAAATGCCCAA-3′
PKD2_Ex3-6	PKD2_NGS_Ex3-6_F	5′-GAGAAGACCTTGTGTGAATTTGTCCA-3′	10.9	chr4:88957246-88968207
	PKD2_NGS_Ex3-6_R	5′-TCATACTCAGCAAAGTTACTCATGCAAA-3′
PKD2_Ex7-10	PKD2_NGS_Ex7-10_F	5′-TCGGGTAAGTATAATGGTGAGCCCT-3′	10.3	chr4:88973018-88983323
	PKD2_NGS_Ex7-10_R	5′-CATCAAGACTCCAAGATAGGGAACATTT-3′
PKD2_Ex11-15	PKD2_NGS_Ex11-15_F	5′-CACGTACTTGTTGAATGGCCAATGT-3′	10.8	chr4:88986401-88997197
	PKD2_NGS_Ex11-15_R	5′-ATGAAACTCAGAAGCCCTTTGACAGTT-3′

Library Preparation and Indexing

For each patient, 4 μg of LR-PCR products were pooled together in a total reaction volume of 210 μL of Tris-EDTA buffer and were fragmented to approximately 300 bp using adaptive focused acoustics (Covaris S2; LGC Ltd., Teddington, UK) with the following settings: duty cycle, 20%; intensity, 5; and cycles per burst, 200. After shearing, the fragments underwent end repair using the NEBNext end repair module (New England BioLabs Inc., Ipswich, MA) by adding 3′ dA overhangs to the blunt-ended DNA. After purification on AMPure XP beads, unique indexed adaptors were ligated to each patient pool using the NEBNext quick ligation module (New England BioLabs Inc.). Twenty-five different adapters with 5-nt bar codes were used for indexing and library preparation. The bar codes located in the 3′ end of each adapter were designed using a published Python script (create_index_sequences.py)²¹ with a minimum edit distance, or mutation tolerance, of three. Each adapter sequence began with the paired-read oligonucleotide sequences as specified by Illumina Inc. (San Diego, CA): 5′-GATCGGAAGAGCGGTTCAGCAGGAATGCCGAG-3′ and 5′-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′. Annealing of the indexed adaptors was performed at 95°C for 2 minutes, followed by a cooldown to room temperature at a rate of 0.1°C per second using a thermal cycler (Biometra GmbH, Goettingen, Germany). The ligation products were then purified with AMPure XP beads and subjected to size selection using 2% E-Gel SizeSelect (Invitrogen). The selected 500-bp libraries were then amplified by PCR using AccuPrime Taq high fidelity (Invitrogen) with the PCR primers (Illumina Inc.) 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′ and 5′-CAAGCAGAAGACGGCATACGAGATCGGTCTCGGCATTCCTGCTGAACCGCTCTTCCGATCT-3′ and the following PCR conditions: 98°C for 3 minutes, 10 cycles at 98°C (80 seconds), and 65°C (90 seconds), followed by a final extension step at 65°C for 10 minutes using a thermal cycler (Biometra GmbH).

Library Pooling and Illumina Sequencing

PCR library products from each patient were purified with AMPure XP beads, quantified, and pooled together at equimolar amounts in groups of 25 samples. The finished libraries were quantified using the PicoGreen method (Invitrogen) and were analyzed using the DNA high-sensitivity chip on an Agilent 2100 Bioanalyzer system (Agilent Technologies Inc., Santa Clara, CA) for quality control purposes and for assessing the library size. Each library pool was then diluted to 10 nmol/L and was loaded onto a flow cell of the MiSeq system (Illumina Inc.) and subjected to cluster generation and sequencing using a paired-end 150-bp cycle protocol according to the manufacturer's instructions. Typically, this read length enables the detection of indels of approximately 1 to 51 bp using the computational pipeline. An average MiSeq run generates 6.14 million reads with 5.74 million pass filter reads, and 77.1% of reads have a high quality score >Q30.

Data Analysis Pipeline

For data analysis, Illumina sequencing reads in FASTQ format were first subjected to quality control checks using the FastQC program (http://www.bioinformatics.babraham.ac.uk/projects/fastqc, last accessed October 25, 2013). The reads were then sorted according to their bar code by FastqMultx²² (https://code.google.com/p/ea-utils/wiki/FastqMultx, last accessed November 3, 2013) or the FASTQ/A barcode splitter program in the FASTX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/index.html, last accessed October 25, 2013). Reads were then paired-end aligned using the Burrows-Wheeler Aligner (BWA) program version 0.5.9rc1,²³ to a modified version of the human genome assembly hg19, where all the nucleotides outside the PKD1/PKD2 loci are masked and replaced with Ns. This procedure has the advantage of maintaining the genomic coordinates of the variants and enabling subsequent variant annotation using standard NGS analysis software (eg, ANNOVAR). Because the PKD1 pseudogenes are effectively masked, reads cannot be erroneously mapped to these genomic regions. Sequence variants were called using the Genome Analysis Toolkit (GATK) software package version 1.6 (Broad Institute, Cambridge, MA),²⁴ carefully following the best practice guidelines recommended by GATK,²⁵ including initial read mapping, local realignment particularly around indels, followed by base quality score recalibration. Single nucleotide variations (SNVs) and INDELs were called simultaneously on all 25 samples with the default setting of GATK Unified Genotyper on the realigned and recalibrated reads, followed by SNV and INDEL filtering to eliminate false-positive calls. The GATK uses the Phred scaled probability that a reference/alternative (ALT) polymorphism exists at a given site (given sequencing data) as the main metric for calling polymorphic sites. This metric is called QUAL in the GATK output, and we require that QUAL ≥50 together with additional filters based to define a site as PASS. The GATK command line we used is shown in Supplemental Table S2. For each patient, genotyping likelihood was then defined for the AA, A/B, and BB genotypes, and A/B and BB calls were used as variant calls. We did not specify a percentage of ALT allele threshold (defined as second number allelic depth tag in the VCF file divided by depth of coverage); however, we determined in postvariant calling analyses that the percentage of ALT allele was >12% for all variants called. The minimum number of reads supporting the ALT allele for the called variants was 10×. The minimum depth of coverage for the variants was 39×. Given that we obtained excellent sensitivity and specificity using current parameters, as noted in Results, we did not introduce any additional filters based on ALT allele percentage but continue to rely on GATK genotype likelihoods. A variant call was filtered out with any of the following criteria met: i) SNVs in clusters (three SNVs within 10 bp of each other), ii) more than four reads with mapping quality of zero and >10% of reads with mapping quality of zero, iii) strand bias higher than or equal to −1.0, iv) SNV quality score <30, v) quality-by-depth score <1.5, vi) largest contiguous homopolymer run of variant allele >10, and vii) depth of coverage less than fivefold. After filtering, variant calls were annotated using ANNOVAR software version 2012,²⁶ and the final genotyping reports were then generated. The entire data analysis pipeline was automated using Unix shell scripts, with raw lllumina sequencing as input and final genotyping reports as output. In the present study, we restricted analysis to the coding exons and 20 bp of the flanking intronic regions. Statistical computations were performed using R version 3.0.1 software (The R Foundation for Statistical Computing, http://www.r-project.org).

