key: cord-0507355-xtx8hl14
authors: Martin, Alexandra; Storto, Alexandre; Andre, Barbara; Mallory, Allison; Dangla, Remi; Visseaux, Benoit; Gossner, Olivier
title: High-sensitivity COVID-19 group testing by digital PCR
date: 2020-06-03
journal: nan
DOI: nan
sha: 90ea0d87e6c90695278dd75b2fcd78dda9eabf28
doc_id: 507355
cord_uid: xtx8hl14

Background: Worldwide demand for SARS-CoV-2 RT-PCR testing is increasing as more countries are impacted by COVID-19 and as testing remains central to contain the spread of the disease, both in countries where the disease is emerging and in countries that are past the first wave but exposed to re-emergence. Group testing has been proposed as a solution to expand testing capabilities but sensitivity concerns have limited its impact on the management of the pandemic. Digital PCR (RT-dPCR) has been shown to be more sensitive than RT-PCR and could help in this context. Methods: We implemented RT-dPCR based COVID-19 group testing on commercially available system and assay (Naica System from Stilla Technologies) and investigated the sensitivity of the method in real life conditions of a university hospital in Paris, France, in May 2020. We tested the protocol in a direct comparison with reference RT-PCR testing on 448 samples split into groups of 3 sizes for RT-dPCR analysis: 56 groups of 8 samples, 28 groups of 16 samples and 14 groups of 32 samples. Results: Individual RT-PCR testing identified 25 positive samples. Using groups of 8, testing by RT-dPCR identified 23 groups as positive, corresponding to 26 true positive samples including 2 samples not initially detected by individual RT-PCR but confirmed positive by further RT-PCR and RT-dPCR investigation. For groups of 16, 15 groups tested positive, corresponding to 25 true positive samples identified. 100% concordance is found for groups of 32 but with limited data points.

As the first wave of the COVID-19 is fading in most countries of the Northern hemisphere, many of them have started implementing extensive monitoring policies to prevent the apparition of new clusters potentially leading to a second wave. These policies all require important testing capabilities, as was exemplified in Wuhan where all 11 Million citizens were recently tested in 10 days during May 2020. In some Southern countries, the amount of new cases is still increasing and wide population testing cannot be achieve everywhere. Thus, scaling up and maintaining large testing capacities worldwide remains a challenge, with limited reagent production and scarcity of testing equipment likely to remain limitations.

Group testing or pooling, first suggested by Dorfman in 1943, is a protocol through which individual samples are combined together before running the test (1) . The advantage of the method is an overall saving in the number of tests required to screen a given population (2) , and thereby an increase in testing capabilities for fixed reagent and instrumentation availability. Savings depend on key parameters such as the disease prevalence and the group size. Group testing protocols using real-time reverse-transcriptase PCR (RT-PCR) have been evaluated and implemented for Covid-19 screening around the world in several experiments using RT-PCR detection techniques, notably in Israel, Germany, California, Nebraska, NY State, and Italy (3) (4) (5) (6) (7) (8) (9) .

Although these studies show that positive individuals can be detected in pooled samples, the number of amplification cycles needed to detect (Ct value) those is often increased by dilution and perhaps by inhibition effects (3, 5, 7, 9) . This can prevent weakly positive specimen from being detected in group samples (3, 8) . Concerns about the sensitivity of group testing have been raised by French medical authorities, leading to a negative recommendation on their use in France (10) .

Digital PCR (or RT-dPCR) is a PCR technique known for its higher sensitivity and precision over classical RT-PCR (11, 12) . Digital PCR has also been shown to be more resistant to PCR inhibitors (13) . Recent studies have confirmed high sensitivity of RT-dPCR for the detection of SARS-CoV-2 (14) (15) (16) .

In this study we propose a novel group testing protocol using a commercially available RT-dPCR assay and compare empirically the sensitivity of individual identification through RT-PCR with group testing by RT-dPCR for three groups sizes of 8, 16 and 32 samples. We find that, in our condition of evaluation, group testing by RT-dPCR has a better or similar sensitivity than the reference individual RT-PCR testing for groups of 8 and 16.

