Comparison of Whole Exome Capture Products Coverage ...
1
Comparison of Whole Exome Capture Products – Coverage & Quality vs Cost B Marosy, J Gearhart, B Craig, KF Doheny Center for Inherited Disease, Johns Hopkins Genomics, McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins School of Medicine, Baltimore, MD Figure 5 – Pooling Balance. Compares the ‘evenness’ of pools in the sequencing data between samples (prior to downsampling) that underwent pre-capture pooling (IDT, Roche & Twist) vs samples that were post-capture pooled (Agilent). Colored stripes within each bar denote an individual sample within the pool, size of each bar represents amount of raw sequencing data in Gb. Results – Pooling Balance Obtaining ‘even’ pooling is a critical step in the pre-capture workflow. If a sample fails at sequencing it is more difficult to re-queue since samples which may require additional sequencing cannot be broken out of the pool. This would require re-hybridization of the sample, thus increasing sample costs. Figure 5 depicts the pooling evenness observed between post-capture and pre-capture pooling. Post-capture pooling more consistently produces even pools than pre-capture pooling. Figure 1a-c: Plots Mean Target Coverage (x-axis) vs Percent on Target (y-axis) with increasing depth (10x=top; 20x=middle; 30x=bottom) for each sample. All samples are downsampled to 5 Gb and each panel is trellised by BED File ( 4-way Intersection = left; UCSC coding exons = middle; vendor specific targets = right). Points are colored by Capture product (Blue=Agilent, Red=IDT, Yellow=Roche, Green=Twist). Percent on Target @ 10x Percent on Target @ 20x Mean Target Coverage Percent on Target @ 30x Results – Target Coverage Figure 1 plots coverage vs percent on Target by 10x, 20x and 30x trellised by BED file. At 10x coverage all captures are effective in capturing targets >90%, with only 5 Gb of sequencing data. As depth increases, with the same amount of Gb, coverage at 90% drops across products. Captures with higher efficiency are able to maintain coverage as depth increases with the same amount of sequencing yield. Methods – Sequencing & Data Analysis Captured libraries were clustered and sequencing was performed on the Illumina HiSeq2500 platform using on-board clustering and SBS chemistry with either 2x100 or 2x125 read lengths. Sequencing data was trimmed to 100bp, if applicable, then downsampled (Picard; DownsampleSam) to 5 Gigabases (Gb) of raw data yield and three different BED files (UCSC coding exons, vendor specific, intersection of all 4 vendor BED files) were applied independently to ensure equal comparison between methods and capture content. Samples that fell below 5 Gb were processed ‘as is’ but removed from the final analysis. Subsequent data analysis was performed using CIDRs in-house Research Whole Exome pipeline. Introduction Next Generation Sequencing technology has been a driving force for reducing sequencing costs and enhancing genome technology, allowing investigators and clinicians the opportunity to better delineate an individuals’ genetic makeup. For facilities providing genetic services, maintaining the balance of generating high quality data at the lowest cost is a continuous effort requiring recurrent evaluation and optimization of protocols. In preparation for higher throughput of exome sequencing using the NovaSeq, we evaluated uniformity, coverage and cost of sequencing to achieve 90% on target at 10x, 20x and 30x depth across four different exome capture products: • SureSelect Human All Exon v7 (Agilent Technologies) • xGEN Exome Research Panel v1.0 (Integrated DNA Technologies) • Prime Exome (Roche Sequencing) • Human Core Exome (Twist Bioscience) Methods – Library Preparation To reduce experimental variables and minimize bias between libraries, all libraries were prepared using the Kapa Hyper prep library prep kit. Table 1 summarizes the parameters used. 50ng of DNA from 8 different HapMap samples was sheared using the Covaris LE220. End Repaired/A-Tailed