Assay Analytical Characteristics

An evaluation of the assay analytical characteristics was performed using only variants located in regions analyzed by both NGS and Sanger sequencing and compared herein. Positive results were defined as all variant alleles detected by NGS in the 25-patient cohort. Negative results were defined as genotypes identical to the reference sequence at genomic sites where variants are found across the 25 patients. The new assay was evaluated for sensitivity and specificity using the following formulas:

Sensitivity = number of true-positives/(number of true-positives + number of false-negatives),

Specificity = number of true-negatives/(number of true-negatives + number of false-positives).

Reference Sequences, Variant Nomenclature, and Variant of Uncertain Significance Scoring

NCBI RefSeq sequences were used for reference sequence: PKD1, NM_000296.3; PKD2, NM_000279.3. The standard nomenclature recommended by the Human Genome Variation Society (http://www.hgvs.org/mutnomen, last accessed October 25, 2013)²⁷ was used to number nucleotides and name mutations or variants. All sequence variant descriptions were checked for accuracy using the Mutalyzer 2.0 program (http://www.mutalyzer.nl/2.0, last accessed October 25, 2013). Variants of uncertain significance were classified based on computational analysis scores as previously described.¹⁸

Results

LR-PCR–Based NGS Analysis of Pooled Control Samples

To improve ADPKD testing and eliminate the need for Sanger sequencing, we used an LR-PCR NGS strategy to specifically amplify and directly sequence the entire coding region of both PKD1 and PKD2 genes for up to 25 patients in a single sequencing reaction. An overview of the NGS workflow is shown in Figure 1, which includes a diagram of the PKD genes and the relative primer pair positions, LR-PCR enrichment and product analysis, and data analysis steps. The unique LR-PCR oligonucleotides have been carefully designed to cover approximately 67.8 kb of genomic sequence, including all exonic sequences and flanking intronic regions (see Materials and Methods). Agarose gel electrophoresis with ethidium bromide staining of the LR-PCR products demonstrated specific fragments ranging in size from 1.4 to 10.8 kb, each covering 1 to 11 exons corresponding to the sequence of PKD1 and PKD2 (Figure 1). For evaluating the NGS workflow, we pooled individually bar-coded sequencing libraries from up to 25 patients onto a single flow cell of the MiSeq system (Illumina Inc.) and subjected it to cluster generation and sequencing using paired-end sequencing. We analyzed the sequencing results using an automated NGS data analysis pipeline combining the FastQC, FASTX-Toolkit, BWA, GATK, and ANNOVAR software packages, as illustrated in Figure 2. The Sanger sequencing–verified gene variations allowed a detailed analysis of read depth (number of reads per known variant), coverage (percentage of the regions of interest adequately covered), sensitivity (proportion of true-positives), and specificity (proportion of true-negatives).

Figure 1.

Visualization of the NGS workflow. PKD1 and PKD2 genes were individually amplified as 10 locus-specific LR-PCR products (1.4 to 10.9 kb in size), with all coding regions and most intronic regions covered, in total, an approximately 68.0 kb genomic region. A: Map of the PKD1 and PKD2 genes showing the position of the 10 pairs of primers used for LR-PCR amplification of the coding regions. The highlighted green and yellow regions correspond to the duplicated and single-copy sequences of PKD1, respectively. B: Amplification quality was verified using agarose gel electrophoresis. C: LR-PCR products from each patient were pooled together at equimolar ratio, followed by fragmentation and library preparation. The finished libraries were quantified and batched together at equimolar amounts in groups of 25 patient samples and were assessed for quality by a high-sensitivity chip using an Agilent Bioanalyzer instrument. D: The pooled libraries were sequenced on an Illumina MiSeq platform. The raw sequencing reads were sorted by bar code first and then were subjected to quality control analysis before proceeding with the mutation analysis. The quality score (Phred-like score) is shown at each position of the reads. E: Reads were then mapped back to the PKD1/PKD2 loci of human genome assembly hg19 using the BWA program. In this example, PKD1 sequencing coverage is shown using the Integrative Genomics Viewer (Broad Institute, Cambridge, MA). Red areas, reads from the plus DNA strands; blue areas, reads from the minus strands. F: Variant callings were made by the GATK software package and were visualized using the Integrative Genomics Viewer. Ex, exon; FU, fluorescence unit.

Figure 2.

Illustration of the data analysis pipelines. Listed are the analysis steps (A) and the corresponding software/application programs involved (B).