Summary of the method of the comparative study Overall, 448 patient samples are tested for SARS-CoV-2 by i) individual RT-PCR (local gold standard method), ii) RT-dPCR in 56 groups of 8 samples, iii) RT-dPCR in 28 groups of 16 samples and iv) RT-dPCR in 14 groups of 32 samples and results are compared between all four test protocols. In case of discordance between the results of individual RT-PCR testing and group testing in RT-dPCR, samples were re-analyzed individually by RT-dPCR, by the gold-standard RT-PCR and a confirmatory RT-PCR assay. The whole protocol is illustrated in Figure 1 . 

Nasopharyngeal swabs of 448 patients screened for Covid-19 at the Bichat university hospital (Paris, France), one of the two Paris reference centers for emerging diseases, between May 6 th and May 26 th , 2020 were included. All samples were collected in universal transport medium (UTM) (Virocult® or eSwab TM ) and tested, within 15 hours maximum upon collection, for SARS-CoV-2 detection according to the local standards. Briefly, 400 µL of transport medium were tested by RT-PCR (Cobas SARS-CoV-2 test, Roche). All the remaining volume of transport medium of all specimens were kept at +5°C.

No later than 24 hours after routine screening, all samples with a leftover UTM volume above 600 µL were systematically included in the group testing analysis. Thus, 125 µL of each included specimen were sampled and randomly mixed with seven others to generate a total of 56 groups of 8 specimens with a final volume of 1 mL per group. Nucleic acids were extracted from each group prior to viral titration by RT-dPCR. The remaining volume of transport medium was stored at +5°C for further investigations if required.

According to the current French ethical laws, samples used in the current study were only included after the completion of all analysis required for the patient's care.

All 448 specimens were analyzed individually on a Cobas® 6800 system (Roche, Switzerland) for Covid-19 screening using the Cobas® SARS-CoV-2 Test kit targeting conserved regions for ORF-1a/b and Egene. For each specimen, 400 µL of transport medium were mixed with 400 µL of Cobas® lysis buffer and loaded on the robot. During the run, extracts were eluted in 50 µL of which 27 µL were used in the RT-PCR amplification of E and ORF-1a/b. A sample was considered positive for routine screening of COVID-19 ("RT-PCR+") if either target had a Ct value below 40 PCR cycles.

Within 11 days maximum (and 20 days for "Sample_25659") upon storage at +5°C, some samples which had different results for RT-PCR and RT-dPCR were reassessed on the Cobas® 6800 system. To compensate for the low remaining amounts of transport medium, the nasal swabs were vortexed once more into the remaining transport medium diluted 1 to 10 with new transport medium.

All nucleic acids extractions for RT-dPCR assays were performed on a MagNA Pure LC 2.0 (Roche, Switzerland) using the MagNA Pure LC Total Nucleic Acid Isolation Kit (Roche, Switzerland) following manufacturer's instructions. For all sample groups, the total volume of 1 mL was used for the extraction. For individual samples, 200 µL was diluted with 800 µL of buffer before extraction. Nucleic acids were eluted from 1mL to 50 µL of the elution buffer provided with the kit and stored at +5°C for a maximum of 12 hours before analysis.

After extraction of the 56 groups of 8 specimen (P8 extracts) and prior to viral titration by RT-dPCR, 28 groups of 16 individual samples (P16 groups) were obtained by mixing 15 µL of 2 P8 extracts and 14 groups of 32 (P32 groups) were obtained by mixing 10 µL of 2 P16 groups.

Detection of SARS-CoV-2 by grouped RT-dPCR testing SARS-CoV-2 titration of the grouped samples by RT-dPCR was performed on the Naica system (Stilla Technologies, France) within the next three hours after extraction, using the COVID-19 Multiplex Digital PCR Detection Kit (Stilla Technologies, France/Apexbio, China). This one-step reversetranscription PCR kit is a triplex PCR allowing amplification, detection and quantification of one sequence in the N gene, one sequence in the ORF1ab region of SARS-CoV2 and an endogenous internal control (IC) to assess the quality of the sample and extraction. These sequences are targeted by three TaqMan probes respectively labelled with a FAM, HEX and Cy®5 fluorophore.

