key: cord-0254265-pdpity08
authors: Kazmer, S.; Hartel, G.; Robinson, H.; Richards, R. S.; Yan, K.; Van Hal, S. J.; Chan, R.; Hind, A.; Bradley, D.; Zieschang, F.; Rawle, D. J.; Le, T. T.; Reid, D. W.; Suhrbier, A.; Hill, M. M.
title: Infrared spectroscopy enables rapid, robust, portable COVID-19 saliva screening based on pathophysiological response to SARS-CoV-2
date: 2021-12-24
journal: nan
DOI: 10.1101/2021.12.22.21268265
sha: ecaeca790dc712916e679b5279dbd7a57b911998
doc_id: 254265
cord_uid: pdpity08

Fourier-transform infrared (FTIR) spectroscopy provides a (bio)chemical snapshot of the sample, and was recently proposed for COVID-19 saliva screening in proof-of-concept cohort studies. As a step towards translation of this technology, we conducted controlled validation experiments in multiple biological systems. SARS-CoV-2 or UV-inactivated SARS-CoV-2 were used to infect Vero E6 cells in vitro, and K18-hACE2 mice in vivo. Potentially infectious culture supernatant or mouse oral lavage samples were treated with ethanol or Trizol to 75% (v/v) for attenuated total reflectance (ATR)-FTIR spectroscopy, or RT-PCR, respectively. The control condition, UV-inactivated SARS-CoV-2 elicited strong biochemical changes in culture supernatant/oral lavage despite lack of replication determined by RT-PCR or cell culture infectious dose 50%. Crucially, we show that active SARS-CoV-2 infection induced additional FTIR signals over the UV-inactivated SARS-CoV-2 infection, which correspond to innate immune response, aggregated proteins, and RNA. For human patient cohort prediction, we achieved high sensitivity of 93.48% on leave-on-out cross validation (n=104 participants) for predicting COVID-19 positivity using a partial least squares discriminant analysis model, in agreement with recent studies. However, COVID-19 patients negative on follow-up (RT-PCR on day of saliva sampling) were poorly predicted in this model. Importantly, COVID-19 vaccination did not lead to mis-classification of COVID-19 negatives. Meta-analysis revealed SARS-CoV-2 induced increase in Amide II band in all arms of this study and recent studies, indicative of altered {beta}-sheet structures in secreted proteins. In conclusion, ATR-FTIR is a robust, simple, portable method for COVID-19 saliva screening based on detection of pathophysiological responses to SARS-CoV-2.

Queensland, Australia, 4029 and 4072. Abstract 32 Fourier-transform infrared (FTIR) spectroscopy provides a (bio)chemical snapshot of the sample, 33 and was recently proposed for COVID-19 saliva screening in proof-of-concept cohort studies. As As an alternative to antibody-based rapid testing, Fourier transform infrared (FTIR) 68 spectroscopy was recently reported as a promising, point-of-care technology for COVID-19 69 detection using pharyngeal swab or saliva (3) (4) (5) . FTIR provides a biochemical snapshot of the 70 sample by measuring the vibration of chemical bonds (6). FTIR spectra collected from saliva of 71 COVID-19 patients and healthy controls were used to develop prediction algorithms that 72 demonstrated high predictive accuracy in cross-validation of the same cohort (4, 5) or in an 73 independent cohort (3). FTIR sampling using either transflection (slide mount), or attenuated 74 total reflectance (ATR, directly deposited on highly reflective crystal) was able to distinguish 75 healthy controls from confirmed COVID-19 cases with high specificity and sensitivity (3-5).

76 77

These recent cross-sectional cohort studies provided promising proof-of-concept for the 78 use of FTIR in COVID-19 screening using saliva as a non-invasive sample that can be self-79 collected. To further develop this technology towards point-of-care application, we generated 80 comprehensive data on the pathobiological basis underpinning the SARS-CoV-2/COVID-19 81 FTIR signal using a rapid and biosafe processing method. We utilized three different biological . CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

