Development and Evaluation of a Multiplex SARS-CoV-2 RT-ddPCR, Applicable to Respiratory and Wastewater Samples

The worldwide emergence and spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) since 2019 has highlighted the importance of rapid and reliable diagnostic testing to prevent and control the viral transmission. The monitoring of the viral spread can be performed using individual diagnostics in clinical samples and global detection of SARS-CoV-2 in wastewater samples. Reverse-transcription droplet digital PCR (RT-ddPCR) offers an attractive platform for absolute RNA quantication in comparison to the gold standard reverse transcription quantitative PCR (RT-qPCR) that only yields relative quantication results. To avoid inaccurate results due to false negatives caused by polymorphisms or point mutations that can possibly compromise the accuracy of the diagnostic tests, a multiplex RT-ddPCR method was developed by combining previously published SARS-CoV-2 detection RT-qPCR assays targeting well-conserved ORF1a and RdRp genes. These assays were successfully evaluated in silico for their inclusivity using GISAID sequences from the beginning of November 2020 to the end of February 2021. The sensitivity and specicity of the new RT-ddPCR method was evaluated and its applicability on both clinical and wastewater samples was illustrated. In conclusion, a reliable and sensitive multiplex one-step RT-ddPCR assay for reliable quantication of SARS-CoV-2 RNA in clinical and wastewater samples was developed and assessed, providing a key tool for the current health crisis.


Introduction
The ongoing "coronavirus disease 2019" (COVID-19) pandemic is caused by the "severe acute respiratory syndrome coronavirus 2" (SARS-CoV-2), a positive-sense single-stranded RNA virus. Together with SARS-CoV, this virus belongs to the species "Severe acute respiratory syndrome-related coronavirus" from the subgenus Sarbecovirus of the family Coronaviridae 1 . The symptoms of COVID-19 include cough, respiratory problems, fever, aches and pains, fatigue, diarrhea and taste and smell disorders 2 . SARS-CoV-2 can also cause severe complications, including death, mostly in the elderly or in people suffering from comorbidities, such as diabetes mellitus, obesity, cardiovascular diseases, hypertension, cancer, chronic kidney disease or immunosuppression 3,4 . Due to the virus' capacity for human-to-human transmission and a lack of immunity in the population, many governments decided to implement a variety of sanitary restrictive measures 5 , such as curfews, lockdowns and travel bans. Viral RNA and viable SARS-CoV-2 are shed in bodily excreta, including sputum, saliva and faeces, with respiratory droplets as primary viral transmission route. Hence, the measures meant to control the disease were aimed at diminishing close person-to-person contact and people's movement 5,6 . Additionally, many governments implemented intensive contact tracing, testing and isolation 7-10 , allowing to monitor the spread of COVID-19 epidemic and to reduce transmission. This set of measures has been supplemented since the end of 2020 by mass vaccination campaigns.
The gold standard for the detection of SARS-CoV-2 is reverse transcription quantitative polymerase chain reaction (RT-qPCR) on extracted RNA from nasopharyngeal swabs for individual diagnostics. By using RT-qPCR for individual diagnostics, in May 2021 already more than 30 300 000 positive cases were detected in EU since the start of the pandemic in 2019 11 . However, the number of con rmed positive cases is likely an underestimation because this depends on the willingness of the people to get tested. Additionally, testing such a large population results in a very high cost and during some periods of high virus prevalence, the number of COVID-19 cases exceeded the testing capacity of public health systems. Furthermore, some people are asymptomatic or pre-symptomatic while still being able to transmit the virus 12,13 and consequently are often not tested. Therefore, in order to rationalize the monitoring of the virus spread at the level of a country or region, monitoring of wastewater was proposed for surveillance of SARS-CoV-2 14-16 based on previous experience for early surveillance of disease prevalence, such as poliomyelitis 17,18 . Indeed, SARS-CoV-2 genomes can be detected also in faeces 19,20 with reported RNA loads ranging from 0.55-1.21x10 2 copies/µL 21 and consequently may be found back in wastewater. Additionally, it was shown that SARS-CoV-2 genomes in faeces can still be detected several weeks after respiratory samples tested no longer positive 22 . This suggests that the viral excretion may last longer in faeces. The presence of SARS-CoV-2 has been reported in wastewater and an association was observed between