The ability to confidently detect low-frequency variants from NGS-based assays has been steadily rising in importance. More than ever, researchers in a wide range of fields are interested in pushing the limits of sensitivity and specificity in variant detection, an area that has previously been limited by sample preparation, amplification artifacts, and sequencing errors. Accurate detection of variants has recently been refined by the incorporation of unique molecular identifiers (UMIs) and unique dual indexes (UDIs) on Illumina platforms. As library preparation has a direct impact on the quality of sequencing results, ThruPLEX Tag-Seq HV chemistry has been engineered and optimized to produce highly diverse libraries with reproducible sequencing performance from 5 to 200 ng of input DNA. Its single-tube workflow (Figure 1) is the simplest in the industry and enables the addition of adapters containing UMIs and Illumina-compatible UDIs in three short steps. The sample never leaves the tube, ensuring accurate sample tracking, minimizing handling errors, and preventing loss of valuable samples.

Figure 1. ThruPLEX Tag-Seq HV single-tube library preparation workflow. This protocol consists of three simple steps that take place in the same PCR tube or well, thus eliminating the need to purify or transfer the sample material.

The ThruPLEX adapters have been redesigned to include discrete UMIs (Figure 2, left panel). Each pool of adapters contains 144 unique sequence combinations with a Hamming distance above 6. The adapters were carefully balanced to obtain equal representation. The seven base UMIs are located at the beginning of the reads, ensuring easy demultiplexing of the samples to simplify analysis (see Methods for details). The UMIs used to “tag” DNA molecules are then processed to identify consensus sequences and reduce the false-positive variants introduced by amplification or sequencing errors (Figure 2, right panel).

Figure 2. ThruPLEX Tag-Seq HV includes 144 discrete UMIs. This workflow is designed to eliminate the ambiguity in variant calling by reducing the false-positive calls resulting from DNA polymerase and sequencing errors.

Design of discrete UMIs for optimal performance

The ThruPLEX Tag-Seq HV UMI sequences were carefully selected for their high Hamming distance, providing high dissimilarity from each other. Since the UMI sequences are known and distinct from each other, it is possible to correct potential sequencing errors before grouping the reads by UMI family. (See Methods for more details.) The UMI sequences were also selected for their color balance in order to enable high-quality sequencing on Illumina platforms (Figure 3, Panel A). The concentration of each UMI adapter was optimized to produce an even representation of the 144 UMI combinations at every DNA input level. The percentage of each UMI combination was calculated from the total number of reads. As expected, the average representation of a given UMI was 0.7% (100%/144 combinations) and deviated less than 50% of the mean (±0.35%) (Figure 3, Panel B).

Figure 3. ThruPLEX Tag-Seq HV UMI design strategy. Panel A. UMI sequences were selected to ensure high-quality sequencing on Illumina platforms. Panel B. Concentration was optimized to obtain even representation, with DNA inputs of 50 ng (purple) and 5 ng (green) showing the same UMI distribution.

Improved genome coverage

The human genome contains an average of 41% GC content, mostly ranging from 20% to 65%. However, promoter regions are known for their high GC content and their overlap with CpG islands. ThruPLEX Tag-Seq HV has been optimized to capture all regions of the genome, enabling high sensitivity in mutation detection, even in extreme GC content regions. To demonstrate this benefit, the system was used to perform low-pass whole-genome sequencing, and Picard’s CollectGCBiasMetrics was used to assess the GC bias. ThruPLEX Tag-Seq HV showed a normalized coverage close to the expected theoretical value of 1, even at AT- and GC-rich windows (Figure 4).

Figure 4. Low GC bias across the human genome. Libraries were prepared from 5 ng and 50 ng of sheared human genomic DNA (Horizon Discovery Quantitative Multiplex Reference Standard, Cat. # HD701) using ThruPLEX Tag-Seq HV. Libraries were sequenced on an Illumina MiSeq® instrument, and reads were aligned to the human genome (HG19) using bowtie2. The alignment metrics and GC bias were calculated using Picard tools.

Reliable detection of variants

Enrichment of regions of interest within the human genome is essential for the detection of low-frequency variants at an affordable cost. Additionally, multiple samples can be processed together in a single enrichment and multiplexed on a sequencer. To test the ability of ThruPLEX Tag-Seq HV to detect low-frequency variants, replicate libraries were produced from reference DNA presenting characterized variants inside cancer-related genes at known allele frequencies. Cancer-related genes were enriched using an IDT xGEN Pan Cancer Panel. The panel targets 800 kb of the human genome—more specifically, 127 genes frequently mutated in solid tumors, including KRAS, EGFR, and others. The variants were identified from the aligned BAM files before deduplication, based on the read pair start and end coordinates, and by using the UMI consensus reads (see Methods for details). As depicted in Table 1, the average coverage of the targeted regions from the reads deduplicated by coordinates or by UMI consensus was similar and concordant with the PCR duplication reported by Picard MarkedDuplicates. It is also important to note that the UMI representation was unaffected by the hybridization capture process, showing the unbiased ligation of the adapters (Figure 5).