98 secretome ATR-FTIR spectra 99 100 As a first step, a standard Vero cell in vitro infection model was used to investigate the 101 secretory host response to SARS-CoV-2 infection. Two controls were used: a media control and 102 ultraviolet light (UV)-inactivated SARS-CoV-2, which cannot replicate as UV destroys RNA. 103 RT-qPCR of SARS-CoV-2 RNA of the culture supernatant confirmed the lack of infectivity for 104 both controls while the active infection demonstrated an increased SARS-CoV-2 RNA load at 24 105 and 48 hrs (Fig 1a) . 106 107 Interestingly, despite the 9 log increase in SARS-CoV-2 RNA at 24 hr, the FTIR spectra 108 of the supernatant showed minimal change at this time point, apart from increased absorbance at 109 Amide I band (1700-1600 cm -1 ) in the active SARS-CoV-2 infected sample (Fig 1b) . At 48 hr, 110 the FTIR profiles of UV-inactivated SARS-CoV-2 and active SARS-CoV-2 infected 111 supernatants showed increased bands at 2970 cm -1 , 2924 cm -1 , 2874 cm -1 , 1590 cm -1 , 1415 cm -1 , 112 and decreased at 1373 cm -1 , 1309 cm -1 , 1042 cm -1 , 988 cm -1 compared to media control (Fig 1b   113 and S1a Fig) . However, at 48 hr active SARS-CoV-2 infected secretome displayed separation 114 from both controls in Amide I/II (1700-1470 cm -1 ) and fingerprint (FP) region (1450-600 cm -1 ) 115 (Fig 1b) , as well as right shifting to a lower wavenumber at 1668 cm -1 to 1595 cm -1 (Fig 1d and An FDR LogWorth analysis confirmed significance in a number of these wavelengths 119 from both controls, shown as regions above the dotted line in Fig. 1c (p<0.001) . To clarify the 120 spectral changes per each condition over time, averaged spectra of the 48 hr time point were 121 subtracted from those at 24 hr (Fig 1d and Fig S1) . The greatest separations of spectra between 122 active SARS-CoV-2 and UV-inactivated SARS-CoV-2 occurred at 2977 cm -1 , 2920 cm -1 , 1668-123 1665 cm -1 , 1595 cm -1 , 1418 cm -1 , 1298 cm -1 , 1122 cm -1 , 1021 cm -1 , 854 cm -1 (Fig 1d and S1 124 Fig). Active SARS-CoV-2 infection demonstrated separation from media control at 1600 cm -1 , 125 1304 cm -1 , 1124 cm -1 , 1042 cm -1 , and 1023 cm -1 (Fig 1c,d) . These features notably included 126 increased absorbance at 1124 cm -1 , a region considered to reflect symmetric stretching of 127 phosphodiester linkages of RNA (8, 9)(ν s PO 2 -).

. CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Oral lavage was collected from anaesthetized mice prior to infection (day 0), and then on days 2 155 and 4 post inoculation (Fig 2a) . Body mass of active SARS-CoV-2 infected mice started 156 declining on day 3, reaching minus 10-15% on day 4 (Fig 2b) . Comparison of the average oral 157 lavage ATR-FTIR spectra showed more significant changes on day 4 compared to day 2 (S2 To further elucidate the pathophysiology, we conducted untargeted proteomics on the lavage. 172 The protein concentrations of the SARS-CoV-2 POS lavage was elevated in comparison to the 173 SARS-CoV-2 UV-I group, indicating strong secretory response to SARS-CoV-2 infection (Fig 2e) . 174 Proteomic analysis on equal amount of lavage protein revealed upregulation of several 175 kallikreins, and proteins involved in immune modulation such as lectin galactoside-binding 176 soluble 3 binding protein (Lgals3bp) and progranulin (Grn) (Fig 2f) . Furthermore, a number of 177 proteins were comparatively down regulated, notably including Calmodulin-3 (Calm3).

. CC-BY-NC 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. Table) .

The acquired spectra (n=3-6 technical replicates per biological sample) were baseline corrected 208 and normalized, then the technical variance in the dataset was assessed using pairwise Euclidean Of the 14 COVID.POS FU.NEG participants, only 2 (14.3%) were predicted correctly with the 232 remaining 10 (71.4%) and 2 (14.3%) predicted as COVID.NEG or COVID.POS (Fig 3d) . This is 233 not surprising as viral clearance dynamics are highly variable in individual post infection onset.

. CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10. 1101 /2021 Although visualisation of the average spectra revealed little separation between these subgroups, 235 a region we previously reported to correlate with COVID-19 disease severity (7) Comparison of the Logworth FDR analysis with the VIP analysis of the PLS-DA model (Fig 4c) 281 revealed that most of the predictive peaks overlap with the significant peaks from the 282 COVID.POS FU.POS vs COVID.NEG analysis (Fig 4b) .