Using this approach, 85% of all sequenced reads mapped to the PKD1/2 reference genome. Of these mapped reads, 100% mapped back to the targeted PKD1 and PKD2 regions. Overall, a very high read depth was obtained across all PKD gene target regions, with >93% of targeted sequences covered with >30× for the entire amplicon. Focusing on exonic regions, we calculated the minimum, maximum, and average read depth for all PKD1 and PKD2 exons extending 20 bp from each end. This analysis demonstrated that except for two patients with a failed PCR product encompassing a single exon and PKD1 exon 1 (see below for explanation), all the exons had minimum coverage ≥11×, average coverage ≥103×, and maximum coverage ≥127×. For most exons (53 of 60), minimum coverage was ≥50×. The minimum, maximum, and average read depths as well as quality metrics for all the exons are provided in Supplemental Tables S3 and S4, respectively. For PKD1 exon 1, which contains a highly GC-rich region centered approximately on the start codon, there was a low minimum coverage in several of the patients (down to 0× at one to two nucleotides). The length of the poorly covered region (defined as a region in which one or more patients had <5× coverage) was 156 bp (chr16:2185623-2185778), suggesting that, except for this sequence, accurate variant calling is possible across all the PKD1 and PKD2 exons. Representative read depth and coverage analysis results are shown in Figure 3. For this typical sample, 100% of the targeted region was covered >15-fold depth, and 93% of the targeted region was covered >30-fold depth, with PKD1 exon 1 having the lowest read depth because of its extremely high GC content (approximately 85%) and low complexity.

Figure 3.

Read depth and coverage analysis results. Plot shows the base coverage (y axis) of each LR-PCR amplicon of the PKD1 and PKD2 genes of one patient. The x axis represents the genomic interval. The average read depth for each fragment is indicated under each amplicon. The plots were generated using the Integrative Genomics Viewer. Ex, exon.

Based on these results, SNVs and small insertion/deletion variants (three deletions of 10, 4, and 2 bp in length and one 24-bp-long insertion) (Figure 4) were detected, indicating that multiplexing of LR-PCR libraries did not compromise sensitivity or specificity rates.

Figure 4.

Visualization of typical PKD1 NGS gene variation calls. A nonsense mutation (A) and a 10-bp deletion variant (B) are shown. NGS reads were piled up and are shown on the Integrative Genomics Viewer on top; Sanger sequencing confirmations are shown below.

NGS Variant Analysis

Using this method, we validated a cohort of 25 patients who have been previously genotyped by Sanger sequencing. The samples selected harbored gene variations spanning the entire genetic sequence of PKD1 and PKD2. Data mining has correctly identified all 250 Sanger sequencing changes (corresponding to 83 gene variations), indicating that the NGS method has a diagnostic performance comparable with that of the direct sequencing approach currently used in our laboratory (Weill Cornell Medical College Molecular Pathology Research Laboratory, New York, NY). However, in one of these patients, two changes (PKD1 c. 7165T>C:p.= and PKD1 c. 6598 C>T:p.R2200C) were incorrectly identified as homozygous by NGS rather than as heterozygous. Further investigation using Sanger sequencing of the NGS LR-PCR product revealed that both variants were located in the same amplicon and that the miscalling resulted from allele dropout of the reference allele (data not shown).

Investigation of the distribution of the ALT alleles showed two sharp peaks centered on 50% ALT frequency and near 100% frequency (Figure 5). Most variants (90.5%) were within the expected 40% to 60% or >90% ALT allele frequency range. We identified 16 variants with <40% ALT allele frequency. Of these, two variants had relatively low coverage (56× and 39×), which could have led to inaccurate ALT frequency estimation. The 14 remaining low ALT percentage variants had coverage similar to the other variants and were spread across six distinct patients. Three low ALT percentage variants were repeatedly found in four patients. These three variants are known single nucleotide polymorphisms (SNPs) (rs10960, rs3087632, and rs7203729) and are found in other patients at frequencies in the 40% to 60% or >90% ALT allele frequency range. All three variants are within 700 bp of each other on chromosome 16 (positions 2140010, 2140454, and 2140680). Visual inspection of these low ALT percentage variants did not reveal any sequence artifacts; eg, the regions around these SNPs were neither GC poor nor GC rich, suggesting preferential amplification of the normal allele during LR-PCR. A detailed list of all PKD1 and PKD2 genetic variations identified in this study and an assessment of the pathogenic potential of missense variants in ANNOVAR software are shown in Tables 2 and 3, respectively. Of these 250 variants, 221 (88.4%) were in PKD1 and 29 (11.6%) were in PKD2, and 16 variants were pathogenic (Table 2). The analytic sensitivity and specificity were calculated for 2075 sites in aggregate across the 25 samples (corresponding to 83 target locations in 25 patients), including 250 variants and 1825 normal alleles, that matched the reference genes. The sensitivity of the NGS method was 99.2% (95% CI, 96.8%–99.9%) and the specificity was 99.9% (95% CI, 99.7%–100.0%) compared with Sanger sequencing results (Table 4). Taken together, these results suggest that pooling 25 bar-coded samples in a single Illumina run is feasible, with an expected read depth of approximately 300-fold.

Figure 5.

Distribution of the ALT allele frequencies. The analysis was performed using R software version 3.0.1 (http://www.r-project.org). The results are shown as a function of the ALT allele frequency percentage.

Table 2.

Details of PKD1 and PKD2 Variants Analyzed in the Cohort of 25 Sanger Sequencing–Confirmed Patients with ADPKD Analyzed by NGS

