Blood transcriptional biomarkers of acute viral infection for detection of pre-symptomatic SARS-CoV-2 infection

medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 1 Title 2 Blood transcriptional biomarkers of acute viral infection for detection of pre-symptomatic SARS-CoV-2 infection. 3 Authors 4 Rishi K. Gupta1,2†, Joshua Rosenheim2†, Lucy C. Bell2†, Aneesh Chandran2, Jose A. Guerra-Assuncao2, Gabriele 5 Pollara2, Matthew Whelan2, Jessica Artico3, George Joy3, Hibba Kurdi3, Daniel M. Altmann4, Rosemary Boyton5, 6 Mala Maini2, Aine McKnight6, Jonathan Lambourne7, Teresa Cutino-moguel8, Charlotte Manisty3,9, Thomas A. 7 Treibel3,9, James C Moon3,9, Benjamin Chain2 and Mahdad Noursadeghi2* on behalf of the COVIDsortium 8 Investigators. 9 Affiliations 10 1Institute of Global Health, University College London, London, UK 11 2Division of Infection and Immunity, University College London, London, UK 12 3Barts 13 4Department 14 5Lung 15 16 6Blizard 17 7Department of Infection, Barts Health NHS Trust, London, UK 18 8Department of Virology, Barts Health NHS Trust, London, UK 19 9Institute 20 *To whom correspondence should be addressed: Division of Infection & Immunity, Cruciform Building, University 21 College London, London WC1E 6BT, UK. Email: [email protected] (https://orcid.org/0000-0002-4774- 22 0853) 23 †Contributed equally. Heart Centre, St Bartholomew's Hospital, Barts Health NHS Trust, London, UK, London, UK of Immunology and Inflammation, Imperial College London, London, UK Division, Royal Brompton & Harefield NHS Foundation Trust, London, UK Institute, Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London, UK. of Cardiovascular Sciences, University College London, London, UK NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 24 Abstract 25 We hypothesised that host-response biomarkers of viral infections may contribute to early identification of SARS- 26 CoV-2 infected individuals, critical to breaking chains of transmission. We identified 20 candidate blood 27 transcriptomic signatures of viral infection by systematic review and evaluated their ability to detect SARS-CoV-2 28 infection, compared to the gold-standard of virus PCR tests, among a prospective cohort of 400 hospital staff 29 subjected to weekly testing when fit to attend work. The transcriptional signatures had limited overlap, but were 30 mostly co-correlated as components of type 1 interferon responses. We reconstructed each signature score in 31 blood RNA sequencing data from 41 individuals over sequential weeks spanning a first positive SARS-CoV-2 32 PCR, and after 6-month convalescence. A single blood transcript for IFI27 provided the highest accuracy for 33 discriminating individuals at the time of their first positive viral PCR result from uninfected controls, with area under 34 the receiver operating characteristic curve (AUROC) of 0.95 (95% confidence interval 0.91–0.99), sensitivity 0.84 35 (0.7–0.93) and specificity 0.95 (0.85–0.98) at a predefined test threshold. The test performed equally well in 36 individuals with and without symptoms, correlated with viral load, and identified incident infections one week before 37 the first positive viral PCR with sensitivity 0.4 (0.17–0.69) and specificity 0.95 (0.85–0.98). Our findings strongly 38 support further urgent evaluation and development of blood IFI27 transcripts as a biomarker for early phase SARS- 39 CoV-2 infection, for screening individuals such as contacts of index cases, in order to facilitate early case isolation 40 and early antiviral treatments as they emerge. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 41 Introduction 42 Rapid and accurate testing is central to effective public health responses to COVID-19. Infectivity, measured by 43 upper respiratory tract SARS-CoV-2 titres, peaks during the first week of illness (1). Early case detection followed 44 by rapid isolation of index cases, alongside contact tracing and quarantine, are therefore key interventions to 45 interrupt onward transmission. As a proportion of individuals with SARS-CoV-2 shed virus while asymptomatic or 46 pauci-symptomatic (2, 3), there is also global interest in screening tests for at-risk individuals who do not fulfil case 47 definition criteria and in mass testing for early case detection among the general population (4). 48 Effective screening tests must be accurate and reliable (5). However, current tools such as lateral flow assays 49 (LFAs) for SARS-CoV-2 antigens appear to have inadequate sensitivity to effectively rule out active infection and 50 may have limited value for contact and general population screening (6). Reverse-transcriptase PCR (RT-PCR) 51 tests, which identify viral RNA, are the current gold standard for diagnosis of SARS-CoV-2 infection but pose 52 different challenges including test speed and the requirement of a skilled laboratory operator (7). LAMP (loop- 53 mediated isothermal amplification) assays improve on RT-PCR timings but with an associated reduction in 54 sensitivity (8). All current viral detection tests rely on swabbing of nasopharyngeal and/or oropharyngeal mucosa, 55 the effectiveness of which is operator-dependent and prone to sampling variability. While positive SARS-CoV-2 56 test results are useful in clinical management and infection control settings, all available tests have false negative 57 rates which impact on interpretation, particularly in the context of high pre-test probability in high transmission 58 settings (9). 59 There is therefore a clear need to expand the portfolio of tests available for identification of SARS-CoV-2 infection, 60 for both screening and diagnostic purposes. Measurement of the host response, as opposed to viral targets, is 61 one potential diagnostic strategy. Numerous studies have demonstrated whole-blood transcriptional perturbation 62 during other acute viral infections (10–13). A range of blood transcriptomic signatures have therefore been 63 proposed as candidate diagnostic biomarkers for purposes including discrimination of viral from bacterial infection 64 or no infection (10–21), diagnosis of pre-symptomatic viral infection in known contacts (22), diagnosis of specific 65 viral infections (23, 24), or prognostication of severity (25). These signatures have not yet been evaluated for early 66 diagnosis of pre-symptomatic or mild SARS-CoV-2 infection. We present a systematic evaluation of the potential 67 for existing candidate whole-blood transcriptomic signatures of viral infection to predict nasopharyngeal SARS- 68 CoV-2 PCR positivity in healthcare workers undergoing weekly testing with paired blood RNA sampling. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 69 Results 70 Overview of study cohort 71 We included a total of 169 blood RNA samples from 96 participants in a nested case-control study (Fig. S1) derived 72 from an observational healthcare worker cohort (26–28). Of these, 114 samples (including 16 convalescent 73 samples collected 6 months after infection) were obtained from 41 incident cases with SARS-CoV-2 infection, and 74 55 samples from uninfected controls. Participant baseline characteristics are shown in Table S1. 