As recommended by the kit manufacturer, the PCR mix for a single reaction contained 12.5 µL of dPCR MasterMix 1, 1 µL of dPCR Mix 2, 1 µL of COVID-19 Assay and 10.5 µL of either, P8, P16, P32, positive control, negative control or individual extract. The 25 µL of this PCR mix were loaded in the inlet ports of the Sapphire chips (Stilla Technologies, France). The chips were placed in the Naica Geode (Stilla Technologies, France) for droplets generation, reverse transcription and PCR amplification following the kit manufacturer's instructions.

After amplification, the chips were transferred to the Naica Prism3 (Stilla technologies, France) for fluorescence reading in the three detection channels and data were analyzed with Crystal Miner Software (Stilla Technologies, France) following the kit manufacturer's instructions.

An illustration of the resulting fluorescence dot-plots used to quantify the SARS-CoV-2 virus by RT-dPCR is shown in Supplementary Materials.

In case of discrepancies between individual RT-qPCR and grouped RT-dPCR, RT-dPCR results were confirmed by extracting individually each sample of the group and retesting them individually according to the previously described RT-dPCR protocol. Confirmatory RT-qPCR test were also performed on individual samples as previously described with the Cobas SARS-CoV-2 assay and a third method, RealStar® SARS CoV-2 RT-PCR Kit (Altona Diagnostics, Germany), was also performed on an ABI 7500 thermocycler (ThermoFisher Scientific, United-States) from the same individual extraction as for the confirmatory RT-dPCR assay. Briefly, this latter assay targets all lineage B-betacoronaviruses by amplifying a sequence of the E gene and a sequence of the S gene specific to SARS-CoV-2 as well as a heterologous internal control. A sample was considered positive to SARS-CoV-2 if the Ct value of either target is below 40 PCR cycles.

The Limit of Blank (LoB) and Limit of Detection (LoD) were evaluated for SARS-CoV-2 detection using the group testing approach used in the study on a cohort of 256 pre-epidemic nasal swab samples (negative control samples) that were collected between December 1 st 2019 and January 31 st 2020 and for which transport medium was stored at -20 °C within 48h after sampling.

Specimens were randomly grouped into 32 groups of 8 negative controls which were co-extracted and analyzed by RT-dPCR using the same protocol described above. The results for all 32 groups are given in Supplementary Materials. The LoB at 95% confidence level for the N target and the ORF1ab target is determined to be of 2 and 0 positive droplets respectively. The LoD at 95% confidence level for each target is of 0.49 copies / µL (7 copies/PCR) and 0.24 copies/µL (3 copies/PCR), respectively.

Consequently, a threshold of a least 3 positive droplets in aggregate between both the N target and ORF1ab target is used to classify a sample as positive to SARS-CoV-2 by RT-dPCR in this study.

Using routine RT-PCR testing, 25 samples were identified as positive out of the 448 samples tested, corresponding to an average test positivity rate of 5.5%. The positivity rate decreased during the study period from 8.5% for the first 224 samples to 2.5% for the last 224 samples, in correlation with the decreasing disease prevalence observed during the month of May in France. 

All results for the detection of SARS-CoV-2 by RT-dPCR for the grouped extracts (P8) and the subsequent groups of 16 (P16) and 32 (P32) are presented in Table 1 0  35  32  3  12  11  1  2  2  0  1  18  1  17  10  2  8  6  0  6  2  2  0  2  4  0  4  3  0  3  3  1  0  1  1  0  1  1  0  1 Total  56  33  23  28  12  16  14  2 12 Because sample pooling was performed systematically as samples came in the laboratory for routing RT-PCR testing, the groups contain variable numbers of RT-PCR positive samples ("RT-PCR+" samples).

