A flexible and efficient template format for circular consensus sequencing and SNP detection

Abstract

A novel template design for single-molecule sequencing is introduced, a structure we refer to as a SMRTbell template. This structure consists of a double-stranded portion, containing the insert of interest, and a single-stranded hairpin loop on either end, which provides a site for primer binding. Structurally, this format resembles a linear double-stranded molecule, and yet it is topologically circular. When placed into a single-molecule sequencing reaction, the SMRTbell template format enables a consensus sequence to be obtained from multiple passes on a single molecule. Furthermore, this consensus sequence is obtained from both the sense and antisense strands of the insert region. In this article, we present a universal method for constructing these templates, as well as an application of their use. We demonstrate the generation of high-quality consensus accuracy from single molecules, as well as the use of SMRTbell templates in the identification of rare sequence variants.

Figure 1.

Schematic of a SMRTbell™ template. (A) A SMRTbell template consists of a double-stranded region (the insert) flanked by two hairpin loops. The hairpin loops present a single-stranded region to which a sequencing primer can bind (orange). (B) As a strand-displacing polymerase (gray) extends a primer from one of the hairpin loops, it uses one strand as the template strand and displaces the other. When the polymerase returns to the 5′-end of the primer, it begins strand displacement of the primer and continues to synthesize DNA (moving in the direction of the blue arrow). Therefore, the length of sequence obtained from these templates is not limited by the insert length. Furthermore, the resulting sequence is derived from both sense- and anti-sense strands.

Figure 2.

Method of construction of SMRTbell templates. (A) Method for template generation. The boxes on the left depict the process for SMRTbell generation from a PCR fragment. The boxes on the right illustrate the process for randomly generated fragments of DNA. Whereas PCR products are produced in a defined length and, if digested with a restriction enzyme, contain defined overhangs, genomic DNA must be sheared down to an appropriate size, end polished to generate blunt ends and then extended by 1 nt to generate a single A overhang. This is represented schematically with a generic ‘N’ overhang. Hairpin loops with an overhang complementary to the overhang on the DNA fragments are ligated to the ends of the insert in the final step. (B) The single-nucleotide polymorphism-containing constructs used in this work. The two templates contain an insert of ∼140 bp, with either a T/A or a G/C base pair at the site of the polymorphism (indicated in bold).

Figure 3.

Bulk extension products of two SMRTbell templates. Lanes 1–4 contain the product from a template derived from strain FDA 209 (the ‘T’ allele) at timepoints of 2, 10, 30 and 60 min. Lanes 5–8 contain the product from a template derived from strain Mu50 (the ‘C’ allele) at timepoints of 2, 10, 30 and 60 min. Approximate sizes of products were determined relative to a 1 kb molecular weight ladder (indicated on the left). The position of the sequencing primer is indicated with an arrow.

Figure 4.

Demonstration of single-molecule traces that include both sense and antisense strands of a single molecule. The pulses within the sequencing traces are colored according to location within the template, with blue corresponding to the sense strand, orange corresponding to the antisense strand, and light blue and green corresponding to the hairpin adapters. (A) A representative trace from an aroE132 template. In this example, incorporations are observed on four complete passes of the template, generating four-fold coverage of the sense strand and four-fold coverage of the antisense strand. (B) A representative trace from a 1000-bp PCR product derived from the PhiX174 genome. In this case, there are two sub-reads that correspond to the sense strand and one to the antisense strand.

Figure 5.

Comparison of measured empirical quality values to predicted consensus quality values. In the test set, data are binned around the predicted consensus QVs and the numbers of errors are tallied for calculation of the empirical consensus base call QV. We normalize the number of total base calls of each bin to 10 000. We repeat this sampling procedure 10 times for each predicted QV to derive the error bar to represent sampling errors.

Figure 6.

Comparison of the expected SNP frequency to the measured SNP frequency. The SMRTbell templates derived from each of the alleles were mixed in ratios of 0:100, 2.5:97.5, 5:95, 10:90, 25:75, 50:50 and 100:0 (listed as T:C). The frequency of calls for all four possible bases are shown.

Figure 7.

The unfiltered distributions of EQV as a function of single-molecule coverage (each full synthesis of a SMRTbell corresponds to 2× single-molecule coverage representing the forward and reverse strands). The upper and lower error bars represent the 75th and 25th quartile, respectively.