32/41 individuals 75 with incident virus PCR positive infection denied any disease defining symptoms at the time of their positive PCR 76 test, whilst 9/41 described one or more of cough, fever, or anosmia. A further 22 individuals developed symptoms 77 during subsequent follow up. 78 Overview of candidate RNA signatures for viral infection 79 Our systematic literature search identified 1150 titles and abstracts; 61 studies were shortlisted for full-text review. 80 A total of 18 studies, describing 20 distinct transcriptional signatures for viral infection, met the eligibility criteria for 81 inclusion in the final analysis (Table 1 and Fig. S2). Signatures comprised between 1 and 48 component genes 82 and were discovered in a range of populations including children and adults with acute viral infections, and adults 83 experimentally challenged with viruses including influenza, respiratory syncytial virus (RSV) and rhinovirus. The 84 majority of signatures (12/20) were discovered with the objective of discriminating viral infection from bacterial or 85 other inflammatory presentations (10–21). Three aimed to discriminate viral infection from healthy individuals (29, 86 30) and two were discovered with a specific objective of diagnosing influenza infection (23, 24). One signature 87 aimed to predict the severity of RSV infection in children (25). One study evaluated a pre-existing signature with 88 the objective of identifying pre-symptomatic viral infection in individuals who were close contacts of index cases 89 with acute viral respiratory tract infections (22). 90 In most cases there was little overlap between the constituent genes in each signature, but most signatures 91 showed moderate to strong correlation, which was only partly explained by overlapping constituent genes 92 (Fig. 1A–C). Bioinformatic analysis of the integrated list of constituent genes to identify upstream regulators using 93 Ingenuity Pathway Analysis was consistent with type 1 interferon regulation of these genes to explain the strong 94 correlation between signatures despite limited overlap of their constituents. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 95 Diagnostic accuracy of RNA signatures for SARS-CoV-2 infection 96 Among all the signatures, the transcript for Interferon Alpha Inducible Protein 27 (IFI27) by itself provided the best 97 discrimination of contemporaneous SARS-CoV-2 infection by nasopharyngeal PCR, compared to uninfected 98 controls, achieving an AUROC of 0.95 (95% confidence interval 0.91–0.99). Using a pre-specified Z2 cut-off based 99 on two standard deviations above the mean of the uninfected control samples, IFI27 had a sensitivity of 0.84 100 (0.70–0.93) and specificity of 0.95 (0.85–0.98). Three other candidate signatures (Sweeney7, Zaas48 and 101 Pennisi2) had statistically equivalent accuracy to IFI27 using paired DeLong tests (Table 2). Fig. S3 shows 102 constituent genes for these four best performing signatures; only one of these (Pennisi2) did not include IFI27. 103 Exclusion of participants with contemporaneous case-defining symptoms at the time of SARS-CoV-2 infection 104 (n=9) had no significant impact on the primary analysis (Table S2). Scores for each of the four best performing 105 signatures were inversely correlated with SARS-CoV-2 RT-PCR cycle thresholds, also independent of symptoms, 106 suggesting that higher viral loads were associated with higher signature scores (Figure 2B; Spearman rank 107 correlation coefficients −0.61 to −0.69). 108 Longitudinal expression of the four best performing signatures peaked at the week of first virus PCR-positivity and 109 had normalised at the point of convalescent sampling (Fig. 2). Importantly, however, measurements in the week 110 preceding the first positive virus PCR were higher than uninfected controls and convalescent samples. AUROCs 111 for discrimination between samples taken in the week prior to first SARS-CoV-2 detection and uninfected controls 112 revealed statistically significant discrimination, but were lower than those for contemporaneous virus PCR positivity 113 (Table S3). For illustration, IFI27 predicted infection one week before a positive virus PCR test with an AUROC of 114 0.79 (0.6–0.98). At a Z2 cut-off this achieved sensitivity of 0.4 (0.17–0.69) and specificity of 0.95 (0.85–0.98). 115 Discussion 116 To our knowledge, this is the first evaluation of host transcriptomic signatures for detection of pre-symptomatic 117 SARS-CoV-2 infection. Using a longitudinal blood transcriptomic dataset prospectively collected from London 118 healthcare workers during the first wave of the COVID-19 pandemic, we systematically compared the diagnostic 119 accuracy of 20 candidate transcriptional signatures originally discovered in a wide range of viral infection cohorts. 120 We found that four candidate signatures had high accuracy (AUROCs 0.91–0.95) for discriminating individuals 121 with acute contemporaneous SARS-CoV-2 infection from uninfected controls. Three of the four signatures medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 122 contained the interferon-stimulated gene IFI27, which was by itself the top-performing biomarker, originally 123 discovered in a paediatric cohort (30) to discriminate RSV infections from controls. 124 The candidate signatures evaluated in the present study are collectively associated with type 1 IFN responses, 125 which are a canonical feature of antiviral host defences. The importance of this response in SARS-CoV-2 infection 126 is highlighted by the association of severe COVID-19 with both of loss-of-function genetic variation in various 127 components of type 1 IFN pathways and with anti-type 1 IFN antibodies (31, 32). IFI27 is best characterised for 128 its functional role in type 1 IFN mediated apoptosis as a component of anti-tumour effects of IFNs (33). Differential 129 regulation of IFN inducible genes might explain why expression of IFI27 transcripts outperforms other type 1 IFN 130 signatures and merits investigation in future work to evaluate its significance in the anti-viral response. 131 A key feature of our study is that all participants self-declared as fit to work when attending study visits, including 132 at the time of their first positive SARS-CoV-2 PCR test, when most were asymptomatic. We also found detectable 133 expression of the signatures in blood transcriptomes collected at the study visit one week prior to virus PCR test 134 positivity among a subset of participants. Our data therefore demonstrate that measurable type 1 IFN-stimulated 135 responses to SARS-CoV-2 precede the onset of symptoms, and in some individuals may predate detectable viral 136 RNA on RT-PCR testing. We propose that novel diagnostic tests that detect transcripts (or associated protein 137 targets) from the top-performing candidate signatures could be valuable tools in the rapid detection and isolation 138 of individuals in the very earliest stages of pre-clinical infection. Importantly, these signatures also correlated with 139 viral load independently of symptoms, indicating that they have strong potential to identify the most infectious 140 individuals, critical to breaking chains of transmission. 