Introduction

Avian influenza viruses (IAV) are enveloped segmented negative-sense RNA viruses belonging to the Orthomyxoviridae family. They are worldwide present in a large diversity of bird species1,2. IAV are classified into subtypes based on their two surface proteins antigenicity, hemagglutinin (HA), and neuraminidase (NA). H5 and H7 subtypes can mutate from low pathogenicity avian influenza viruses (LPAIV) to high pathogenicity avian influenza viruses (HPAIV)3. During the low to high pathogenicity conversion, the HA cleavage site mutate from dibasic to multi-basic, enabling its cleavage by ubiquitous furin-like proteases present in the entire bird organism instead of respiratory and digestive-restricted trypsin-like proteases4.

Contrary to LPAIV, HPAIV infections are systemic and characterized by high mortality and transmission rates between birds and from one farm to another. An HPAIV epizootic can cause millions of birth deaths and induces major losses to the poultry industry. Even if LPAIV have a lesser impact on the poultry industry, their surveillance is also essential to prevent reassortment and emergence of new HPAIV subtypes5,6. Early detection and surveillance strategies must be developed to help control the spread of the viruses. In particular, rapid, simple, non-invasive, real-time, cost-effective, and field-deployable strategies must be investigated.

Human exhaled breath Volatile Organic Compounds (VOCs) analysis have successfully discriminated influenza A virus infections from other respiratory diseases such as human metapneumovirus, rhinovirus, or enterovirus in patients with flu-like symptoms7,8,9. VOCs changes were associated with oxidative stress and inflammatory biomarkers, but a specific virus-related VOCs signature was found. VOCs changes in exhaled breath of influenza A vaccinated patients have been established within the first two days following the vaccination7,10,11. Exhaled breath has also been studied in animals, such as swine12. In an experimental setting, H1N1-infected swine showed early VOCs pattern modification during asymptomatic infection. Individual or flock animal VOCs monitoring can also be used to detect infection13,14. All these studies strongly suggest that exhaled breath VOCs analysis upon influenza A infection could be an excellent diagnosis strategy. However, breath, animal, and flock VOCs analysis are complex with high host variability. To overcome these difficulties, studies have investigated tissues such as hair, fur, skin, or easy-to-collect samples like urine or feces15,16,17. In vitro studies with infected cell analysis have been the most used strategy to overcome host complexity and high variability for proof-of-concept studies18,19. Cell analysis allowed for rapid and more straightforward biomarker investigation in controlled experimental settings. In fact, VOCs from cell headspace successfully discriminated H1N1 IAV VOCs signature pattern from SARS-CoV-2, hCoV-NL63, or rhinovirus20,21. Prediction model based on selected VOCs signature can present high specificity and sensitivity for the characterization of new samples but also can predict the infection timeline20,21. Finally, it was shown that viruses present unique time-evolving VOCs signature22.

In the last decades, gas analysis of VOCs has been developed as an innovative diagnostic strategy15. VOCs analysis has been based on Mass Spectrometry (MS) methods, which allows for the detection of specific biomarkers or infection signatures. VOCs analysis from infected cells headspace to patient exhaled breath have been largely investigated in the last years, with a focus on the exhaled breath as it presents a non-invasive, easy-to-sample detection matrix of infection and fast detection strategy7,8,12,16,22.

Virus detection through VOCs biomarkers or infection signature analysis has been largely investigated using gas chromatography MS (GC–MS) strategies as it possesses high sensitivity biomarker detection. However, GC–MS often rely on solid phase microextraction (SPME), implying VOCs concentration and possible selective sampling depending on SPME characteristics23. Also, it is expensive and requires time-consuming steps from sampling preparation to biomarker quantification and identification24. Selected ion flow tube MS (SIFT-MS) technology enables high sensitivity, specificity, and repeatable large VOCs signature from complex gas mixture with no or limited sampling preparation25,26,27. Full scan mode allows for an untargeted large screen of variables which make this strategy well suited for the detection of unknown biomarkers or signatures on the contrary to the more common SIM mode strategy. In fact, SIM mode targets and quantify known compounds by selecting specific reagent and product ions while in full scan mode all precursor ions are used for the analysis. Over the past years, SIFT-MS has been used for real-time and online gas analysis26,27,28,29,30 and even in field utilization as the instrument is transportable31.

