key: cord-351864-zozrj7w5
authors: Chappleboim, A.; Joseph-Strauss, D.; Rahat, A.; Sharkia, I.; Adam, M.; Kitsberg, D.; Fialkoff, G.; Lotem, M.; Gershon, O.; Schmidtner, A.-K.; Oiknine-Djian, E.; Klochendler, A.; Sadeh, R.; Dor, Y.; Wolf, D.; Habib, N.; Friedman, N.
title: ApharSeq: An Extraction-free Early-Pooling Protocol for Massively Multiplexed SARS-CoV-2 Detection
date: 2020-08-13
journal: nan
DOI: 10.1101/2020.08.08.20170746
sha: 
doc_id: 351864
cord_uid: zozrj7w5

The global SARS-CoV-2 pandemic led to a steep increase in the need for viral detection tests worldwide. Most current tests for SARS-CoV-2 are based on RNA extraction followed by quantitative reverse-transcription PCR assays that involve a separate RNA extraction and qPCR reaction for each sample with a fixed cost and reaction time. While automation and improved logistics can increase the capacity of these tests, they cannot exceed this lower bound dictated by one extraction and reaction per sample. Multiplexed next generation sequencing (NGS) assays provide a dramatic increase in throughput, and hold the promise of richer information on viral strains and host immune response. Here, we establish a significant improvement of existing RNA-seq detection protocols. Our workflow, ApharSeq, includes a fast and cheap RNA capture step, that is coupled to barcoding of individual samples, followed by sample-pooling prior to the reverse transcription, PCR and massively parallel sequencing. Thus, only one step is performed before pooling hundreds of barcoded samples for subsequent steps and further analysis. We characterize the quantitative aspects of the assay, and test ApharSeq on dozens of clinical samples in a robotic workflow. Our proposed workflow is estimated to reduce costs by 10-50 fold, labor by 5-100 fold, automated liquid handling by 5-10 fold, and reagent requirements by 100-1000 fold compared to existing testing methods.

A novel Coronavirus, SARS-CoV-2, has infected almost 20 million, as of early August 2020, with more than 700,000 related deaths caused by the virus associated COVID-19 1 . There is a consensus that large-scale testing is paramount for the containment of the pandemic 2 , yet widespread testing in most countries is lacking. At the time of writing this manuscript, multiple countries are undergoing a "second wave" of infections, highlighting the need for population-level screens for prolonged durations, effectively requiring a billions of tests world-wide.

The current benchmark for SARS-CoV-2 testing is a panel of RT-qPCR tests approved by various institutions applied to nasopharyngeal swab samples 3 . The swabs are mixed into a lysis buffer or transport buffer and subsequently mixed with a lysis buffer, and undergo RNA extraction and RT-qPCR. Generally, in these tests, samples with cycle threshold (Ct) lower than 37 are considered positive. While this test is sensitive and specific, qualified labor and the shortage of required equipment and reagents have proven to be limiting factors at different stages of the pandemic 4 . Specifically, the testing capacity is limited since each sample is treated as a separate qPCR reaction with a fixed reaction time.

In the last decade, next generation sequencing (NGS) has replaced RT-qPCR and microarrays as the assay of choice for quantifying RNA molecules. Amidst the pandemic, different groups suggested NGS-based assays to measure the existence and abundance of the viral genome in samples 5 , 6 , 7 . In addition to detection and quantification, these assays can provide strain-specific sequence information as events are unfolding, providing epidemiologists with data to analyze the contagion propagation through the population 8, 9 . Similarly, by assaying the RNA from host cells, aspects of the immune response in infected individuals can be unmasked 10 , providing potentially crucial information for patient treatment, research, and policymakers.