an increase of the RNA concentration in raw wastewater 14-16 and an increase in reported COVID-19 cases 15 . This renders wastewater-based epidemiology as an important early-warning tool to monitor the circulating viruses in a community. Wastewater-based epidemiology also provides opportunities to estimate the genetic diversity, geographic distribution and prevalence 23,24 . Furthermore, wastewater surveillance could offer an unbiased method not limited by the asymptomatic nature of the viral infections leading to the under-diagnosis of positive cases compared to the clinical surveillance 25 . Finally, this surveillance makes it possible to assess the spread of infection in different areas, even areas with limited resources for clinical diagnosis or delays in test reporting 26 . However, there are several limitations to wastewater surveillance. The excretion rate during the course of the infection determines the viral load in the sample. Consequently, the correlation between the viral load and the speci c number of positive SARS-CoV-2 cases may be challenging. Additionally, inconsistent capture of spatial variability makes the correlation with the number of positive SARS-CoV-2 cases di cult. This is the consequence of travel and use of multiple wastewater systems in time. It is also due to temporal delays, inactivation during the wastewater transport process and/or dilution due to rainfall. Additionally, infrequent or absent clinical testing of possible positive SARS-CoV-2 cases also complicates the correlation 27 . Furthermore, the virus detection and quanti cation can be limited due to the instability of the genome in wastewater, low e ciency of virus concentration methods and the lack of sensitive detection assays 14 .
Although RT-qPCR methods are the standard for clinical and consequently often used in wastewater samples due to the availability of these methods, many drawbacks were reported related to the use of this technology. First, the tests are expressed in cycle quanti cation (Cq). The Cq represents the PCR cycle at which the sample produced a uorescent signal above the background. These Cq values are laboratory-and instrument-speci c and a calibration to a quantitative standard is necessary to determine the absolute viral load. Furthermore, Cq values are not directly comparable across assays or technology platforms due to differences in nucleic acid extraction methods, viral targets and other parameters 28 , thereby affecting inter-laboratory harmonization in interpretation of the test results. Finally, RT-qPCR is not adapted for wastewater samples that often contain inhibitors that might in uence the Cq values. This could affect the accuracy of viral quanti cation 29 .
Reverse-transcriptase droplet digital PCR (RT-ddPCR), may offer an interesting alternative for the detection and quanti cation of SARS-CoV-2 RNA 30,31 . Similarly to RT-qPCR, a target-speci c uorescent probe coupled with primers are used, which makes adaptation of existing RT-qPCR assays straightforward. In a ddPCR, a reaction is emulsi ed into thousands of nanodroplets of which a proportion does not contain the template molecule 32 . The nanodroplets are used as unique and small bioreactors to amplify the template [33][34][35][36] . At end-point, the number of positive droplets are digitally counted relative to the total number of droplets. Furthermore, their known volume while owing through micro uidic devices allows absolute target quanti cation using Poisson statistics 37,38 , which enables an easier comparison between different laboratories and tests compared to RT-qPCR. To the best of our knowledge, eight RT-ddPCR methods designed to detect SARS-CoV-2 were published, of which two are commercial kits designed by BioRad 31,39−45 . The performance of these methods was tested using reference standards, and four of the methods were tested on clinical samples of infected patient's throat and nasopharyngeal samples. Three of these methods were tested on wastewater samples. Moreover, four of these RT-ddPCR methods were tested on respiratory samples, and in some cases were found positive compared to the negative RT-qPCR results 31,41 . Additionally, the sensitivity of the RT-ddPCR methods for the detection of SARS-CoV-2 has been described previously as comparable or even higher compared to RT-qPCR methods 31,39−41 . This makes this technology interesting in case of a low viral load. Furthermore, inhibition can be encountered in some matrices, like wastewater. RT-ddPCR separates DNA, inhibitors and reagents in droplets and is an end-point measurement, only measuring after the PCR ampli cation.
Consequently, a reduction in the biases linked to the inhibitors are often observed in RT-ddPCR 46 , which makes RT-ddPCR an interesting method for wastewater surveillance.
