Validation of a rapid, saliva-based, and ultra-sensitive SARS-CoV-2 screening system for pandemic-scale infection surveillance

Abstract

Without any realistic prospect of comprehensive global vaccine coverage and lasting immunity, control of pandemics such as COVID-19 will require implementation of large-scale, rapid identification and isolation of infectious individuals to limit further transmission. Here, we describe an automated, high-throughput integrated screening platform, incorporating saliva-based loop-mediated isothermal amplification (LAMP) technology, that is designed for population-scale sensitive detection of infectious carriers of SARS-CoV-2 RNA. Central to this surveillance system is the "Sentinel" testing instrument, which is capable of reporting results within 25 min of saliva sample collection with a throughput of up to 3840 results per hour. It incorporates continuous flow loading of samples at random intervals to cost-effectively adjust for fluctuations in testing demand. Independent validation of our saliva-based RT-LAMP technology on an automated LAMP instrument coined the "Sentinel", found 98.7% sensitivity, 97.6% specificity, and 98% accuracy against a RT-PCR comparator assay, confirming its suitability for surveillance screening. This Sentinel surveillance system offers a feasible and scalable approach to complement vaccination, to curb the spread of COVID-19 variants, and control future pandemics to save lives.

Figure 1

The presence of mineral oil markedly reduces the rate of production of false positive RT-LAMP reactions. Two assay chemistries were compared: NEB WarmStart Colorimetric LAMP with UDG (M1804) and Hayat Rapid Colorimetric & Fluorometric One Step LAMP SARS-CoV-2 Test Kit, each set up with and without 15 μl mineral oil overlay. Twenty-one identical replicate negative control reactions were set up per condition with a single saliva sample negative for SARS-CoV-2 diluted in VTM and AviSal at a ratio of 1:1:2 and heat inactivated for 10 min at 95 °C. The sample was added to give 5% (NEB) and 3.75% (Hayat) final concentrations of crude saliva in a 25 μl reaction volume. + M.O. with mineral oil overlay; − M.O. without mineral oil. In a typical 30-min reaction runtime, only Hayat chemistry resulted in no false positives and 100% specificity. For the NEB chemistry, false positives were observed after 20 min, even with a mineral oil overlay. Mineral oil overlay markedly reduces the false positive rate.

Figure 2

Process flow from saliva collection to data reporting. (A) Schematic diagram of the main steps of the process flow. (B) Sentinel robot for LAMP reaction setup, incubation, and detection. Individual process steps: 1. Saliva collection via microswabs into test tubes containing inactivation solution. 2. Sample tubes with saliva dilutions were heat-inactivated for 10 min at 95 °C. 3. Automated uncapping of a complete sample tube rack. 4. Sample tube barcode reading during Sentinel robot loading. 5. The automated pipetting system adds a selected sample volume to the RT-LAMP master mix in clear reagent plates. 6. Plate sealing by automated heat sealer. 7. Plate crane transport of the reaction plate onto a 65 °C incubation field. 8. LAMP reactions were incubated at 65 °C for 25 min; digital, parallel scanning of multiple 96-well microplates. 9. Real-time, secure reporting of deidentified data and analysis. 10. Disposal of scanned microplates.

Figure 3

Stability of saliva samples during short-term storage determined by RT-PCR and RT-LAMP. Five sample pools were prepared from ten saliva samples to give a range of viral loads corresponding to predicted Ct values (from extracted RNA) from 24.6 to 35.0 for the N-gene and 25.8 to 39.1 for ORF1ab. Sample pools were diluted with an equal volume of AviSal buffer and stored short-term under the following test conditions prior to heat inactivation: room temperature (RT) for 0, 6, 24 and 48 h (t = 0, 6 h, 24 h and 48 h) or 4 °C for 48 h. Samples were then heat inactivated at 95 °C for 10 min and frozen. Upon thawing, samples were stored post inactivation at either 4 °C or RT for 2 h or 4 °C for 48 h (t = 0 only). The 4 °C incubation was taken as baseline as a proxy for no post-inactivation storage. RT-PCR (BGI Real-Time Fluorescent RT-PCR Kit for Detecting SARS-CoV-2), targeting ORF1ab (A) and the N gene (B), and RT-LAMP (Hayat Genetics) (C,D) assays were performed on samples subjected to each of these conditions. To understand the effect of freezing samples prior to heat inactivation, for the t = 0 timepoint, a fresh versus frozen comparison was performed. Viral loads (as measured by RT-PCR on extracted RNA) of sample pools #3, 4 and 5 are too low for 100% sensitivity of detection by RT-LAMP—all have Ct ≥ 33 for N gene. Hence, not all reactions were positive within the 30 min runtime (D). All Ct (Cycle Threshold) and TTT (Time to Threshold) values were calculated as the mean of technical duplicates.

Figure 4

Comparison of sensitivity of different assay technologies grouped by Ct values. Five Light blue—Hayat/Avicena test, grey—direct RT-LAMP, dark blue—RAT/LFD/LFA. Direct LAMP test had initial sensitivity 68% and lost it significantly from Ct 25. RAT/LFD/LFA tests had initial sensitivity 50% and lost it completely from Ct 25 and above.