A cluster of new pneumonia cases was reported to the World Health Organization (WHO) in late 2019 from Wuhan, Hubei Province of China. The causative agent was named as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and led to a global pandemic [1
]. While the major impact of SARS-CoV-2 was attributed to frail and elderly people with co-morbidities, coronavirus disease 2019 (CoVID-19) was mainly spread by asymptomatic or mildly symptomatic patients [2
]. Due to their high mutation rates and recombination events, coronaviruses can infect a range of animal species including humans, avian, rodents, carnivores, chiropters and other mammals [4
]. Prior to the emergence of SARS-CoV-2, a total of six different coronaviruses had been reported to infect humans, including HCoV-229E, HCoV-OC43, HCoV-NL63, HCoV-HKU1, MERS and SARS-CoV-1 (also known as classical SARS). The SARS-CoV-2 belongs to the β-coronavirus of the group 2B within the family of Coronaviridae
The SARS-CoV-2 shares a high level of genetic similarity (up to 96%) with coronaviruses originating from bats [3
]. The genome of β-coronavirus encodes for the replicase complex (ORF1ab), spike (S), envelope (E), membrane (M) and nucleoprotein (N) genes in addition to the several non-structural and accessory proteins in the order from 5′-untranslated to 3′-untranslated regions [3
]. Owing to the nature of viral genetics, the N gene is the most transcribed and highly conserved gene within the Coronaviridae
family and has been a major target for both antigen and antibodies diagnostics. Across the genome, the RNA-dependent RNA polymerase (RdRP), encoded by the ORF1b gene segment, presents a high level of intra-group conservation and therefore is an ideal target for a diagnostic application [5
As is evident by previous coronaviruses associated pandemics and other viral diseases, a highly specific, sensitive and easily deployable diagnostic tool is critical for identification, tracing, rationalizing control measures and documentation of symptomatic and asymptomatic carriers [7
]. Additionally, due to the unavailability of the registered vaccines or effective therapeutics, rapid and reliable diagnostics are of paramount importance to curtail SARS-CoV-2 infection. Because of shortcomings associated with the virus isolation (time consuming and required containment) and cross-reactivities of antigen and antibodies assay, several real-time reverse transcription-polymerase chain reactions (qRT-PCR) and reverse-transcription loop mediated isothermal amplification (RT-LAMP) assays have been developed, validated and commercialized as useful laboratory diagnostics for the detection of SARS-CoV-2 [13
]. However, the majority of these assays are time-consuming and require laboratory-intense instrumentation. Furthermore, they are unable to meet the current unprecedented rapid growth and demand for testing a large proportion of the population, identification of asymptomatic carriers and contact tracing.
Though qRT-PCR remains the gold standard for the diagnosis of SARS-CoV-2, RT-LAMP assays have been demonstrated to produce diagnostic results with increased sensitivity and specificity [14
]. Furthermore, its ability to tolerate PCR inhibitors eliminates the need for laborious RNA extraction and purification methodologies [15
]. Several platforms capable of performing LAMP assays in the field have previously been documented [17
]. However, most platforms have employed fluorescence detection with integrated optical units or a smart phone dock to achieve detection [18
]. Similarly, for colorimetric LAMP assays, smart phone cameras or user interpretation of the colour changes were used to achieve detection [20
]. The fully integrated real-time fluorescence-based platforms are expensive and the smartphone-based platforms are only designed for specific smartphone models. Therefore, to fulfil the need for a standalone colorimetric isothermal nucleic acid amplification platform [22
], we have developed an ultra-low-cost molecular diagnostic device with an integrated single-board computer, imaging camera, artificial intelligence-based image processing algorithm and mobile app.
In this study, we developed a high-resolution comparative genomics analysis-guided novel RT-LAMP assay for the specific and sensitive detection of SARS-CoV-2 in comparison to WHO recommended qRT-PCR assays. In order to provide a simple “sample-to-answer workflow”, an ultra-low-cost and user-friendly diagnostic platform was engineered and further enhanced with a module for automated image acquisition and processing. Artificial intelligence-guided assessment of the LAMP assay provided faster detection of colour changes in the LAMP reaction, further enhancing the assay performance and thus reducing the potential for human error in results interpretation. Finally, the assay was validated on RNA extracted from clinical samples from SARS-CoV-2 suspected patients to demonstrate its real-life applicability.
