COV-ID: A LAMP sequencing approach for high-throughput co-detection of SARS-CoV-2 and influenza virus in human saliva

The COVID-19 pandemic has created an urgent need for rapid, effective, and low-cost SARS-CoV-2 diagnostic testing. Here, we describe COV-ID, an approach that combines RT-LAMP with deep sequencing to detect SARS-CoV-2 in unprocessed human saliva with high sensitivity (5-10 virions). Based on a multi-dimensional barcoding strategy, COV-ID can be used to test thousands of samples overnight in a single sequencing run with limited labor and laboratory equipment. The sequencing-based readout allows COV-ID to detect multiple amplicons simultaneously, including key controls such as host transcripts and artificial spike-ins, as well as multiple pathogens. Here we demonstrate this flexibility by simultaneous detection of 4 amplicons in contrived saliva samples: SARS-CoV-2, influenza A, human STATHERIN, and an artificial SARS spike-in. The approach was validated on clinical saliva samples, where it showed 100% agreement with RT-qPCR. COV-ID can also be performed directly on saliva adsorbed on filter paper, simplifying collection logistics and sample handling.


INTRODUCTION 46
Within the first year of the COVID-19 pandemic SARS-CoV-2 has swept across the world, leading 47 to more than 130 million infections and over 2.8 million deaths worldwide (as of April 2021). In 48 many countries, non-pharmaceutical interventions, such as school closures and national 49 lockdowns, have proven to be effective, but could not be sustained due to economic and social 50 impact 1, 2 . Regularly performed population-level diagnostic testing is an attractive solution 3 , 51 particularly as asymptomatic individuals are implicated in rapid disease transmission, with a 52 strong overdispersion in secondary transmission 4 . Maintenance of population-level testing can be 53 successful in isolating asymptomatic individuals and preventing sustained transmission 5, 6 ; 54 however, considerable barriers exist to the adoption of such massive testing strategies. Two such 55 barriers are cost and supply constraints for commercial testing reagents, both of which make it 56 impractical to test large numbers of asymptomatic individuals on a recurrent basis. A third major 57 barrier is the lack of "user-friendly" protocols that can be rapidly adopted by public and private 58 organizations to establish high-throughput surveillance screening. In addition, while COVID-19 59 testing of symptomatic individuals might be effective during the summer season, when other 60 respiratory infections are rare, new strategies are needed to facilitate rapid differential diagnosis 61 between SARS-CoV-2 and other respiratory viruses in winter. 62 . 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) 4 internal sequences, are incorporated in opposite orientation across the target sequence in the 6 We used COV-ID-adapted primer sets for N2 and STATH (Table S1) in multiplex on inactivated 158 saliva spiked with a range of SARS-CoV-2 from 5 to 10,000 virions/μL. Subsequently, each  LAMP reaction was separately amplified via PCR using a unique P5 and P7 index combination, 160 pooled, quantified, and deep-sequenced to an average depth of 6,000 reads per sample. After 161 read trimming, alignment, and filtering (see Methods), 76% of reads from saliva COV-ID reactions 162 were informative (Fig. S2C). In order to differentiate SARS-CoV-2 positive and negative samples, 163 we calculated the ratio between N2 reads and reads mapping to the human STATH control. Using 164 the highest N2/STATH read ratio in control (SARS-CoV-2 negative saliva) as a threshold, 95% 165 (19/20) of samples with spiked-in virus were correctly classified as positives (Fig. 2D). Using 166 COV-ID, we consistently detected SARS-CoV-2 in saliva samples containing as low as 5 virions 167 per µL, a sensitivity comparable and in some cases superior to those of established testing 168 protocols 32 . 169 Scaling COV-ID to handle higher sample numbers requires pooling samples immediately 170 following RT-LAMP, prior to the PCR step (Fig. 1A). We designed 32 unique 5-nucleotide 171 barcodes for several target LAMP amplicons ( Fig. S2D and Table S2). We first individually 172 validated each barcode and primer combination by real-time fluorescence and PCR efficiency. 173 Certain barcodes inhibited the RT-LAMP reaction, possibly due to internal micro-homology and 174 primer self-hybridization 33 . Nonetheless, out of 32 barcodes tested in 3 separate  reactions (N2, ACTB, and STATH), 25 successfully amplified all three target RNAs (Fig. S2D). 176 Saliva samples spiked with various concentrations of inactivated SARS-CoV-2 were amplified via 177 barcoded RT-LAMP, then optionally pooled prior to PCR and sequencing (Fig. S2E).  2/STATH ratios demonstrated no loss of sensitivity or specificity in the pooled samples compared 179 to the individual PCRs. 180 To test the potential of COV-ID on patient samples, we tested saliva specimens, collected and 181 previously analyzed at the Hospital of the University of Pennsylvania (see Methods). We carried 182 out multiplex barcoded RT-LAMPs on each sample (COV-ID step I, Fig. 1B), pooled the reactions 183 and then constructed libraries via PCR (COV-ID step II, Fig. 1D). After deep sequencing, analysis 184 of N2/STATH ratios showed 100% (8/8) concordance with viral copy numbers generated by a 185 standard clinical test (RNA purification followed by RT-qPCR) (Fig. 2E), demonstrating the 186 effectiveness of the COV-ID approach. 187 Taken together, our data show that COV-ID can be utilized to detect viral and human amplicons 188 in multiplex directly from saliva. The samples that can be batch amplified and deconvoluted after 189 deep sequencing. 190 . 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 April 23, 2021. ; 7

Calibration of COV-ID using an artificial spike-in 191
Existing deep sequencing approaches for massively parallel COVID-19 testing based on  incorporate artificial spike-ins, which serve as an internal calibration controls and allow for better 193 estimates of viral loads by end-point PCR 7, 8 . At the same time, adding to the reactions an artificial 194 substrate for amplification helps minimizing spurious signals as it can "scavenge" viral 195 amplification primers in negative samples. Finally, by providing a baseline amplification even in 196 empty samples, a properly designed spike-in strategy can reduce variance in total amounts of 197 final amplified products across samples, which compresses the dynamic-range of sequence 198 coverage for each patient in a complex pool and, therefore, reduces the risk of inconclusive 199 samples due to low sequencing coverage 8 . 200 We reasoned that a spike-in approach for LAMP-based quantification would provide similar 201 benefits in the context of COV-ID. To generate a SARS-CoV-2 spike-in, we synthesized a 202 fragment of the N2 RNA that retained all primer-binding regions for RT-LAMP and contained a 203 divergent 7-nt stretch of sequence to distinguish reads originating from the spike-in from those 204 originating from the natural virus (Fig. S3A). After confirming that the spike-in template was 205 efficiently amplified via RT-LAMP with the N2 primer set (Fig. S3B), we performed pooled COV-206 ID on virus-containing saliva in the presence of 20 fg of N2 spike-in RNA. As expected 8 , addition 207 of a constant amount of viral spike-in across reactions reduced the variability in total read numbers 208 for individual samples in the final pool (Fig. S3C). As discussed above, a narrower range in 209 sequencing output across samples in a pool optimizes the utilization of sequencing reads, and 210 ultimately lowers the cost per sample. Because the spike-in provides an internal calibration that 211 is independent of the RNA quality found in saliva, in several cases normalization against the 212 spike-in resulted in lower levels of false positive signal from negative samples (Fig. S3D). This is 213 likely because in cases where very few STATH reads were obtained, possibly due to degradation 214 of host RNA in the saliva sample, the resulting small denominator inflated the N2/STATH ratio 215 even for SARS-CoV-2 signal that was low in absolute terms and likely spurious. 216 Thus, these data show that spike-in strategies are compatible with the COV-ID workflow and 217 provide a means to stabilize total amplification and read allocation per sample while also offering 218 an additional calibration control to better estimate the viral load in samples where the endogenous 219 STATH mRNA might be below detection due to improper collection or handling. 220 221 222 . 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021

Simultaneous detection of SARS-CoV-2 and influenza A by COV-ID 223
Given the challenge of distinguishing early symptoms of COVID-19 from other respiratory 224 infections, we evaluated COV-ID for the simultaneous detection of more than one viral pathogen. 225 Multiple distinct products can be simultaneously amplified by RT-LAMP in the same tube by 226 providing the appropriate primer sets in multiplex, as we demonstrated above by co-amplifying 227 N2 and STATH in the same COV-ID reaction (see Fig. 2). In fact, simultaneous detection of 228 SARS-CoV-2 and influenza virus by RT-LAMP was previously achieved, albeit in a fluorescent-229 based, low-throughput type of assay 34 . We reasoned that the sequencing-based readout of COV-230 ID would allow extending this approach to the simultaneous detection of multiple pathogens as 231 well as endogenous (host mRNA) and artificial (spike-in) calibration standards, all in a single 232 reaction. 233 To test the ability of COV-ID to simultaneously detect multiple viral templates, we selected and 234 validated a generic "flu" RT-LAMP primer set that recognizes several strains, including influenza 235 A virus (IAV) and influenza B 34, 35 , and modified the BIP and FIP sequence to introduce the COV-236 ID barcodes and handles for PCR ( Fig. S2D and Table S1). We added inactivated SARS-CoV-2 237 virus (BEI resources) and IAV strain H1N1 RNA (Twist Biosciences) to saliva according to a 3 x 238 4 matrix of (10 4 , 10 3 , or 0 copies per µL) SARS-CoV-2 RNA against H1N1 RNA (10 5 , 10 4 , 10 3 , or 239 0 copies per µL) (Fig. 3A), as well as the N2 spike-in control. We performed multiplex COV-ID on 240 these samples using primers sets for STATH, N2 (to detect SARS-CoV-2), and IAV (to detect 241 H1N1) and sequenced to an average depth of 21,000 reads per sample. Both H1N1 and SARS-242 CoV-2 were detected above background and the signal correlated with the amount of the 243 respective template added to saliva ( Fig. 3B-C). Overall, multiplex COV-ID correctly identified 244 samples that contained only SARS-CoV-2 (7/8) or H1N1 (6/8). For samples that contained both 245 pathogens we observed reduced sensitivity (11/16 identification of both pathogens), which was 246 also observed in a previous multiplexing attempt 34 . However, in practice individuals who are 247 simultaneously infected with both viruses presumably would be rare, and for these cases the 248 ability to detect at least one virus successfully would allow to follow up with further diagnostic 249 testing. We found that of the samples containing both viruses, 16/16 showed positive detection of 250 at least one pathogen (SARS-CoV-2 or H1N1), suggesting the reduced sensitivity of the multiplex 251 assay is due to interference between amplification of both viral templates. This also demonstrates 252 that COV-ID can be used as an effective screening approach for multiple viral templates. 253 254 . 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 Paper-based saliva sampling for COV-ID 255 As an additional step toward increasing the throughput of the COV-ID approach, we explored 256 avenues to simplify collection, lower costs, and expedite processing time. Absorbent paper is an 257 attractive alternative to sample vials for collection, given its low cost, wide availability, and smaller 258 environmental footprint. In fact, paper has been used as a means to isolate nucleic acid from 259 biological samples for direct RT-PCR testing 36 as well as 38 . 260 We sought to determine whether the COV-ID workflow would be compatible with saliva collection 261 on absorbent paper. First, we immersed a small square of Whatman filter paper into water 262 containing various dilutions of inactivated SARS-CoV-2. After 2 min, the paper was removed and 263 transferred to PCR strip tubes followed by heating at 95ºC for 5 minutes to air-dry the sample 264 ( Fig. 4A). Next, we added the RT-LAMP mix containing the N2 COV-ID primer set directly to the 265 tubes containing the paper squares and let the reaction proceed in the usual conditions. COV-ID 266 PCR products of the correct size were evident in all samples containing viral RNA, with sensitivity 267 of at least 100 virions / μL (Fig. 4B) and in none of the controls, demonstrating that the presence 268 of paper does not interfere with the RT-LAMP reaction and subsequent PCR amplification with 269

Illumina adapters. 270
To assay direct COV-ID detection from saliva on paper, we saturated Whatman filter paper 271 squares with saliva containing different amounts of inactivated SARS-CoV-2 virus, which, we 272 reasoned, would be equivalent to a patient collecting their own saliva by chewing on a small piece 273 of absorbent paper. Next, we placed the paper squares into reaction tubes containing 274 TCEP/EDTA inactivation buffer (see Methods) similar to that used for the in-solution samples 275 used in our previous experiments (see Fig. 1A). We dried the paper at 95ºC and performed RT-276 LAMP followed by PCR (Fig. 4C), which resulted in the appearance of COV-ID products of the 277 correct size starting from saliva spiked with as few as 50 virions / μL (Fig. 4D). We then performed 278 COV-ID sequencing on saliva collected on paper using primers N2 and STATH in the presence 279 of the N2 spike-in RNA. The sequence data showed more variability and limited coverage of the 280 control amplicons compared to in-solution COV-ID likely due to the more challenging reaction 281 conditions; therefore, we normalized viral reads using both STATH and spike-in. This paper-282 based COV-ID proof-of-principle experiment detected the presence of viral RNA in samples with 283 as little as 320 copies / µL (Fig. 4E), a lower sensitivity compared to that of in-solution COV-ID 284 but still well within the useful range 39 to detect infections. 285 Taken together, these data show that the RT-LAMP step of COV-ID is compatible with the 286 presence of paper in the reaction tube and suggest that self-collection of saliva by patients directly 287 . 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) 10 on absorbent paper could provide a simple and cost-effective strategy to collect and test 288 thousands of saliva samples for multiple pathogens (Fig. 4F). 289

DISCUSSION 290
Testing strategies are vital to an effective public health response to the COVID-19 pandemic, 291 particularly with the spread of the disease by asymptomatic individuals. An ongoing challenge to 292 COVID-19 testing is the need for massive testing strategies for population-level surveillance that 293 are needed for efficient contact tracing and isolation. Most FDA-approved clinical SARS-CoV-2 294 diagnostic tests are based on time-consuming and expensive protocols that include RNA 295 purifications and RT-PCR 32 and must be performed by trained personnel in well-equipped 296 laboratories. Point-of-care antigen tests provide a much faster turnaround time and require little 297 manipulation, but there remains limited data on their specificity in real-world applications 40 . 298 Because of reagent limitations and diagnostic testing bottlenecks, prioritization of COVID 299 diagnostic testing continues to be for symptomatic individuals and individuals who are particularly 300 vulnerable for infection after exposure 41 . Private organizations, including colleges and 301 universities, have circumvented some of these challenges by contracting with private laboratories 302 to establish asymptomatic surveillance testing protocols; this is a costly option for population-level 303 surveilling of asymptomatic SARS-CoV-2 infections. 304 Several effective COVID-19 vaccines have been developed and there is a concerted ongoing 305 global vaccination effort, providing a concrete means to end the pandemic. Despite this progress 306 there are several potential risks that require vigilance: possible COVID-19 transmission in 307 vaccinated individuals, emergence of vaccine-resistant viral variants, and public skepticism of 308 vaccines or faltering compliance with social distancing guidelines 42 . For these reasons ongoing 309 testing and surveillance efforts will remain important for the foreseeable future, both to monitor 310 the progress of vaccination in reducing symptomatic cases and to detect emerging variants. 311 In order to scale testing to an effective volume and frequency, surveillance tests must 312 demonstrate the following qualities: 1) sensitivity, to identify both asymptomatic and symptomatic 313 carriers; 2) simplicity in methodology, to be performed in a number of traditional diagnostic 314 laboratories, without specialized equipment; 3) low cost and easily accessible reagents; 4) ease 315 of collection method; 5) rapid turnaround time to allow for isolation and contract tracing; and 6) 316 ability to co-detect multiple respiratory viruses, given the overlap in patient symptoms. To this 317 end, we have developed COV-ID, an RT-LAMP-based parallel sequencing SARS-CoV-2 318 . 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 April 23, 2021. 11 detection method that can provide results from tens of thousands of samples per day at relatively 319 low cost to simultaneously detect multiple respiratory viruses. 320 COV-ID features several key innovations that make it well-suited to high-throughput testing. First, 321 COV-ID uses a two-dimensional barcoding strategy 8 , where the same 96 barcodes are used in 322 each RT-LAMP plate, making it possible to pre-aliquot barcodes in 96-well plates ahead of time 323 and store them at -20ºC, simplifying execution of the assay and shortening turnaround times. 324 Second, since RT-LAMP does not require thermal cycling, tens of thousands of samples can be 325 run simultaneously in a standard benchtop-sized incubator or hybridization oven held at 65ºC. 326 Third, individual samples are pooled immediately following RT-LAMP; therefore, a single 327 thermocycler has the potential to process up to 96 or 384 RT-LAMP plates, generating 9,216 or 328 36,864 individually barcoded samples, respectively ( Fig. 1A, 4F, 4G). Only 96 unique FIP 329 barcodes are required for this scaling; here, we show that 28 out of 32 LAMP barcodes tested 330 were functional for both N2 and STATH. This proof-of-principle experiment demonstrates the 331 feasibility of generating the library of barcodes required to apply COV-ID to a large population. An 332 additional advantage of sequencing-based approaches, such as COV-ID is that with carefully 333 designed primers it would be possible to recover information about viral variants directly from the 334 sequencing reads 43 . Finally, COV-ID can generate ready-to-sequence libraries directly from saliva 335 absorbed onto filter paper, which would allow for major streamlining of the often-challenging 336 logistical process of sample collection (Fig. 4). Thus, COV-ID libraries for thousands and tens of 337 thousands of samples can be generated with relatively minimum effort in biological laboratories 338 with basic equipment and easily accessible reagents. 339 With the average throughput of an Illumina NextSeq 500/550, a relatively affordable next-340 generation sequencer up to 9,216 (96 RT-LAMPs x 96 pools) can be sequenced at a depth of 341 ~48,000 reads per sample, and up to 36,864 (96 RT-LAMPs x 384 pools) can be sequenced at a 342 depth of ~12,000 reads, which, we showed, is more than sufficient to obtain information about 343 multiple viral and control amplicons. Considering that reagents for one NextSeq run cost ~1,500 344 U.S. dollars, the theoretical sequencing cost per sample could be as low as $0.04 (Fig. 4G). While 345 sequencing instruments are relatively specialized and not ubiquitous, amplified COV-ID DNA 346 libraries could be shipped to remote facilities for sequencing in a cost-effective manner as 347 previously proposed by the inventors of LAMP-seq 14 . Finally, because of the limited sequence 348 space against which reads must be aligned, computational analysis of the resulting data can be 349 performed in a matter of minutes with optimized pipelines, providing results shortly after the 350 sequencing run has completed. 351 . 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 April 23, 2021. ; 12 COV-ID has sensitivity of 5-10 virions of SARS-CoV-2 per μL in contrived saliva samples (Fig. 352 2D) and at least 300 virions / μL in saliva collected from patients in a clinical setting (Fig. 2E). 353 However, this was sufficient to properly classify 100% of the clinical samples analyzed, given that 354 all positive samples had an estimated viral load > 300 virions / µL. Importantly, this was also the 355 apparent limit of sensitivity of paper-based COV-ID (Fig. 4E), suggesting that even in these 356 settings COV-ID would be capable of accurately classifying the majority of patient-derived 357 samples. 358 In conclusion, COV-ID is a flexible platform that can be executed at varying levels of scale with 359 additional flexibility in sample input, making it an attractive platform for surveillance testing. 360 Population-level monitoring of SARS-CoV-2 infections will be critical while vaccines are being 361 distributed to the global population, and continued surveillance will likely remain an effective 362 strategy to protect immune-compromised and unvaccinated members in society and within 363 entities and organizations where regular monitoring is critical to social isolation strategies. To that 364 end, effective, low-cost, multiplexed, and readily-implementable strategies for surveillance 365 testing, such as COV-ID, are important to mitigate the effects of the current and future pandemics. 366

RT-LAMP primer design 368
Primers against ACTB were designed using PrimerExplorerV5 (https://primerexplorer.jp/e/) using 369 default parameters and including loop primers (Table S1). 370 For COV-ID, priming sequences for PCR were inserted in FIP and BIP primers between the target 371 homology regions (F1c and F2, and B1c and B2, respectively, see Fig. S1). After testing, we 372 determined that 12 nts and 11 nts were most effective for the P5 and P7 binding regions, 373 respectively, being the shortest insertion that allowed reliable PCR amplification from LAMP 374 products without impacting LAMP efficiency. In addition a 5 nt barcode sequence was inserted at 375 the immediate 3' end of the P5-binding region of the FIP primer. 376

LAMP barcode design 377
Starting from the total possible 1,024 unique 5-nt barcodes, we removed those that matched any 378 sequence within the RT-LAMP primers used in this study (Table S1) in either sense or anti-sense 379 orientation. From the remaining pool, we selected 32 barcodes with hamming distance of at least 380 2 between all candidates. We tested FIPs incorporating candidate barcodes for ACTB, STATH, 381 . 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 April 23, 2021. ; https://doi.org/10.1101/2021.04.23.21255523 doi: medRxiv preprint 13 N2, and IAV primer sets on saliva RT-LAMP with 1,000 copies target amplicon. Primers that failed 382 to show LAMP signal by real time fluorescence monitoring or generate expected PCR product 383 were discarded. Final usable barcodes are provided in (Table S2). 384

RT-LAMP 399
All RT-LAMP reactions were set up in clean laminar flow hoods and all steps before and after 400 LAMP were carried out in separate lab spaces to avoid contamination. RT-LAMP reactions were 401 set up on ice as follow: for each amplicon 5 or 6 LAMP primers were combined into 10x working 402 stock at established concentrations: 16 μM FIP, 16 μM BIP, 4 μM LF, 4 μM LB, 2 μM F3, 2 μM 403 B3. For multiplexed COV-ID reactions 10x working primer mixes for each amplicon were either 404 added proportionally so that the total primer content remained constant, or mixed so that BIP and 405 FIP primers were scaled down depending on amplicon number while remaining primers (LF 406 and/or LB, F3, B3) were kept at same concentration as in single reactions. 407 Each 10 μL RT-LAMP reaction mix consisted of 1x Warmstart LAMP 2x Master Mix (NEB Cat. 408 E1700S), 0.7 μM dUTP (Promega Cat. U1191), 1 μM SYTO-9 (Thermo Cat. S34854), 0.1 μL 409 Thermolabile UDG (Enzymatics Cat. G5020L), 1 μL of saliva template and optionally 20 fg of N2 410 Spike RNA. Reactions were prepared in qPCR plates or 8-well strip tubes, sealed, vortexed and 411 centrifuged briefly, then incubated in either a QuantStudio Flex 7 or StepOnePlus instrument 412 . 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 April 23, 2021. ; https://doi.org/10.1101/2021.04.23.21255523 doi: medRxiv preprint (Thermo) for 65ºC 1 hr. Real-time fluorescence measurements were recorded every 30 sec to 413 monitor reaction progress but were not used for data analysis. Following LAMP the reactions were 414 heated at 95ºC 5 min to inactivate LAMP enzymes. 415

Library construction by PCR amplification 416
All post-LAMP steps were carried out on a clean bench separate from LAMP reagents and 417 workspace. For individual LAMP samples, LAMP amplicons were diluted either 1:100 or 1:1,000 418 in water. For pooling of individually barcoded LAMP reactions, equal amounts of all LAMP 419 reactions were combined and then either diluted 1:1000 or purified via SPRIselect beads 420 (Beckman Coulter Cat. B23317) using a bead-to-reaction ratio of 0.1x. Purified material was 421 diluted to final 100-fold dilution relative to LAMP. 422 1 μL of diluted LAMP material was used as a template for PCR using OneTaq DNA polymerase 423 (NEB Cat. M0480L) with 100 nM each of custom dual-indexed Illumina P5 and P7 primers in 424 either 10 or 25 μL reaction (Table S1). PCR reactions were incubated as follows: (25 cycles  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 April 23, 2021. ; 15 coordinator, patients were instructed to self-collect saliva into a sterile specimen container which 444 was then placed on ice until further processing for analysis. 445 The saliva used in the remaining experiments was donated by one of the authors. Because it was 446 only used for protocol optimization the Penn IRB has determined that it did not constitute human 447 subjects research and therefore approval was not required. 448

Paper COV-ID 449
Squares of Whatman no. 1 filter paper (2 mm x 2 mm) were cut using a scalpel on a clean surface 450 under a laminar flow hood and stored at room temperature until used. Using ethanol-sterilized 451 fine-nosed tweezers a single square was dipped twice into unprocessed, freshly collected saliva 452 with or without added SARS-CoV-2 (BEI Resources Cat. NR-52286) until saliva was saturated on 453 paper by eye. Paper was then transferred to well of 96-well plate containing 10 ul of 1x 454 TCEP/EDTA buffer (2.5 mM TCEP, 1 mM EDTA, 1.15 NaOH). Plate was placed on heat block 455 inside laminar flow hood or inside open thermocycler and incubated at 95ºC x 10 min. 456 10 ul RT-LAMP mixture was prepared as described above in the absence of the N2 Spike RNA. 457 10 ul of RT-LAMP reaction mixture was added to each paper strip, then plate was sealed and 458 incubated 65ºC x 1 hr, 95ºC x 5 min in QuantStudio Flex 7 (Thermo). 1 ul of each reaction was 459 either diluted 1:100 or purified via SPRIselect beads and PCR amplified as described above. 460

Sequencing 461
Libraries were sequenced on one of the following Illumina instruments: MiSeq, NextSeq 500, 462 NextSeq 550, NovaSeq 6000 and sequenced using single end programs with a minimum of 40 463 cycles on Read 1 and 8 cycles for index 1 (on P7) and index 2 (on P5). 464

Sequence Analysis 465
Reads were filtered for optical quality using FASTX-toolkit utility fastq_quality_filter 466 sequences using bowtie2 46 with options --no-unal and --end-to-end. Alignments with greater than 472 1 mismatch were removed and the number of reads mapping to each target for all barcodes were 473 extracted and output in a matrix. Barcodes with fewer than 25 total mapped reads were discarded. 474 . 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 April 23, 2021. . 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 April 23, 2021. . 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 April 23, 2021. . 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 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021    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 April 23, 2021. ;

22
(E) SARS-CoV-2 virus was added to saliva and prepared as in (C). RT-LAMP and sequencing 650 was carried out in presence of SARS spike-in RNA. Viral reads are presented as ratio against the 651 sum of STATH and N2 spike-in reads. Positive threshold was set as 2x maximum value in 652 negative saliva and indicated by dashed horizontal line. 653 (F-G): Paper-based COV-ID workflow (F) and cost calculations (G). Saliva is collected orally on 654 a precut strip of paper, from which a 2 mm square would be cut out and added to a reaction vessel 655 containing TCEP/EDTA inactivation buffer and processed as shown in (C). 656 657 658 . 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 April 23, 2021. ;

Figure S1. Detailed COV-ID mechanism 660
Steps of COV-ID protocol are depicted, showing RT-LAMP mechanism and ultimate amplicon 661 that is sequenced. For clarity only selected steps of RT-LAMP reaction are shown and loop primer 662 intermediates are not depicted. For full LAMP mechanism see 21 . 663 (E) COV-ID primers targeting ACTB mRNA were used for RT-LAMP with HeLa total RNA. LAMP was diluted 1:100, amplified via PCR and resolved on 2% agarose gel. iii. v.
COV-ID step 2: library construction by PCR . 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021  (B) RT-LAMP followed by COV-ID PCR performed directly on saliva. Saliva with and without addition of 1,000 copies of inactivated SARS-COV-2 templates was inactivated as described in (A), then used as template.
(C) Alignment of sequenced reads against SARS-COV-2 genome from COV-ID of inactivated saliva spiked with without 1,280 virions SARS-COV-2 per µL. All SARS-COV-2 reads align exclusively to expected region of the N gene. Open reading frames of viral genome are depicted via gray boxes below alignment. Inset: scale shows reads per 1,000.
(D) Scatter plot for the ratio of SARS-CoV-2 / (STATH + 1) reads obtained by COV-ID (y axis) versus the number of virions per µL spiked in human saliva (x axis). The threshold was set above the highest values scored in a negative control (dashed line).
(E) COV-ID performed on clinical saliva samples. The scatter plot shows the SARS-CoV-2 / (STATH + 1) read ratio (y axis) versus the viral load in the sample estimated by a clinically approved, qPCR-based diagnostic test. The threshold was set based on the negative controls shown in (D).  . 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021 Reads ( . 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 April 23, 2021. ; (B) PCR reactions from paper samples immersed in water with indicated viral concentrations then amplified with N2 COV-ID primers.
(C) Scheme for COV-ID on saliva spiked with viral and RNA and absorbed on paper.
(D) Same as (B) but on saliva absorbed on paper.
(E) SARS-CoV-2 virus was added to saliva and prepared as in (C). RT-LAMP and sequencing was carried out in presence of SARS spike-in RNA. Viral reads are presented as ratio against the sum of STATH and N2 spike-in reads. Positive threshold was set as 2x maximum value in negative saliva and indicated by dashed horizontal line.
(F-G): Paper-based COV-ID workflow (F) and cost calculations (G). Saliva is collected orally on a precut strip of paper, from which a 2 mm square would be cut out and added to a reaction vessel containing TCEP/EDTA inactivation buffer and processed as shown 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.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 23, 2021. ; https://doi.org/10.1101/2021.04.23.21255523 doi: medRxiv preprint Figure S1. Detailed COV-ID mechanism The steps of the COV-ID protocol are depicted, showing RT-LAMP mechanism and the final barcoded amplicon that is sequenced. For clarity only, selected steps of RT-LAMP reaction are shown and loop primer intermediates are not depicted. For full LAMP mechanism see (Nagamine et al., 2002).    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 April 23, 2021. ; https://doi.org/10.1101/2021.04.23.21255523 doi: medRxiv preprint Figure S2. Optimization of COV-ID in human saliva (A) Saliva COV-ID sequence validation. Single saliva COV-ID reaction using N2 primers was sequenced by the Sanger method.
(B) Validation of control human amplicons for RT-LAMP on saliva. RT-LAMP of TCEP/EDTA inactivated saliva was performed with conventional RT-LAMP primer sets for ACTB and STATH in the presence or absence of RNase A.
(C) Characterization of COV-ID sequencing libraries. Breakdown of reads for sequence data presented in Fig. 2D. Samples without added template consist of predominantly adapter dimers.
(D) Validation of COV-ID LAMP barcodes. 32 potential barcodes were tested for LAMP primer sets indicated, incompatible barcodes are marked in red.
(E) Validation of pooled PCR. COV-ID was performed on saliva samples using unique LAMP barcodes. The RT-LAMP reactions were then amplified either by individual PCR or by first pooling and then performing a single PCR on the pool.   CTCTTCCGATCTCCTGTGGAACTGATTACAAACATTGGCCGCAAATTGCACAATTTGCCCCCAGCGCTTCAGCGTTCTTCGGAATGTCGCGCATTGGCATGGAAGTCACATCTCCGAGC . 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 April 23, 2021. ; https://doi.org/10.1101/2021.04.23.21255523 doi: medRxiv preprint Figure S3. Spike-in strategy for COV-ID( A) Synthetic N2 Spike RNA. SARS-CoV-2 N2 RNA fragment was synthesized including 7 nt divergent sequence inside the forward loop primer-binding site, maintaining all other LAMP primer binding sites and identical GC content.
(B) RT-LAMP using COV-ID N2 primers was carried out on indicated amounts of spike-in RNA, showing rapid amplification down to picogram quantities of added template.
(C) Total number of reads per barcode in COV-ID pool obtained by including (+) or omitting (-) the N2 spike-in.
(D) Spurious COV-ID signal for the N2 amplicon in negative control samples after normalization either to the STATH control in absence of spike-in (left) or to the N2 spike-in control. . 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 April 23, 2021. ;https://doi.org/10.1101https://doi.org/10. /2021