On January 10, 2020, the first SARS-CoV-2 genome sequence was released online (7). That day, we designed 2 complementary panels of primers to amplify the virus genome for sequencing.

For the first panel, we used the PRIMAL primer design tool (5) to design multiplex PCRs to amplify the genome by using only a few PCRs (Appendix). The final design consists of 6 pools of primers optimized for sensitivity and assay flexibility. The amplicons average 550 bp with 100-bp overlaps to enable sequencing on either the Oxford MinION or Illumina MiSeq.

For the second panel, we designed sets of primers to generate nested, tiling amplicons across the SARS-CoV-2 genome (Appendix) for enhanced sensitivity in samples with lower viral loads. Each amplicon is 322–1,030 bp with an average overlap of 80 bp. These amplicons are designed to be amplified and sequenced individually on Sanger instruments but can also be pooled for sequencing on next-generation sequencing platforms.

To determine the sensitivity of each sequencing strategy, we generated a set of 6 ten-fold serial dilutions of a SARS-CoV-2 isolate (8). Virus RNA was diluted into a constant background of A549 human cell line total nucleic acid (RNaseP cycle threshold [C_t] 29). We quantitated each dilution by using the Centers for Disease Control and Prevention SARS-CoV-2 real-time reverse transcription PCR for the nucleocapsid 2 gene (9). The 6 dilutions spanned C_t values from 22 to 37, corresponding to ≈2 × 10⁰ to 1.8 × 10⁵ copies. We amplified triplicate samples at each dilution by using the multiplex PCR pools. Next, we pooled, barcoded, and made libraries from amplicons of each sample by using the ligation-based kit and PCR barcode expansion kit (Appendix). MinION sequencing was performed on an R9.4.1 or R10.3 flow cell (Oxford) until we obtained >1–2 million raw reads. From those reads, 50%–60% of them could be demultiplexed. In addition, we sequenced these amplicons by using the Illumina MiSeq for comparison (Appendix).

Figure 1

Figure 1. Limits of detection for sequencing severe acute respiratory syndrome coronavirus 2. Triplicate serial dilutions of virus isolate A12 (8) were amplified by using the singleplex or multiplex primer set....

For MinION sequencing, the reads were basecalled and analyzed by using an in-house read mapping pipeline (Appendix). For samples with C_t <29, we obtained >99% SARS-CoV-2 reads and >99% genome coverage at 20× depth, decreasing to an average of 93% genome coverage at C_t 33.2 and 48% at C_t 35 (Figure 1, panels A, B). Furthermore, we were able to obtain full genomes at >20× reading depth within the first 40–60 min of sequencing (Figure 1, panel C).

Consensus accuracy, including single-nucleotide polymorphisms and indels, is critical for determining coronavirus lineage and transmission networks. For high-consensus–level accuracy, we filtered reads based on length, mapped them to the reference sequence (GenBank accession no. RefSeq NC_045512), trimmed primers based on position, and called variants with Medaka ((https://github.com/nanoporetech/medaka) (Appendix). Each Medaka variant was filtered by coverage depth (>20×) and by the Medaka model-derived variant quality (>30). We used the variant quality score as a heuristic to filter remaining noise from the Medaka variants compared with Sanger-derived sequences. After these steps, the data approaches 100% consensus accuracy (Table 1). Identical results were found by using the R9.4.1 pore through samples with C_t values through 33.2. The larger deletions in some of the samples with C_t values >33.2 (Table 1) do not appear to be sequencing errors because they are also detected as minor populations within higher-titer samples.

In the MiSeq data, we observed a similar trend in percent genome coverage at 100× depth, and a slightly lower percentage mapped reads compared with Nanopore data (Figure 1, panels A, B). Increased read depth using the MiSeq potentially enables increased sample throughput. However, the number of available unique dual indices limits actual throughput.

For the nested, singleplex PCR panel, we amplified the same serial dilutions with each nested primer set (Appendix). The endpoint dilution for full-genome coverage is a C_t ≈35 (Figure 1, panel B). At the C_t 37 dilution, we observed major amplicon dropout; at this dilution, there are <10 copies of the genome on average/reaction.

These protocols enabled rapid sequencing of initial clinical cases of infection with SARS-CoV-2 in the United States. For these cases, we amplified the virus genome by using PCR and sequencing the amplicons by using the MinION and Sanger instruments to validate MinION consensus accuracy. The MinION produced full-length genomes in <20 min of sequencing, and Sanger data was available the following day.

Figure 2

Figure 2. Sequencing of severe acute respiratory syndrome coronavirus 2 clinical samples. A, B) Percentage mapped (A) and percentage genome coverage (B) for 167 clinical severe acute respiratory syndrome coronavirus 2 samples amplified...

We used the multiplex PCR strategy for subsequent SARS-CoV-2 clinical cases (n = 167) with C_t values ranging from 15.7 to 40 (mean 28.8, median 29.1). In cases with a C_t <30, we observed an average of 99.02% specific reads and 99.2% genome coverage at >20× depth (Figure 2, panels A, B). Between C_t 30 and 33, genome coverage varied by sample, and decreased dramatically at higher C_t values, analogous to the isolate validation data. For these samples, we multiplexed 20–40 barcoded samples/flowcell. Enough data are obtained with 60 min of MinION sequencing for most samples, although for higher titer samples, 10–20 min of sequencing is sufficient (Figure 2, panel C).

Up-to-date primer sequences, protocols, and analysis scripts are available on GitHub (https://github.com/CDCgov/SARS-CoV-2_Sequencing/tree/master/protocols/CDC-Comprehensive). Data from this study is deposited in the National Center for Biotechnology Information Sequence Read Archive (BioProjects PRJNA622817 and PRJNA610248).

Full-genome sequencing is a critical tool in understanding emerging viruses. Initial sequencing of SARS-CoV-2 showed limited genetic variation (10,11). However, some signature variants have been useful for describing the introduction and transmission dynamics of the virus (12,13; T. Bedford et al., unpub. data, https://doi.org/10.1101/2020.04.02.20051417; M. Worobey et al., unpub. data, https://doi.org/10.1101/2020.05.21.109322).

We provide 2 validated PCR target-enrichment strategies that can be used with MinION, MiSeq, and Sanger platforms for sequencing SARS-CoV-2 clinical specimens. These strategies ensure that most laboratories have access to >1 strategies.

The multiplex PCR strategy is effective at generating full genome sequences up to C_t 33. The singleplex, nested PCR is effective up to C_t 35, varying based on sample quality. The turnaround time for the multiplex PCR MinION protocol is ≈8 hours from nucleic acid to consensus sequence and that for Sanger sequencing is ≈14–18 hours (Table 2). The multiplex PCR protocols offer an efficient, cost-effective, scalable system, and add little time and complexity as sample numbers increase (Table 2). Results from this study suggest multiplex PCR might be used effectively for routine sequencing, complemented by singleplex, nested PCR for low-titer virus samples and confirmation sequencing.