283 284

. CC-BY-NC 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) preprint . CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10. 1101 /2021 Delineation of COVID-19 spectral signature across diverse models 296 297 Finally, we sought to establish the most characteristic COVID POS spectral signature across 298 multiple models and studies by comparing the significant results from all three study arms, as 299 well as the three recent publications (Table 1) , bearing in mind the differing nature of the models 300 and methodologies across studies. This analysis revealed several consistent spectral changes due 301 to SARS-CoV-2 infection across multiple models/studies (Table 1) . Most strikingly, a change in 302 the structure of proteins was indicated by Amide II increase in all studies, indicative of β -sheet 303 structures. In all human cohorts (but not in vitro or mouse models), Amide III, aliphatic, 304 phosphodiester asymmetric stretching (ν as PO 2 -) and saccharide bands were also increased. In 305 contrast, saccharide bands were decreased in in vitro and mouse models, with VIP values 1.2-306 1.43 over the range 1067-1006 cm -1 (Fig. 3b/e, Fig. S5b ). This range of wavenumbers was 307 previously recognized for having significance in predicting severe COVID-19 outcomes when 308 evaluating blood plasma, including an elevated AUC at 1592-1588 cm -1(7 The COVID-19 saliva FTIR signature shares many biochemical features with amyloid deposits 332 (aliphatic, amide, and phosphodiester ν as ) and lipofuscin (19, 30) . is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 with the ATR-FTIR bands for extracted SARS-CoV-2 RNA(46), in agreement with the PCR 368 results; however, this difference in saccharide absorbance may also indicate recovery from the 369 previously described, hyperglycemic state This group also showed reduced signal in the finger 370 print region, proposed by to represent immunoglobulins IgG, IgM and 371 IgA. As these COVID.POS FU.NEG patients are likely to have continued immunoglobulin 372 expression/secretion (47), our results suggest that the Amide I/II and fingerprint regions more 373 likely correlates with clearance of protein aggregates (β-sheet) and aliphatic amino acids as seen 374 by significant decreases at 1688 cm -1 and 1373 cm -1, respectively (Fig. 4b) . In contrast to previous fasting requirements of >8-hr prior to saliva collection (4), we took a 390 pragmatic approach of only 20-30 minutes abstinence from food prior to testing. Our results 391 support this time interval between sample collection and testing making a point-of-care rapid 392 testing application more feasible. It is unlikely that this time interval can be shortened further as 393 saliva is likely to be "contaminated" with food particles interfering with FTIR signals. We did 394 notice, however, excessive precipitation while mixing saliva with ethanol, secondary to the 395 initial high postprandial cephalic secretion. Adding a low-speed centrifugation of raw saliva 396 prior to inactivation with ethanol circumvented this problem. Our simple, inactivation procedure 397 with ethanol removes any possible biosafety concerns. All these features make 398 future development for point-of-care application feasible. However, saliva collection and 399 processing methods would require additional refinement (e.g. use of a capillary action sampling 400 cartridge).

In conclusion, ATR-FTIR technology with saliva self-collection provides a simple, rapid and 403 biosafe sample processing, which has high potential as a non-invasive, low-resource method for CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10. 1101 /2021 414 The SARS-CoV-2 isolate (hCoV-19/Australia/QLD02/2020) was kindly provided by Dr Alyssa 415 Pyke (Queensland Health Forensic & Scientific Services, Queensland Department of Health, 416 Brisbane, Australia). Virus stocks were prepared in Vero E6 cells as described (48) CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Samples from cohorts i) and ii) were stored on ice and processed within 30 minutes. After brief 542 vortex, an aliquot of raw saliva was transferred to a 1.5 ml Eppendorf tube and centrifuged for 10 543 minutes at 500x g at 4 C to remove particulates. Clarified saliva was transferred to a cryotube 544 containing ethanol to obtain 75% v/v ethanol and incubated at room temperature for 30 minutes.

. CC-BY-NC 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) preprint

The copyright holder for this this version posted December 24, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Inactivated saliva samples were stored at -80°C. For the TPCH COVID.POS samples, another 546 tube was prepared to 75% v/v Trizol for RT-PCR using the protocol described above.