Chromosome	Position	dbSNP ID	REF	ALT	Gene	Exon/Intron	Function	Variant nomenclature	QUAL	FILTER
16	2139875	rs62038811	G	A	PKD1	Exon46	Synonymous SNV	NM_000296.3:c.12762C>T:p.=	4925.23	PASS
16	2140010	rs7203729	A	G	PKD1	Exon46	Synonymous SNV	NM_000296.3:c.12627T>C:p.=	11,757.62	PASS
16	2140177		A	C	PKD1	Exon46	Nonsynonymous SNV	NM_000296.3:c.12460T>G:p.F4154V	5601.09	PASS
16	2140321	rs79899502	G	A	PKD1	Exon45	Synonymous SNV	NM_000296.3:c.12406C>T:p.=	10007.98	PASS
16	2140454	rs3087632	T	C	PKD1	Exon45	Synonymous SNV	NM_000296.3:c.12273A>G:p.=	28352.74	PASS
16	2140554	rs3209986	G	A	PKD1	Exon45	Nonsynonymous SNV	NM_000296.3:c.12173C>T:p.A4058V	11352.25	PASS
16	2140680	rs10960	T	C	PKD1	Exon44	Nonsynonymous SNV	NM_000296.3:c.12130A>G:p.I4044V	28827.19	PASS
16	2141454		G	A	PKD1	Exon42	Synonymous SNV	NM_000296.3:c.11679C>T:p.=	2940.32	PASS
16	2141522		C	A	PKD1	Exon42	Stopgain SNV	NM_000296.3:c.11611G>T:p.E3871X	4450.2	PASS
16	2142573		C	G	PKD1	Exon39	Nonsynonymous SNV	NM_000296.3:c.11174G>C:p.W3725S	4102.69	PASS
16	2144026		GA	G	PKD1	Intron35	Intronic deletion	NM_000296.3:c.10616-13delT	6754.83	PASS
16	2144123		G	A	PKD1	Exon35	Stopgain SNV	NM_000296.3:c.10585C>T:p.Q3529X	2175.31	PASS
16	2144176	rs34197769	G	A	PKD1	Exon35	Nonsynonymous SNV	NM_000296.3:c.10532C>T:p.A3511V	6720.4	PASS
16	2147518		G	A	PKD1	Intron32	Intronic SNV	NM_000296.3:c.10218-14C>T	6194.41	PASS
16	2150323	/	A	G	PKD1	Intron27	Intronic SNV	NM_000296.3:c.9569-13T>C	6960.57	PASS
16	2150489		GCT	G	PKD1	Exon27	Frameshift deletion	NM_000296.3:c.9474_9475del:p.3158_3159del	7103.23	PASS
16	2152129	/	A	G	PKD1	Exon26	Synonymous SNV	NM_000296.3:c.9330T>C:p.=	64581.14	PASS
16	2152387	/	A	G	PKD1	Exon25	Nonsynonymous SNV	NM_000296.3:c.9196T>C:p.F3066L	49993.66	PASS
16	2152388	/	C	G	PKD1	Exon25	Synonymous SNV	NM_000296.3:c.9195G>C:p.=	50653.93	PASS
16	2152619	/	C	T	PKD1	Exon25	Synonymous SNV	NM_000296.3:c.8964G>A:p.=	7886.09	PASS
16	2152651	/	T	C	PKD1	Intron24	Intronic SNV	NM_000296.3:c.8949-17A>G	41146.96	PASS
16	2152847	/	G	A	PKD1	Exon24	Synonymous SNV	NM_000296.3:c.8916C>T:p.=	7036.67	PASS
16	2153272		A	See below∗	PKD1	Exon23	Nonframeshift insertion	See below∗	33209.17	PASS
16	2153618	/	C	T	PKD1	Exon23	Nonsynonymous SNV	NM_000296.3:c.8440G>A:p.G2814R	3477.3	PASS
16	2153619	/	G	A	PKD1	Exon23	Synonymous SNV	NM_000296.3:c.8439C>T:p.=	4679.38	PASS
16	2153765	/	G	A	PKD1	Exon23	Nonsynonymous SNV	NM_000296.3:c.8293C>T:p.R2765C	4200.12	PASS
16	2154537	/	G	A	PKD1	Exon22	Nonsynonymous SNV	NM_000296.3:c.8123C>T:p.T2708M	3605.52	PASS
16	2154565		G	A	PKD1	Exon22	Stopgain SNV	NM_000296.3:c.8095C>T:p.Q2699X	3154.68	PASS
16	2155426	/	T	C	PKD1	Exon21	Nonsynonymous SNV	NM_000296.3:c.7913A>G:p.H2638R	32433.16	PASS
16	2156021	/	A	G	PKD1	Exon20	Synonymous SNV	NM_000296.3:c.7708T>C:p.=	45989.17	PASS
16	2156447	/	G	A	PKD1	Exon18	Synonymous SNV	NM_000296.3:c.7441C>T:p.=	39665.05	PASS
16	2156623		G	T	PKD1	Exon18	Nonsynonymous SNV	NM_000296.3:c.7265C>A:p.T2422K	1634.58	PASS
16	2156850	/	A	G	PKD1	Exon17	Synonymous SNV	NM_000296.3:c.7165T>C:p.=	90944.82	PASS
16	2158022	/	G	A	PKD1	Exon16	Synonymous SNV	NM_000296.3:c.6927C>T:p.=	5649.07	PASS
16	2158570	/	G	A	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.6598C>T:p.R2200C	28297.7	PASS
16	2158871	/	C	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.6297G>T:p.=	7173.93	PASS
16	2159313		C	T	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.5855G>A:p.G1952D	8717.56	PASS
16	2159321	/	G	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.5847C>T:p.=	3863.12	PASS
16	2159391		G	A	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.5777C>T:p.A1926V	2041.5	PASS
16	2159405	/	C	T	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.5763G>A:p.=	6637.77	PASS
16	2159557	/	C	T	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.5611G>A:p.A1871T	4010.9	PASS