141 A key strength of our study was the weekly longitudinal follow-up of study participants which enabled detailed 142 characterisation of the study cohort, including contemporaneous capture of blood RNA samples at the point of 143 SARS-CoV-2 PCR positivity in pre-symptomatic and asymptomatic infection. In addition, we performed a 144 comprehensive systematic literature search to identify candidate blood transcriptional signatures for viral infection. 145 This enabled direct head-to-head assessments of their diagnostic accuracy for SARS-CoV-2 infection and will 146 provide a framework for future systematic evaluations of blood transcriptional biomarkers for viral infections. 147 Our findings focus on early pre-symptomatic infection and may not be generalisable in moderate to severe 148 COVID-19. In addition, we have not sought to evaluate discrimination between SARS-CoV-2 and other acute viral 149 infections. In view of their discovery in a range of viral infections, we expect these signatures to serve as medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 150 non-specific biomarkers of acute viral infection. Nonetheless, their sensitivity for detecting pre-symptomatic 151 infection offers potential clinical utility for screening contacts of index cases of SARS-CoV-2 in order to inform 152 infection control management, and stratify the need for confirmatory viral PCR testing. 153 In summary, we have shown that a single transcript (IFI27) detects SARS-CoV-2 infection with high accuracy. If 154 translated to a near-patient diagnostic test (34, 35), this transcript could have significant clinical utility by facilitating 155 early case detection. 156 Materials and Methods 157 Ethical approval 158 The study was approved by a UK Research Ethics Committee (South Central - Oxford A Research Ethics 159 Committee, reference 20/SC/0149). All participants provided written informed consent. 160 Study design 161 We undertook a case-control study nested within our COVIDsortium health care worker cohort. Participant 162 screening, study design, sample collection, and sample processing have been described in detail previously (26– 163 28) and the study is registered at ClinicalTrials.gov (NCT04318314). Briefly, healthcare workers were recruited at 164 St Bartholomew’s Hospital, London, UK in the week of lockdown in the United Kingdom (between 23 rd and 31st 165 March 2020). Participants underwent weekly evaluation using a questionnaire and biological sample collection 166 (including serological assays) for up to 16 weeks when fit to attend work at each visit, with further follow up samples 167 collected at 6 months. 168 Participants with available blood RNA samples who had PCR-confirmed SARS-CoV-2 infection (Roche cobas® 169 diagnostic test platform) at any time point were included as ‘cases’. A subset of consecutively recruited participants 170 without evidence of SARS-CoV-2 infection on nasopharyngeal swabs and who remained seronegative by both 171 Euroimmun anti-S1 spike protein and Roche anti-nucleocapsid protein throughout follow-up were included as 172 uninfected controls. 173 Systematic search for candidate transcriptional signatures 174 We performed a systematic literature search of peer-reviewed publications in order to identify concise blood 175 transcriptional signatures discovered or applied with a primary objective of diagnosis or assessment of severity of 176 viral infection from blood or PBMC samples. We searched Medline on 12/10/2020 using comprehensive MeSH medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 177 and key word terms for “viral infection”, “transcriptome”, “biomarker” and “blood” (full search strategy shown in 178 Table S4). Additional studies were identified in reference lists and from expert consultation. Titles and abstracts 179 were initially screened by two independent reviewers; full-text review was performed for shortlisted articles to 180 determine eligibility and conflicts were resolved through discussion and arbitration by a third reviewer where 181 required. We focused on ‘concise’ signatures that might be more amenable to translation to diagnostic tests and 182 defined this as any signature discovered using a defined approach to feature selection to reduce the number of 183 constituent genes, as previously (36). We required that gene names that comprised the signature were publicly 184 available, along with the corresponding signature equation or model coefficients, and that the signature must be 185 validated in at least one independent test or validation set in order to prioritise signatures discovered from higher 186 quality studies. Where multiple signatures were discovered for the same intended purpose and from the same 187 discovery cohort, we included the signature with highest discrimination (as defined by the AUROC) in the validation 188 data, or the signature with fewest number of genes where accuracy was equivalent. 189 For each signature that met our eligibility criteria, we extracted constituent genes, modelling approaches and 190 coefficients to enable independent reconstruction of signature scores. Extraction was performed by a single 191 reviewer and was verified by a second reviewer. 192 Blood RNA sequencing 193 For ‘cases’, we included all available RNA samples, including convalescent samples at week 24 of follow-up for a 194 subset of participants. For uninfected controls, we included baseline samples only. Genome wide mRNA 195 sequencing was performed as previously described (37), resulting in a median of 26 million (range, 19.8–32.4 196 million) 41 bp paired-end reads per sample. RNAseq data were mapped to the reference transcriptome (Ensembl 197 Human GRCh38 release 100) using Kallisto (38). The transcript-level output counts and transcripts per million 198 (TPM) values were summed on gene level and annotated with Ensembl gene ID, gene name, and gene biotype 199 using the R/Bioconductor packages tximport and BioMart (39, 40). 200 Data analysis 201 For each eligible signature, we reconstructed signature scores as per the original authors’ descriptions. For logistic 202 and probit regression models, we calculated scores on the linear predictor scale by summing the expression of 203 each constituent gene multiplied by its coefficient. Scores for each signature were standardised to Z scores using 204 the mean and standard deviation among the uninfected control population. Scores that were designed to decrease medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 205 in viral infection were multiplied by −1 in order to ensure that higher scores should be associated with higher risk 206 of viral infection across all candidate signatures. 