In this study, a new multiplex RT-ddPCR method speci c for the detection of SARS-CoV-2 was developed. This method targets two different parts of the genome of the virus based on sequences used in the RT-qPCR methods developed by Institute Pasteur 47 and Lu et al. 48 In silico inclusivity of the target was veri ed using 154 489 whole genome sequences, including several circulating SARS-CoV-2 variants. This novel RT-ddPCR method was in-house validated, including speci city and sensitivity assessments. Additionally, the applicability of the proposed RT-ddPCR method was investigated using clinical and wastewater samples.

Selection and evaluation of key target for PCR detection of SARS-CoV-2 using WGS data
For the development of the RT-ddPCR method, two sets of primers and probe were selected from publicly available RT-qPCR assays, namely RdRp_IP4 assay from Institut Pasteur (Paris) 47 , and the ORF1a assay from Lu et al., 2020, 48 that target two separate locations speci c to the SARS-CoV-2 genome (Table 1). These assays were evaluated in silico 49 for their inclusivity and exclusivity in a previous study in May 2020 which determined the RdRp_IP4 assay 47 , S assay from Chan et al., 2020 50 and ORF1a assay 48 as the most speci c assays. However, due to the emergence of the B1.351 lineage in South Africa, a mismatch located in the probe sequence of the S assay was identi ed, which could lead to a lower sensitivity 51 .Therefore, from the three previously described, only the ORF1a and RdRp_IP4 assays were retained in this study. Table 1 Primer and probe sets included in the multiplex RT-ddPCR assay.
ORF1a The in silico inclusivity of ORF1a and RdRp_IP4 assays was evaluated using the bioinformatics tool SCREENED v1.0 52 , previously used for in silico SARS-CoV-2 assay assessment 49,51 , and recent whole genome SARS-CoV-2 sequences. A total of 296 187 SARS-CoV-2 genomes, coming from samples collected between November 1st, 2020 and February 28th, 2021 were obtained from the GISAID database 53 on March 7th, 2021. Only complete genomes with high coverage for which the collection date was available were selected, and genomes with low coverage were excluded. Additionally, genomes containing undetermined nucleotides "N" and degenerate nucleotides were excluded from the dataset to retain only high quality genomes (154 489 genomes) (Supplementary File S1, Supplementary File S2). These genomes were divided per month according to their collection date (November: 13 678 genomes; December: 41 128 genomes; January: 58 484 genomes; February: 41 199 genomes). From these datasets, SCREENED performed a two-step BLAST approach to nd in each genome the complete amplicon sequence targeted by the ORF1a and RdRp_IP4 primers and probe sets, and subsequently produced mismatch statistics from the hybridization between the nucleotides of the primers and probes and their corresponding annealing sites in the amplicon. Based on these mismatch scores, SCREENED considered that a theoretical positive RT-ddPCR signal was produced if no mismatch in the rst ve nucleotides of the 3' end of the primers was reported, if the total number of reported mismatches did not exceed 10% of the oligonucleotide length and if at least 90% of the oligonucleotide sequence aligned correctly to their targets. For the primers and probes evaluated here, this resulted in no more than one or two mismatches being tolerated. These criteria were selected because it has been previously reported that two or more mismatches can lead to potential total test failure, especially if located at the 3' end 54, 55 . Two mismatches or less can result in potential loss of sensitivity but less likely in total test failure. For each analyzed SARS-CoV-2 genomes, a negative SCREENED detection signal was considered as a theoretical false negative (FN) result, which was used for the in silico inclusivity evaluation (Eq. 1): Inclusivity (%) = (1 -(Number of FN / Total Number of high quality SARS-CoV-2 genomes)) x 100 (1) FASTA les for November, December, January and February containing 13 678, 41 128, 58 484 and 41 199 SARS-CoV-2 genomes respectively (Accession ID: Supplementary File S1) and a tab-delimited text le (Supplementary File S3), containing the primer and probe sequences and their corresponding amplicon sequence to be mined in the genomes, were used as input for SCREENED.