2. Materials andMethods
2.1. Ethics Statement
This study was conducted in accordance with and approved by the Faculty of Health and Medicine Research Ethics Committee (FHMREC) of Lancaster University. The FHMREC has approved the research application on 8 June 2020 and is available under the reference number FHMREC19112. The study was exempt from requiring specific patient consent as it only involved the use of extracted RNA and existing collections of data or records that contained non-identifiable data about human patients.
2.2. Cells and Viruses
Vero cells and MDCK cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, Carlsbad, CA, USA) supplemented with 10% inactivated foetal bovine serum (FBS) (Gibco, Carlsbad, CA, USA), 2 mM l-glutamine (Gibco) and 100 U/mL penicillin/streptomycin (Gibco) at 37 °C in 5% CO2. Influenza A virus (A/chicken/Pakistan/UDL-01/2008(H9N2), Newcastle disease virus strain LaSota and infectious bronchitis virus strain H120, Vesicular stomatitis virus (VSV) and Sendai virus (SeV) were propagated and used to determine the specificity of the LAMP. All viruses except influenza were titrated on Vero and MDCK cells, respectively, by the standard plaque assay.
2.3. In Silico Nucleotide Sequence Comparisons and Primer Design
To design specific LAMP primer sets for the detection of SARS-CoV-2, all available complete genome sequences were downloaded from GISAID Initiative (https://www.gisaid.org/
), aligned and the conserved part was selected and used as the template of the LAMP primer design. To identify an efficient primer set, three sets of specific LAMP primers were hand-picked and validated using PrimerExplorer V5 software (http://primerexplorer.jp/elamp4.0.0/index.html
). Primers were validated using online BLAST program (http://www.ncbi.nlm.gov/BLAST
) to ensure their specificity.
2.4. Cloning and In Vitro Transcription of RdRP Target Gene
The coding sequence of SARS-CoV-2 RdRp gene was chemically synthesized and cloned into pVAX1 plasmid (Invitrogen, Carlsbad, CA, USA) between KpnI and NotI restriction sites. The plasmid was propagated in DH5α cells and purified using MiniPrep Qiagen Kits (Qiagen, Manchester, UK). The linearized plasmid with pVAX1-RdRP was used for in vitro transcription using the T7 RiboMAX™ Express Large-Scale RNA Production System (Promega, Madison, WI, USA). The copy number of in vitro transcribed RNA was calculated from RNA concentration measured with NanoDrop™ 2000c Spectrophotometers (Thermo-Fisher Scientific, Waltham, MA, USA) in triplicate. RNA products were then purified using the RNeasy MinElute Cleanup Kit (Qiagen, Valencia, CA, USA). A standard curve was generated using dilutions of the standard in vitro transcribed RNAs using SuperScript III Platinum One-Step qRT-PCR Kit as per the manufacturer’s protocol (Invitrogen, Carlsbad, CA, USA) using the CFX96 Touch Real-Time PCR Detection System (BioRad Laboratories, Watford, UK).
2.5. Clinical Sample Processing and Spiking with miR-cel-miR-39-3p RNA
A total of 199 nasopharyngeal swabs were individually collected from CoVID-19 suspected patients through the routine NHS collection procedure for COVID-19 screening. These samples were stored and transported in the virus transport media (VTM) to the NHS diagnostic laboratory at Lancaster University, UK. All samples were individually spiked with 50 pmol/L of synthesized Caenorhabditis elegans miR-cel-miR-39-3p (Thermo-Fisher Scientific, Waltham, MA, USA). The miR-cel-miR-39-3p RNA lacked any sequence homology to the human or viral gene and thus presents an ideal RNA extraction control. Total RNA including miRNAs was extracted using 140 µL of the spiked-VTM by the commercial QIAamp Viral RNA Mini kit (Qiagen, Valencia, CA, USA). The miR-cel-miR-39-3p RNA was used to serve as an internal control to monitor extraction efficiency and was used for data normalization. The final RNA yield and purity were determined by the A260/A280 ratio measured by a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies/Thermo-Fisher Scientific, Waltham, MA, USA) with a ratio of 1.80 to 2.00, which is indicative of good RNA purity. The isolated RNA was stored at −80 °C for further use.