Samples from cohort iii) were initially transported to the laboratory at room temperature. 549 Aliquots of raw saliva were frozen at -80 o C, subsequently thawed on ice, inactivated with 75% 550 v/v ethanol and shipped on dry-ice to QIMR Berghofer for FTIR analysis. A nasal pharyngeal 551 swab was collected on the same day as saliva, and was analysed by RT-PCR using TaqPath 552 COVID-19 Combo Kit (Thermo Fisher Scientific) according to manufacturer instructions. (Table S1 ). 556 557 ATR-FTIR spectra acquisition and processing 558 Samples in 75% ethanol were thawed on ice and homogenised by high speed vortexing. An 559 aliquot of 2 µL was applied to the crystal of an ATR-FTIR instrument (Agilent Cary 630). and 560 allowed to air dry (~30 sec) before spectral acquisition occurred over the wavenumber range, 561 4000-650 cm -1 . Background was collected without sample, i.e. ambient room air at 21 between 562 each measurement following cleaning of the crystal with 80% ethanol. Settings included 64 563 scans (Sample/Background) with a resolution of 8 cm -1 . All spectra were baseline adjusted with 564 baseline estimated using regions 2031-1865 cm -1 and 3971-3799 cm -1 . Spectra were then 565 normalised by adjusting to area under the curve (AUC) as 1.

Statistical analysis 568 Euclidean distance was calculated for each pairwise comparison of normalised spectra to 569 determine intra-and inter-sample variability. Each comparison was grouped into a "intra-570 sample" (spectra from same biological replicate, 1,970 comparisons) or "inter-sample" (spectra 571 from different biological replicate, 89,253 comparisons) category and represented as a violin 572 plot.

Clustering of the samples was explored using discriminant analysis to create a canonical plot to 575 display clustering of clinical groups. LogWorth statistic was applied to identify spectral regions 576 that significantly deviate between two sample groups. The false discovery rate p-value cut-off for 577 each comparison was chosen in a data-dependent manner accounting for the differences from CC-BY-NC 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) preprint 

Brisbane) for providing the SARS-CoV-2 598 isolates. We thank Clive Berghofer and the Brazil Family Foundation (and many others) for 599 their generous philanthropic donations to support SARS-CoV-2 research at QIMR Berghofer

We thank Peter Vardon, research nurse for coordinating the biological sample collection at The 601

Prince Charles Hospital. A.S. holds an Investigator grant from the National Health and Medical 602 Research Council (NHMRC) of Australia (APP1173880)

This project was supported by Agilent Technologies Applications & Core Technology -605

QIMR Berghofer Medical Research Institute COVID 606 Research (AS, DR) and The Prince Charles Hospital Research Foundation (DR)

FZ are employees of Agilent Technologies. All other authors declare that they have no

Culture supernatant ATR-FTIR spectra and subtractive analysis of Vero cell SARS-CoV

Full ATR-FTIR spectra and subtraction analysis of mouse oral lavage

Significant features of in vivo infection mouse model using LogWorth FDR analysis

Acceptable technical variance between replicates using pairwise Euclidean distancing 616 analysis of human samples

Human subject ATR-FTIR close-up of averaged spectra per three groups

AS 621 Data curation: GH, HR 622 Formal analysis: GH 623 Funding acquisition: DWR, AS, MMH 624 Investigation

SARS-CoV-2 Tests: Bridging the 643 Gap between Laboratory Sensors and Clinical Applications

point-of-care antigen and 645 molecular-based tests for diagnosis of SARS-CoV-2 infection

Ultrarapid On-Site 647 Detection of SARS-CoV-2 Infection Using Simple ATR-FTIR Spectroscopy and an Analysis Algorithm: High 648 Sensitivity and Specificity

ATR-FTIR spectrum analysis of saliva samples from COVID-19 positive patients

Infrared Based 653 Saliva Screening Test for COVID-19

ATR-FTIR spectroscopy as the future of diagnostics: a systematic review of the 655 approach using bio-fluids

Rapid Classification of 657 COVID-19 Severity by ATR-FTIR Spectroscopy of Plasma Samples

Application of Fourier Transform Infrared Spectroscopy for Tumor 659 Diagnosis

Spectroscopy as a Discriminatory Tool for Myotonic Dystrophy Type 1 Metabolism: A Pilot Study