16	2159996	/	G	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.5172C>T:p.=	47423.27	PASS
16	2160280		G	A	PKD1	Exon15	Stopgain SNV	NM_000296.3:c.4888C>T:p.Q1630X	6618.24	PASS
16	2160494	/	C	T	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.4674G>A:p.=	19874.71	PASS
16	2160503	/	T	G	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.4665A>C:p.=	49280.48	PASS
16	2160622	/	C	T	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.4546G>A:p.A1516T	3201.27	PASS
16	2160673	/	G	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.4495C>T:p.=	3420.25	PASS
16	2160716		C	T	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.4452G>A:p.=	3964.01	PASS
16	2160973	/	A	G	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.T4195C:p.W1399R	32607.04	PASS
16	2161244		CTGCGTGGGGA	C	PKD1	Exon15	Frameshift deletion	NM_000296.3:c.3914_3923del:p.1305_1308del	7343.71	PASS
16	2161443		G	A	PKD1	Exon15	Nonsynonymous SNV	NM_000296.3:c.3725C>T:p.T1242M	3864.26	PASS
16	2161655	/	G	C	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.3513C>G:p.=	2905.22	PASS
16	2161793	/	G	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.3375C>T:p.=	33353.67	PASS
16	2161796	/	G	A	PKD1	Exon15	Synonymous SNV	NM_000296.3:c.3372C>T:p.=	33778.82	PASS
16	2161874		T	G	PKD1	Intron14	Splicing SNV	NM_000296.3:c.3296-2A>C	3192.03	PASS
16	2162361	/	A	G	PKD1	Exon14	Nonsynonymous SNV	NM_000296.3:c.3275T>C:p.M1092T	31648.89	PASS
16	2162839	/	T	C	PKD1	Exon13	Synonymous SNV	NM_000296.3:c.3111A>G:p.=	11160.46	PASS
16	2162887	/	A	G	PKD1	Exon13	Synonymous SNV	NM_000296.3:c.3063T>C:p.=	22392.19	PASS
16	2164294	/	G	A	PKD1	Exon11	Synonymous SNV	NM_000296.3:c.2730C>T:p.=	11471.01	PASS
16	2164324	/	C	T	PKD1	Exon11	Synonymous SNV	NM_000296.3:c.2700G>A:p.=	11,448.82	PASS
16	2164330	/	T	G	PKD1	Exon11	Synonymous SNV	NM_000296.3:c.2694A>C:p.=	4682.76	PASS
16	2164808	/	C	T	PKD1	Exon11	Nonsynonymous SNV	NM_000296.3:c.2216G>A:p.R739Q	279070.57	PASS
16	2165630	/	T	C	PKD1	Intron9	intronic SNV	NM_000296.3:c.1850-4A>G	50257.02	PASS
16	2166061		A	T	PKD1	Exon9	Nonsynonymous SNV	NM_000296.3:c.1781T>A:p.F594Y	4476.37	PASS
16	2167874		G	A	PKD1	Exon5	Synonymous SNV	NM_000296.3:c.1119C>T:p.=	268925.84	PASS
16	2167970		G	A	PKD1	Exon5	Synonymous SNV	NM_000296.3:c.1023C>T:p.=	12263.35	PASS
16	2169178		C	A	PKD1	Exon3	Nonsynonymous SNV	NM_000296.3:c.296G>T:p.S99I	8791.34	PASS
16	2185509		G	A	PKD1	Exon1	Nonsynonymous SNV	NM_000296.3:c.182C>T:p.P61L	750.13	PASS
16	2185584		G	T	PKD1	Exon1	Nonsynonymous SNV	NM_000296.3:c.107C>A:p.P36H	127.59	PASS
4	88928968	rs1805044	G	C	PKD2	Exon1	Nonsynonymous SNV	NM_000297.3:c.83G>C:p.R28P	41620.06	PASS
4	88929080		G	GGACC	PKD2	Exon1	Frameshift insertion	NM_000297.3:c.195_196insGACC:p.R65fs	10644.57	PASS
4	88929082		A	AC	PKD2	Exon1	Frameshift insertion	NM_000297.3:c.197_198insC:p.D66fs	4694.47	PASS
4	88929305	rs2728118	G	A	PKD2	Exon1	Synonymous SNV	NM_000297.3:c.420G>A:p.=	5295.98	PASS
4	88929453	rs117078377	G	A	PKD2	Exon1	Nonsynonymous SNV	NM_000297.3:c.568G>A:p.A190T	4083.65	PASS
4	88940594	rs62310565	C	T	PKD2	Intron1	Intronic SNV	NM_000297.3:c.596-16C>T	2722.03	PASS
4	88959475		C	T	PKD2	Exon4	Stopgain SNV	NM_000297.3:c.916C>T:p.R306X	5216.94	PASS
4	88959479		GT	G	PKD2	Exon4	Frameshift deletion	NM_000297.3:c.921delT:p.S307fs	10175.19	PASS
4	88959517		C	T	PKD2	Exon4	Stopgain SNV	NM_000297.3:c.958C>T:p.R320X	3946.43	PASS
4	88959653		CGTAA	C	PKD2	Intron4	Splicing deletion	NM_000297.3:c.1094+1_1094+4del	8740.89	PASS
4	88964610		G	A	PKD2	Intron5	Splicing SNV	NM_000297.3:c.1319+1G>A	8073.4	PASS
4	88967919	rs75762896	T	G	PKD2	Exon6	Nonsynonymous SNV	NM_000297.3:c.1445T>G:p.F482C	6455.92	PASS
4	88977424		G	A	PKD2	Intron8	Intronic SNV	NM_000297.3:c.1898+5G>A	7426.45	PASS
4	88996055		C	T	PKD2	Exon14	Stopgain SNV	NM_000297.3:c.2614C>T:p.R872X	3914.06	PASS

