DISCUSSION
Here, we evaluated the use and suitability of the RT-LAMP assay for the detection of SARS-CoV-2 infection. We also developed LAMP-sequencing as a fully scalable alternative to colorimetric or fluorometric analysis of DNA amplification reactions. Our results indicate that whereas the RT-LAMP assay using the N-A primer set is not sensitive enough to replace RT-qPCR in all applications, it does hold promise as a method for testing large numbers of samples.
We tested the RT-LAMP primer sets suggested by Zhang et al. (11) and found that the N-A primer set for the N gene worked better than the 1a-A primer set for ORF1a. For samples with a CT ≤ 30 as measured by RT-qPCR with E-Sarbeco primers, we found overall satisfactory sensitivity and specificity values for SARS-CoV-2 RNA detection by the RT-LAMP assay using RNA samples isolated from pharyngeal swab specimens (Fig. 3 and Table 1). For samples with CT > 30, the RT-LAMP assay was much less sensitive. However, there is debate about which CT value for a positive RT-qPCR result should be considered clinically relevant. Vogels et al. (16) indicate that a CT value above 36 corresponds to less than 10 molecules of RNA.
On the basis of our data, we conclude that the colorimetric RT-LAMP assay would be suitable for identifying individuals with a high or moderate SARS-CoV-2 viral load. On the other hand, for those with a low viral load (at the onset of illness or during later stages of the disease), the sensitivity of the RT-LAMP assay, in its current implementation using the N-A primer set, is insufficient to detect a SARS-CoV-2 infection. A number of other LAMP primer sets have been proposed and initially tested (21, 27, 28), showing that optimized primers and the use of combinations of primer sets hold promise to further increase the sensitivity of the RT-LAMP assay for detecting viral genomes. Furthermore, alternative sample types, e.g., sputum or stool (29), might be more reliable. One promising lead for future applications is the exploration of the hot swab–to–RT-LAMP assay using saliva specimens, although the relative sensitivity compared to using pharyngeal swab specimens is currently unclear (30–33). Compatibility of the RT-LAMP assay with direct saliva specimens has been shown using spike-in experiments (22, 34).
Although faster and more convenient, the direct swab–to–RT-LAMP assay was less sensitive and less robust than the RT-LAMP assay using isolated RNA. To increase robustness, various treatments of crude swab samples have been described previously [reviewed in (35)], many of which require additional processing of the samples, for example, by pipetting or by adding proteinase K to degrade contaminating proteins. Rabe and Cepko (22) have suggested using cheap silica preparations and new sample inactivation protocols to enrich the RNA before the RT-LAMP assay, but this would complicate the simple swab–to–RT-LAMP assay workflow.
Last, our analysis found that a short heat treatment of 5 min at 95°C, which poses minimal additional handling steps, did not destroy the RNA but rather stabilized it and this improved the sensitivity and specificity of the swab–to–RT-LAMP assay (Fig. 5). The heat likely helped to homogenize the sample, to inactivate ribonucleases (RNAses), and to break up the viral capsid to release the viral RNA. Overall, our data demonstrate the feasibility of using a swab–to–RT-LAMP test and suggest applications especially in scenarios where RNA isolation is not available, e.g., in resource-poor settings. In such cases, the hot swab–to–RT-LAMP assay seems a good option given that the direct swab–to–RT-LAMP assay yields a number of false positives due to spurious amplification (14).
Although spike-in experiments with IVT RNA can be informative, we have experienced clear differences when comparing such experiments to those using clinical RNA samples isolated from swab specimens (figs. S6 and S7, and data file S1). We therefore recommend validating any new proposed rapid SARS-CoV-2 diagnostic test using “real-life” clinical samples including a large fraction of negative clinical samples. To overcome the problem of spurious amplification, an expanded oligonucleotide set that incorporates sequence-specific probes (34) or a CRISPR/Cas12a–based approach (36) could be used. However, these applications have yet to be tested with large numbers of diverse clinical samples.
There are several differences between the RT-LAMP assay and RT-qPCR. First, RT-qPCR requires a thermocycler to conduct the DNA amplification reaction, which is an expensive instrument, whereas isothermal incubation of RT-LAMP reactions can be conducted using a simple water bath or a heating block. This makes the RT-LAMP assay more amenable for point-of-care applications. Second, the reagents for the RT-LAMP assay are different from the ones used for RT-qPCR and are supplier independent. According to the supplier of the RT-LAMP reagents used in this study (New England Biolabs), production of RT-LAMP reagents can be easily ramped up to satisfy high demand. Third, the RT-LAMP assay, when combined with LAMP-sequencing, is suitable for analyzing large numbers of RT-LAMP reactions owing to the fully scalable DNA barcoding strategy. In contrast, there are several hurdles to scaling up RT-qPCR assays, the major hurdle being the need for a large number of thermocyclers. The RT-LAMP assay overcomes this problem and therefore will be a more scalable method for mass testing.