207 The primary outcome was the AUROC for discriminating participants with PCR-confirmed SARS-CoV-2 infection 208 during their first week of PCR positivity, from uninfected control samples. The secondary outcome was the AUROC 209 for discriminating participants with PCR-confirmed SARS-CoV-2 infections in the week prior to first PCR positivity, 210 from uninfected control samples. We calculated corresponding sensitivities and specificities for each signature for 211 the primary and secondary outcomes using pre-specified cut-offs based on two standard deviations above the 212 mean of the uninfected controls, as previously (36). In order to identify the subset of best performing signatures, 213 we performed pairwise DeLong tests to the signature with the highest AUROC for the primary outcome (or most 214 parsimonious in the event of equal performance), with adjustment for multiple testing using a Benjamini-Hochberg 215 correction (41). Signatures were considered to have statistically inferior accuracy to the best performing signature 216 if adjusted p<0.05. We also performed a sensitivity analysis for the primary outcome, excluding participants with 217 positive SARS-CoV-2 swabs who reported contemporaneous case-defining symptoms at the time of sampling. 218 Upstream analysis of transcriptional regulation of the constituent genes in the candidate signatures was performed 219 using Ingenuity Pathway Analysis (Qiagen, Venlo, The Netherlands) and visualized as network diagrams in Gephi 220 v0.9.2, depicting all statistically overrepresented molecules predicted to be upstream >2 target genes, as 221 previously (36). We evaluated pairwise Spearman rank and Jaccard indices between each candidate signature in 222 order to quantify correlations and proportions of intersecting genes between signatures. 223 Role of the funding source 224 The funder had no role in study design, data collection, data analysis, data interpretation, writing of the report, or 225 decision to submit for publication. The corresponding authors had full access to all the data in the study and had 226 final responsibility for the decision to submit for publication. 227 Supplementary Materials 228 Fig. S1. CONSORT (Consolidated Standards of Reporting Trials) flow diagram. 229 Fig. S2. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow-chart of systematic 230 review process. 231 Fig S3. Constituent genes of best-performing RNA signatures. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 232 Table S1. Search strategy for systematic Medline search. 233 Table S2. Validation metrics of whole-blood RNA signatures at week of first week of PCR-positivity. 234 Table S3. Validation metrics of whole-blood RNA signatures at week prior to first week of PCR-positivity. 235 Table S4. Baseline characteristics of the study cohort. 236 Table S5. Full list of COVIDsortium investigators. 237 References and Notes: 238 1. M. Cevik, M. Tate, O. Lloyd, A. E. Maraolo, J. Schafers, A. Ho, SARS-CoV-2, SARS-CoV, and MERS-CoV viral 239 load dynamics, duration of viral shedding, and infectiousness: a systematic review and meta-analysis, The Lancet 240 Microbe 2 (2020), doi:10.1016/S2666-5247(20)30172-5. 241 2. M. R. Kasper, J. R. Geibe, C. L. Sears, A. J. Riegodedios, T. Luse, A. M. V. Thun, M. B. McGinnis, N. Olson, 242 D. Houskamp, R. Fenequito, T. H. Burgess, A. W. Armstrong, G. DeLong, R. J. Hawkins, B. L. Gillingham, An 243 Outbreak 244 doi:10.1056/NEJMoa2019375. 245 3. A. A. Sayampanathan, C. S. Heng, P. H. Pin, J. Pang, T. Y. Leong, V. J. Lee, Infectivity of asymptomatic versus 246 symptomatic COVID-19, The Lancet 397 (2020), doi:10.1016/S0140-6736(20)32651-9. 247 4. T. Burki, Mass testing for COVID-19, The Lancet Microbe 1, e317 (2020). 248 5. M. J. Dobrow, V. Hagens, R. Chafe, T. Sullivan, L. Rabeneck, Consolidated principles for screening based on 249 a systematic review and consensus process, CMAJ : Canadian Medical Association Journal 190, E422 (2018). 250 6. J. J. Deeks, A. E. Raffle, Lateral flow tests cannot rule out SARS-CoV-2 infection, BMJ 371 (2020), 251 doi:10.1136/bmj.m4787. 252 7. N. Younes, D. W. Al-Sadeq, H. AL-Jighefee, S. Younes, O. Al-Jamal, H. I. Daas, Hadi. M. Yassine, G. K. 253 Nasrallah, Challenges in Laboratory Diagnosis of the Novel Coronavirus SARS-CoV-2, Viruses 12 (2020), 254 doi:10.3390/v12060582. of Covid-19 on an Aircraft Carrier, New England Journal of Medicine 383 (2020), medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 255 8. M. Mustafa Hellou, A. Górska, F. Mazzaferri, E. Cremonini, E. Gentilotti, P. De Nardo, I. Poran, M. M. Leeflang, 256 E. Tacconelli, M. Paul, Nucleic acid amplification tests on respiratory samples for the diagnosis of coronavirus 257 infections: a systematic review and meta-analysis, Clin Microbiol Infect (2020), doi:10.1016/j.cmi.2020.11.002. 258 9. J. Watson, P. F. Whiting, J. E. Brush, Interpreting a covid-19 test result, BMJ 369 (2020), 259 doi:10.1136/bmj.m1808. 260 10. J. A. Herberg, M. Kaforou, V. J. Wright, H. Shailes, H. Eleftherohorinou, C. J. Hoggart, M. Cebey-López, M. J. 261 Carter, V. A. Janes, S. Gormley, C. Shimizu, A. H. Tremoulet, A. M. Barendregt, A. Salas, J. Kanegaye, A. J. 262 Pollard, S. N. Faust, S. Patel, T. Kuijpers, F. Martinón-Torres, J. C. Burns, L. J. M. Coin, M. Levin, IRIS Consortium, 263 Diagnostic Test Accuracy of a 2-Transcript Host RNA Signature for Discriminating Bacterial vs Viral Infection in 264 Febrile Children, JAMA 316, 835–845 (2016). 265 11. E. C. Lydon, R. Henao, T. W. Burke, M. Aydin, B. P. Nicholson, S. W. Glickman, V. G. Fowler, E. B. 266 Quackenbush, C. B. Cairns, S. F. Kingsmore, A. K. Jaehne, E. P. Rivers, R. J. Langley, E. Petzold, E. R. Ko, M. 267 T. McClain, G. S. Ginsburg, C. W. Woods, E. L. Tsalik, Validation of a host response test to distinguish bacterial 268 and viral respiratory infection, EBioMedicine 48, 453–461 (2019). 269 12. T. E. Sweeney, H. R. Wong, P. Khatri, Robust classification of bacterial and viral infections via integrated host 270 gene expression diagnostics, Science Translational Medicine 8, 346ra91-346ra91 (2016). 271 13. D. Sampson, T. D. Yager, B. Fox, L. Shallcross, L. McHugh, T. Seldon, A. Rapisarda, R. A. Hendriks, R. B. 272 Brandon, K. Navalkar, N. Simpson, S. Stafford, E. Gil, C. Venturini, E. Tsaliki, J. Roe, B. Chain, M. Noursadeghi, 273 Blood transcriptomic discrimination of bacterial and viral infections in the emergency department: a multi-cohort 274 observational validation study, BMC Medicine 18, 185 (2020). 275 14. A. Gómez-Carballa, M. Cebey-López, J. Pardo-Seco, R. Barral-Arca, I. Rivero-Calle, S. Pischedda, M. J. 276 Currás-Tuala, J. Gómez-Rial, F. Barros, F. Martinón-Torres, A. Salas, A qPCR expression assay of IFI44L gene 277 differentiates viral from bacterial infections in febrile children, Sci Rep 9, 1–12 (2019). 278 15. R. Lopez, R. Wang, G. Seelig, A molecular multi-gene classifier for disease diagnostics, Nature Chem 10, 279 746–754 (2018). medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 280 16. D. L. Sampson, B. A. Fox, T. D. Yager, S. Bhide, S. Cermelli, L. C. McHugh, T. A. Seldon, R. A. Brandon, E. 281 Sullivan, J. J. Zimmerman, M. Noursadeghi, R. B. Brandon, A Four-Biomarker Blood Signature Discriminates 282 Systemic Inflammation Due to Viral Infection Versus Other Etiologies, Sci Rep 7, 1–17 (2017). 