Pathogenic mutations are denoted in bold.

REF, reference; /, there was entry in dbSNP but was removed because it was located in PKD1 duplicated region thus could be contamination from pseudogenes.

^∗AGCAGCGTATAGTTGAGCTGCAGAT; variant nomenclature, NM_000296.3:c.8786_8787insATCTGCAGCTCAACTATACGCTGC:p.L2929delinsHLQLNYTLL.

Table 3.

Assessment of the Pathogenic Potential of Missense Variants in the ANNOVAR Program

Gene	Amino acid change	Frequency in 1000 Genome Project	dbSNP132	SIFT prediction	PolyPhen-2 prediction	MutationTaster prediction	Overall prediction
PKD1	NM_000296.3:c.296G>T:p.S99I			D	D	D	PP
PKD1	NM_000296.3:c.7265C>A:p.T2422K			D	D	D	PP
PKD1	NM_000296.3:c.8123C>T:p.T2708M			D	D	D	PP
PKD1	NM_000296.3:c.12460T>G:p.F4154V			D	D	D	PP
PKD1	NM_000296.3:c.11174G>C:p.W3725S			D	D	D	PP
PKD1	NM_000296.3:c.7913A>G:p.H2638R			T	B	N	NE
PKD1	NM_000296.3:c.4546G>A:p.A1516T			T	B	N	NE
PKD1	NM_000296.3:c.12130A>G:p.I4044V	0.23	rs10960	T	B	N	NE
PKD1	NM_000296.3:c.2216G>A:p.R739Q			T	B	N	NE
PKD1	NM_000296.3:c.10532C>T:p.A3511V	0.06	rs34197769	T	B	N	NE
PKD1	NM_000296.3:c.3275T>C:p.M1092T			T	B	N	NE
PKD1	NM_000296.3:c.5777C>T:p.A1926V			T	B	N	NE
PKD1	NM_000296.3:c.4195T>C:p.W1399R			T	B	N	NE
PKD1	NM_000296.3:c.8440G>A:p.G2814R			T	P	N	PN
PKD1	NM_000296.3:c.107C>A:p.P36H			T	NA	N	NE
PKD1	NM_000296.3:c.5855G>A:p.G1952D			D	P	N	PN
PKD1	NM_000296.3:c.182C>T:p.P61L			D	P	N	PN
PKD1	NM_000296.3:c.12173C>T:p.A4058V	0.03	rs3209986	T	D	N	PN
PKD1	NM_000296.3:c.1781T>A:p.F594Y			T	D	N	PN
PKD1	NM_000296.3:c.5611G>A:p.A1871T			T	D	N	PN
PKD1	NM_000296.3:c.9196T>C:p.F3066L			T	D	N	PN
PKD1	NM_000296.3:c.8293C>T:p.R2765C			D	D	N	PN
PKD1	NM_000296.3:c.3725C>T:p.T1242M			D	D	N	PN
PKD1	NM_000296.3:c.6598C>T:p.R2200C			D	D	N	PN
PKD2	NM_000297.3:c.1445T>G:p.F482C	0.003	rs75762896	D	D	D	D
PKD2	NM_000297.3:c.568G>A:p.A190T	0.07	rs117078377	T	P	N	NE
PKD2	NM_000297.3:c.83G>C:p.R28P	0.14	rs1805044	D	P	N	PN

B, benign; D, deleterious; N, neutral; NA, not available; NE, neutral (sequence changes scored as benign by all three software applications¹⁸); P, possibly damaging; PP, probably pathogenic (sequence changes scored as deleterious by all three computational analysis tools¹⁸); PN, probably neutral (sequence changes scored as benign by only one or two of the software applications¹⁸); T, tolerant.

Table 4.

NGS Analytic Sensitivity and Specificity (Variants Detection)

NGS	Sanger sequencing
	Variant alleles (positive)	Reference alleles (negative)	Total
Variant alleles (positive)	248	0	248
Reference alleles (negative)	2	1825	1827
Total	250	1825	2075

Compared with the Sanger sequencing assay: sensitivity = 99.2% (95% CI, 96.8%–99.9%); specificity = 99.9% (95% CI, 99.7%–100.0%).

Mutation Analyses of Pathogenic Mutation-Negative and Novel Patients with ADPKD by Bar-Coded and Multiplexed NGS

Based on these proof-of-principle results, we analyzed an additional 24 ADPKD cases that tested negative for a pathogenic mutation by Sanger sequencing. The NGS method identified PKD1 pathogenic variants in three patients for whom variants were not previously detected (NM_000296.3:c.3296-2A>T; NM_000296.3:c. 7288 C>T:p.R2430X; and NM_000296.3:c. 1937 G>A:p.W646X), which were then further confirmed by Sanger sequencing. One of these variants, NM_000296.3:c.3296-2A>T, was previously missed because of a technologist's error, whereas the other two were mainly due to allele dropout during the LR-PCR step. Careful analysis of the original Sanger sequencing data demonstrated very low signal (<10% to 20%) for the mutant alleles, leading to miscalled sequences.

To further evaluate the sensitivity and specificity of the NGS assay, we analyzed an additional cohort of 25 patients with ADPKD that had not been previously genotyped. Of the 25 patients, 16 (64%) had a pathogenic or probably pathogenic mutation, whereas no mutations were identified in the remainder of the patient cohort (Table 5). All the positive results and the mutation-negative cases were confirmed by Sanger sequencing, suggesting sensitivity and specificity of 100% for detecting ADPKD mutations in this sample.

Table 5.

Details of PKD1 and PKD2 Mutations in a Cohort of 25 Novel Patients

No.	Chr.	Position	REF	ALT	Exon	Gene	Exonic function	Amino acid change	QUAL	FILTER
1	16	2141440	CAGCG	C	42	PKD1	Frameshift deletion	NM_000296.3:c.11690_11693del:p.3897_3898del	11,229.66	PASS
2	16	2157900	CT	C	16	PKD1	Frameshift deletion	NM_000296.3:c.7049delA:p.E2350fs	16754.66	PASS
3	16	2144151	GCCCCAGCTCC	G	35	PKD1	Frameshift deletion	NM_000296.3:c.10548_10557del:p.3516_3519del	32720.66	PASS
4	16	2168287	G	A	5	PKD1	Stopgain SNV	NM_000296.3:c. 706C >T:p.Q236X	6574.71	PASS
5	16	2164185	G	A	11	PKD1	Stopgain SNV	NM_000296.3:c. 2839 C>T:p.Q947X	9480.71	PASS
6	16	2156600	G	A	18	PKD1	Stopgain SNV	NM_000296.3:c. 7288 C>T:p.R2430X	4314.5	PASS
7	16	2140782	G	A	44	PKD1	Stopgain SNV	NM_000296.3:c. 12028 C>T:p.Q4010X	13180.82	PASS
8	16	2160674	G	T	15	PKD1	Stopgain SNV	NM_000296.3:c. 4494 C>A:p.Y1498X	9530.5	PASS
9	16	2166531	T	A	8	PKD1	Nonsynonymous SNV∗†	NM_000296.3:c. 1721 A>T:p.E574V	6856.71	PASS
10	16	2156912	A	G	17	PKD1	Nonsynonymous SNV∗	NM_000296.3:c. 7103 T>C:p.L2368S	16,133.71	PASS
11	16	2164844	A	G	11	PKD1	Nonsynonymous SNV∗	NM_000296.3:c. 2180 T>C:p.L727P	7176.71	PASS
12	4	88929082	A	AC	1	PKD2	Frameshift insertion	NM_000297.3:c.197_198insC:p.D66fs	10197.45	PASS
13	4	88959475	C	T	4	PKD2	Stopgain SNV	NM_000297.3:c. 916 C>T:p.R306X	11,046.71	PASS
14	4	88959517	C	T	4	PKD2	Stopgain SNV	NM_000297.3:c. 958 C>T:p.R320X	18,191.71	PASS
15	4	88929145	G	A	1	PKD2	Stopgain SNV	NM_000297.3:c. 260 G>A:p.W87X	4583.71	PASS
16	4	88959536	T	G	4	PKD2	∗Nonsynonymous SNV	NM_000297.3:c. 977 T>G:p.V326G	11,941.71	PASS

Chr, chromosome; REF, reference.