Here, we propose a general modification of current RNA-seq protocols that allows for pooling of barcoded samples prior to reverse transcription -A mplicon P ooling by H ybridization A nd R NA-Seq (ApharSeq). This improvement is especially pertinent for large scale testing as it potentially reduces the need for labor and reagents by orders of magnitude. Briefly, we show ( Figure 1 ) that we can introduce barcoded and target-specific reverse transcription primers to the samples, allowing them to hybridize to target RNA molecules already in the lysis buffer, or after a brief RNA cleanup step. After several minutes of hybridization with the barcoded primers, we capture the sample RNA onto beads, washing away excess primers and the chaotropic lysis buffer. Importantly, the primer-RNA hybrids are preserved during this step. The bead-bound RNA is isolated, and samples are pooled to a single tube to undergo reverse transcription from the primers that remain hybridized to their targets. Finally, the pool undergoes library PCR and sequencing. Viral molecular counts per sample are determined by the sample-specific barcodes and unique molecular identifiers introduced at the very beginning of the protocol. We . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint demonstrate that cross-sample contamination in this workflow is negligible, and determine sensitivity to be ~450-900 copies/ml, comparable to existing approved tests 11 .

A) Barcoded and uniquely-identifiable RT primers are hybridized to samples in transport/lysis buffer for 10 minutes, paramagnetic beads are used to extract RNA and wash excess primers.

Beads are pooled and RNA undergoes an RT/PCR reaction with pre-hybridized target-specific primers to generate a sequencing library.

Libraries are sequenced and analysed, PCR duplicates are collapsed to molecular counts for detection and potentially more elaborate analyses, e.g. contact tracing by viral sequence analysis, host physiological state by mRNA analysis.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint

A simple and quick RNA capture step Lysis buffers contain protein denaturation and degradation reagents, typically the chaotropic guanidium thiocyanate, as well as detergents. Therefore, enzymatic reactions such as reverse transcription require RNA extraction from the lysis buffer in which the samples arrive. We tested two home-made bead-based RNA cleanup protocols, SPRI and polyT extraction 12 , that are cheap, simple, amenable to automation, and take less than 30 minutes for 96 samples. In terms of RNA extraction, the performance of both approaches is within a ±50% range of a widely used commercial kit (see supplementary note on RNA extraction; Figure 2A ; Figure S1 ). Preliminary tests showed that both approaches can be used for ApharSeq ( Figure S1D ), but we focused on the polyT-based extraction variant. Note that we do not elute the RNA from the polyT beads, rather we use the beads to capture the RNA and continue with the bead-bound material to the next step of the protocol.

We designed barcoded RT primers for the viral E gene (reverse), as it appears in the WHO panel 13 . The primer includes a 10bp barcode and a 10bp unique molecular identifier to allow for single-molecule counting 14 ( Figure   1A , methods). Each sample is hybridized to primers with a different barcode ( Figure 1A ), effectively identifying the RNA for the remainder of the protocol. The bead-bound RNA is then washed and reverse transcribed to generate sample-labeled cDNA.

To evaluate the efficacy of primer-RNA hybrid formation and stability through the cleanup stage, we designed a qPCR reaction targeting the generic PCR handle on the RT primer and the amplicon target sequence. This assay allows us to quantify the number of hybrids that survive the washes and succeed in generating a cDNA molecule.

Using the qPCR assay, we established that RT primers indeed remain hybridized during the RNA extraction and initiate reverse transcription reactions ( Figure S2 ). Furthermore, we used this simple assay to run several optimizations for the first steps of the protocol and improve the yield significantly ( Figure S2 ).

The next step in the ApharSeq protocol is to generate sequencing libraries. This is achieved in a single PCR step, amplifying the RT primer from the generic sequencing handle and from the forward amplicon-specific primer, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint extending the amplicon with Illumina-compatible handles ( Figure 1B ). When we applied this PCR to positive and negative samples we consistently obtained amplicon-specific libraries only in SARS-CoV-2-positive samples ( Figure 2B ). When sequenced on a next generation sequencing platform, these libraries yield highly specific results with >95% of reads aligning to expected viral target sequence in positive samples ( Figure 2C ). Importantly, sequencing the insert provides direct evidence for the presence of the virus in samples, bolstering confidence in the assay's results. Observing the target sequence directly allowed us to identify viral sequence variations in some cases ( Figure 2D ).Cross-Sample Contamination is minimal When pooling samples early on in the protocol, the main concern is that RNA molecules will be erroneously tagged due to residual free primers, or due to other artifacts during RT, PCR, or sequencing. To test cross-contamination levels in the RT stage, we hybridized positive (Ct 26) and negative samples with two differently barcoded primers, pooled them, performed RT and tested the amount of cross-contamination by barcode-specific qPCR ( Figure 3A and 3B). We find that the pooled negative sample is indistinguishable from the unpooled negative sample, suggesting cross contamination is negligible.