libraries were ligated with IDT dual indexed adapters and amplified using the Kapa HiFi DNA Polymerase for 8 cycles. Agilent captures were hybridized as single sample reactions using 750ng of library as input. IDT, Roche and Twist captures were hybridized as pools of 8 samples using 750ng, 625ng, 187.5ng of library input, respectively. All Hybridization and Post-hyb capture & washes were performed according to each respective manufacturer’s protocol. Post Hyb PCR was performed using the Kapa HiFi DNA Polymerase for 10 cycles, regardless of capture product. Table 1 – Library Prep Methods Results – Sequencing Data QC Table 3 compares QC metrics generated during sequencing when using vendor specific target BED files. Data quality (TiTv, Percent SNV OnTarget SNP138 and Count SNV OnTarget) across captures is consistent. Insert size, duplication rates and library complexity vary between capture products when using the same library prep methods and amplification, indicating that capture also impacts these metrics. Table 3 – Sequencing QC metrics based on vendor specific target BED files. Figure 3: GC dropout (x-axis) vs AT dropout (y-axis) as generated by Picard, which measures how under covered a low/high % GC region is relative to the mean. Points are colored by Capture product (Blue = Agilent, Red = IDT, Yellow = Roche, Green = Twist). Vendor specific target BED files were used. GC Dropout AT Dropout Figure 2: Depicts uniformity across capture products. Data generated from Depth of Coverage is plotted as the count of bases (y-axis) at a given depth (x-axis) for all bases captured from NA12878. Each panel varies by BED File. Lines are colored by capture product (Blue=Agilent, Red=IDT, Yellow=Roche, Green=Twist). UCSC Vendor Depth Count Results – Coverage Uniformity Coverage uniformity was analyzed in several ways. Figure 2 represents sample statistic summaries generated from GATK - Depth of Coverage (DOC). UCSC exon coding and vendor specific BED files were used, however, the 4-way intersection BED file results were comparable to the UCSC plot and not shown here. Tight/narrow peaks at the highest depth represent captures with the most uniformity. In addition to DOC metrics, we analyzed GC/AT dropout which measures how under covered a low/high %GC region is relative to the mean (Figure 3). Regions of low coverage in clinically relevant genes were viewed in IGV and are shown in Figure 4. Figure 4a-c : Integrative Genomics Viewer (IGV) plot of NA12878 at regions of low coverage identified within clinically relevant genes. In lower panel: gene track, target BED files for Agilent, IDT, Roche and Twist, respectively. Agilent IDT Roche Twist ALX3 – Frontonasal dysplasia Agilent IDT Roche Twist SKI – Shprintzen-Goldberg Craniosynostosis Syndrome Agilent IDT Roche Twist PDE4D – Acrodysostosis 2 Results – Coverage Uniformity cont’d Discussion & Conclusions • Captures with higher efficiencies are able to maintain coverage as depth increases with the same amount of sequencing yield. Determining which capture product to use may depend on project specific needs (i.e. coverage of specific targets, depth requirements, available automation etc.) and cost requirements. • Library Prep Methods – for purposes of this experiment to minimize bias, we utilized a single library prep method that has been optimized in our facility. Vendor specific library prep reagents/protocols may provide additional optimizations that are not addressed here. • GC/AT rich & Low coverage regions – PCR enzyme performance is typically considered a ‘culprit’ of these difficult to sequence regions. However, capture products may also influence coverage of these regions. • Pre-capture pooling can provide a ~2.5 fold reduction in reagent costs. However, the unbalanced pools obtained post capture increases the redo rates as much as ~30%. For higher through-put facilities, this adds considerably to the labor costs and overall time to completion for projects. Capture products that maintain high efficiency will decrease