^∗Classified as probably pathogenic based on SIFT, Polyphen-2, and MutationTaster predictions as specified in Materials and Methods.

^†Predicted to affect exon splicing by computational analysis by distrusting an exonic splice enhancer.

NGS Cost Analysis

We also evaluated the costs of reagents and sequencing for the NGS-based approach. By pooling 25 patient samples in a single NGS Illumina MiSeq run, the cost of the test was reduced by approximately 70% compared with Sanger sequencing, from approximately $271 per patient to approximately $82 per patient (Table 6). Moreover, although the hands-on time required for setting up the LR-PCR reactions, preparing the library, and sequencing still requires approximately 1 week, data analysis is considerably faster with NGS than with the Sanger method and can be completed in only a few hours. Therefore, a 25-patient cohort could be analyzed in 1 to 2 weeks compared with ≥4 weeks with the Sanger method.

Table 6.

Comparison of Reagents, Sequencing Costs, and Time of Labor for Sanger Sequencing and NGS

Method	Purpose	Quantity	Cost ($)			Labor time (days)
			Per sample	Per run	Per subject
Sanger sequencing (N = 25)	LR-PCR (PKD1)	250	2.40	600.00	24.00	5
	Standard PCR (PKD2)	400	1.50	600.00	24.00	4
	Purification	200	2.40	480.00	19.20	1
	Sequencing primers	3050	0.10	305.00	12.20	NA
	Sanger sequencing	1600	3.00	4800.00	192.00	5
	Data analysis	NA	NA	NA	NA	4
	Total			6785.00	271.40	19
NGS (N = 25)	LR-PCR (PKD1 and PKD2)	250	1.45	362.50	14.50	2
	LR-PCR product quantification	250	0.12	30.00	1.20	0.5
	DNA fragmentation	25	6.50	162.50	6.50	0.5
	Library preparation	25	20.00	500.00	20.00	3
	Library quality assessment	25	0.20	5.00	0.20	0.25
	NGS sequencing (MiSeq)	1	990.00	990.00	39.60	1
	Data analysis	NA	NA	NA	NA	1
	Total			2050.00	82.00	8.25

NA, not applicable.

Discussion

NGS technology has revolutionized genomic and genetic research and the field of clinical genomics. Sample bar coding and multiplexing capabilities and the availability of simpler workflows and faster turnaround time instruments, such as Illumina MiSeq, have made this method attractive to clinical laboratories. Rossetti et al¹⁷ recently developed an original approach for PKD gene mutation detection in large cohorts by coupling LR-PCR with NGS analysis with sensitivity of 78% and specificity of 100%. Herein, we describe a new NGS PKD genotyping approach with analytical sensitivity of 99.2% and specificity of 99.9% compared with the Sanger sequencing method. This approach is based on LR-PCR amplification of both the PKD1 and PKD2 genes using 10 pairs of carefully designed PCR primers covering approximately 68.0 kb of PKD genomic region, corresponding to 31.9 kb (68.8%) and 35.8 kb (51.0%) of the PKD1 and PKD2 genomic regions, respectively, particularly tailored to relatively small cohorts and clinical diagnostic applications. This improvement in sensitivity is mainly due to two factors. First, we individually bar coded LR-PCR products by patient rather than pooling of DNA samples or LR-PCR amplicons from different patients before sample bar coding.¹⁷ In contrast, Rossetti et al¹⁷ pooled DNA samples or PCR products before library generation, leading to a substantial loss of sensitivity and specificity due primarily to lack of coverage. The strategy used in the present study allows for improved coverage of individual patient samples by increasing read depth and decreasing background noise, thus allowing variant calls with high confidence. Second, we used longer reads on the MiSeq platform (150 bp × 2 as opposed to 101 bp/75 bp × 2) compared with Rossetti et al,¹⁷ ensuring higher genome mapping accuracy by reducing mapping errors.

We used the GATK software package to refine the BAM (Binary Alignment/Map) file generated by the BWA program; this step can improve accuracy in variant calls, particularly for indels. Variant calling of NGS data is prone to a high error rate owing to the following factors: many reads spanning insertion/deletion sites are misaligned because each read is aligned independently, and the raw base quality scores often vary with instrument features, such as the platform used, machine cycle, and sequence context, and, thus, cannot reflect the true base calling error rates.²⁵ These errors in alignments and base calling will be translated into variant and genotype inference, leading to false sequence calls. The local realignment around indels and base quality score recalibration function in the GATK package can decrease the false-positive calls around indels, increasing base quality at the end of the reads and enabling overall higher accuracy of the sequence data.²⁵ Furthermore, the multiple sample–calling feature of GATK enabled processing of multiple samples simultaneously, allowing the use of sequencing information across all samples, further increasing the accuracy of variant call. Liu et al²⁸ conducted a systematic assessment of several variant calling packages and found that GATK in combination with the BWA aligner performed better compared with other software applications for high coverage of Illumina data (≥20×). The present NGS data had average coverage of 103×, far above the threshold, making it suitable for these analyses.