K18-hACE2 mice 664 develop respiratory disease resembling severe COVID-19

SARS-CoV-2 infection in K18-ACE2 transgenic mice replicates human pulmonary 666 disease in COVID-19

Post-exposure protection of 668 SARS-CoV-2 lethal infected K18-hACE2 transgenic mice by neutralizing human monoclonal antibody

COVID-19 treatments and 671 pathogenesis including anosmia in K18-hACE2 mice

COVID-19 673 vaccine candidates based on modified vaccinia virus Ankara expressing the SARS-CoV-2 spike induce robust T-674 and B-cell immune responses and full efficacy in mice

A Potently Neutralizing Antibody 676 Protects Mice against SARS-CoV-2 Infection

A SARS-CoV-2 Infection 678 Model in Mice Demonstrates Protection by Neutralizing Antibodies

BET inhibition blocks 680 inflammation-induced cardiac dysfunction and SARS-CoV-2 infection

Contribution of Ribonucleic Acid 682 (RNA) to the Fourier Transform Infrared (FTIR) Spectrum of Eukaryotic Cells

In situ characterization of 684 protein aggregates in human tissues affected by light chain amyloidosis: a FTIR microspectroscopy study

On the Protein Fibrillation Pathway: Oligomer Intermediates 687 Detection Using ATR-FTIR Spectroscopy

The importance of hydration and DNA conformation in interpreting infrared spectra of cells and 689 tissues

Fourier transform infrared (FTIR) spectroscopy of biological tissues

Application of FTIR Spectroscopy to Analyze RNA Structure

Observation of Potential Contaminants in Processed 695

A library of IR bands of nucleic acids in solution

Infrared Spectroscopy, Gas Chromatography/Infrared in Food Analysis

Saliva as a Noninvasive Specimen for 700 Detection of SARS-CoV-2

Self-Collected Oral Fluid Saliva 702

Is Insensitive Compared With Nasal-Oropharyngeal Swabs in the Detection of Severe Acute Respiratory Syndrome 703 Coronavirus 2 in Outpatients

Comparison of Saliva and Nasopharyngeal Swab Nucleic Acid 705

Amplification Testing for Detection of SARS-CoV-2: A Systematic Review and Meta-analysis

EGCG Inhibited Lipofuscin Formation Based 708 on Intercepting Amyloidogenic beta-Sheet-Rich Structure Conversion

Likelihood of amyloid formation in COVID-19-induced ARDS

Presence of a SARS-CoV-2 Protein Enhances Amyloid 712 Formation of Serum Amyloid A

Serum amyloid A concentrations, COVID-19 severity 714 and mortality: An updated systematic review and meta-analysis

Seeding Brain Protein Aggregation by SARS-CoV-2 as a Possible 716 Long-Term Complication of COVID-19 Infection

SARS-CoV-2 spike protein interactions with amyloidogenic proteins: Potential clues to 718 neurodegeneration

SARS-CoV-2 ORF8 Forms Intracellular 720 Aggregates and Inhibits IFNgamma-Induced Antiviral Gene Expression in Human Lung Epithelial Cells. Front 721 Immunol

Roles of Factor XII in Innate Immunity

Expression) as Well as Potential Anti-Corona Viral Activity of the Marine Secondary Metabolite Polyphosphate on 727 A549 Cells

Coagulation factor XII in thrombosis and inflammation

Platelet polyphosphates are 730 proinflammatory and procoagulant mediators in vivo

Polyphosphate as a Target for Interference With Inflammation 732 and Thrombosis

Lectin Galactoside-734 binding Soluble 3 Binding Protein (LGALS3BP) Is a Tumor-associated Immunomodulatory Ligand for CD33-735 related Siglecs

Acute and long-term 737 disruption of glycometabolic control after SARS-CoV-2 infection

Hyperglycemia in acute 739 COVID-19 is characterized by insulin resistance and adipose tissue infectivity by SARS-CoV-2

A simple and 741 fast spectroscopy-based technique for Covid-19 diagnosis

Delayed production of neutralizing 743 antibodies correlates with fatal COVID-19

A versatile reverse genetics 745 platform for SARS-CoV-2 and other positive-strand RNA viruses

Heat shock protein 10 747 inhibits lipopolysaccharide-induced inflammatory mediator production

 594 We thank all participants for donating their time and saliva samples to this research. From 595 QIMR Berghofer Medical Research Institute we thank Dr Itaru Anraku for managing the PC3 596