Figure 5. UMI representation before and after hybridization capture. The UMI distribution was unaffected by the hybridization capture process.

All characterized variants were determined from the reads corrected using the UMI algorithm at the expected allele frequencies in the replicate libraries processed with HD701, as depicted in Table 2 below. 

Detection of multiple low-frequency variants from cfDNA surrogates

The sensitivity of the assay was evaluated by calculating the detection rate of low-frequency variants in a reference standard. ThruPLEX Tag-Seq HV libraries were generated from 10 ng of Quan-Plex Patient-Like ctDNA Reference Standard (AccuRef) containing a series of characterized mutations occurring at 0%, 1%, and 5%. Cancer-related genes were enriched using an IDT xGEN Pan Cancer Panel. The hybridization resulted in high enrichment of the targets, as ~79% of the reads were on or near targets for 10-ng inputs. As depicted in Table 3, the average coverage of the targeted regions from the reads deduplicated by coordinates or by UMI consensus was similar and concordant with the PCR duplication reported by Picard MarkedDuplicates.

Table 4 below displays two replicate measurements of the expected variants at the frequency identified in the reads collapsed using the UMI algorithm detailed in the Methods. The variants were identified around their respective expected allele frequency of 1% or 5%. No variants were detected in the negative control (0%) at the positions of the characterized variants.

A large number of inconsistent variants with allele frequencies above 0.5% are observed in not deduplicated files or files deduplicated solely by coordinates. However, the number of variants identified in the aligned files generated from the consensus sequences from the UMI correction was reproducibly noticeably lower, demonstrating the power of the UMIs in correcting for amplification and sequencing errors. More specifically, a significant number of variants of low or random allele frequencies were reported in near proximity of the expected EGFR L858R variant in the files not deduplicated or deduplicated by coordinate only. The consensus sequence reads from the UMI collapsing and filtering cleaned up false-positive variants but retained expected variants.

Figure 6. False-positive variants are filtered out when using the fgbio UMI collapsing algorithm. Panel A. The number of variants identified over 0.5% in files deduplicated by coordinates and reads collapsed using UMI algorithm. Panel B. Visualization of the variants above 0.5%.

The low GC bias observed in ThruPLEX Tag-Seq HV empowers the even coverage of GC- and AT-rich regions. The 5'UTR of CEBPA is known for its very high GC content overlapping a CpG island. As depicted in Figure 7, the coverage is not affected by the varying GC content from replicate 10-ng DNA inputs.

Figure 7. Even coverage of GC-rich regions. Constant read coverage is observed along the CEBPA gene. The BAM files were visualized using the Integrated Genome Viewer (IGV, Broad Institute).

The ThruPLEX Tag-Seq HV kit has been engineered and optimized to generate DNA libraries with high molecular complexity and balanced GC representation from input volumes of up to 30 µl. The entire three-step workflow takes place in a single tube or well, enabling the conservation of valuable samples while ensuring accurate sample tracking. No intermediate purification steps and no sample transfers are necessary, thus preventing handling errors and loss of valuable samples. With the incorporation of unique molecular identifiers, ThruPLEX Tag-Seq HV offers added power in correcting for polymerase and sequencing errors—key advantages when detecting low-frequency variants. The UMIs included in the system have been carefully selected for even representation and the best possible performance both in error correction and sequencing results on Illumina platforms.

DNA preparation

Horizon Discovery (HD701) reference gDNA was sheared at 250 bp using a Covaris M220. Sheared input material was evaluated for size distribution using an Agilent 2100 BioAnalyzer. The concentration of these samples was measured using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific).

Library preparation and hybridization capture

Libraries were prepared following the ThruPLEX Tag-Seq HV kit user manual. Library size profiles were assessed on the Agilent 2100 BioAnalyzer and quantified using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific). The ThruPLEX Tag-Seq HV libraries were sequenced on an Illumina MiSeq for low-pass whole-genome sequencing or enriched by hybridization capture using the xGEN Pan Cancer Panel v1.5 (IDT, Cat. # 1056205) according to the manufacturer's instructions, and then sequenced on an Illumina NextSeq 500.

Data analysis

Illumina adapter trimming was performed using Trimmomatic, and reads were downsampled using setk. The reads were aligned to the human genome assembly HG19 using bowtie2. Duplicates, alignment metrics, insert size, and GC bias were calculated using Picard MarkDuplicates, CollectAlignmentSummaryMetrics, CollectInsertSizeMetrics, and CollectGcBiasMetrics, respectively. Hybridization capture metrics were determined using Picard CollectHsMetrics. Consensus reads based on UMI algorithm were obtained using fgbio tools, and variant calling was performed using Vardict. Additional information can be found here.