Given the positivity rate of RT-PCR at the time of the study, the majority of P8 extracts had no RT-PCR+ samples (35 out of 56) and 21 groups contained at least one RT-PCR+ sample, including 18 that had one single RT-PCR+ sample.

For the largest group size of 32 samples, only 2 P32 groups had no RT-PCR+ samples. 12 P32 groups had at least one RT-PCR+ sample, including 6 with only one single RT-PCR+ sample.

Detailed results for RT-dPCR in groups of 8

The results for SARS-CoV-2 detection by RT-dPCR in groups of 8 samples, detailed in Tables 1 and 2 , are in concordance with the reference individual RT-PCR testing for 52 groups (corresponding for 416 samples), out of which 32 were RT-PCR negative groups and 20 groups contained at least one RT-PCR+ sample.

For the remaining 4 groups with discording results, three RT-PCR negative groups tested positive by RT-dPCR ("RT-PCR-/dPCR+" discordances -group IDs: P8_20, P8_28 and P8_39) and one RT-PCR+ group was found negative by RT-dPCR (RT-PCR+/dPCR-discordance -group ID: P8_02). The Ct values for the sample associated with the RT-PCR+/dPCR-discordance was of 34 and 32.3 for the E gene and ORF1ab with the Cobas® SARS-CoV-2 assay, respectively. This sample is referred to as Sample 25659 for later discussion.

Complementary analysis of the discordances are depicted below. Detailed results for RT-dPCR in groups of 16

The results for SARS-CoV-2 detection by RT-dPCR in groups of 16 samples, detailed in Tables 1 and 3 , are in concordance with individual RT-PCR testing for 25 groups (corresponding for 400 samples), out of which 11 are RT-PCR-groups and 14 are RT-PCR+ groups. Among the three groups with discording results, one presented a RT-PCR-/dPCR+ discordance (Group ID: P16_14) and two RT-PCR+/dPCRdiscordances (Group IDs: P16_13 and P16_28). Investigation of RT-PCR-/dPCR+ discordances To further investigate these RT-PCR-/dPCR+ discordances, confirmatory testing RT-dPCR was performed on all individual samples from corresponding groups. For each group, one sample tested positive by individual RT-dPCR, with measured concentrations of viral RNA ranging from 128 copies per reaction to 2 copies per reaction for the N gene, and from 106 to 1 copies for the ORF1ab gene.

The three corresponding dPCR+ samples (sample IDs: 52042, 56075 and 60401) were retested on the Investigation of the sample RT-PCR+/dPCR-One group of 8 (P8_02) was tested negative by RT-dPCR and contained one RT-PCR+ sample (Sample 25659) with Ct values of 34 and 32.3 for the E gene and ORF1ab respectively. Sample 25659 was subsequently retested by RT-PCR on the Cobas protocol and was also re-extracted individually and retested by individual RT-dPCR.

Sample_25659 was found to be borderline negative by RT-dPCR (N=2; ORF1ab=0) but Ct values of 37.3 and 34.9 were found for E and ORF respectively in the second Cobas® assessment.

Based on these results and for further sensitivity discussions, sample 25659 is considered as a true positive samples. 

In this work, we assessed the sensitivity and specificity of group testing combined with digital PCR for SARS-CoV-2 detection. Three different group sizes were investigated using a commercially available digital PCR assay, the COVID-19 Multiplex Digital PCR Detection Kit (Stilla Technologies, France/Apexbio, China). This assay demonstrated a low LoB (at 2 and 0 positive droplets per PCR for N and ORF1ab genes, respectively) and low LoD (at 5 and 3 copies/PCR for N and ORF1ab genes, respectively). This LoD is lower than most estimation for WHO and other reference RT-PCR assays typically ranging between 5 to 500 copies/PCR (17, 18) .

For our analysis, we proposed a protocol of group screening performed by RT-dPCR with secondary individual re-testing of positive groups as illustrated in Figure 3 . We assessed this protocol by assessing in real-life condition 448 consecutive samples grouped by 8, 16 and 32 samples.

