Highly Sensitive Lineage Discrimination of SARS-CoV-2 Variants through Allele-Specific Probe PCR

Study design.

Samples used for method development and assessment in this study were collected as part of two clinical trials conducted during periods of VoC and VoI emergence in their study locations. The variants were hypothesized to have an impact on outcomes seen within these trials; thus, highly sensitive methods for identifying specific variants were sought and developed. All samples were processed by ASP-PCR, NGS, or both methods, and no sample results were excluded from this study.

Sample collection.

The REMAP-CAP Immunological Domain was an international, open-label, randomized convalescent plasma trial enrolling patients aged 18 years or older receiving intensive care-level organ support. Patients were eligible given admission to intensive care within 48 h of hospital admission and a positive SARS-CoV-2 microbiological test. A full trial protocol and efficacy results are available elsewhere (9, 10). Oropharyngeal or nasopharyngeal swabs were taken prior to randomization, transported to a central academic hospital and frozen at −80°C, shipped on dry ice to a central testing laboratory, and processed as described below.

COV003 is a participant-blinded, randomized, controlled phase 3 multisite trial that began in Brazil on 23 June 2020 assessing the efficacy of the ChAdOx1 nCoV-19 vaccine against symptomatic SARS-CoV-2 infection. Efficacy, safety data, the full study protocol, and exploratory analysis of lineage-specific efficacy are available elsewhere (8, 14, 15). Volunteers recruited for the trial were 18 years or older and at high risk of exposure (e.g., health care workers). Volunteers who developed primary COVID-19 symptoms were asked to contact their study site. Nasopharyngeal swab samples that were separately confirmed SARS-CoV-2 positive using commercial nucleic acid amplification test (NAAT) assays at local laboratories were shipped to a central testing laboratory and processed as described below.

Ethics statement.

REMAP-CAP and COV003 were conducted according to the principles of the latest version of the Declaration of Helsinki (version Fortaleza 2013). REMAP-CAP was performed in accordance with regulatory and legal requirements (EudraCT number: 2015-002340-14) and was approved by London-Surrey Borders Research Ethics Committee London Centre (18/LO/0660). COV003 was approved by the Brazilian National Research Ethics Committee (ref: 32604920.5.0000.5505) and the Oxford Tropical Research Ethics Committee (ref: 20-36).

Nucleic acid extraction.

REMAP-CAP samples and research reagent 19/304 (National Institute of Biological Standards and Control [NIBSC]) containing encapsulated, quantified full-length SARS-CoV-2 RNA were extracted using either the QIAamp viral RNA minikit (Qiagen) as described previously or the Quick-DNA/RNA viral kit (Zymo Research) (7). COV003 samples were extracted using the Quick-DNA/RNA viral kit as described previously (8).

Real-time quantitative PCR.

SARS-CoV-2 viral RNA was detected and quantified by real-time quantitative PCR (RT-qPCR) as previously described using oligonucleotides listed in Table S1 in the supplemental material (ATDBio) (7). SARS-CoV-2 RNA was quantified using a standard curve of research reagent 19/304 serially diluted from 10,000 copies/reaction to 100 copies/reaction (REMAP-CAP) or 1,000 copies/reaction to 10 copies/reaction (COV003). RT-qPCR cycle threshold (C_T) values were converted to copy number/reaction by use of the standard curve and to international units (IU)/milliliter by the conversion rate in the product sheet.

ASP-PCR.

SNP sites targeted for lineage discrimination were chosen based on their lineage-specific predictive value estimated using publicly available sequence data published on GISAID (16).

ASP-PCR was performed using the QuantiTect probe RT-PCR kit (Qiagen) with 5 μL of extracted RNA in a 25-μL reaction volume on an Applied Biosystems StepOnePlus real-time PCR system using the genotyping program. SNPs were designated based on their clustering with discrimination controls. Serially diluted cDNA aliquots of sequence-confirmed samples were used as discrimination controls; ultrapure water served as negative controls. Samples that failed to achieve a change in signal in either probe greater than those of the no-template controls or lacked evidence of amplification were designated “undetermined.” Reaction conditions (annealing and extension temperature and time and oligonucleotide concentrations) were optimized using serially diluted cDNA generated from samples of known lineage. The RT-PCR settings for each ASP-PCR are described in Table S2.

To test the effects of modified oligonucleotides, the D1118H/Alpha ASP oligonucleotide set was redesigned using locked nucleic acids (LNA) over the SNP site (Table S1). To measure the impact of low-molecular-weight amides on differentiation, 2-pyrrolidinone (Merck) was added to the qPCR master mix at a reaction concentration of 0.4 M as recommended in the work of Chakrabarti and Schutt (17).