283 17. T. E. Sweeney, A. Shidham, H. R. Wong, P. Khatri, A comprehensive time-course–based multicohort analysis 284 of sepsis and sterile inflammation reveals a robust diagnostic gene set, Science Translational Medicine 7, 287ra71- 285 287ra71 (2015). 286 18. S. Trouillet-Assant, S. Viel, A. Ouziel, L. Boisselier, P. Rebaud, R. Basmaci, N. Droz, A. Belot, S. Pons, K. 287 Brengel-Pesce, Y. Gillet, E. Javouhey, A. S. Group, M. Mommert, A. Guichard, F. Bartolo, L. Generenaz, A. 288 Pachot, C. Capella, L. Hees, E. Mezgueldi, C. Toumi, C. Bouchiat-Sarabi, J. Casalegno, A. Portefaix, R. D. de 289 Cambronne, M. Perret, Type I Interferon in Children with Viral or Bacterial Infections, Clin Chem 66, 802–808 290 (2020). 291 19. E. L. Tsalik, R. Henao, M. Nichols, T. Burke, E. R. Ko, M. T. McClain, L. L. Hudson, A. Mazur, D. H. Freeman, 292 T. Veldman, R. J. Langley, E. B. Quackenbush, S. W. Glickman, C. B. Cairns, A. K. Jaehne, E. P. Rivers, R. M. 293 Otero, A. K. Zaas, S. F. Kingsmore, J. Lucas, V. G. Fowler, L. Carin, G. S. Ginsburg, C. W. Woods, Host gene 294 expression classifiers diagnose acute respiratory illness etiology, Science Translational Medicine 8, 322ra11- 295 322ra11 (2016). 296 20. I. Pennisi, J. Rodriguez-Manzano, A. Moniri, M. Kaforou, J. A. Herberg, M. Levin, P. Georgiou, Translation of 297 a host blood RNA Signature distinguishing bacterial from viral infection into a platform suitable for development as 298 a point-of-care test, JAMA Pediatrics (2021), doi:doi:10.1001/jamapediatrics.2020.5227. 299 21. A. K. Zaas, T. Burke, M. Chen, M. McClain, B. Nicholson, T. Veldman, E. L. Tsalik, V. Fowler, E. P. Rivers, R. 300 Otero, S. F. Kingsmore, D. Voora, J. Lucas, A. O. Hero, L. Carin, C. W. Woods, G. S. Ginsburg, A Host-Based 301 RT-PCR Gene Expression Signature to Identify Acute Respiratory Viral Infection, Science Translational Medicine 302 5, 203ra126-203ra126 (2013). 303 22. M. T. McClain, F. J. Constantine, B. P. Nicholson, M. Nichols, T. W. Burke, R. Henao, D. C. Jones, L. L. 304 Hudson, L. B. Jaggers, T. Veldman, A. Mazur, L. P. Park, S. Suchindran, E. L. Tsalik, G. S. Ginsburg, C. W. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 305 Woods, A blood-based host gene expression assay for early detection of respiratory viral infection: an index- 306 cluster prospective cohort study, The Lancet Infectious Diseases 0 (2020), doi:10.1016/S1473-3099(20)30486-2. 307 23. M. Andres-Terre, H. M. McGuire, Y. Pouliot, E. Bongen, T. E. Sweeney, C. M. Tato, P. Khatri, Integrated, Multi- 308 cohort Analysis Identifies Conserved Transcriptional Signatures across Multiple Respiratory Viruses, Immunity 43, 309 1199 (2015). 310 24. S. E. Henrickson, S. Manne, D. V. Dolfi, K. D. Mansfield, K. Parkhouse, R. D. Mistry, E. R. Alpern, S. E. 311 Hensley, K. E. Sullivan, S. E. Coffin, E. J. Wherry, Genomic Circuitry Underlying Immunological Response to 312 Pediatric Acute Respiratory Infection, Cell Reports 22, 411–426 (2018). 313 25. H. K. Brand, I. M. L. Ahout, D. de Ridder, A. van Diepen, Y. Li, M. Zaalberg, A. Andeweg, N. Roeleveld, R. de 314 Groot, A. Warris, P. W. M. Hermans, G. Ferwerda, F. J. T. Staal, Olfactomedin 4 Serves as a Marker for Disease 315 Severity in Pediatric Respiratory Syncytial Virus (RSV) Infection, PLOS ONE 10, e0131927 (2015). 316 26. J. B. Augusto, K. Menacho, M. Andiapen, R. Bowles, M. Burton, S. Welch, A. N. Bhuva, A. Seraphim, C. Pade, 317 G. Joy, M. Jensen, R. H. Davies, G. Captur, M. Fontana, H. Montgomery, B. O’Brien, A. D. Hingorani, T. Cutino- 318 Moguel, Á. McKnight, H. Abbass, M. Alfarih, Z. Alldis, G. L. Baca, A. Boulter, O. V. Bracken, N. Bullock, N. 319 Champion, C. Chan, X. Couto-Parada, K. Dieobi-Anene, K. Feehan, G. Figtree, M. C. Figtree, M. Finlay, N. 320 Forooghi, J. M. Gibbons, P. Griffiths, M. Hamblin, L. Howes, I. Itua, M. Jones, V. Jardim, V. Kapil, W.-Y. J. Lee, V. 321 Mandadapu, C. Mfuko, O. Mitchelmore, S. Palma, K. Patel, S. E. Petersen, B. Piniera, R. Raine, A. Rapala, A. 322 Richards, G. Sambile, J. C. de Sousa, M. Sugimoto, G. D. Thornton, J. Artico, D. Zahedi, R. Parker, M. Robathan, 323 L. M. Hickling, N. Ntusi, A. Semper, T. Brooks, J. Jones, A. Tucker, J. Veerapen, M. Vijayakumar, T. Wodehouse, 324 L. Wynne, T. A. Treibel, M. Noursadeghi, C. Manisty, J. C. Moon, Healthcare Workers Bioresource: Study outline 325 and baseline characteristics of a prospective healthcare worker cohort to study immune protection and 326 pathogenesis in COVID-19, Wellcome Open Research 5, 179 (2020). 327 27. T. A. Treibel, C. Manisty, M. Burton, Á. McKnight, J. Lambourne, J. B. Augusto, X. Couto-Parada, T. Cutino- 328 Moguel, M. Noursadeghi, J. C. Moon, COVID-19: PCR screening of asymptomatic health-care workers at London 329 hospital, The Lancet 395, 1608–1610 (2020). medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 330 28. C. J. Reynolds, L. Swadling, J. M. Gibbons, C. Pade, M. P. Jensen, M. O. Diniz, N. M. Schmidt, D. K. Butler, 331 O. E. Amin, S. N. L. Bailey, S. M. Murray, F. P. Pieper, S. Taylor, J. Jones, M. Jones, W.-Y. J. Lee, J. Rosenheim, 332 A. Chandran, G. Joy, C. D. Genova, N. Temperton, J. Lambourne, T. Cutino-Moguel, M. Andiapen, M. Fontana, 333 A. Smit, A. Semper, B. O’Brien, B. Chain, T. Brooks, C. Manisty, T. Treibel, J. C. Moon, Covid. Investigators$, M. 334 Noursadeghi, Covid. immune correlates Network$, D. M. Altmann, M. K. Maini, Á. McKnight, R. J. Boyton, 335 Discordant neutralizing antibody and T cell responses in asymptomatic and mild SARS-CoV-2 infection, Science 336 Immunology 5 (2020), doi:10.1126/sciimmunol.abf3698. 337 29. A. Roers, H. K. Hochkeppel, M. A. Horisberger, A. Hovanessian, O. Haller, MxA Gene Expression after Live 338 Virus Vaccination: A Sensitive Marker for Endogenous Type I Interferon, J Infect Dis 169, 807–813 (1994). 339 30. J. Yu, D. R. Peterson, A. M. Baran, S. Bhattacharya, T. N. Wylie, A. R. Falsey, T. J. Mariani, G. A. Storch, 340 Host Gene Expression in Nose and Blood for the Diagnosis of Viral Respiratory Infection, J Infect Dis 219, 1151– 341 1161 (2019). 342 31. E. Pairo-Castineira, S. Clohisey, L. Klaric, A. D. Bretherick, K. Rawlik, D. Pasko, S. Walker, N. Parkinson, M. 343 H. Fourman, C. D. Russell, J. Furniss, A. Richmond, E. Gountouna, N. Wrobel, D. Harrison, B. Wang, Y. Wu, A. 344 Meynert, F. Griffiths, W. Oosthuyzen, A. Kousathanas, L. Moutsianas, Z. Yang, R. Zhai, C. Zheng, G. Grimes, R. 345 Beale, J. Millar, B. Shih, S. Keating, M. Zechner, C. Haley, D. J. Porteous, C. Hayward, J. Yang, J. Knight, C. 346 Summers, M. Shankar-Hari, P. Klenerman, L. Turtle, A. Ho, S. C. Moore, C. Hinds, P. Horby, A. Nichol, D. Maslove, 347 L. Ling, D. McAuley, H. Montgomery, T. Walsh, A. Pereira, A. Renieri, X. Shen, C. P. Ponting, A. Fawkes, A. 348 Tenesa, M. Caulfield, R. Scott, K. Rowan, L. Murphy, P. J. M. Openshaw, Malcolm G. Semple, A. Law, V. Vitart, 349 J. F. Wilson, J. K. Baillie, Genetic mechanisms of critical illness in Covid-19, Nature , 1–1 (2020). 