2.6. Real-Time Fluorescent-Based Quantitative PCR
Suspected SARS-CoV-2 clinical samples were tested for positivity by qRT-PCR. Briefly, RNA was extracted from Viral Transport Media using the QIAamp Viral RNA Mini kit (Qiagen, Valencia, CA, USA) following the manufacturer’s instructions. The qRT-PCR was conducted using the SuperScript III Platinum One-Step qRT-PCR Kit as per the manufacturer’s protocol (Invitrogen Carlsbad, CA, USA) in the CFX96 Touch Real-Time PCR Detection System (BioRad Laboratories, Watford, UK), according to the cycling protocol. The reaction was performed using the specific primer set RdRpF, RdRpR and FAM-labelled probe or NP-F and NP-R and ROX labelled probes designed to detect SARS-CoV2. The 25-µL PCR reaction consists of 12.5 µL 2X Reaction Mix, 0.2 µM of each primer and 0.1 µM probe, 0.5 µL of SuperScript® III RT/Platinum® Taq Mix, 5 µL of RNA sample and nuclear free water. The cycling program was performed in the CFX96 Touch Real-Time PCR Detection System (Applied Biosystems, Madison, WI, USA), according to the cycling protocol. The amount of viral RNA in each sample was estimated by comparing the cycle threshold values (Ct) to the standard curve made by 10-fold dilutions of previously titrated in vitro transcribed RdRP gene.
2.7. AI-LAMP Assay Performance
All experiments for LAMP assay were run in triplicate. The LAMP reactions were performed using WarmStartTM Colorimetric LAMP 2X Master Mix (New England Biolabs, Hitchin, UK). A 10X primer mix (FIP, 16 µM; BIP, 16 µM; F3, 2 µM; B3, 2 µM; LF, 4 µM; LB, 4 µM) was prepared. A 25 µL reaction mixture (12.5 µL 2X MasterMix; 2.5 µL 10X primer mix; 2.5 µL RNA and 7.5 µL DNase & RNase-free molecular grade water) was mixed homogeneously and centrifuged. The LAMP assays were performed in a thermocycler (MJResearch) at 65 °C for 30 min or in the engineered device. Colour change was observed directly by the naked eye or through AI image processing and agarose gel electrophoresis was performed to confirm the results. The completion of amplification was indicated by the colour in the tube, wherein yellow was considered positive and pink was regarded as negative. All amplicons were confirmed by 2% agarose gel electrophoresis.
2.8. Artificial Intelligence Based Test-Tube Colour Detection
A loop-mediated isothermal amplification (LAMP) assay based COVID-19 test device was developed to capture the COVID-19 test results in 30 min, based on colour changes. Artificial intelligence (AI) based colour detection was proposed to identify colour changes considering different lighting issues and to reduce the test running time to less than 30 min. Images were acquired from the COVID-19 test kit, which carried 8 test-tubes including NTC (negative test control) and PTC (positive test control) for every 20 s during the test operation. Each image was cropped into separate tubes using the template matching approach and labelled manually based on their colour.
2.9. Analytical Specificity and Sensitivity of the LAMP Assay
The designed RdRp primer sets for LAMP to detect SARS-CoV-2 were validated for their specificity by testing the cross-reactivity with other viruses, including influenza a virus, Vesicular stomatitis virus (VSV), Sendai virus (SeD), infectious bronchitis virus (IBV) and Newcastle disease virus (NDV). Likewise, the developed LAMP assay was evaluated to test the primers set sensitivity in a serially diluted standard RNA template prepared by 10-fold dilutions. The amplification patterns were observed for each dilution to determine the lowest amount of absolute RNA template required for detectable amplification. The degree of colour intensity of the amplified product as well as the observed electrophoretic pattern during gel electrophoresis was used for the analysis of LAMP amplification.