The alignment algorithm in the analysis pipeline, BWA, is a fast and memory-efficient short read aligner,²³ allowing implementation on a standard desktop computer without the need for expensive computer clusters and making it attractive to the standard diagnostic laboratory. In our experience, data from 25 patients could be efficiently analyzed (within several hours) using an iMac desktop computer with two processors and a 3.06-GHz CPU and 16 GB of memory (Apple Inc., Cupertino, CA). The data analysis pipelines have been fully automated with Unix shell scripts, significantly reducing the hands-on time required for quality checks of the sequencing data. This is a major improvement compared with the labor-intensive pipeline required for analyzing Sanger sequencing data, even when using automated applications, such as Mutation Surveyor (SoftGenetics LLC., State College, PA),¹³ for variant calling. Moreover, ANNOVAR genomic annotator can readily access several prediction programs, including SIFT,²⁹ PolyPhen-2,³⁰ and Mutation Taster (http://www.mutationtaster.org, last accessed October 25, 2013),³¹ used for evaluating the pathogenic potential of missense change, providing important information about the pathogenicity of the variants analyzed. In this study, of 27 missense variants, six were classified as deleterious by all three prediction algorithms, which is in agreement with the results obtained by the present standard bioinformatics procedures (Table 3).¹⁸

Taken together, the workflow reported herein can be easily adapted to a routine clinical diagnostic setting. The LR-PCR and library preparation steps have been automated using a liquid handler (Corbett 1200; Qiagen Inc.), standardizing the reaction setup process. Similarly, the individual bar coding of patient samples and the simultaneous processing of 25 samples, together with the automated data analysis pipeline, have greatly reduced the number of test reactions, decreasing technologist's errors and increasing the overall precision and accuracy of the sequencing data obtained. Consequently, up to 25 patients can be genotyped in 1 to 2 weeks at reagent costs of $82 per patient compared with $271 per patient with the current Sanger sequencing method.

Overall, the present method detected all 250 Sanger sequencing–verified gene variations, including single nucleotide changes, splice site alterations, and indel mutations, except for two PKD1 changes residing in the same amplicon that were ascertained as homozygous by NGS rather than as heterozygous. In contrast, of the 24 patients with Sanger sequencing–negative ADPKD, 3 were found to harbor a pathogenic mutation in PKD1 by the NGS method, subsequently confirmed by Sanger sequencing, using different primers. Except for PKD1 c.3296-2A>T, previously missed by Sanger sequencing owing a technologist's error, the two other discrepancies were due to allele dropout during the LR-PCR amplification step. Allele dropout or reduction to homozygosity has been well documented, particularly for PKD1, which is highly polymorphic, and it is likely due to the presence of an SNP in the primer binding sites in one of the two alleles, leading to unequal PCR amplification of the two heterozygous alleles.^17,18 This phenomenon is less likely to occur with NGS because deep sequencing has a significantly higher level of analytical sensitivity of at least 5%^32,33 compared with the Sanger method for detection of mutations in impure populations of DNAs. Preferential amplification of one allele versus the other can also explain the lower-than-expected (50%) percentage ALT alleles observed for approximately 10% of the variants in this study. In addition, NGS requires only a limited number of primer pairs, as opposed to >80 pairs with Sanger sequencing, thereby considerably reducing the chances for mispriming. Nevertheless, in all cases with homozygous mutations, subsequent confirmatory sequencing analysis must be performed to rule out false-positive results.¹⁸ Note that the greater analytical sensitivity achieved by high-depth sequencing can be particularly suitable for detecting low-abundance mutations in rare cases of mosaicism. Although false-positive results have not been seen in this study, we propose that as a part of a routine clinical workflow, all mutations should be confirmed by Sanger sequencing. The low coverage of the GC-rich exon 1 found in this study for multiple samples may lead to false-negative results. To overcome this problem, we suggest that follow-up Sanger sequencing be used as an alternative method for negative cases with low coverage (<5×).³⁴

One potential limitation of this NGS-based ADPKD genotyping approach is that the target enrichment process still requires LR-PCR, and setting up LR-PCR is cumbersome and complicated, especially when large numbers of samples are analyzed. Qi et al³⁵ used hybridization-based exon capture as an approach to ADPKD genotyping. However, although target enrichment is considerably easier to perform compared with LR-PCR, this approach is not suitable for distinguishing PKD1 from the pseudogenes, with a true-positive mutation detection rate of only 28.6%.³⁵ Low coverage of GC-rich DNA regions due to the unexpected secondary structure of the DNA template may also be a limitation and an underestimated cause for missed variants, particularly in exon 1 of PKD1, which is 85% GC rich.^17,36 This phenomenon of base composition bias has been well documented for Illumina sequencing and is primarily attributed to the enrichment PCR step during library construction.³⁶ Amplification with AccuPrime Taq high fidelity enzyme blend at a low primer extension temperature of 65°C has been shown to improve the overall coverage of high–GC-rich areas.³⁶ In our hands, there was still very good average read coverage of PKD1 exon 1 (15×), allowing for confident variant calling. However, we identified a 156-bp-long region in exon 1 (chr16:2185623-2185778) in which we observed <5× coverage at one nucleotide or more and in at least one patient in the cohort. Unequal and sometimes low coverage in this region indicates that variants occurring in this region might be missed, suggesting that this amplicon should be added in an excess molar ratio compared with all other amplicons when pooling to provide sufficient read depth for a confident mutation call.¹⁷ We note, however, that Sanger sequencing did not detect any variants in this region in the present cohort despite the high quality of the Sanger traces (not shown).

Finally, large deletions have been shown to play an important role in ADPKD in 1% to 3% of cases.^37,38 Although the paired-end mapping feature in the genotyping protocol makes it possible to detect such mutations, it requires that the deletion is located in the LR-PCR amplicon (Figure 6). The method will not detect any deletion that is outside the LR-PCR primer range, requiring the continuing need for methods such as multiplex ligation-dependent probe amplification.³⁸

Figure 6.

Coverage plot illustrating the identification of a large-sized deletion in PKD2. The patient had an approximately 6-kb deletion that included PKD2 exon (Ex) 5 and adjacent intronic regions compared with the control sample.

In summary, NGS-based ADPKD genetic analysis is a highly accurate and reliable approach for mutation analysis, achieving high sensitivity and improved intronic coverage with a faster turnaround time and lower cost. Optimization of the workflow and the stepwise process quality control metrics for data analysis will likely become routine for clinical genetic testing, and NGS would be an appropriate new standard for clinical genetic testing of ADPKD.