350 32. Q. Zhang, P. Bastard, Z. Liu, J. L. Pen, M. Moncada-Velez, J. Chen, M. Ogishi, I. K. D. Sabli, S. Hodeib, C. 351 Korol, J. Rosain, K. Bilguvar, J. Ye, A. Bolze, B. Bigio, R. Yang, A. A. Arias, Q. Zhou, Y. Zhang, F. Onodi, S. 352 Korniotis, L. Karpf, Q. Philippot, M. Chbihi, L. Bonnet-Madin, K. Dorgham, N. Smith, W. M. Schneider, B. S. 353 Razooky, H.-H. Hoffmann, E. Michailidis, L. Moens, J. E. Han, L. Lorenzo, L. Bizien, P. Meade, A.-L. Neehus, A. 354 C. Ugurbil, A. Corneau, G. Kerner, P. Zhang, F. Rapaport, Y. Seeleuthner, J. Manry, C. Masson, Y. Schmitt, A. 355 Schlüter, T. L. Voyer, T. Khan, J. Li, J. Fellay, L. Roussel, M. Shahrooei, M. F. Alosaimi, D. Mansouri, H. Al-Saud, 356 F. Al-Mulla, F. Almourfi, S. Z. Al-Muhsen, F. Alsohime, S. A. Turki, R. Hasanato, D. van de Beek, A. Biondi, L. R. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 357 Bettini, M. D’Angio’, P. Bonfanti, L. Imberti, A. Sottini, S. Paghera, E. Quiros-Roldan, C. Rossi, A. J. Oler, M. F. 358 Tompkins, C. Alba, I. Vandernoot, J.-C. Goffard, G. Smits, I. Migeotte, F. Haerynck, P. Soler-Palacin, A. Martin- 359 Nalda, R. Colobran, P.-E. Morange, S. Keles, F. Çölkesen, T. Ozcelik, K. K. Yasar, S. Senoglu, Ş. N. Karabela, 360 C. Rodríguez-Gallego, G. Novelli, S. Hraiech, Y. Tandjaoui-Lambiotte, X. Duval, C. Laouénan, C.-S. Clinicians†, 361 C. Clinicians†, I. C. Group†, F. C. C. S. Group†, C.-C. Cohort†, A. U. C.-19 Biobank†, C. H. G. Effort†, N.-U. C. I. 362 Group†, A. L. Snow, C. L. Dalgard, J. D. Milner, D. C. Vinh, T. H. Mogensen, N. Marr, A. N. Spaan, B. Boisson, S. 363 Boisson-Dupuis, J. Bustamante, A. Puel, M. J. Ciancanelli, I. Meyts, T. Maniatis, V. Soumelis, A. Amara, M. 364 Nussenzweig, A. García-Sastre, F. Krammer, A. Pujol, D. Duffy, R. P. Lifton, S.-Y. Zhang, G. Gorochov, V. Béziat, 365 E. Jouanguy, V. Sancho-Shimizu, C. M. Rice, L. Abel, L. D. Notarangelo, A. Cobat, H. C. Su, J.-L. Casanova, 366 Inborn errors of type I IFN immunity in patients with life-threatening COVID-19, Science 370 (2020), 367 doi:10.1126/science.abd4570. 368 33. S. Rosebeck, D. W. Leaman, Mitochondrial localization and pro-apoptotic effects of the interferon-inducible 369 protein ISG12a, Apoptosis 13, 562–572 (2008). 370 34. E. L. Tsalik, R. Henao, M. Aydin, C. Bullard, J. Montgomery, J. Nawrocki, M. Deneris, C. Gritzen, J. Jones, R. 371 Crisp, M. T. Mcclain, T. Burke, G. S. Ginsburg, A. Hemmert, C. W. Woods, 2012. FilmArray® Measurement of 372 Host Response Signatures RapidLy Discriminates Viral, Bacterial, and Non-infectious Etiologies of Illness, Open 373 Forum Infect Dis 5, S586 (2018). 374 35. R. R. Miller III, B. K. Lopansri, J. P. Burke, M. Levy, S. Opal, R. E. Rothman, F. R. D’Alessio, V. K. Sidhaye, 375 N. R. Aggarwal, R. Balk, J. A. Greenberg, M. Yoder, G. Patel, E. Gilbert, M. Afshar, J. P. Parada, G. S. Martin, A. 376 M. Esper, J. A. Kempker, M. Narasimhan, A. Tsegaye, S. Hahn, P. Mayo, T. van der Poll, M. J. Schultz, B. P. 377 Scicluna, P. K. Klouwenberg, A. Rapisarda, T. A. Seldon, L. C. McHugh, T. D. Yager, S. Cermelli, D. Sampson, 378 V. Rothwell, R. Newman, S. Bhide, B. A. Fox, J. T. Kirk, K. Navalkar, R. F. Davis, R. A. Brandon, R. B. Brandon, 379 Validation of a Host Response Assay, SeptiCyte LAB, for Discriminating Sepsis from Systemic Inflammatory 380 Response Syndrome in the ICU, American Journal of Respiratory and Critical Care Medicine 198 (2018), 381 doi:10.1164/rccm.201712-2472OC. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 382 36. R. K. Gupta, C. T. Turner, C. Venturini, H. Esmail, M. X. Rangaka, A. Copas, M. Lipman, I. Abubakar, M. 383 Noursadeghi, Concise whole blood transcriptional signatures for incipient tuberculosis: a systematic review and 384 patient-level pooled meta-analysis, The Lancet Respiratory Medicine 8, 395–406 (2020). 385 37. C. T. Turner, R. K. Gupta, E. Tsaliki, J. K. Roe, P. Mondal, G. R. Nyawo, Z. Palmer, R. F. Miller, B. W. Reeve, 386 G. Theron, M. Noursadeghi, Blood transcriptional biomarkers for active pulmonary tuberculosis in a high-burden 387 setting: a prospective, observational, diagnostic accuracy study, The Lancet Respiratory Medicine 8, 407–419 388 (2020). 389 38. N. L. Bray, H. Pimentel, P. Melsted, L. Pachter, Near-optimal probabilistic RNA-seq quantification, Nat 390 Biotechnol 34, 525–527 (2016). 391 39. C. Soneson, M. I. Love, M. D. Robinson, Differential analyses for RNA-seq: transcript-level estimates improve 392 gene-level inferences, F1000Research 4, 1521 (2016). 393 40. S. Durinck, Y. Moreau, A. Kasprzyk, S. Davis, B. De Moor, A. Brazma, W. Huber, BioMart and Bioconductor: 394 a powerful link between biological databases and microarray data analysis, Bioinformatics 21, 3439–3440 (2005). 395 41. E. R. DeLong, D. M. DeLong, D. L. Clarke-Pearson, Comparing the areas under two or more correlated 396 receiver operating characteristic curves: a nonparametric approach, Biometrics 44, 837–845 (1988). 397 42. A. K. Zaas, M. Chen, J. Varkey, T. Veldman, A. O. Hero, J. Lucas, Y. Huang, R. Turner, A. Gilbert, R. Lambkin- 398 Williams, N. C. Øien, B. Nicholson, S. Kingsmore, L. Carin, C. W. Woods, G. S. Ginsburg, Gene Expression 399 Signatures Diagnose Influenza and Other Symptomatic Respiratory Viral Infections in Humans, Cell Host & 400 Microbe 6, 207–217 (2009). 401 43. O. Ramilo, W. Allman, W. Chung, A. Mejias, M. Ardura, C. Glaser, K. M. Wittkowski, B. Piqueras, J. 402 Banchereau, A. K. Palucka, D. Chaussabel, Gene expression patterns in blood leukocytes discriminate patients 403 with acute infections, Blood 109, 2066–2077 (2007). 404 44. R. Pescarmona, A. Belot, M. Villard, L. Besson, J. Lopez, I. Mosnier, A.-L. Mathieu, C. Lombard, L. Garnier, 405 C. Frachette, T. Walzer, S. Viel, Comparison of RT-qPCR and Nanostring in the measurement of blood interferon 406 response for the diagnosis of type I interferonopathies, Cytokine 113, 446–452 (2019). medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 407 45. Y. Huang, A. K. Zaas, A. Rao, N. Dobigeon, P. J. Woolf, T. Veldman, N. C. Øien, M. T. McClain, J. B. Varkey, 408 B. Nicholson, L. Carin, S. Kingsmore, C. W. Woods, G. S. Ginsburg, A. O. Hero, Temporal dynamics of host 409 molecular responses differentiate symptomatic and asymptomatic influenza A infection, PLoS Genet 7, e1002234 410 (2011). 411 Funding 412 Funding for COVIDsortium was donated by individuals, charitable Trusts, and corporations including Goldman 413 Sachs, Citadel and Citadel Securities, The Guy Foundation, GW Pharmaceuticals, Kusuma Trust, and Jagclif 414 Charitable Trust, and enabled by Barts Charity with support from UCLH Charity. RKG is funded by National 415 Institute for Health Research (DRF-2018-11-ST2-004). JCM, CM and TAT are directly and indirectly supported by 416 the University College London Hospitals (UCLH) and Barts NIHR Biomedical Research Centres and through the 417 British Heart Foundation (BHF) Accelerator Award (AA/18/6/34223). TAT is funded by a BHF Intermediate 418 Research Fellowship (FS/19/35/34374). MN is supported by the Wellcome Trust (207511/Z/17/Z) and by NIHR 419 Biomedical Research Funding to UCL and UCLH. RJB/DMA are supported by UKRI/MRC Newton 420 (MR/S019553/1, MR/R02622X/1 and MR/V036939/1MR/S019553/1 and MR/R02622X/1), NIHR Imperial 421 Biomedical Research Centre (BRC):ITMAT, Cystic Fibrosis Trust SRC, and Horizon 2020 Marie Curie Actions. 