2.2 Development of RT-ddPCR method for the detection of SARS-CoV-2 The RT-ddPCR assay was evaluated using puri ed RNA from the SARS-CoV-2 virus (Vircell, MBC137-R). The RT-ddPCR was performed using the One-Step RT-ddPCR Advanced Kit for Probes (Bio-Rad). All the components from the kit were thawed on ice for 30 minutes and thoroughly mixed by vortexing each tube at maximum speed for 30 seconds. The reagents were made into larger master mixes and then aliquoted into individual reactions. Each reaction had a total volume of 22 µL that was set up on ice, including 0.99 µL of each primer with an initial concentration of 20 µM and 0.55 µL of each probe with an initial concentration of 10 µM, 1.1 µL of 300 mM DTT, 0.14 µL of dH 2 O, 2.2 µL Reverse Transcriptase, 5.5 µL One-Step Supermix and 8 µL of sample. The primers were obtained from Eurogentec, while the ZEN-probes were supplied by Integrated DNA Technologies. According to manufacturer's instructions, 20 µL of the reaction mix and 70 µL of Droplet Generation Oil for Probes were loaded into a QX200™ droplet generator (Bio-Rad) and to increase the number of droplets, the cartridge was kept for two minutes at room temperature. After the droplet generation, 40 µL of droplets were recovered per reaction. The ampli cation was performed in a T100™ Thermal Cycler (Bio-Rad) with the following conditions: one cycle at 25°C for 3 minutes, one cycle at 50°C for 60 minutes (RT), one cycle at 95°C for 10 minutes (Taq polymerase activation); 40 cycles at 95°C for 30 seconds (denaturation), 55°C for 60 seconds (annealing); one cycle at 98°C for 10 minutes (enzyme inactivation) and nally one cycle at 4°C for 30 minutes (stabilization). Next, the plate was transferred to the QX200 reader (Bio-Rad) and results were acquired using the HEX and FAM channel, according to the manufacturer's instructions. The QuantaSoft software v1.7.4.0917 (Bio-Rad) was used for the interpretation of the results and the threshold was set manually.
2.3 Validation of the speci city of the RT-ddPCR assay for SARS-CoV-2 The speci city of the method was experimentally established using a set of DNA and RNA controls from Bacillus subtilis

Validation of sensitivity of the RT-ddPCR assay for SARS-CoV-2
The evaluation of the sensitivity was done using serial dilutions of puri ed RNA from the SARS-CoV-2 virus. Seven serial dilutions were prepared ranging from 0.5 to 200 copies/µL and each dilution was tested in 12 replicates. The limit of detection (LOD 95% ) was calculated using the web application Quodata, as the number of copies of the target that is required to ensure a probability of detection (POD) of 95% 57 .

Applicability assessment
To assess the applicability of this RT-ddPCR assay on non-arti cial samples, ve samples collected from patients showing clinical signs of COVID-19 were collected. From these ve samples, three samples (Clinical samples 1, 2, 3) tested previously with RT-qPCR positive for SARS-CoV-2 with a high, moderate and low Cq while two tested negative for SARS-CoV-2 (Clinical samples 4, 5) (Supplementary File S4). The clinical samples were obtained from a biobank (allowed by Biobank compendium of the "Federaal Agentschap voor Geneesmiddelen en Gezondheidsproducten" 58 ). All experiments were performed in accordance with relevant guidelines and regulations. In addition, three wastewater samples (Wastewater sample 1, 2, 3) were included that also previously tested positive for the SARS-CoV-2 virus with RT-qPCR with a high, moderate and low Cq (see Supplementary File S4). Due to the high concentration of the clinical sample 3, the sample was diluted 80 times. Consequently 0.1 µL of sample and 7.9 µL of dH 2 O was used in the reaction (dilution: 80X).

In silico inclusivity evaluation for the ORF1a and RdRp_IP4 assays using SCREENED
The ORF1a and RdRp_IP4 assays were evaluated for their inclusivity with four datasets corresponding to the months November 2020, December 2020, January 2021 and February 2021 (Table 2) using 13 678, 41 128, 58 484 and 41 199 SARS-CoV-2 genomes, respectively. Both for the ORF1a and RdRp_IP4 assays, excellent inclusivity was obtained for the four datasets because all assays had an inclusivity of more than 99.5%. The little variation observed between the months can mainly be attributed to random and rare mutation events that did not spread in the viral population. In addition, it was veri ed that when a FN result was obtained for a given genome, this was limited to either only the forward or reverse primer or the probe. Moreover, if a FN result was obtained for a genome for one of the assays, a positive signal was obtained for the other assay. Consequently, the inclusivity of the multiplex method using the combination of the ORF1a assay and RdRp_IP4 assay is 100%.