We observed a better sensitivity than individual RT-PCR testing for groups of 8 samples. Indeed, a total of 23 groups of 8 samples tested positive and included 26 true positive samples. Only 25 samples were identified as positive using reference individual RT-PCR testing, corresponding to an +4% improvement in sensitivity. Two among the three samples associated with RT-PCR-/dPCR+ discordances were confirmed as true positive by RT-PCR using a SARS-CoV-2 specific assay from Altona. The last one was not confirmed as positive by RT-PCR and is undergoing further investigations. One sample was associated to a RT-PCR+/dPCR-discrepant results and is also considered a true positive sample as confirmed by RT-PCR retesting. Further investigation by sequencing is underway on these 4 samples in an effort to rigorously assess the existence and nature of nucleic acids variations responsible for the discording results. Testing in the 14 groups of 32 samples by RT-dPCR has 100% concordance with the reference RT-PCR testing. Re-testing positive groups by RT-dPCR would have likely led to better sensitivity than RT-PCR. However, we are careful in drawing conclusions for groups of 32 given the limited data points in this case as only 14 groups, including only 2 RT-PCR negative groups.

An alternative and even more cost-effective group testing protocol could be to perform the re-testing steps using RT-PCR with Cobas or Altona assays. In these protocols, the sensitivity becomes dependent on the RT-PCR kit used, leading to potential discrepancies with RT-dPCR as observed in our results for groups of 8 samples. However, our data suggest that performing the individual tests for groups of 8 and 16 with a RT-PCR assay would still have a similar sensitivity.

Overall, our data indicates that COVID-19 group testing combined with digital PCR for large group sizes of 8 and 16 samples, followed by individual re-testing of positive groups, has better or similar sensitivity than individual RT-PCR testing. Groups of 32 samples could also be considered, but our analysis needs to be confirmed in the future as, due to the epidemiological situation at the time, a low number of negative groups and groups containing a single positive sample were assessed.

The gain in sensitivity of the proposed method is likely due to a combination of i) a concentration effect due to performing the pooling prior extraction and performing the extraction step from a large volume of 1 mL of pooled transport medium and ii) the intrinsic superior sensitivity of digital PCR compared to RT-PCR, as demonstrated previously for SARS-CoV-2 (14-16) and other viruses (12, 19) detection.

Below standard sensitivity is one of the main reasons why group testing has not been widely adopted for COVID-19 testing, whilst research groups have advocated for its implementation as a solution to the world-wide un-met demand for tests and reagent shortage(2-9).

The current study suggests that the high sensitivity of group testing could be achieved using by digital PCR instead of RT-PCR in the first group screening step. It makes the approach viable for large-scale, low cost patient screening with minimum reagent consumption. For a test positivity rate of 1%, our group testing protocol would save 79%, 80% and 69% of reagents compared to individual testing, for group sizes of 8, 16 and 32 samples. Group size does not significantly impact cost or reagent consumption for group sizes above 8 samples and test positivity rates below 1%. Consequently, the choice of group size (8, 16 or 32) is mostly a balance of sensitivity and test capacity. Our results indicate that group testing by digital PCR and a group size of 16 would increase testing capabilities by more than 10-fold while having a sensitivity comparable to the current standard of individual testing by RT-PCR.

Group testing can be used in various context where testing is not currently put in place due to testing capacity or with strong economics constraints and where SARS-CoV-2 prevalence is low. In countries where the pandemic is not yet under control or could re-emerge, enhancing testing capacity is essential to winning the battle against COVID-19. In countries where the first wave is fading, the fight against COVID-19 goes through a combination between a test / trace / isolate strategy and social distancing. Increasing the range of people tested amongst contacts with positive cases, but also periodic testing of population in frequent contact with others (e.g. nurses, transportation workers, clerks, etc…) as well as in fragile populations such as nursing homes can be part of a strategy against COVID-19 while allowing a relaxation of social distancing measures at the same time. Group testing can help in all of these situations.