422 MKM is supported by the UKRI/NIHR UK-CIC grant, a Wellcome Trust Investigator Award (214191/Z/18/Z) and a 423 CRUK Immunology grant (26603) AM is supported by Rosetrees Trust, The John Black Charitable Foundation, 424 and Medical College of St Bartholomew’s Hospital Trust. 425 Author contributions 426 CM, TAT, JCM and MN designed the study. JR, AC, MJ, GJ, MA, JL, TCM, CM and TAT acquired the data. RKG, 427 JR, LB and MN analysed the data. RKG, LB and MN designed and performed the systematic literature search. 428 RKG, JR, LB, DA, RB, MM, AM, BMC, CM, TAT, JCM and MN interpreted the data. RKG, JR, LB and MN wrote 429 the manuscript with input from all the authors. 430 Competing interests 431 The authors declare no conflict of interests. medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 432 Data sharing statement: 433 Applications for access to the individual participant de-identified data (including data dictionaries) and samples 434 can be made to the access committee via an online application https://covid-consortium.com/application-for- 435 samples/. Each application will be reviewed, with decisions to approve or reject an application for access made 436 on the basis of (i) accordance with participant consent and alignment to the study objectives (ii) evidence for the 437 capability of the applicant to undertake the specified research and (iii) availability of the requested samples. The 438 use of all samples and data will be limited to the approved application for access and stipulated in the material and 439 data transfer agreements between participating sites and investigators requesting access. 440 Data and materials availability 441 Open access to RNAseq data and associated essential metadata are available under accession no E-MTAB- 442 10022 at ArrayExpress (https://www.ebi.ac.uk/arrayexpress/). medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 443 Figures 444 Fig. 1. Correlation and Jaccard indices for all RNA signatures for viral infection included in analysis. (A) 445 Jaccard index intersect of constituent genes for all pairs of signatures clustered by Euclidean distance. (B) 446 Spearman rank correlation coefficients for all pairs of signatures clustered by 1-Spearmean rank distance (C) 447 relationship between pairwise Jaccard indices and Spearman rank correlation coefficients. (D) Network plot of 448 significantly enriched predicted upstream regulators by cytokine (red nodes), transmembrane receptors (purple 449 nodes), kinase (dark blue nodes) and transcription factors (light blue nodes) of all constituent genes in any 450 signature (black nodes). Size of upstream regulator nodes proportional to statistical enrichment (-log10 FDR). 451 Figure 2. Four best performing RNA signatures for discriminating participants with contemporaneous PCR-confirmed SARS-CoV-2 infection, 452 compared to uninfected controls. Shown as (A) Z scores stratified by interval to positive SARS-CoV-2 PCR; and (B) Z scores vs. contemporaneous 453 PCR cycle threshold for SARS-CoV-2 ORF1. NIC = non-infected controls; Conv = convalescent samples, collected at study week 24. AUC = area under 454 the receiver operating characteristic curve (95% CI) for discriminating participants with contemporaneous PCR-confirmed SARS-CoV-2 infection 455 (PCR+ve_0 group), compared to uninfected controls. R = Spearman rank correlation coefficient. 456 Tables Signature(s) Model& Discovery population(s) AndresTerres11 (23) Geometric mean of all genes (“influenza metasignature”) Five cohorts -children and adults with influenza; adults challenged with influenza; adults with bacterial pneumonia Henrickson16 (24) Difference in geometric means between upregulated and downregulated genes (“Influenza paediatric signature score”) Disease risk score† Four cohorts -children with influenza-like illness USA Children with viral bacterial infection or UK, USA Spain and IFI44L (14) N/A Children with viral bacterial infection (10) or UK, USA Spain and IFIT3; RSAD2* (22) N/A Three cohorts of adults challenged with rhinovirus, influenza or RSV (42) UK and USA Lopez7 (15) Sum of weighted gene expression values (“Bacterial versus viral classifier”) Logistic regression (“Viral classifier”)§ USA Herberg2 (10) Discovery setting(s) UK, USA Australia Discovery approach Validation population(s) Intended application Differential expression followed by leave-one-cohort-out strategy and filtering for heterogeneity of effect size, using genome-wide data Eight cohorts - children or adults with influenza or bacterial infection; adults challenged with influenza; adults vaccinated against influenza Two cohorts – children or adults with influenza. Influenza versus bacterial or other viral infection Elastic net followed by forward selection–partial least squares, using significantly differentially expressed transcripts Elastic net followed by forward selection–partial least squares, using significantly differentially expressed transcripts Sparse latent factor regression analysis on genome-wide data (42) followed by regularised logistic regression on the resulting 30-gene signature SVM using genome-wide data Children with bacterial or viral infection, inflammatory disease, or indeterminate diagnosis Children with bacterial or viral infection Viral versus bacterial infection in febrile children Close contacts of students with acute upper respiratory viral infections Pre-symptomatic viral infection versus healthy Children with acute viral or bacterial infections (43) Viral versus bacterial respiratory infection USA LASSO using 87 selected target genes from previously derived signatures (19, 21) Patients with viral/bacterial co-infection or suspected bacterial infection Viral versus bacterial respiratory infection N/A Pre-selected plausibility Viral infection verus healthy Differential expression and PAM classifier training using genomewide data Elastic net followed by forward selection–partial least squares, using significantly differentially expressed transcripts (10), followed by selection of an adequately expressed transcript for use in RTLAMP Adults challenged with the live yellow fever virus vaccine A second cohort of children with RSV infection Children with bacterial or viral infection Viral versus bacterial infection in children and Meta-analysis and leave-one-out strategy to identify common genes using genome-wide data MX1 (29) N/A Children and adults with viral, bacterial or noninfectious acute respiratory illness (19) Adolescents and adults with viral, bacterial or noninfectious acute respiratory illness N/A OLFM4 (25) N/A Children with RSV infection The Netherlands Pennisi2 (20) Disease risk score† Children with viral bacterial infection (10) UK, USA Spain Lydon15 (11) or and due to biological Influenza infection versus healthy Viral versus bacterial infection in febrile children Severity of RSV infection in children Sampson10 (13) Disease risk score (“Combined SeptiCyte score”) Eight cohorts of