Next, we examined potential cross-contamination during PCR or sequencing 15 . We subjected four samples -Ct18, Ct33, and two negative controls (ddw) to ApharSeq. We hybridized the barcoded primers, and then pooled the samples prior to RT and PCR ( Figure 3C ). Again, we find that barcodes that were hybridized to negative samples have at least 50,000-fold less reads than those that were hybridized to the high Ct positive sample. These results are not unique to the polyT-based capture, and were qualitatively replicated using SPRI-based RNA cleanup ( Fig   S1D) . We conclude that cross-sample contamination is a minor issue in ApharSeq.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 13, 2020. . C) Libraries are highly specific : >75% of libraries have more than 95% of reads mapped to the expected sequence. D) Observed sequences conform to the reference genome in >99.9% of reads (not shown). However, in at least one sample we observe a sequence variation in reads from the E amplicon at 26,353 (G to A). Observed SNP is in more than 90% of reads of each UMI shown (~200 reads each), excluding the possibility of sequencing errors.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint

To evaluate the dynamic range of ApharSeq, we titrated a positive sample into lysis buffer and generated samples that span the Ct range~23-31 in 320 µl. We applied ApharSeq to these samples in a pool and as individual samples ( Figure 4A ). We find that the number of unique molecules scales linearly with the input (p-value < 0.001; Figure 4B ). Accounting for observed background in negative samples, we predict the limit of detection to be~Ct 35.7. Importantly, the linear titration curve in the case of the pooled samples highlights the minimal interaction and contamination between pooled samples.

We next tested the limit of detection directly, by performing another pooled titration experiment with highly diluted samples with Ct range 30-42 ( Figure 4C ), while adding a quantified reference (methods). Using these quantified controls, we estimate the end-to-end capture rate of ApharSeq at~1.5% (observed 33 and 14 molecules out of an input of~2000 and~1000 molecules, respectively). Similarly, we could also calibrate the titration curve from Ct units to molecular counts, and found the limit of detection to be 450-900 molecules/ml, depending on sequencing depth ( Figure 4D ), threshold selection, and input volume used (methods).

A major advantage of sequencing-based assays is the capacity to capture and readout a large number of targets from the same sample 16 . As a first step towards a multi-target assay, we aimed at multiplexing two targets. We started by designing RT and PCR primers for the viral N1 amplicon, as it appears in the CDC panel 17 , and used it in conjunction with the E amplicon RT and PCR primers. We applied ApharSeq to a positive sample with each primer separately or with both primers together ( Figure 5A ). The results of individual and multiplexed amplicons are almost identical ( Figure 5B ), suggesting that the viral targets are largely independent and can be probed simultaneously to improve confidence, and potentially improve sensitivity. Notably, the N1 amplicon yields roughly 2-3 fold more molecules, consistent with previous reports 18 .

Next, as an internal control, we designed primers for several human transcripts with varying expression levels 19 and after a preliminary test ( Figure 5C , Figure S3 ) we decided to continue with a β-Actin amplicon as it is also used in an approved detection kit 20 . Subjecting positive and negative samples to the ApharSeq pipeline with primers targeting viral E and human ActB amplicons produced sequencing libraries, albeit with slightly reduced yields ( Figure 5D ). Importantly, qPCR tests on mixed libraries showed that titration of the human-specific primer in the PCR affects the human/viral amplicon ratio accordingly, allowing for calibration of the number of reads allocated to each target in a multi-target library ( Figure S3 ).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint Samples are subjected to PCR with different barcodes (red/green) to distinguish their treatments. B) Assay is linear in both cases (p value < 0.001). The linearly extrapolated LoD for pooled samples is~Ct 35.7 and is 4 times lower than the LoD for the single samples (methods). C) Low target titration experiment , in which we also introduced a viral target control for quantification (methods) shows the actual LoD is ~680 molecules/ml (Ct ~37.4).