Speci city assessment
The speci city of the RT-ddPCR method was experimentally tested for each positive and negative material (Table 3). SARS-CoV-2 RNA was used as a positive control, while four closely related corona viruses, 10 other viruses, human DNA, plant (Zea mays), two bacteria and two fungi were used as negative controls. Excellent exclusivity was observed because no ampli cation was observed for all negative controls, while the positive control presented an ampli cation (Table 3). The absence and presence of ampli cation is symbolized by a "-" or "+", respectively. The RT-ddPCR method was performed in duplicate on each sample. As positive control SARS-CoV-2 RNA was included.

Sensitivity assessment
The sensitivity of the designed RT-ddPCR method was tested using SARS-CoV-2 RNA with different estimated target copy numbers, namely 200, 50, 25, 10, 5, 1, 0.5 and 0 copies/µL. An ampli cation for all 12 replicates was observed until 5 estimated target copies/µL ( µL. Notably, in 4/12 and 9/12 replicates for the ORF1a assay and RdRp_IP4 assay, respectively, it also tested positive for samples with an estimation of 0.5 and 1 copies/µL (  The absence and presence of ampli cation are indicated by "-" or "+", respectively. For each estimated target copy number, 12 replicates were tested and the number of positive replicates is indicated between brackets at the middle line of each box. In addition, the average of the observed copies/µL (± the standard deviation, as obtained with the RT-ddPCR measurement, is indicated between brackets at the lower line.

Applicability assessment
The presence and quantity of SARS-CoV-2 was investigated in ve clinical (nasopharyngeal swabs) and three wastewater samples. Among the ve clinical samples, three samples tested positive for both the ORF1a and RdRp_IP4 assay ( Table 5).
The three wastewater samples also tested positive for SARS-CoV-2 (Table 5) Table 5 and Supplementary File S7. The sample name and the kind of sample are given in addition to the results of the detection of SARS-CoV-2 using the ORF1a assay and the RdRp_IP4 assay. The presence or absence of PCR ampli cation is symbolized by "+" or "-" respectively. For each RT-ddPCR, the observed copies/µL is given between brackets. Detailed results from the RT-qPCR can be found in Table 5.

Discussion And Conclusion
A novel RT-ddPCR multiplex assay was developed for the detection of SARS-CoV-2, by combining two already existing simplex RT-qPCR assays targeting the conserved regions of ORF1a and RdRp genes. The main advantage of targeting two regions is to anticipate false negative results that could occur due to mutations that lead to possible mispriming of the primers and/or probe and consequently to a lack of viral detection. Such false negative results have been reported previously in clinical samples 59 . These can be the results of the sample source, the sample quality and sampling time that in uence the viral load 60-62 or due to the genetic evolution of the virus 51,63,64 . The use of multiple targets for the detection of the viral genome 65 can reduce the impact of false negative results related to viral mutations in the region of the annealing of the primers and/or probe. The failure of one region can be compensated by the detection of the other as was shown in this study for the in silico evaluation.
During the development of a new method for pathogen detection, it is of utmost importance to carefully asses its speci city, i.e. inclusivity and exclusivity. For inclusivity, a large number of various strains belonging to the targeted organism should ideally be tested. However, in the case of SARS-CoV-2, it is di cult to obtain a representative collection of circulating strains, and to test it experimentally. Fortunately, this kind of speci city evaluation can be done in silico using bioinformatics and the large number of SARS-CoV-2 sequences publicly available, as previously performed for ORF1a and RdRp_IP4 assays (ref 48,49). Nevertheless, constant monitoring of detection assays is needed because the virus evolves and a mutation could be introduced within these targets. Especially now that several new SARS-CoV-2 variants emerged carrying an unusually high number of mutations, assessing all assays for false negatives is important. Therefore, in the present study, the latest WGS published data of SARS-CoV-2 was used to perform an in silico analysis of ORF1a and RdRp_IP4 assays, which both showed excellent results, i.e an inclusivity of more than 99.5% from the beginning of November 2020 to the end of February 2021.