SIFT-MS analysis is a high throughput technology that can generate a large number of product ions26 for a limited number of samples. Data analysis strategy must be chosen carefully to avoid ill-posed problem (small number of samples but large number of variables)32 observable when the number of variables is much greater than the number of samples33. In addition, univariate methods analysis, such as ANOVA, which considers each variable individually, can lack precise biological phenomena description by missing out variable relation between each other. Multivariate analysis offers the possibility to obtain a more holistic description of the variable's relationship34,35. Supervised multivariate discriminant analysis such as sparse partial least square discriminate analysis (sPLS-DA) allowed for the investigation of variable signatures that describe the best categorical output of samples and enable categorical prediction of external samples based on selected variables36.

This study used a clade 2.3.4.4b H5N8 HPAIV/LPAIV pair to infect chicken cells for cell supernatant headspace VOCs analysis by SIFT-MS coupled with sPLS-DA analysis. Chicken cells were either mock-infected as a control or infected with the HPAIV or LPAIV virus in a total of five independent experiments (Fig. 1). Cell supernatants were collected at various time points from 1 to 72 h post inoculation (HPI), and VOCs were analyzed by SIFT-MS full scan mode. sPLS-DA, from the mixOmics package, was then applied in the first four experiments for sample discrimination based on their infection status and selected VOCs signature34. Product ion signatures were used for the prediction of the infection status of the fifth experiment samples (Fig. 1).

Figure 1
figure 1

Overview of the experimental protocol.

Full size image

We aimed to assess the ability of cell supernatant SIFT-MS coupled with sPLS-DA analysis to (1) discriminate IAV infected cells based on their infection status, (2) investigate the precocity of samples discrimination, and (3) predict the status of external samples based on selected VOCs signature. SIFT-MS could potentially represent a new strategy for real-time monitoring of viral infection.

Material and methods

Viruses

Segment sequences of the A/mulard duck/France/171201g/2017 (H5N8) HPAIV field isolate were used to produce a reversed genetic virus (H5N8/HP) (PUBMED accession number: MK859904 to MK859911). Based on these sequences information, a LPAIV was engineered using a site-directed mutagenesis (H5N8/LP) (PUBMED accession number: MK859926 to MK859933). Briefly, a 9-nucleotide deletion was performed on the HA cleavage site flanked with two single nucleotide polymorphisms37. Both viruses were propagated on 9–11 day-old specific-pathogen-free (SPF) embryonated chicken eggs (INRAE, PFIE, Nouzilly, France) by allantoic sac inoculation.

Both viruses were genetically identical, except for the sequence encoding the HA cleavage site, allowing us to compare two pathotypes of the same strain.

These viruses were produced and used only in biosafety level 3 laboratories at the National Veterinary School of Toulouse (ENVT) France.

Viral infection of chicken cell lines

DF-1 cell line of chicken embryo fibroblast were provided by the Friedrich Loeffler Institut. DF-1 were cultivated in Dulbecco’s modified Eagle’s medium (DMEM) high glucose with 10% fetal bovine serum (FBS) and 1% antibiotics (penicillin–streptomycin) at 37 °C with 5% CO2.

24H prior to infection, 9T-75 flasks were set with 7.106 cells/flask. On the day of infection, cell confluency and homogeneity between flasks were assessed. Flasks were randomly divided in three groups of three for the H5N8/HP, H5N8/LP, and the control group. Infection media was prepared using Opti-MEM supplemented with 0.5 mg/mL TPCK-treated trypsin. Infection media was used to prepare the inoculum and to mock viral infection for the control group. Before inoculation, flasks were washed once with phosphate-buffered saline (PBS) and then infected at a multiplicity of infection (MOI) of 10−5. After 1H, all inoculum were removed, and 25.5 mL of fresh infection media was added to each flask.

The experiment was repeated in an independent manner five times. The first four were used as training experiments, and the last was used as a test experiment. The test experiment with new cells was performed two months after the training experiment.

Cell supernatant collection

5.5 mL of cell supernatant was collected at 1H, 6H, 10H, 24H, 48H and 72H. 5 mL were placed in 20 mL glasses vial (ref 180420, BGB Analytik AG, Rohrmattstrasse, Switzerland) with silicone/PTFE septa (Ref 180301, BGB Analytik AG, Rohrmattstrasse, Switerland) and 0.5 mL in 1.5 mL collection tubes (Eppendorf, Hamburg, Germany).

Glasses vial were immerged 10 min at 60 °C in a water bath for viral decontamination before being stored at − 20 °C before being used for VOC’s analysis. Collection tubes were stored at − 80 °C before being used for vRNA detection and quantification.