Highlighted bar (~25,000 reads per sample) corresponds to data in C.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint The one positive sample that yielded no viral reads at all was diluted twice and did yield many N1 reads in the other instance. We therefore suspect that we made an error during sample preparation, and this is not a false negative instance.

Overall, we conclude that the robotic protocol works efficiently, with minimal to none cross-sample contamination, and we can clearly separate positive from negative samples.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint

The results presented in this manuscript establish the ApharSeq pipeline as a SARS-CoV-2 detection test. We believe that ApharSeq is uniquely poised to allow tests at a massive scale, for both the low cost and minimal labor it requires.

The cost for a single test is currently between 2$-4$ (when processing 8000-1000 samples, see supplementary note on costs), and the bulk of the costs are due to primer synthesis and beads (50%) and consumable plastics (tips, plates, etc, 30%). These prices can probably be reduced by a factor of 10-20 when the process is streamlined, and reagents are manufactured and purchased in bulk.

In terms of labor, the bottleneck is the first step that involves per-test hybridization and washes. While this can be performed manually in roughly 45' for 96 samples with a multichannel pipette, the optimal setup includes an automated liquid handling station, in which case this step requires 20-40 minutes, depending on the sample volume tested and the specifics of the robotic setup.

The sequencing requirement for the test is reading a single pool barcode (8 bp) and at least 50 bases from read1 or 20 bp from read1 and 30 bp from read2. When we down-sampled the data in the titration experiment, we found that a change of x1000 in sequencing depth only incurs a 2-fold reduction in sensitivity. Additionally, using the Ct distribution observed in the clinic, we performed a simulation to estimate the false positive and false negative rate as a function of sequencing depth ( Figure S5 ). In these simulations, when sequencing depth ranged from 10,000 to 100,000 reads per sample the false negative rate ranged from 4.5% to 0.2% respectively. We conclude that 50,000 reads per sample on average should suffice ( Figure 4D, figure S5 ), especially considering that most samples are negative and "consume" much less sequencing reads (from the human target only). This means that a single NextSeq run with 400x10 6 reads suffices for processing~8,000 samples, and a NovaSeq S2 100bp run with 8x10 9 reads for processing ~160,000 samples.

While highly informative and rich, the main drawbacks of sequencing-based assays are twofold -the requirement for specialized and expensive equipment, and the slow generation of data (e.g. a NextSeq 55bp run requires~6.5 hours). While a 12-18 hour turnaround time is probably prohibitive for emergency testing, we believe that for large scale and routine population screens it is reasonable. Further, there are potential optimizations and workarounds that can reduce the sequencing runtime by a factor of 2-3 21 .

Importantly, we have preliminary results (not shown) that pooled RNA can be maintained for at least 24 hours in a standard RNA preserving buffer, and therefore samples can be processed in different facilities and collected to a central sequencing site for the last steps of the protocol (Figure 7) .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . Since the main equipment required for the application of ApharSeq already exists in many diagnostic facilities, we envision a gradual transition from the qPCR technology to sequencing-based assays by partial re-purposing of labs. For example, a facility with five liquid handlers can process upto~2000 samples in two hours 12 , and then get back to its usual testing qPCR program. The pools generated during these two hours from four such facilities can be collected to a central sequencing center.

Once such advancements are made and ApharSeq proves useful and informative it can be used as an alternative avenue in case of supply shortages, or diversify the types of tests a single facility can provide for different use cases, namely high-throughput non-urgent population screens vs. lower-throughput urgent testing. We propose a division of labor scheme where any facility that has an automated liquid handler and a PCR block can function as a first step in the processing pipeline. A single 96 plate is processed and pooled every~30 minutes in every such station. These pooled batches can be collected to a central sequencing facility that pools them further for cost-effective sequencing.