2.10. Quantitative Real Time PCR for miR-cel-miR-39-3p RNA
In order to determine the RNA extraction efficiency, the extracted RNA was reverse transcribed using a commercially available kit (Applied Biosystems/Thermo-Fisher Scientific, Warrington, UK) using miR-specific stem-loop primers as per the manufacturer’s instructions. A total of 5 µL of the sample was added to a 96-well plate together with 10 µL reaction mixture (MasterMixTM), along with MultiscribeTM
reverse transcriptase (50 U/µL) and 0.19 µL RNAase inhibitor (20 U/µL). The RT reaction was performed at 16 °C for 30 min, followed by 42 °C for 30 min and 85 °C for 5 min and was finally kept at 4 °C. A NTC was considered in every individually run reaction to identify any unspecific amplification. The RT products were quantified immediately by qPCR using TaqManTM
MicroRNA assays (Applied Biosystems/Thermo-Fisher Scientific, UK) in a 96 well plate using the 7900HT Fast Real-Time PCR System (Applied Biosystems, Warrington, UK), as we described before [23
]. The quantification cycle (Cq) was determined with instrument default threshold settings (10 SDs above the mean fluorescence of the baseline cycle).
2.11. Statistical Analysis
A total sample size of 200 was calculated to assess the performance of the LAMP assay. GraphPad Prism Software version 6.01 for Mac (GraphPad Software, La Jolla, CA, USA) was used for graphs generation. The LAMP detection sensitivity and specificity were calculated using the chi-squared test. TPR (true positive rate), TNR (true negative rate), FPR (false positive rate) and FNR (false negative rate) were calculated according to the following equations:
where TP: total number of true positives, TN: total number of true negatives, TN: total number of true negatives, FN: total number of false negatives.
The SARS-CoV-2 is now a global pandemic. Over 216 countries are currently reporting active infections and the number of daily infections and deaths is continuing to rise, especially in the Americas and South East Asia through a series of multiphasic spread [29
]. Currently, there is no licensed vaccine or any registered drugs, leaving timely identification of CoVID-19 patients, contact tracing and isolation of positive contacts as the most effective means of containing the pandemic. Among different molecular diagnostic chemistries, LAMP technology provides a promising approach for the rapid and reliable detection in resource-limited settings [17
]. Recently, the LAMP technology has been widely applied for the identification of West Nile virus, influenza virus, yellow fever virus, Marburg virus, Ebola virus, Zika virus and other myriads of viruses [30
]. Here, we demonstrated the specificity, sensitivity and utility of a novel ai-LAMP assay for SARS-CoV-2.
The genome of SARS-CoV-2 is approximately 30kb in size with a coding capacity of 9860 amino acids. All of the β-coronaviruses encode for structural (replicases, S, E, M and N) genes in the order of 5′ to 3′ in the positive sense genome [5
]. A range of qRT-PCRs have been proposed and are referred to by the World Health Organization [29
]. While diagnostic assays can be designed on the most conserved region of the viral genome, most of the routinely applied RT-PCR and RT-LAMP have been targeting the S, N, RdRP, E and ORF1a/b genes, mainly due to their high level of transcription and abundance in expression compared to other genes of the SARS-CoV-2 [5
]. For the detection of SARS-CoV-2, Chan et al. [23
] have targeted and developed a standard RT-LAMP with LoD of 11.2 RNA copies/reaction using in vitro RNA transcripts. Yan et al. [16
] have adapted the ORF1ab to develop RT-LAMP assay with a detection limit of sensitivities of 2 × 101
copies per reaction. The majority of these diagnostic assays carry a high level of sensitivity, specificity and repeatability; however, these primarily lack clinical validation and/or optimization on the synthetic targets.