Next-generation sequencing.

Samples were sequenced using the veSeq NGS protocol as described previously (2). An extended NGS method description is available in the supplemental material. Two approaches were used to assess lineage assignment by veSeq. NGS-SNP evaluated variant calls at the same sites of interest targeted by ASP-PCR. Consensus sequences with no coverage over the site of interest were deemed “undetermined.” NGS-Pangolin assessed the ability of the widely used Pangolin tool (v3.1.11 for REMAP-CAP and v.2.4.2 for COV003) to make a lineage designation (specifically, testing whether sequences meet the default quality control) (18). Degraded samples were defined as those with high viral loads (≥10⁶ IU/mL) but poor genome coverages (≤20,000 bases with read depth of at least two reads).

Data analysis.

Probit regression models were generated using base function ‘glm’ in R version 4.0.4 (19). Model outputs are available in the supplemental material. Confidence intervals were generated in consultation with open-source code and methods published by Gavin Simpson on his website (20).

Predictive performance.

Method performance on representative data sets was estimated by summing the probit function output for each individual sample. IU/milliliter, N gene C_T value (Tso et al. [21]), or mean ORF1ab C_T value (Choudhuri et al. [22]) were used as the model input variable depending on metadata availability.

Data availability.

Data used to generate figures and statistical analysis in this paper are available from the corresponding author upon request. Broader clinical trial data are available from the corresponding authors of the cited manuscripts.

SNP positive and negative percent agreement.

To assess the suitability of different SNP targets for lineage specification in the ASP-PCR platform, global sequencing data published as part of the GISAID repository were utilized, leveraging the collation performed by https://www.Outbreak.info (16, 23). The country-specific positive percent agreement (PPA), negative percent agreement (NPA), positive predictive value (PPV), and negative predictive value (NPV) for various lineage-defining SNPs were calculated (Table 1). Despite being designed when few sequencing data were available, the SNPs used in this study were shown to be robust and highly accurate for lineage discrimination. Within the United Kingdom, the S:D1118H SNP is present in 99.90% of Alpha infections and detection of the SNP gives a 99.93% chance that the sample is Alpha. The ORF1a:L3468V SNP proved to be overall the best SNP to target P.2 with the highest PPV of analyzed SNPs of 99.69%. For Gamma, the S:K417T SNP demonstrated the lowest PPA among the 10 SNPs analyzed (93.72%) but still maintained very high PPV (99.42%). SNP selection for ASP-PCR cannot solely be on the measure of PPV and NPV and must also consider target suitability for ASP-PCR design (see Discussion).

Analyses of potential target SNPs at a global scale for Alpha, Gamma, P.2, and other VoC/VoI are included in the supplemental material as reference (see Table S3).

ASP-PCR and NGS performance.

RNA was extracted from primary oropharyngeal or nasopharyngeal samples from SARS-CoV-2-infected participants in the REMAP-CAP and COV003 trials and processed via either ASP-PCR, NGS, or both methods. For REMAP-CAP, this included 717 samples for ASP-PCR and 1,130 samples for NGS (661 by both methods). For COV003, this included 365 samples for ASP-PCR and 641 samples for NGS (353 by both methods). NGS results were assessed either for their ability to produce a lineage using the Pangolin software (18) (NGS-Pangolin) or for base calling over the SNPs targeted by ASP-PCR (NGS-SNP).

In terms of raw performance, ASP-PCR, NGS-Pangolin, and NGS-SNP successfully typed 61.8%, 33.8%, and 39.5% of samples tested in REMAP-CAP and 81.9%, 65.8%, and 59.8% in COV003, respectively (Fig. 1A to C and Fig. 2A to C, respectively). For direct comparison, the NGS-Pangolin and ASP-PCR results for samples tested by both methods were compared. NGS-Pangolin was used for comparison as a more standard method. The most common combination of results was being successfully typed by ASP-PCR and not typed by NGS-Pangolin (37.4% in REMAP-CAP, 48.4% in COV003, Table 2). Samples were not randomized to screening method, so direct comparison of raw performances may not be appropriate. High-viral-load samples successfully sequenced early in screening were diverted from ASP-PCR; samples with insufficient viral load for sequencing were tested only by ASP-PCR. This selection bias led to significantly different viral load distributions between screening methods (P = 0.015 for REMAP-CAP and P = 6.3e−6 for COV003, Kolmogorov-Smirnov test; see Fig. S1 in the supplemental material). REMAP-CAP samples tested by ASP-PCR had significantly lower median viral load than NGS (4.40e4 IU/mL versus 1.08e5 IU/mL, P = 0.002, Mann-Whitney U test) as did those in COV003 (4.72e6 IU/mL versus 2.01e7 IU/mL, P = 8.85e−6, Mann-Whitney U test).