RT-qPCR analysis

Total RNA from all collected samples was extracted using the magnetic bead-based kit ID Gene Mag Fast Extraction (Innovative Diagnostics, Grabel, France) associated with the Ideal 96 automated extraction robot (Innovative Diagnostics, Grabel, France) following manufacturer’s instructions. Detection and quantification of viral RNA (vRNA) were performed in a 10µL final volume using the Itaq SYBR green one-step RT-qPCR kit (Bio-Rad, Hercules CA, USA) following manufacturer instructions with the 5′-GACCTCTGTTACCCAGGGAGCCT-3′ and 5′-GGACAAGCTGCGCTTACCCCT-3′ forward and reverse primers, respectively, specific to both H5 hemagglutinin37. The absence of viral contamination was assessed by performing Itaq SYBR green one-step RT-qPCR (Bio-Rad, Hercules CA, USA) using HPAIV and LPAIV specific primers37.

vRNA statistical analysis

Welsh-adapted t-test was performed for vRNA data analysis.

SIFT-MS analysis

Sample preparation

Cell supernatant collected at 1H, 10H, 24H 48H, and 72H were selected for SIFT-MS analysis. Samples from all five experiments were injected in separate days. Each experiment samples were injected within 48H, or 72H. The injection order was randomly selected for each experiment.

12H before the start of the injection, samples were stored at 6 °C in a refrigerated room. 1H before their injection, samples were placed 30 min at room temperature to allow the VOCs to equilibrate in the vial headspace. Then, they were heated at 37 °C for 30 min in a dry block heater (Ohaus, Parsippany, USA). SIFT-MS injection was made while samples were still heated at 37 °C.

SIFT-MS analysis

VOCs measurements were performed using a Selected Ion Flow Tube Mass Spectrometer (SIFT-MS) voice 200 ultra (SYFT Technologies, Christchurch, New Zealand) equipped with a dual polarity source of positive and negative precursor ions (H3O+, NO+, O2+, O−, OH−, O2−, NO2−, and NO3−) set to full scan mode (from 15 to 250 m/z). Each full scan includes four successive full mass scans for a duration of 18 min. For each full mass scan, precursor ions were successively selected by a first quadrupole mass filter and injected to the flow tube with Nitrogen as carrier gas (Alphagaz, Air liquid, 99.9999%, Paris, France). Nitrogen flow rate was set at 2.0 Torr.L/s. Air samples were introduced in the instrument with a flow rate of 0.3 Torr.L/s. Dwell time limit was set to 5 ms. Precursor ions and air samples analytes reaction happened in a 199 °C flow tube kept at 0.006 kPa. The reaction generated product ions with specific mass-to-charge ratios (m/z), which were detected by a second quadrupole mass spectrometer. SIFT-MS was calibrated daily before sample analysis with a standard gas (Air liquid America, Specialty Gases LLC, Plumsteadville PA, USA). The calibration is validated by the instrument if the standard gas (i.e. 1,2,3,4 tetrafluoro benzene, benzene, ethylene, isobutane, octafluorotoluene, p xylene, perfluorobenzen, and toluene) concentration are detected at 2 ppm. Data acquisition and analysis were performed by the LabSyft 1.6.2 software (Syft Technologies, Christchurch, New Zealand).

Headspace from the 20 mL glasses vials containing cell supernatants were injected in the SIFT-MS through a customized injection line connected to a 5 mL sterile syringe with a 20G needle. Each sample was also connected to a Nitrogen filled Tedlar bag (ref 22050, Restek, Centre County PA, USA). Three Tedlar bags were used for H5N8/HP, H5N8/LP, and control cells group samples, respectively. Even though the Tedlar bags were not changed for the training experiment samples, the absence of contamination was assessed before each experiment. New ones were used for the test experiment samples.

Each day, after the instrument calibration, three blanks were performed, one for each sample group, using the appropriate Tedlar bags and empty 20 mL glasses vials.

VOCs analysis

General data pre-processing

Before complete statistical analysis, all samples from the five experiments were pre-processed altogether using R Statistical Software version 4.1.138. As full scan mode was set to four full mass scans, each sample presented four signal intensities for each product ion. To insure correct analysis, the first of the four full scan was removed as significantly different form the others. Then, the mean of the three last mass scans was calculated for each sample and each production. Therefore, we obtained only one signal intensity value for each product ions per sample. Secondly, background noise was removed by subtracting each corresponding blank from the samples. Negative obtained values were set to zero.

This resulted in an XY matrix with X being the samples (n = 225) and Y the product ions defined based on their precursor ions and m/z value (n = 1888). On these data, 28 ions were removed due to a significant clustering effect (Supp data Table 1).