If collection into plates is not feasible, there are dedicated automated liquid handler stations for re-organizing tubes into plates, as is already the case in most testing facilities.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint

Here we propose ApharSeq, an early-pooling protocol for the detection of the SARS-CoV-2 virus in clinical samples using next generation sequencing. While sensitive RT-qPCR assays are the backbone of testing in the current pandemic, they are lacking sequence information that may be crucial to trace infection chains, and more importantly, they are difficult to scale up.

The current second wave of infections highlights that global testing and tracing efforts require an orders-of-magnitude scale-up. We believe that our approach, and future improvements and modifications thereof, might form the basis for such an improvement of several orders of magnitude relative to current testing strategies.

Specifically, the ability to pool hundreds or thousands of samples as early as possible, without losing the sample-specific information, is a crucial improvement on all currently published NGS-based tests 6 , 22 , 23 , 24 , 25 , 26 . The early pooling approach proposed here, which might be applicable to other protocols, elicits proportional reductions in costs and labor that directly translate to higher throughput in testing.

While we established key properties of our approach, namely linearity, high-sensitivity, low cross-reactivity, and the potential for multi-target testing, there's still much to be done. Specifically, more optimization can improve efficiencies, shorten durations, reduce background noise and increase the sensitivity. Similarly, it is straightforward to introduce multiple internal synthetic controls that will provide clinicians with reproducible and informative assay measures and reliable results 7 .

Once the detection hurdle is surpassed and the relevant infrastructure and logistics are set into place, ApharSeq should be easy to extend to other more advanced applications in the context of the disease. Testing for co-morbid or confounding pathogens like the flu, amplifying viral variable regions to identify infection chains 27 , 28 , and monitoring the host immune response with key transcripts 29 are just some of the potential applications of our generic approach. While these will incur some additional costs, we believe that for many applications the benefits will significantly outweigh the costs.

Finally, we want to note that the approach developed and validated here -hybridization of barcoded primers followed by early pooling -is of a general nature and can potentially be used to enhance existing protocols, including single cell and bulk RNA-seq protocols.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. 

There are two different PCR strategies we employed during the development process -one-step and two-step PCR. The two step PCR is composed of a first step that amplifies the target molecules with an extendable handle, and in the second step barcode and the remaining Illumina sequences are introduced. The one-step reaction performs everything in a single reaction with a single long primer (~90 bp). A one-step reaction is more convenient, and is less prone to contaminations (see supplementary note on contaminations), however, it's less modular. Specifically -the long primer contains a target specific sequence and a barcode, which means that a barcoded primer collection must be synthesized per target. The two step PCR decouples this dependency, which means that a single collection of barcoded primers can be used on any target, assuming a simple target-specific primer is used in the first step. Both approaches yielded similar results, and we are currently using the single-step reaction to avoid contaminations.

The PCR amplifies the generic handle on the RT primer on one side, and a target-specific sequence on the other side.

Additionally, the PCR extends the amplicons to a sequencing library by adding the relevant flanking Illumina sequences.

Including an 8bp barcode that marks the pool of samples amplified in the PCR reaction:

Reverse primer (generic):

CAAGCAGAAGACGGCATACGAGATGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT Forward primer (target-specific, and indexed ) -N1

gaccccaaaatcagcgaaa -E acaggtacgttaatagttaatagcgt -ActB caccaactgggacgacat TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG Forward primer (B6 -N1 PCR primer):

The second PCR step extends the handles to a complete library with the Ad1.x and Ad2.x indexed primers as published 30 .

The detailed and complete protocol was published separately 31 . For convenience we briefly describe the protocol below.

Since the protocol stabilized with time, some experiments are slightly modified relative to the current protocol. The supplementary material contains a list of experimental modifications per experiment shown indexed by figure panel.

We tested commercially available polyT beads (ThermoFisher dynabeads cat# 61002 ), or conjugated carboxylate coated beads (GE healthcare Sera-Mag SpeedBeads cat# 65152105050250), and followed the manufacturer conjugation protocol .