the amount of sequencing required and when combined with the lower cost of sequencing on the NovaSeq will offset the added cost of pre-capture pooling making this a more feasible workflow. Results – Design Coverage Coverage summaries were generated using Galaxy v1.0.0 coverage tool and summarized in Table 2. Each vendors target BED file and the 4-way intersection was used to determine design coverage across the UCSC Coding exons by region (exonic coordinates) and by base. The total target footprint for each vendor is also listed. Table 2 - UCSC Exon Coding Coverage Summary by Capture Product UCSC Coding Exon (35.20 Mb) BED File Coverage Summary Agilent (49.48 Mb) IDT (50.84 Mb) Roche (37.24 Mb) Twist (33.05 Mb) 4-way Intersection (26.91 Mb) % Regions not covered 1.74 3.72 5.76 5.20 6.69 % Regions covered <50% 1.03 0.46 0.89 0.67 1.93 % Regions covered >50% & <100% 38.19 3.35 3.97 2.71 37.97 % Regions covered 100% 59.04 92.47 89.38 91.43 53.42 % Bases not covered 7.77 5.02 7.89 6.49 13.95 % Bases covered 92.23 94.98 92.11 93.51 86.05 Vendor DNA Input Library Prep Pre Capture PCR Cycles Capture Pooling Hyb Input Post-Capture PCR Cycles Post-Capture PCR Enzyme Agilent 50 ng Kapa Hyper Prep 8 Single 750 ng 10 Kapa HiFi IDT 50 ng Kapa Hyper Prep 8 8-plex 750 ng 10 Kapa HiFi Roche 50 ng Kapa Hyper Prep 8 8-plex 625 ng 10 Kapa HiFi Twist 50 ng Kapa Hyper Prep 8 8-plex 187.5 ng 10 Kapa HiFi Sequencing QC Metric Agilent IDT Roche Twist n = Downsampled to 5 Gb 6 3 4 3 Percent Selection 73.0 87.0 85.5 82.7 Percent Duplication 4.1 4.3 7.6 2.7 Estimated Library Size (millions) 317 316 157 429 Mean Insert Size 250 241 287 216 Percent Total Concordance 99.75 99.82 99.72 99.60 Sensitivity 2 Het 98.82 99.13 99.24 99.38 Percent SNV on Target dbSNP138 98.55 99.14 99.35 99.23 Count SNV on Target 41072 39114 27483 26453 Known TiTv Ratio 3.07 3.07 3.09 3.11 Known TiTv Count 21668 20370 21261 22743
Transcript of Comparison of Whole Exome Capture Products Coverage ...
Comparison of Whole Exome Capture Products – Coverage & Quality vs CostB Marosy, J Gearhart, B Craig, KF Doheny
Center for Inherited Disease, Johns Hopkins Genomics, McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins School of Medicine, Baltimore, MD
Figure 5 – Pooling Balance. Compares the ‘evenness’ of pools inthe sequencing data between samples (prior to downsampling)that underwent pre-capture pooling (IDT, Roche & Twist) vssamples that were post-capture pooled (Agilent). Coloredstripes within each bar denote an individual sample within thepool, size of each bar represents amount of raw sequencingdata in Gb.
Results – Pooling BalanceObtaining ‘even’ pooling is a critical step in the pre-captureworkflow. If a sample fails at sequencing it is more difficult tore-queue since samples which may require additionalsequencing cannot be broken out of the pool. This wouldrequire re-hybridization of the sample, thus increasing samplecosts. Figure 5 depicts the pooling evenness observed betweenpost-capture and pre-capture pooling. Post-capture poolingmore consistently produces even pools than pre-capturepooling.
Figure 1a-c: Plots Mean Target Coverage (x-axis) vs Percent on Target(y-axis) with increasing depth (10x=top; 20x=middle; 30x=bottom) foreach sample. All samples are downsampled to 5 Gb and each panel istrellised by BED File ( 4-way Intersection = left; UCSC coding exons =middle; vendor specific targets = right). Points are colored by Captureproduct (Blue=Agilent, Red=IDT, Yellow=Roche, Green=Twist).
Pe
rce
nt
on
Tar
get
@ 1
0x
Pe
rce
nt
on
Tar
get
@ 2
0x
Mean Target Coverage
Pe
rce
nt
on
Tar
get
@ 3
0x
Results – Target CoverageFigure 1 plots coverage vs percent on Target by 10x, 20x and 30xtrellised by BED file. At 10x coverage all captures are effective incapturing targets >90%, with only 5 Gb of sequencing data. As depthincreases, with the same amount of Gb, coverage at 90% drops acrossproducts. Captures with higher efficiency are able to maintain coverageas depth increases with the same amount of sequencing yield.