neonates, children and adults with bacterial infections UK, USA, Estonia and Australia Regression analysis of transcript pairs using the 6000 most highly expressed genes from each dataset Sampson4 (16) Disease risk (“Septicyte VIRUS”) USA, Finland Australia Brazil, and Regression analysis of transcript pairs using the 6000 most highly expressed genes from each dataset Sweeney11 (17) Difference in geometric means between upregulated and downregulated genes, multiplied by ratio of counts of positive to negative genes (“Sepsis metascore”) Difference in geometric means between upregulated and downregulated genes, multiplied by ratio of counts of positive to negative genes (“Bacterial/viral metascore”) Median expression of 6 interferon-stimulated genes (“Interferon score (44)”) Logistic regression (“Viral ARI classifier”)§ Ten cohorts - children and adults with viral infections. Two cohorts - adults challenged with influenza. Two cohorts - macaques challenged with Lassa or LCMV. Nine cohorts of patients with sepsis or trauma USA, Australia, Europe, Scandinavia, Canada, UK Greedy forward search of 82 differentially expressed genes identified by multicohort analysis Eight cohorts of children and adults with viral and bacterial infections USA, Australia, UK N/A N/A Children and adults with viral, bacterial or noninfectious acute respiratory illness, and healthy controls USA Children with acute respiratory illness and a positive result for a viral infection on a nasopharyngeal swab Two cohorts of adults challenged with influenza A H3N2 or H1N1 USA Modified supervised principal component analysis using all expressed transcripts USA Elastic net using 48 selected genes comprised of 29 derived as a signature in a previous study (42) , 7 shown to be downregulated in analysis of influenza challenge time course data (45) and 12 control genes Sweeney7 (12) TrouilletAssant6 (18) Tsalik33 (19) score Yu3; IFI27 (30) 1. Mean expression (“nonRSV infections vs. controls”)¶¶ 2. N/A Zaas48 (21) Probit regression classifier”)§ (“Viral Unselected consecutive patients presenting to the emergency department with febrile illness Seven human cohorts and six non-human mammal cohorts, infected or challenged with viruses across all seven of the Baltimore virus classification groups Twelve cohorts of adults with viral or bacterial sepsis, or trauma Viral versus bacterial in febrile patients Greedy forward search of 72 differentially expressed genes identified by multicohort analysis Twenty-four cohorts of children and adults with viral or bacterial infections, or healthy controls Viral versus infection Differential expression using 15 preselected interferon-stimulated genes LASSO using the 40% of microarray probes with the largest variance after batch correction Febrile children with bacterial or viral infection Viral versus bacterial infection in febrile children Five cohorts – children or adults with viral, bacterial or non-infectious respiratory illness, or viral/bacterial coinfection Children with RSV or rhinovirus infection Viral versus bacterial acute respiratory illness Adults presenting to the emergency department with fever and healthy controls Viral versus bacterial acute respiratory illness Viral versus conditions non-viral Viral or bacterial sepsis versus sterile inflammation Viral versus children bacterial healthy in 457 Table 1. Characteristics of whole-blood RNA signatures for viral infection included in analysis. Signatures are referred to by combining the first author's name 458 of the corresponding publication as a prefix, with number of constituent genes as a suffix. Log 2-transformed transcripts per million data used to calculate all signatures. 459 *Study by McClain et al. sought to validate a 36-transcript signature for detection of respiratory viral infections. Model coefficients for the 36-transcript model are not 460 provided; we therefore included the two best performing single transcripts from the study in the current analysis, since they demonstrated similar performance to the 461 full model in the original publication. 462 downregulated genes subtracted from the sum of upregulated genes. §Logistic and probit regression models were calculated on the linear predictor scale using model 463 coefficients from original publications. Abbreviations: RSV, respiratory syncytial virus. PAM, prediction analysis of microarrays. SVM, support vector machine. LASSO, 464 Least Absolute Shrinkage Selector Operator. RT-LAMP, Reverse Transcription Loop-mediated Isothermal Amplification. LCMV, lymphocytic choriomeningitis virus. &Where applicable, the name of the signature from the original publication is indicated in brackets. †Defined as the sum of medRxiv preprint doi: https://doi.org/10.1101/2021.01.18.21250044; this version posted January 20, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. 465 Signature AUROC Sensitivity Specificity p IFI27 0.95 (0.91 - 0.99) 0.84 (0.7 - 0.93) 0.95 (0.85 - 0.98) NA Sweeney7 0.95 (0.91 - 0.99) 0.82 (0.67 - 0.91) 0.95 (0.85 - 0.98) 0.851 Zaas48 0.93 (0.88 - 0.98) 0.61 (0.45 - 0.74) 0.95 (0.85 - 0.98) 0.088 Pennisi2 0.91 (0.86 - 0.96) 0.58 (0.42 - 0.72) 0.95 (0.85 - 0.98) 0.088 IFI44L 0.9 (0.84 - 0.96) 0.55 (0.4 - 0.7) 0.95 (0.85 - 0.98) 0.039 AndresTerre11 0.89 (0.83 - 0.95) 0.55 (0.4 - 0.7) 0.95 (0.85 - 0.98) 0.021 Henrickson16 0.89 (0.82 - 0.96) 0.55 (0.4 - 0.7) 0.93 (0.83 - 0.97) 0.009 TrouilletAssant6 0.87 (0.8 - 0.94) 0.53 (0.37 - 0.68) 0.93 (0.83 - 0.97) 0.008 Lydon15 0.86 (0.79 - 0.94) 0.58 (0.42 - 0.72) 0.95 (0.85 - 0.98) 0.005 Herberg2 0.84 (0.76 - 0.92) 0.5 (0.35 - 0.65) 0.93 (0.83 - 0.97) 0.003 Sampson4 0.84 (0.76 - 0.92) 0.5 (0.35 - 0.65) 0.93 (0.83 - 0.97) 0.003 Sampson10 0.83 (0.74 - 0.92) 0.5 (0.35 - 0.65) 0.95 (0.85 - 0.98) 0.002 RSAD2 0.83 (0.74 - 0.91) 0.47 (0.32 - 0.63) 0.93 (0.83 - 0.97) 0.002 MX1 0.82 (0.74 - 0.91) 0.45 (0.3 - 0.6) 0.95 (0.85 - 0.98) 0.002 Tsalik33 0.79 (0.7 - 0.89) 0.39 (0.26 - 0.55) 0.98 (0.9 - 1) 0.001 Lopez7 0.79 (0.69 - 0.88) 0.37 (0.23 - 0.53) 0.98 (0.9 - 1) 0.001 IFIT3 0.75 (0.64 - 0.86) 0.45 (0.3 - 0.6) 0.93 (0.83 - 0.97) 0.000 OLFM4 0.62 (0.51 - 0.74) 0.03 (0 - 0.13) 0.98 (0.9 - 1) 0.000 Sweeney11 0.6 (0.48 - 0.73) 0.16 (0.07 - 0.3) 0.96 (0.88 - 0.99) 0.000 Yu3 0.59 (0.47 - 0.71) 0.05 (0.01 - 0.17) 1 (0.93 - 1) 0.000 466 Table 2. Validation metrics of whole-blood RNA signatures for discrimination of participants with PCR- 467 confirmed SARS-CoV-2 infection at first week of PCR-positivity (PCR+ve_0). Discrimination is shown as area 468 under the receiver operating characteristic curve (AUROC). Sensitivity and specificity are shown using pre-defined 469 thresholds of 2 standard deviations above the mean of the uninfected control population (Z2). P values show 470 pairwise comparisons to best performing signature with Benjamini-Hochberg adjustment (false discovery rate 471 0.05). All metrics as shown as point estimates (95% confidence intervals). Equivalent table for discrimination of 472 participants with SARS-CoV-2 infection one week prior to PCR-positivity (PCR+ve_-1), compared to uninfected 473 controls, is shown in Table S3.