This RNA purification protocol is based on a protocol for rapid isolation of mRNA 32 with some modifications. Briefly, polyT conjugated beads were washed once and resuspended in binding buffer. The resuspended beads were mixed 1:1 with the sample. After a hybridization period of 10 minutes at room temperature with periodic mixing, the supernatant is removed and the beads are resuspended in a 50 µl 1:1 mix of binding buffer and 10 µM barcoded RT primers. To denature RNA secondary structures, the samples were incubated at 72C for 2 minutes and immediately transferred to ice for at least 2 minutes. Samples were then incubated at room temperature for 10 minutes with periodic mixing to allow hybridization of RNA to the beads and to RT primers. Beads were resuspended in 450 µl wash buffer A and magnetized. Majority (380 µl) of the supernatant was removed and beads were resuspended in the remaining 70 µl of buffer A, and pooled. After pooling samples are washed once in buffer A, twice in buffer B, and can then be kept in RNA later until they are processed further. Preliminary tests show that RNA can be stored on the beads, in RNA later at 4C, for at least a week. Option 2: Purification and hybridization on SPRI beads RNA extraction with SPRI beads, followed our published protocol for RNA extraction 12 with several modifications. Samples in lysis/transfer buffer were mixed with barcoded RT primers, then incubated at 72C for 2 minutes and immediately transferred to ice for at least 2 minutes. Samples were then mixed 1:1 with binding buffer (as above) and incubated at room temperature for 10 minutes with periodic mixing to allow primer hybridization. Next, samples were mixed 1:0.8 with home-made SPRI beads in PEG buffer 12 . Beads were washed twice with freshly made Ethanol 80%, air dried, and eluted in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted August 13, 2020. . DDW. This was followed by a second 0.8x SPRI cleanup to ensure the removal of any excess primers. At this stage samples were pooled to a PCR tube to undergo RT and PCR.

The pooled beads were washed once with 1xRT buffer and RT reaction was performed with SmartScribe enzyme (SMARTScribe Reverse Transcriptase, Takara Bio) at 42C for one hour followed by incubation at 70C for 15 minutes. To elute the cDNA from the beads, the samples were incubated at 98C for 2 minutes, magnetized and the supernatant was transferred to a new tube and cleaned by SPRI beads x2 (Agencourt AMPure XP, Beckman Coulter). NGS sequences were added by PCR (KAPA HiFi HotStart ReadyMix, Kapa Biosystems, 30 cycles), and the DNA was purified using 2x SPRI beads.

Reads were demultiplexed using bcl2fastq (version 2.20.0) and further processed by ad-hoc python scripts that are available as jupyter notebooks. We used the python API of UMI-tools to cluster UMI sequences. Specifically the "directional" option with the edit distance threshold set to three 33 Quantifying target molecules To generate a quantitative polyA viral reference, we extracted RNA from a clinical sample and estimated the amount of molecules in this RNA extract to be 6000 molecules/µl using a synthetic viral sequence ( Twist Bioscience SARS-CoV-2 RNA, MN908947.3) as a reference in a standard RT-PCR kit. We loaded two samples with 10 and 5 µl of this reference RNA in a total of 320 µl lysis buffer, and applied the ApharSeq protocol. Only 1/30 of the material underwent library preparation, which means that at most, we expect to see 2000, or 1000 molecules in the 10, 5 µl samples respectively. After UMI clustering we observe 33 and 14 molecules respectively, suggesting that we capture~1.5% of molecules. In the same experiment, a sample corresponding to cycle 29.3 had a similar UMI count (32 molecules), allowing us to roughly calibrate the Ct units to target molecules / ml at Ct 29.3 = 6150 x 10 / 0.32 = ±190,000 molecules/ml. This number is much higher than the numbers reported in other publications (for example Ct 33.5 = 200 i.e. Ct 29.3 < 3500). If true, this means the LoD of our assay is actually lower than reported in this manuscript.