Methods – Sequencing & Data AnalysisCaptured libraries were clustered and sequencing was performedon the Illumina HiSeq2500 platform using on-board clusteringand SBS chemistry with either 2x100 or 2x125 read lengths.Sequencing data was trimmed to 100bp, if applicable, thendownsampled (Picard; DownsampleSam) to 5 Gigabases (Gb) ofraw data yield and three different BED files (UCSC coding exons,vendor specific, intersection of all 4 vendor BED files) wereapplied independently to ensure equal comparison betweenmethods and capture content. Samples that fell below 5 Gb wereprocessed ‘as is’ but removed from the final analysis. Subsequentdata analysis was performed using CIDRs in-house ResearchWhole Exome pipeline.
IntroductionNext Generation Sequencing technology has been a driving forcefor reducing sequencing costs and enhancing genome technology,allowing investigators and clinicians the opportunity to betterdelineate an individuals’ genetic makeup. For facilities providinggenetic services, maintaining the balance of generating highquality data at the lowest cost is a continuous effort requiringrecurrent evaluation and optimization of protocols. Inpreparation for higher throughput of exome sequencing using theNovaSeq, we evaluated uniformity, coverage and cost ofsequencing to achieve 90% on target at 10x, 20x and 30x depthacross four different exome capture products:
• SureSelect Human All Exon v7 (Agilent Technologies)
• xGEN Exome Research Panel v1.0 (Integrated DNA Technologies)
• Prime Exome (Roche Sequencing)
• Human Core Exome (Twist Bioscience)
Methods – Library PreparationTo reduce experimental variables and minimize bias betweenlibraries, all libraries were prepared using the Kapa Hyper preplibrary prep kit. Table 1 summarizes the parameters used. 50ng ofDNA from 8 different HapMap samples was sheared using theCovaris LE220. End Repaired/A-Tailed libraries were ligated withIDT dual indexed adapters and amplified using the Kapa HiFi DNAPolymerase for 8 cycles. Agilent captures were hybridized assingle sample reactions using 750ng of library as input. IDT, Rocheand Twist captures were hybridized as pools of 8 samples using750ng, 625ng, 187.5ng of library input, respectively. AllHybridization and Post-hyb capture & washes were performedaccording to each respective manufacturer’s protocol. Post HybPCR was performed using the Kapa HiFi DNA Polymerase for 10cycles, regardless of capture product.
Table 1 – Library Prep Methods
Results – Sequencing Data QCTable 3 compares QC metrics generated during sequencing when usingvendor specific target BED files. Data quality (TiTv, Percent SNVOnTarget SNP138 and Count SNV OnTarget) across captures isconsistent. Insert size, duplication rates and library complexity varybetween capture products when using the same library prep methodsand amplification, indicating that capture also impacts these metrics.
Table 3 – Sequencing QC metrics based on vendor specific target BED files.
Figure 3: GC dropout (x-axis) vs AT dropout (y-axis) as generatedby Picard, which measures how under covered a low/high % GCregion is relative to the mean. Points are colored by Captureproduct (Blue = Agilent, Red = IDT, Yellow = Roche, Green =Twist). Vendor specific target BED files were used.
GC Dropout
AT
Dro
po
ut
Figure 2: Depicts uniformity across capture products. Datagenerated from Depth of Coverage is plotted as the count ofbases (y-axis) at a given depth (x-axis) for all bases capturedfrom NA12878. Each panel varies by BED File. Lines are coloredby capture product (Blue=Agilent, Red=IDT, Yellow=Roche,Green=Twist).
UCSC Vendor
Depth
Co
un
t
Results – Coverage UniformityCoverage uniformity was analyzed in several ways. Figure 2represents sample statistic summaries generated from GATK -Depth of Coverage (DOC). UCSC exon coding and vendorspecific BED files were used, however, the 4-way intersectionBED file results were comparable to the UCSC plot and notshown here. Tight/narrow peaks at the highest depth representcaptures with the most uniformity. In addition to DOC metrics,we analyzed GC/AT dropout which measures how undercovered a low/high %GC region is relative to the mean (Figure3). Regions of low coverage in clinically relevant genes wereviewed in IGV and are shown in Figure 4.