Hence no new mutations impacted the inclusivity, including the mutations linked to the variants of concern that emerged at the end of 2020. Most of the primers and probe sets used in other multi targets RT-ddPCR assays developed for SARS-CoV-2 detection 31,39,40,42, 44 have also been previously analyzed for their inclusivity using the same in-silico approach 49,51 . Most of these sets showed excellent inclusivity results (> 99%), except for the primers and probe set targeting the gene N (June- Therefore, the N target used in these assays should preferably not be chosen for developing SARS-CoV-2 detection methods.
Concerning the exclusivity, this one has also been previously evaluated in silico for ORF1a and RdRp_IP4 assays successfully, with thousands of non-SARS-CoV-2 genomes 49 . Additionally, using a set of DNA and RNA references, the good exclusivity of ORF1a and RdRp_IP4 assays was experimentally con rmed in the present study with no false positives detected for other viral, bacterial, plant and human RNA and DNA, including closely related viruses like SARS-CoV, MERS-CoV and Corona OC43.
In contrast, the other RT-ddPCR methods did not evaluate the speci city in vitro.
As to the sensitivity of our method, a LOD 95% of 4.6 and 1.6 estimated target copies/µL for the ORF1a and RdRp_IP4 assays was established, respectively. Although mostly other targets were used in the other previously published RT-ddPCR methods, similar LODs were observed 66 . When comparing the LOD to RT-qPCR methods, the RdRp_IP4 assay using RT-ddPCR was found to be more sensitive compared to using RT-qPCR for the same target with LOD 95% of 7.9 estimated copies/µL 66 .
Information on the LOD of RT-qPCR could not be found in the literature for the ORF1a assay. compared to RT-qPCR. In this study, no comparison could be made between the RT-qPCR methods used to characterize the clinical and wastewater samples (Supplementary File S4) and the developed ddPCR method because different primers and the probes were used.
The study was designed as a proof of concept, and the number of clinical and wastewater samples in the applicability assessment were limited. Nevertheless, three clinical samples, previously established as positive with the RT-qPCR methods including samples with high Cq, were also determined as positive by the developed RT-ddPCR method suggesting a sensitivity at least as high as the RT-qPCR assays used for these samples. One of the advantages of using RT-ddPCR instead of RT-qPCR is the absolute quanti cation of the viral RNA without calibration, which enables comparison between different assays and laboratories without the necessity of a standard curve. Additionally, the accuracy of the RT-ddPCR methods should be less in uenced by inhibitors that are often present in wastewater samples. However, there are some drawbacks to RT-ddPCR such as the longer turnaround time of the RT-ddPCR compared to RT-qPCR. Additionally, clinical samples may contain a high viral load that would need to be diluted in the RT-ddPCR method. The possible repetition of the detection of the samples that need to be diluted takes more time and makes the RT-ddPCR method a less interesting method for routine surveillance. However, the viral load in wastewater samples is often low, making it most often unnecessary to dilute. Moreover, the lower impact of inhibition on the RT-ddPCR method makes it an interesting method for the wastewater surveillance. Due to its absolute quanti cation, the RT-ddPCR method can also be used to evaluate the performances in different laboratories for the interlaboratory reproducibility and cross-validation of the methods. Because of its potential higher sensitivity, it could also complement the current RT-qPCR diagnostics to improve rapid identi cation of SARS-CoV-2 infections, by detecting the virus before the viral load peak is reached and antibodies appear in a diagnostic sample.
In conclusion, the multiplex RT-ddPCR method provides a reliable molecular detection assay to detect and directly quantify SARS-CoV-2 RNA. The inclusivity of the assays was successfully assessed in silico using GISAID sequences from the beginning of November 2020 to the end of February 2021. Moreover, the in vitro validation of the method shows that the method is speci c and sensitive. The applicability of the method was demonstrated (on both clinical and wastewater samples) and shows its potential use for individual diagnostics as well as for the monitoring of wastewater.