Application of RT-LAMP and LAMP-sequencing for SARS-CoV-2 testing
With its good sensitivity for samples up to CT ≈ 30, the colorimetric RT-LAMP assay has several advantages: It is fast, inexpensive, and it can be evaluated without any equipment. RT-LAMP reactions also appear to be less sensitive to contaminants in the samples than RT-qPCR, but care has to be taken that the samples used do not alter the pH as the colorimetric RT-LAMP assay is performed under conditions of weak pH buffering. Some clinical samples contain contaminants that can lead to acidification of the reaction independent of the presence of a template RNA if too much sample is added. Diagnostic RT-qPCR tests usually include a technical internal control, i.e., another RNA species, which is spiked into all samples and which is detected independent of the gene of interest to safeguard against the possibility of a general reaction failure within a sample tube. It would be desirable to have a similar precaution for the RT-LAMP assay. A multiplexed fluorescence readout might provide this (34) but comes at the expense of the simplicity of a colorimetric readout.
Our particular implementation of deep sequencing to analyze many RT-LAMP reactions simultaneously uses two sets of barcoded primers and is fully scalable so that, in one sequencing run, many thousands of LAMP reactions can be quantitatively analyzed for the presence of viral genomic sequences. Although we used Illumina dye sequencing, more scalable sequencing technologies, such as Oxford Nanopore Technologies sequencing, could be used for amplicon sequencing and counting (37). The workflow shown here uses LAMP-sequencing as a validation and backup procedure to double check the results of the colorimetric RT-LAMP assay. However, LAMP-sequencing could also facilitate scale-up of the workflow for direct analysis of many thousands of samples in an efficient manner, provided that an infrastructure is established that allows the collection of such samples. Thus, LAMP-sequencing could become an important part of workflows for routine testing of large populations.
Schmid-Burgk et al. (38) proposed decentralized RT-LAMP assays using combinatorial primer barcoding and centralized mass analysis of RT-LAMP products by next-generation sequencing as a means to scale-up testing. Although this poses additional challenges in generating the individualized RT-LAMP assay reagents, it would simplify sample handling on the analytical side and it can be easily combined with the barcoding strategy shown here.
There are several limitations to our study. We used surplus RNA sample material from a diagnostic laboratory rather than newly collected clinical samples. The criteria for testing individuals may have influenced cohort characteristics and hence our findings. It is not clear yet how well viral load as indicated by CT values from RT-qPCR assays informs about the degree of infectivity of an individual with a SARS-CoV-2 infection. Therefore, we cannot say how our findings on the sensitivity of the RT-LAMP assay in comparison to RT-qPCR would translate into sensitivity for detecting infectious individuals who are shedding SARS-CoV-2 virus. Moreover, the measured viral load does not indicate the course of a SARS-CoV-2 infection, as even individuals with a very low measured viral load can still develop severe symptoms of COVID-19 disease. This may be, in part, because the viral load in a clinical sample taken from a specific site such as the pharynx is not representative of the overall viral burden that an infected individual carries.
We used LAMP-sequencing to validate the RT-LAMP assay results and did not use it as a diagnostic tool. LAMP-sequencing is dependent on the sensitivity of the RT-LAMP reaction as it cannot detect false negative results caused by a failure of the RT-LAMP assay to amplify viral RNA. Also, reagents such as the primer sets for the RT-LAMP assay may be subject to production-dependent quality fluctuations. Therefore, all reagents must be precisely validated (batch control) before using an RT-LAMP assay diagnostically.
Application of the RT-LAMP assay has great potential, even more so as more sensitive primer sets become available. The RT-LAMP assay and LAMP-sequencing could offer scalable testing that would be difficult to achieve with conventional RT-qPCR–based tests. For example, the RT-LAMP assay could be used for regular testing of a whole workforce or in sentinel testing, ideally combined with simplified sample collection, e.g., in the form of saliva samples. The RT-LAMP assay and LAMP-sequencing extend the range of available test methods and complement individual tests and pooled tests based on RT-qPCR (39) with a faster, simpler, and potentially more cost-effective test method.