Figure 4a-c : Integrative Genomics Viewer (IGV) plot of NA12878 at regions of low coverage identified within clinically relevant genes. In lower panel: gene track, target BED files for Agilent, IDT, Roche and Twist, respectively.
Agilent
IDT
Roche
Twist
ALX3 – Frontonasal dysplasia
Agilent
IDT
Roche
Twist
SKI – Shprintzen-Goldberg Craniosynostosis Syndrome
Agilent
IDT
Roche
Twist
PDE4D – Acrodysostosis 2
Results – Coverage Uniformity cont’d
Discussion & Conclusions• Captures with higher efficiencies are able to maintain coverage as depth increases with the
same amount of sequencing yield. Determining which capture product to use may depend onproject specific needs (i.e. coverage of specific targets, depth requirements, availableautomation etc.) and cost requirements.
• Library Prep Methods – for purposes of this experiment to minimize bias, we utilized a singlelibrary prep method that has been optimized in our facility. Vendor specific library prepreagents/protocols may provide additional optimizations that are not addressed here.
• GC/AT rich & Low coverage regions – PCR enzyme performance is typically considered a ‘culprit’of these difficult to sequence regions. However, capture products may also influence coverageof these regions.
• Pre-capture pooling can provide a ~2.5 fold reduction in reagent costs. However, the unbalancedpools obtained post capture increases the redo rates as much as ~30%. For higher through-putfacilities, this adds considerably to the labor costs and overall time to completion for projects.Capture products that maintain high efficiency will decrease the amount of sequencing requiredand when combined with the lower cost of sequencing on the NovaSeq will offset the addedcost of pre-capture pooling making this a more feasible workflow.
Results – Design CoverageCoverage summaries were generated using Galaxy v1.0.0coverage tool and summarized in Table 2. Each vendors targetBED file and the 4-way intersection was used to determine designcoverage across the UCSC Coding exons by region (exoniccoordinates) and by base. The total target footprint for eachvendor is also listed.
Table 2 - UCSC Exon Coding Coverage Summary by Capture Product
UCSC Coding Exon (35.20 Mb)
BED File Coverage Summary
Agilent
(49.48 Mb)
IDT
(50.84 Mb)
Roche
(37.24 Mb)
Twist
(33.05 Mb)
4-way
Intersection
(26.91 Mb)
% Regions not covered 1.74 3.72 5.76 5.20 6.69
% Regions covered <50% 1.03 0.46 0.89 0.67 1.93
% Regions covered >50% & <100% 38.19 3.35 3.97 2.71 37.97
% Regions covered 100% 59.04 92.47 89.38 91.43 53.42
% Bases not covered 7.77 5.02 7.89 6.49 13.95
% Bases covered 92.23 94.98 92.11 93.51 86.05
Vendor DNA Input Library Prep
Pre Capture
PCR Cycles
Capture
Pooling Hyb Input
Post-Capture
PCR Cycles
Post-Capture
PCR Enzyme
Agilent 50 ng Kapa Hyper Prep 8 Single 750 ng 10 Kapa HiFi
IDT 50 ng Kapa Hyper Prep 8 8-plex 750 ng 10 Kapa HiFi
Roche 50 ng Kapa Hyper Prep 8 8-plex 625 ng 10 Kapa HiFi
Twist 50 ng Kapa Hyper Prep 8 8-plex 187.5 ng 10 Kapa HiFi
Sequencing QC Metric Agilent IDT Roche Twist
n = Downsampled to 5 Gb 6 3 4 3
Percent Selection 73.0 87.0 85.5 82.7
Percent Duplication 4.1 4.3 7.6 2.7
Estimated Library Size (millions) 317 316 157 429
Mean Insert Size 250 241 287 216
Percent Total Concordance 99.75 99.82 99.72 99.60
Sensitivity 2 Het 98.82 99.13 99.24 99.38
Percent SNV on Target dbSNP138 98.55 99.14 99.35 99.23
Count SNV on Target 41072 39114 27483 26453
Known TiTv Ratio 3.07 3.07 3.09 3.11
Known TiTv Count 21668 20370 21261 22743