To facilitate direct comparisons of the methods, probit regression models were derived for both methods and trials to predict the likelihood of producing a lineage designation (Fig. 1D and Fig. 2D). Across the entire range of viral loads seen in either trial, the probit regression models demonstrated the clear superiority of ASP-PCR over both NGS-SNP and NGS-Pangolin for making a lineage designation. This advantage was most pronounced in samples with C_T values in the range of 26 to 30.

Concerns regarding the quality of RNA from several recruitment sites in COV003, motivated by shorter-than-expected average read lengths, prompted investigation of the utility of ASP-PCR for degraded samples. Degraded samples were defined as those that returned a viral load estimate of >10⁶ IU/mL (C_T value ∼26 or less in CDC N1 RT-qPCR assay) but fewer than 20,000 bases with a read depth of at least two. Subsetting analysis to these samples (n = 113 for ASP-PCR, n = 118 for NGS-Pangolin and NGS-SNP) or those that were nondegraded (all samples that did not meet criteria for degradation; n = 252 for ASP-PCR, n = 523 for NGS-Pangolin and NGS-SNP), the advantages of ASP-PCR are clear (Fig. 3). ASP-PCR was minimally impacted by the sample degradation status, successfully typing 95.6% of the 113 total degraded samples tested by ASP-PCR, while the utility of both NGS-Pangolin and NGS-SNP was massively compromised, typing only 11.0% and 0.0% of the 118 degraded samples processed by NGS, respectively.

For the 179 REMAP-CAP samples with paired ASP-PCR and NGS-Pangolin results, lineage designation was concordant for 99% (178/179) of samples. Base calling the sample indicated S:1118D and a false positive for ASP-PCR. For the 122 COV003 samples with paired ASP-PCR and NGS-Pangolin results, designations were concordant for 85% (104/122) of samples. All 18 discordant samples were called Gamma or P.2 by the ASP-PCR. However, this discrepancy was likely due to Pangolin miscalls due to low sequence coverage, as 17/18 of these discordant samples were assigned as a parent lineage of Gamma and P.2 by Pangolin (B.1.1 or B.1.1.28). These samples were subsequently typed as Gamma or P.2 by phylogenetic reconstruction (see methods in the supplemental material), leading to a more accurate assessment of concordance for COV003 of 99% (121/122).

Predicted performance.

The output of a probit regression is a link function describing the likelihood of an outcome given an input variable. To estimate the utility of ASP-PCR over larger, potentially more representative populations, the REMAP-CAP S:D1118H/Alpha ASP-PCR, NGS-Pangolin, and NGS-SNP probit regression models were applied over published data sets containing C_T or viral load data (Table S4) (21, 22). The advantage of ASP-PCR seen in both the hospitalized and community testing cohorts was driven by large proportions of samples in the C_T 26 to 30 range. In the Tso et al. (21) data set of 57,517 samples from community testing, ASP-PCR was predicted to successfully type 96.3% of samples versus only 67.9% for NGS-Pangolin and 62.9% for NGS-SNP.

Effect of molecular modifications.

Altered reaction chemistry that increases the destabilization of a single base mismatch between primer and target may improve the performance of ASP-PCR for low-viral-load samples. To measure the effects of the addition of low-molecular-weight amides, oligonucleotides modified with LNA, or both treatments, a panel of 83 samples of various viral loads from REMAP-CAP were reextracted from the primary samples and used to screen effects on ASP-PCR performance. In samples untreated with 2-pyrrolidinone, modification of oligonucleotides to LNA was associated with a statistically significant increase in typing percentage (P = 0.04, Fisher’s exact test) (Fig. 4). The improvement came from increased likelihood of typing low-viral-load samples (Fig. S2). The general trend was to increase typing percentage with LNA and decrease with 2-pyrolidinone, but the other comparisons did not reach significance. While the addition of 0.4 M 2-pyrrolidinone did decrease the raw signal observed for the discriminating probe, this improvement in discrimination came at the expense of lineage designation for viral load samples of all concentrations (Fig. S2 and S3).