Supplementary Materials S1 -Data format for RT-dPCR as obtained with the Naica System The pre-epidemic groups were used as negative controls for SARS-CoV-2 detection. Only one out of 32 (group 5) had one positive droplet for the N target out of 19881 droplets analyzed. The LOB at 95% confidence level for this RT-dPCR assay was estimated automatically by the Gene-Pi online statistical tool (https://www.genepi.com/statistical-tools/loblod/), considering a droplet volume of 0.548 nL as specified by the manufacturer. The LOD at 95% confidence level is computed from the experimental LOB. The LOB for the N and ORF1ab targets are found to be of 2 and 0 droplets respectively. The LODs are found to be of 0.46 copies/µL and 0.22 copies/µL, or equivalently of 7 and 3 copies per PCR reaction.

Using a joint analysis, a LOB analysis can be performed for the combination of the N and ORF1ab targets. In the case, the sum of the positives for the N and OFR1ab target is used as the measure and the LOB for the sum is 2 positive droplets. Consequently, in this study, a RT-dPCR result is considered as positive when the sum of positive droplets in the N and ORF1ab targets is of 3 droplets or more (strictly greater than 2).

For details on calculation models, see: 

The Detection of Defective Members of Large Populations

Working Papers. Center for Research in Economics and Statistics; 2020 avr

Evaluation of COVID-19 RT-qPCR test in multi-sample pools

FACT-Frankfurt adjusted COVID-19 testing-a novel method enables high-throughput SARS-CoV-2 screening without loss of sensitivity

Pooling nasopharyngeal swab specimens to increase testing capacity for SARS-CoV-2. bioRxiv

Sample Pooling as a Strategy to Detect Community Transmission of SARS-CoV-2

Assessment of Specimen Pooling to Conserve SARS CoV-2 Testing Resources

Assessment of sample pooling for clinical SARS-CoV-2 testing [Internet]. Microbiology; 2020 mai

Pooling of samples for testing for SARS-CoV-2 in asymptomatic people. The Lancet Infectious Diseases

Rapport de l'HCSP. Paris: Haut Conseil de la Santé Publique; 2020 mai

Highly Precise Measurement of HIV DNA by

Next Generation Digital PCR Measurement of Hepatitis B Virus Copy Number in Formalin-Fixed Paraffin-Embedded Hepatocellular Carcinoma Tissue

Tolerance of Droplet-Digital PCR vs Real-Time Quantitative PCR to Inhibitory Substances

Quantitative Detection and Viral Load Analysis of SARS-CoV-2 in Infected Patients

Highly accurate and sensitive diagnostic detection of SARS-CoV-2 by digital PCR

ddPCR: a more accurate tool for SARS-CoV-2 detection in low viral load specimens. Emerging Microbes & Infections

Evaluation of the QIAstat-Dx Respiratory SARS-CoV-2 Panel, the first rapid multiplex PCR commercial assay for SARS-CoV-2 detection

Analytical sensitivity and efficiency comparisons of SARS-COV-2 qRT-PCR assays

HIV-1 remission following CCR5Δ32/Δ32 haematopoietic stem-cell transplantation

In this study, we did not observe any discordance for test results in groups of 32 when compared to the reference individual RT-PCR results. In particular and surprisingly, there were no qPCR+/dPCRdiscordances although we would suspect an increased occurrence of such discordances as group size increases from the dilution effect of group testing.Interestingly, the 2 samples that were associated with the 2 qPCR+/dPCR-discordances for groups of 16 In case of P32_07, the group of 32 is a combination of 2 P16 groups with opposite discordances: P16_13 (qPCR+/dPCR-) and P16_14 (qPCR-/dPCR+). It is likely that the RNA templates from the P16_14 group are those that were detected in the P32_07 group.In the case of P32_14, the number of positive droplets observed in RT-dPCR is right above the threshold of 3 positive droplets (N=2; ORF1ab=1) while for the corresponding group of 16 (P16_28), the number of positive droplets is just below with 2 droplets (N=2; ORF1ab=0). In this case, statistical variations due to sampling error could explain the observation.