For the high load titration ( Figure 4B ), a linear fit (python scipy.stats.linregress) was performed:

x y~b + a . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted August 13, 2020. . https://doi.org/10.1101/2020.08.08.20170746 doi: medRxiv preprint Where y is the log 10 (#UMIs) and x is the calculated Ct of the sample. Given this linear fit, we can extrapolate to the UMI detection threshold, which in this case was set to 3 (a conservative estimate). The fit statistics are: For the low load titration ( Figure 4C and 4D), we perform re-sampling of the data (x500 times):

For factor in (1, 3, 10, 30, 100, 300, 1000):

For each UMI in sample #sampled-reads(UMI) ← Poisson(#reads(UMI) / factor) We then count the number of UMIs per (sample, factor, replicate) as the number of UMIs with #sampled-reads(UMI) > 0.

Given these counts, we set the detection threshold as the minimal number of UMIs that is above 99% of replicates in the negative samples. Therefore this number varies with sequencing depth and the UMI background in the negative samples. We fit each sampled replicate of the data with a Poisson-noised exponent:

Where y is the number of observed UMIs and x is the calculated Ct of the sample. We then set the LoD per factor to be the maximal Ct such that 95% of replicates is above the LoD. 

Coronavirus Pandemic (COVID-19)

Population-scale testing can suppress the spread of

COVID-19 diagnostics in context

Scramble for Coronavirus Supplies, Rich Countries Push Poor Aside

CoronaHiT: large scale multiplexing of SARS-CoV-2 genomes using Nanopore sequencing

LAMP-Seq: Population-Scale COVID-19 Diagnostics Using Combinatorial Barcoding

Fast and accurate diagnostics from highly multiplexed sequencing assays

Full genome viral sequences inform patterns of SARS-CoV-2 spread into and within Israel

The emergence of SARS-CoV-2 in Europe and the US

Host-Viral Infection Maps Reveal Signatures of Severe COVID-19 Patients

Diagnosing COVID-19: The Disease and Tools for Detection

Sars-CoV2 RNA purification with homemade SPRI beads for RT-qPCR test v1 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) (protocols.io.beswjefe)

Detection of 2019 novel coronavirus (2019-nCoV) by real-time RT-PCR

Counting absolute numbers of molecules using unique molecular identifiers

Characterization and remediation of sample index swaps by non-redundant dual indexing on massively parallel sequencing platforms

Simple, Flexible, and Versatile SNP Genotyping by Highly Multiplexed PCR Amplicon Sequencing

US CDC Real-Time Reverse Transcription PCR Panel for Detection of Severe Acute Respiratory Syndrome Coronavirus 2. Emerg

The coding capacity of SARS-CoV-2

The Genotype-Tissue Expression (GTEx) project

Reliable variant calling during runtime of Illumina sequencing

Scalable, rapid and highly sensitive isothermal detection of SARS-CoV-2 for laboratory and home testing

INSIGHT: a scalable isothermal NASBA-based platform for COVID-19 diagnosis

REMBRANDT: A high-throughput barcoded sequencing approach for COVID-19 screening

HiDRA-seq: High-Throughput SARS-CoV-2 Detection by RNA Barcoding and Amplicon Sequencing

Highly multiplexed oligonucleotide probe-ligation testing enables efficient extraction-free SARS-CoV-2 detection and viral genotyping

High-density amplicon sequencing identifies community spread and ongoing evolution of SARS-CoV-2 in the Southern United States

Rapid SARS-CoV-2 whole genome sequencing for informed public health decision making in the Netherlands

Shotgun Transcriptome and Isothermal Profiling of SARS-CoV-2 Infection Reveals Unique Host Responses, Viral Diversification, and Drug Interactions

Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position

SARS-CoV-2 detection with ApharSeq. protocols

In situ isolation of mRNA from individual plant cells: creation of cell-specific cDNA libraries

UMI-tools: modeling sequencing errors in Unique Molecular Identifiers to improve quantification accuracy

We thank Michal Rabani for critical comments and support. We thank Michal Bronstein, Abed Nasereddin, Idit Shiff, Adi Turjeman, Netta Barak, and Moran Yassourfor their help. This work was supported in part by the Rothschild Foundation. AC is an Azrieli scholar and would like to thank the Azrieli Foundation for their support.

Viral RNA was extracted from an in-vitro grown virus (Ct~14) and serially diluted 1:25 in a negative sample. Each dilution was subjected to RNA extraction in three methods: