Abstract
The progress of the COVID-19 pandemic profoundly impacts the health of communities around the world, with unique effects on colleges and universities. Here, we examined the prevalence of SARS-CoV-2 in 1808 asymptomatic individuals on a university campus in California, and compared for the first time the performance of CRISPR- and PCR-based assays for large-scale virus surveillance. Our study revealed that there were no COVID-19 cases in our study population in May/June of 2020. Using the same methods, we demonstrated a substantial shift in prevalence approximately one month later, which coincided with changes in community restrictions and public interactions. This increase in prevalence, in a young and asymptomatic population, indicated the leading wave of a local outbreak, and reflected the rising case counts in the surrounding county. Our results substantiate that large, population-level asymptomatic screening using CRISPR- or PCR-based assays is a feasible and instructive aspect of the public health approach within large campus communities.
Background
Nine months after the first cases of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) were reported in Wuhan, China, the virus has infected more than 25 million people globally and has caused the death of over eight hundred thousand people. The COVID-19 acute respiratory syndrome has brought healthcare systems to the brink of collapse and has disrupted the way of life of countless communities. In an effort to control this pandemic, communities around the world closed businesses, prohibited large social gatherings and adopted non-pharmacological intervention (NPI) measures1-3. Initial restrictions proved to be successful in controlling the epidemic in several countries where new COVID-19 cases, hospitalizations and deaths have declined2,3. However, many other communities relaxed social distancing and restrictions, which led to a resurgence, often following exponential growth. Several metrics, including percent positivity of testing, hospitalizations, and death rates have been used to gain insights into epidemic trends in specific populations. Prevalence among asymptomatic persons has been an important but more elusive metric, largely because of test scarcity and the prioritization of test resources to symptomatic patients or contacts with confirmed cases. Nevertheless, understanding both asymptomatic prevalence and the association between changes in NPI and prevalence has tremendous potential to inform vital decisions that public health leaders and policy makers face in the months ahead.
A fundamental aspect of pandemic control is the careful planning for the return to college campuses with the start of the impending 2020-2021 academic year. While the COVID-19 testing has focused on older, more medically fragile patients with increased mortality risk, an increasing burden of disease has emerged in those aged 19-30, many of whom attend college and university4. Every year since 2017, over 15 million students attend colleges in the US5. Many students reside in dormitories and off-campus housing, frequently in crowded conditions. Most of them will share restrooms, kitchens and common areas6. These living conditions are associated with high morbidities of diseases like meningococcal meningitis, influenza, mumps and measles, among others7-10. Respiratory pathogens like SARS-CoV-2 are easily transmitted to a large number of individuals living in college dormitories and during social contact by exposure to live virus in aerosol droplets11-13. Further complicating transmission of SARS-CoV-2 in university settings is the well-documented infectivity of asymptomatic persons, many of whom are likely to be pre-symptomatic with high viral loads14-20. Those without symptoms, who are either asymptomatic or pre-symptomatic, are thought to be responsible for as many as 44% of new infections21. Identifying these patients early through expanded testing is crucial to mitigate the spread of disease within a community. However, before undertaking large-scale surveillance testing, the prevalence of asymptomatic infection must be ascertained to inform decisions regarding the utility of expanded testing in a university population22.
To better understand the prevalence, model the trajectory, and assess the potential of a Cas13-based test to screen for SARS-CoV2 in asymptomatic persons in a university community, we enrolled healthy volunteers in a virus surveillance study. Oropharyngeal (OP) samples were obtained by self-collection and processed for parallel SARS-CoV2 testing using a newly-developed CRISPR-based assay23 and the US Centers for Disease Control and Prevention (CDC)-recommended RT-qPCR assay24. We compared the results obtained from two time periods. The first collection period occurred during May and June, approximately 2 months into a state-wide stay-at-home mandate. The second collection period occurred during late June and early July, approximately 3 weeks after local restrictions for isolation were removed and the community was allowed to enter the third stage of a planned four-stage reopening. The objectives of our study were to: (i) establish the prevalence of asymptomatic SARS-CoV2 infection in a university population, (ii) assess for any dynamic change associated with the changing community conditions related to NPIs, and (iii) systematically compare the performance of our newly developed CRISPR-based test alongside that of the established, CDC-recommended reference testing by RT-qPCR. Together, our results substantiate the utility of these assays for large-scale surveillance sampling of SARS-CoV-2 in asymptomatic individuals.
Methods
Study population
The population of UCSB includes 26,134 enrolled graduate and undergraduate students and 5,668 staff and faculty. About 38% of the students live in university housing, and 34% live in the nearby community of Isla Vista (23,096 residents, 1.866 mi2, 12,377 people/mi2). Registration for this study was open to all symptom-free individuals who are 18 years of age or older, affiliated with UCSB (student, faculty, staff, direct relatives), or who work/live in Isla Vista. Individuals who exhibited a fever (100.4°F), cough or shortness of breath in the two weeks prior to, or on the day of sample collection, were excluded from the study. From all participants in the study, only five subjects were excluded due to presenting symptoms at the time of collection, and those individuals were referred to local healthcare resources.
Sample collection
Informed consent and demographic data (age, address, telephone, gender and UCSB affiliation) were collected at the specimen sampling locale by a healthcare professional. Samples were assigned a numeric code for deidentification purposes. Samples were acquired as self-collected OP swabs stored in PBS. For the specimen collection, the participants were instructed to swab the tonsillar area 10 times while being monitored by a healthcare worker. Upon receipt in the laboratory, the samples were inactivated at 56 °C for 30 minutes. RNA was extracted from all samples within 24 hours of collection using the QIAamp MinElute Virus Spin Kit (Qiagen cat# 57704) or QIAamp Viral RNA Mini Kit (Qiagen cat# 52906) from 140-200 μL of sample, and eluted in 50 μL. The rest of the sample was stored frozen at -80 °C. All pre- and post-analytical protocols were reviewed and approved by the IRB.
SARS-CoV-2 detection by RT-qPCR (one-step Taqman assay)
Viral RNA was reverse transcribed into cDNA and amplified using the TaqPath one-step cDNA master mix kit (ThermoFisher Cat# 501148245) following the manufacturer’s recommendations. Briefly, a master mix reaction was prepared using the established CDC protocol25, and 15 μL of the master mix were dispensed into a 96-well plate before addition of input RNA. 5 μL of each RNA target dilution was added into the wells containing the corresponding TaqMan primers and probes. For no template controls, 5 μL of nuclease free water were used in place of template RNA and in vitro transcribed RNAs encoding the N1 and N2 sites in the SARS-CoV-2 nucleocapsid sequence were used as positive controls (106 copies/μL). The plate was sealed with plastic self-adhesive film and the well contents were collected at the bottom by centrifugation. The reactions were run on a real time PCR instrument (BioRad CFX96 Touch) using the following thermal profile: 25 °C for 2 minutes; 50 °C for 15 minutes; 45 cycles of 95 °C for 5 seconds followed by 55 °C for 30 seconds and plate read; hold at 4 °C. All data were analyzed using the BioRad CFX Maestro software using a single threshold for Cq determination. To determine the limit of detection, standard curves of in vitro transcribed RNAs, ranging from 106 to 100 copies/μL, were prepared in nuclease free water and a one-way ANOVA with a post-hoc Dunnett’s test was used to determine the Cq value significance from water no template control using Prism v8 software (Graphpad). For N1 and N2 the limit of detection corresponded to 102 copies/μL (Cq of 32.59 and 34.405, respectively) and for RNase P 103 copies/μL (Cq = 34.328).
CREST (Cas13-based, Rugged, Equitable, Scalable Testing)
Purified RNA was reverse transcribed using RevertAid reverse transcriptase (200 U/μL, Thermo Fisher Scientific) in presence of murine RNase inhibitor (New England Biolabs). Reaction components were mixed in 0.2 mL tubes as follows: 5 μL RNA template, 1 μL gene specific primer mix (5 μM each primer), 2 μL 5X reaction buffer, 1μL dNTPs (10 μM each), 0.5 μL RevertAid reverse transcriptase, 0.5μL murine RNase inhibitor (40 U/ μL). All components were mixed gently by pipetting and the tube contents were collected by centrifugation. Negative control reactions were composed of water instead of RNA input, and positive control reactions used 106 copies of in vitro transcribed RNA. The reaction mixtures were heated to 42 °C for 30 minutes in a heat block and placed on ice afterwards. 2 μL of the resulting cDNAs were used as input templates for PCR amplification using a Taq DNA polymerase master mix (New England Biolabs). PCR reactions followed the following thermal profile: 98 °C for 2 minutes; 20 cycles of 98 °C for 15 seconds, 60 °C for 15 seconds, and 72° C for 15 seconds; final extension at 72°C for 5 min. Cas13a was used for site-specific detection using fluorescent probes. The Cas13 reaction was performed in Cas13a cleavage buffer (40 mM Tris pH 7.5, 1 mM DTT) supplemented with 1 mM rNTPs (Thermo Fisher Scientific), 2 U/μL RNase Inhibitor (New England Biolabs), 0.125 μM cleavage reporter (Integrated DNA Technologies), 1.5 U/μL T7 RNA Polymerase (Lucigen), 6.3 ng/μL LwaCas13a, 20 nM Cas13 crRNA and 9 mM MgCl2 unless otherwise indicated. Reactions were composed of 4 μL Cas13a cleavage solution and 1 μL of sample (RT-PCR product) in a single well of a 384 well-plate, with samples run in duplicate or quadruplicate wells. Plates were sealed and fluorescence was acquired every 5 minutes for 30 minutes at 37 °C in a Quantstudio5 qPCR instrument (Applied Biosystems). An initial reading was taken at time = 0 and subtracted from time = 30 to get a DRFU for each well. To determine a threshold for negative and positive results, DRFU from negative control wells were multiplied by 5 and used as a cutoff. Plates were only considered valid if negative control reactions did not increase 3x over the time course of the experiment.
Primer, gRNA and cleavage reporter sequences
Analyses
Correlations between N1 and N2 and between CREST and Taqman assays were calculated using the Pearson correlation coefficient, assuming data are from a bivariate normal distribution, using the R function cor.test(). Percent positive rates (shown in figure 4) were fit using a logistic growth model where 
, with K = 100%, P = 0.03, and r fit by minimizing the error found to be r = 0.101.
Results
To obtain insights on the prevalence of SARS-CoV-2 in our local community, we enrolled 1,808 healthy volunteers in a surveillance study. All participants were asymptomatic for COVID-19 at the time of sample collection. Among the participants, 1,805 reported affiliation with the University of California Santa Barbara (UCSB), and 1,306 (72.2%) were undergraduate and graduate students (Sup. Fig. 1A). This population reflects the composition of the UCSB community which has 26,134 enrolled students (82%) and 5,668 faculty members and staff (18%) (Sup. Fig. 1A).
(A) Affiliations of the UCSB population. (B) Distribution of zip codes reported by the participants. The size of the bubble reflects the size of the population from that zip code. (C) Distribution of the age of participants by cohort.
We acquired self-collected OP swabs from the participants over two time periods, from May 28 to June 11 (Cohort 1), and from June 23 to July 2 (Cohort 2) (Fig. 1A). Over 70% of the subjects in both cohorts self-identified as UCSB students (71% cohort 1; 73% cohort 2) (Fig. 1B), 45-47% were male, 52-54% were female (Fig. 1C), and 67% of all participants reported the UCSB neighboring communities of Goleta and Isla Vista as their place of residence (Sup. Fig. 1B). The average age of our study population was 28.3 and 26.6 years old for cohort 1 and 2 respectively, with a minimum age of 18 years old and a maximum of 73-75 years old (Fig 1D and Sup. Fig. 1C).
(A) 1,808 persons were recruited in two cohorts during the months of May, June and July of 2020. The arrows indicate the days of sample collection. Samples were not collected during the UCSB summer break from June 11 to June 22. (B, C, D). UCSB affiliation, gender and age of study participants. (E) Flow chart of specimen handling (SBCPHD Santa Barbara Public Health Department). Only positive results, confirmed in a CLIA laboratory with diagnostic testing, were reported to the participants and the Santa Barbara County Public Health Department (SBCPHD).
We used two assays to detect SARS-CoV-2 genomes in the collected OP swab samples: a CRISPR-based method we recently developed at UCSB known as CREST23, and the RT-qPCR test recommended by the CDC24 (Sup. Fig. 2). Both methods detect the same two sites in the nucleocapsid gene, N1 and N2, and one site in the host RNase P transcript, which ensured consistency in our analyses. Samples were processed in-house with a turnaround time that ranged from 12-30 hours from the moment of collection. All samples collected in cohort 1 (n=732) were negative by both tests (Fig. 2A, C). In stark contrast, we detected eight positive samples by CREST and nine by RT-qPCR in cohort 2 (n= 1,076) (Fig. 2A, C). We found a good correlation in detecting the nucleocapsid gene using the N1 and N2 primers (RT-qPCR, Pearson correlation coefficient r = 0.566, Fig 2B) and probes (CREST, Pearson correlation coefficient r = 0.872, Fig. 2D). The average age of the positive subjects was 21.7 years old (19-30 years of age), all self-identified as UCSB students (Sup. Table 1). Of the nine positive samples detected by RT-qPCR, eight were independently confirmed by a CLIA-certified laboratory test and the results were reported to the participants and the SBCPHD (Fig. 1E). One sample was positive by RT-qPCR at the threshold of detection, which reflects a low viral copy number (estimated between 38-400 copies/μL, Sup. Table 1, Sup. Table 3). The same sample tested negative by both CREST and CLIA-confirmation. With this single possible exception, RT-qPCR and CREST results had an excellent correlation (Pearson correlation coefficient N1 r= 0.584, N2 r = 0.811) (Sup. Fig. 3).
(A) OP swabs were self-collected by participants and RNA was isolated from the samples. For CREST, RNA was reversed transcribed, and the resulting DNA was amplified by polymerase chain reaction (PCR) using primers for the N1, N2 and RNAse P target regions (see panel B). The amplified region of interest was transcribed in vitro and used as the template for detection by Cas13. The activation of Cas13 following target recognition by the guide RNA (gRNA) was measured using a fluorescent poly-U cleavage reporter. For qPCR, the RNA was reverse transcribed and detected by real time amplification. (B) SARS-CoV-2 genome regions targeted in this study.
Correlation of the signal detected for N1 (A) or N2 (B) in RT-qPCR and CREST. Grey open dots indicate negative samples, solid red dots indicate positive samples. The blue dot indicates one sample that was detected by RT-qPCR, but not confirmed by CREST or in a CLIA laboratory test. Pearson correlation coefficient N1 r = 0.584, N2 r = 0.811.
Our study took place during the transition between stage 2 (Low-risk workplaces) and stage 3 (higher-risk workplaces).
Age of participants, CREST fluorescence signal and RT-qPCR Cq values for each of the positive samples detected in the study. The negative control is the average of signal for all no template sample. The positive control is the average signal for all samples where the in vitro transcribed template was used.
(A) Distribution of the 1/Cq values by cohort, and (B) correlation of N1 and N2 signal detected by RT-qPCR. (C) Distribution of the fluorescence values by cohort, and (D) correlation of N1 and N2 signal by CREST. Grey open dots indicate negative samples, solid red dots indicate positive samples. The blue dot indicates one sample that was detected by RT-qPCR, but not confirmed by CREST or in a CLIA laboratory test. (Note the low level of N2 by CREST for this sample). Dashed line in panel A indicates limit of detection for RT-qPCR (N1 1/Cq 0.0306, N2 1/Cq 0.029).
Following CLIA-confirmation of the positive tests, the individuals were informed of their status and given recommendations by SBCH clinicians. Participants were also offered the opportunity to follow up with clinicians at the UCSB student health service (SHS). Six out of eight individuals provided an update of symptoms to SHS. Two persons reported no symptoms, two persons reported mild symptoms (nasal congestion, sore throat), and two persons reported classical COVID-19 symptoms (fatigue, anosmia). No participants reported fever as a symptom (Sup. Table 2).
Symptoms reported by COVID-19 positive participants. Following CLiA-confirmation of their COVID-19 status, positive individuals were contacted by SBCH clinicians and provided with recommendations to follow. They were also given the opportunity to follow up with the UCSB Student Health Service clinicians.
To estimate the viral load in the asymptomatic or presymptomatic subjects confirmed as positive, we calculated the genome equivalents per μL based on the Cq values for N1 and N2 from the RT-qPCR assay, using linear regression on a standard curve ranging from 100 to 106 gene copies/uL. Our study subjects’ viral load ranged from 286 to 510,000 copies/μL (Sup. Table 3). These viral load levels were not significantly different from those detected in a control set of de-identified nasopharyngeal (NP) swab samples obtained from symptomatic patients in the local community and provided to us by our collaborators at the Santa Barbara County Public Health Department SbCpHD (p = 0.95 for N1, p = 0.497 for N2, Mann-Whitney test) (Sup. Table 3, Fig. 3). Notably, the quality of the self-collected specimens using OP swabs was not significantly different from those collected using NP swabs (positive and negative controls), as measured by the detection of RNase P transcripts (p = 0.63, Kruskal-Wallis test) (Fig. 3).
Viral load (genome equivalents/μL), calculated for the positive samples in this study, and confirmed positive and negative samples from the community of Santa Barbara County.
Viral loads expressed as genome equivalents/μL (A), and RNase P copies (B) were calculated as the genome equivalents per μL using the RT-qPCR data from N1 or N2 from asymptomatic participants in this study (red, N = 8), or confirmed positive (purple, N = 6) and negative patients (green, N = 7). (A) Median, solid line. NS: p = 0.95 for N1, p = 0.497 for N2 Mann-Whitney test. (B) Median, solid line. NS: p = 0.63 for RNase P Kruskal-Wallis test.
Next, we calculated the prevalence of SARS-CoV-2 in our study population using the confirmed cases. The prevalence of cohort 1 was 0% while that of cohort 2 was 0.74%. The daily incidence for the latter ranged from 0 to 1.65% (Fig. 4, Sup. Table 4). The change in prevalence between cohorts was statistically significant (p = 0.013, Fisher exact test). The prevalence dynamics in our study population reflect the increase in COVID-19 cases diagnosed in Goleta and Isla Vista communities, where 67% of all the samples in our study were collected (Fig. 4). The increase in the number of infections detected in our study—and those in Santa Barbara County—coincided with the implementation of the stage three of the California re-opening plan in Santa Barbara County (https://covid19.ca.gov/roadmap/) (Fig. 4, Sup. Fig. 5).
Viral prevalence (% of cases per day) calculated for each collection day in cohorts 1 and 2 in the study. The prevalence varied from 0%-1.653% daily.
The daily viral prevalence was calculated as a % of positive cases (red dots). The trendline indicated by the red dotted line was calculated by finding the r in a logistic growth model that minimized the error while fixing the % prevalence at May 28th to 0.03%. The daily cumulative number of cases in Goleta and Isla Vista, based on official data from the SBCPHD, are shown as grey bars. The timing of the stage two and three of the California re-opening plan in Santa Barbara county are indicated below the graph.
Discussion
As the COVID-19 pandemic ravages the US and the world, it is necessary to increase our understanding of virus transmission dynamics to facilitate the development and implementation of effective mitigation measures. An essential, and until recently overlooked, aspect of the epidemiology of COVID-19 is the contribution of asymptomatic and presymptomatic individuals to the spread of disease. Recent studies indicate that asymptomatic subjects may represent up to 45% of all diagnosed COVID-19 cases19,26,27. This large reservoir population can serve as a vector, kindling for new outbreaks. Therefore, early detection of infection in individuals with mild or absent COVID-19 symptoms is vital to prevent covert outbreaks. The potential of stealth transmission is particularly significant for university communities where dormitories, off-campus residences, and social spaces with high occupation densities favor the rapid transmission of respiratory diseases6.
With the aforementioned considerations in mind, we evaluated the prevalence of SARS-CoV-2 in asymptomatic persons who are part of the UCSB population, including students, faculty, and staff. We procured supervised self-collected OP swabs from all participants, which minimized the risk of exposure for our health care personnel, and maximized sample acquisition efficiency and compliance with self-collection of OP swabs, as outlined in our protocols. Our data support that self-sampling by OP swab is dependable, and thus provides an alternative method for unsupervised or remotely supervised sample acquisition outside of a healthcare setting. OP and other methods of self-sampling could enable sample processing by mail, which can greatly enhance testing coverage28,29. Another advantage of self-collection is that the person can control the swabbing and stop if they experience extreme discomfort. Our study participants reported no pain, nausea, or vomiting induced from the collection, which indicates that OP swab self-collection is less demanding on personnel and on patients. Even though OP swabs have been reported to have a lower detection rate and diminished sensitivity than NP swabs, we found that the viral loads we observed in asymptomatic subjects (ranging from a few hundred to several hundred thousand genome equivalents/μL) were similar to those found in NP samples from symptomatic and previously diagnosed individuals30. Thus, our self-sampling protocol for OP swabs offers sensitivity that is on-par with that of NP swabs with the added benefit of ease of sample collection. As such, our analyses provide proof-of-concept for the use of OP swab self-sampling in viral surveillance studies.
OP swab samples allowed us to compare the efficacy and sensitivity of two alternative SARS-CoV-2 detection protocols. The first one, CREST, is our recently developed CRISPR-based strategy, which uses PCR amplification and Cas13 for the detection of viral genomes with a simple binary outcome. This is the first time that a CRISPR-based method is used for extensive SARS-CoV-2 surveillance in an asymptomatic population. The second method we used is the conventional, and CDC-approved RT-qPCR assay, which is considered the “gold standard”. Our results show that CREST is as efficient at detecting infections in asymptomatic subjects as RT-qPCR, with the added benefit of enabling a precise, easy to interpret, and dependable binary readout. This is not the case for RT-qPCR, which requires highly-trained personnel for the interpretation of non-binary data. CREST showed perfect concordance with the confirmation of positive cases using diagnostic tests in a CLIA certified laboratory (Pacific Diagnostics Laboratory), further attesting to its robustness. Because we designed CREST to be a low-cost and accessible method, it offers a much-sought alternative for communities where access to testing is difficult. In addition, CREST is scalable, enabling high throughput testing, and it uses laboratory generated or off-the-shelf commercially available reagents, thus eliminating the restriction of limiting supply chains. For these reasons, we surmise CREST can offer a solution for places where access to professional laboratories is restrictive and instances in which high volume of repetitive sampling is necessary, including the university setting.
A main challenge for university communities is the potential for covert infections promoted by social and academic gatherings, which are unavoidable in the context of a vibrant university campus. Recent evidence indicates that asymptomatic and presymptomatic individuals can unknowingly transmit the virus and thus can inadvertently fuel covert outbreaks19,26,27. Thus, the early detection of asymptomatic infections, particularly those with high SARS-CoV-2 loads like the ones detected in our analyses, which may underlie super-spreader events, is vital for the mitigation of viral transmission. Moreover, the early identification of these individuals allows containment and contact tracing measures that could guide university directives to make decisions regarding the opening of campuses across the country. This information is essential to ensure the continuity of superior education.
One of our most striking and significant observations is the difference in SARS-CoV-2 prevalence between the two cohorts we analyzed. While we did not detect any SARS-CoV-2 infections in the 732 people tested in May/June, approximately one month later, we demonstrated a substantial shift in prevalence, with eight confirmed cases among 1,076 people surveyed. This significant change in the transmission dynamics coincided with the release of community restrictions and increased public and social interactions due to the implementation of stage three of the California re-opening plan in Santa Barbara County. The increase in prevalence was exclusive to young and asymptomatic individuals (average age 21.7 years old, range 19-30 years old) who may not otherwise have accessed COVID-19 testing. Individuals in this age group are likely to be socially active, which highlights how easily covert infections could result in flare-ups. Our surveillance program detected the initial wave from a local outbreak and coincided with rising case counts in the Goleta and Isla Vista localities, as well as in the Santa Barbara County and the State of California.
Overall, our study provides strong evidence supporting the use of self-collected OP swabs and CRISPR-and PCR-based assays as feasible, rapid and dependable tools for the surveillance of SARS-CoV-2 in asymptomatic individuals. The perfect concordance between gold standard testing (using NP swabs and RT-qPCR) and our strategy of using OP swabs and CREST substantiates the feasibility of using simpler, equally robust approaches for high-volume, recurrent testing, which is desirable in a university setting. Monitoring the population to detect COVID-19 cases before they lead to outbreaks could constitute the paramount mitigation approach within large campus communities and others facing similar challenges.
Data Availability
All data referred to in the manuscript, will be made available upon request
Acknowledgements
We thank the UC Santa Barbara Office of Research for their generous support. We thank Dr. Mary Ferris and the UCSB Student Health Service personnel who helped establish the pre- analytical protocols and collect the samples. We thank Laura Isaac, Erin Ross and Catelynn Kenner for their guidance on the IRB protocols. We thank all the participants of our study. Finally, we thank all essential workers for keeping society running. Without them this work would have never happened.
Footnotes
Included Sup. Table 2 with the list of symptoms reported by participants confirmed to be positive for COVID-19
References