In this study, we have developed and evaluated a novel RT-LAMP in one of the most conserved genes (i.e., RdRP) within the SARS-CoV-2 genome. The RT-LAMP was then directly compared with the currently applied routine diagnostic assays to assess the comparative performance. The RT-LAMP assay developed in this study could detect as low as 100 copies with an in vitro RNA transcript. Importantly, the RT-LAMP has detected the SARS-CoV-2 RNA in 68/199 (34%) and 52/199 (26%) additional specimens that were tested negative by the RdRP-based qRT-PCR and N-based qRT-PCR, respectively. These findings are interesting, both clinically and epidemiologically due to the high proportion of asymptomatic and mildly symptomatic cases of CoVID-19. This higher number of positive sample detection was mainly attributed to the higher sensitivity of the LAMP assay. Moreover, owing to multi-gene amplification in the qRT-PCR, the overall detection of positive cases was limited. The apparently healthy people have been suggested to be a major source of virus propagation and the basis of epidemics within the community [39
]. Therefore, a highly sensitive and specific test is needed to identify cases with a low viral load. The RT-LAMP detected more positive samples that were otherwise negative by routinely applied qRT-PCR assay. In order to assess the potential false positive identification, we run electrophoreses and sequencing of the N gene. The use of a spiked RNA standard that is not expressed in humans (cel-miR-39-3p) helped to confirm the effectiveness of the RNA extraction approach using commercial kits (Qiagen). In addition, we used a fixed total RNA concentration in all experiments, allowing for better comparisons across groups.
The main challenges of using the colorimetric approach are the background which changes the colour perspective, issues in identifying small changes, bubbles in the test tubes, a relatively small area corresponding to colour change and pixel variation due to camera flash and background reflections. In addition to these issues, the SAD approach required manual finding for the suitable threshold due to a different lighting setup and image set with bubbles. In addition to this, we tried different image processing algorithms such as histogram comparisons, conversions to other colour-spaces and calculating colour differences, removing brightness through YUV transformation to get rid of brightness variation, edge detection approach to extract tube sections only and threshold segmentation approach. All the approaches failed to deliver better accuracy in separation due to the presence of bubbles, small tube area and camera light change over time. Out of the image processing approaches mentioned above, SAD worked best, however for its full integration, we had to decide the threshold manually for every image set, which introduces another subjective variable that needs optimization for every experiment. Theoretically, with extensive experimentation in altering and optimizing the parameters, acceptable features can be acquired. Therefore, the CNN based model has been used, which can exploit these underlying patterns and find the best parameters that are robust, keeping in mind those different types of noise that can be present. Although a CNN based approach was used to overcome noise related issues, image processing algorithms were applied to extract tubes before applying the CNN approach. In this hybrid approach, test tubes are extracted using image processing algorithms and then deep learning algorithms are applied to apply fine tuning of images algorithms and classify colours with more accuracy. This approach demanded less expert analysis and fine-tuning, exploiting the tremendous amount of image set we generated through the experiments with different scenarios. The CNN needs to be applied only on extracted test tubes and therefore saves significant computing resources and training efforts compared to processing of the entire image. The trained model, after compression, has been successfully moved to Rpi to identify colour changes in test tubes. The study offered 98% accuracy for images taken and the duration of testing could be dynamically controlled to reduce the length of operating time and heating, with a resulting reduction in energy consumption by the device. Despite the SAD based approach resulting in 82.1% accuracy for the images after 30 min, this approach failed with other datasets containing bubbles and different background lights as different threshold values were manually produced for each image set. Therefore, a CNN approach was utilized in our experiments to generalize the classification for orange and pink test tubes with different background light and bubbles. Although deep learning training complexity and time were higher compared to standard image processing algorithms, testing images with the trained model does not differ substantially on complexity and time compared to pipeline of standard image processing algorithms. Even though the CNN approach gave modest improvement in accuracy (98% from 81.5%), these improvements are significant in the field of CoVID-19 testing.
Collectively, our data show that the newly established ai-LAMP is highly specific for the detection of SARS-CoV-2 RNA from extracted respiratory tract clinical specimens. The application of this novel LAMP assay may be particularly useful for detecting COVID-19 cases with low viral loads and when testing upper respiratory tract specimens (nasal or oral swabs) from patients. Development of ai-LAMP into a multiplex assay that can simultaneously detect other human-pathogenic coronaviruses and respiratory pathogens may further increase its clinical utility in the future.