Timepoint specific analysis

Five subset matrices were created based on sampling time (1H, 10H, 24H, 48H, or 72H) for discrimination and prediction analysis. For each subset matrix, all variables with an intensity value of 0 for all samples in at least one experiment were removed. Then, we used the Combat function39 from the sva packages40 to remove batch effect from the five experiments.

Finally, for all five subsets matrix, data were divided into a training data set and test data sets. The training data set regrouped samples from the first four experiments (n = 36), while the test data tested the samples from the fifth experiment (n = 9).

sPLS-DA analysis

Despite data pre-processing steps, the number of variables is largely more than the number of samples, which is unappreciated for standard analysis. Therefore, we decided to apply dimension reduction analytical methods. Additionally, as the aim of the study was to investigate sample discrimination based on categorical group and assess prediction samples based on selected variables, we decided to apply sparse partial least square discriminant analysis (sPLS-DA) using the user-friendly MixOmics package34,36. sPLS-DA was first used to select the most discriminant variables for group discrimination, and then, selected variables were used to predict the test samples.

The sPLS-DA parameters were tuned for each time point analysis. However, as all timepoint group samples are homogeneous in terms of number per group and number of groups, cross-validation parameters were set identically between all timepoint analyses. Folds were set to three, validation method to ‘Mfold’, distance to ‘max.dist’, nrepeat to 50.

Results

HPAIV and LPAI vRNA replication in cells

For cell supernatant headspace VOCs analysis by SIFT-MS, DF-1 cells were infected using the H5N8/HP and H5N8/LP viruses. A total of five independent experiment were performed. Systematically, each infection was performed in triplicate and a control group (mock cells) was used. The samples from the first four experiments were used as training data, and the fifth experiment as test data.

Following infection, viral replication was monitored and quantified by RT-qPCR. To allow correct replication of the H5N8/LP, all infection media were complemented with TPCK to investigate viruses or infection-dependent VOCs, and not the replication difference between viruses. RT-qPCR results show that the training samples share high similarities, and the vRNA kinetic pattern between viruses is identical (Fig. 2). Viral RNA was detectable from 10H; the maximum load was reached at 48H and stayed constant until 72H. Viral RNA load was significantly different (p < 0.05), by Welch’s adapted t-test, only at 24H for both training and test samples (Supp data Table 2). Viral RNA load differences between H5N8/HP training and test samples and H5N8/LP training and test samples were found significant at 24H, 48H, and 72H.

Figure 2
figure 2

Virus replication monitoring by RT-qPCR over time. Straight lines represent the training data, and the dotted lines represent the test data. For the training and test data, the color code is identical: grey for the control cells, orange for the H5N8/HP samples, and blue for the H5N8/LP samples. Error bars are SEM.

Full size image

sPLS-DA analysis

Sparse Partial Least Square Discriminate Analysis (sPLS-DA) analysis of VOCs was used to assess the possibility of sample discrimination based on their infection status (H5N8/HP, H5N8/LP or control cells), and study the selected variables as a signature (Supp data Table 3) for external samples prediction infection status. Pre-processed data training was used to select the best VOCs signature at each time point.

Training data set analysis

sPLS-DA results show that correct discrimination between groups was possible as early as 1H post-infection (Fig. 3). At 1H, the sPLS-DA three first components all together successfully separated samples (Fig. 3A,B). The control cells group could be discriminated over the infected cells (H5N8/HP and H5N8/LP) using the second components (Fig. 3A,B). At 10H, 24H, 48H, and 72H, the sPLS-DA first two components successfully discriminate the three groups. Systematically, the first component discriminates the H5N8/LP and control cell samples (Fig. 3C–F). H5N8/HP and H5N8/LP samples were discriminated with the first components at 10H, 48H, and, in theory, at 24H, but one sample was out of the 95% confidence ellipse (Fig. 3D). At 72H, H5N8/HP and H5N8/LP samples were discriminated with the second component, while the combination of all two components enabled the discrimination of all three groups. H5N8/HP and control cell groups were discriminated at 10H, 24H, and 48H based on the second component.

Figure 3
figure 3

Sample plots from sPLS-DA analysis performed on collected cell supernatants. (A,B) Cell supernatant samples collected at 1H projected into a 2-dimensional space defined by the first two components and by the first and third components, respectively. (C–F) Cell supernatant samples collected at 10H (C), 24H (D), 48H (E), and 72H (F) respectively. The 2-dimensional space is defined by the first two components of the sPLS-DA analysis. Ellipse represents 95% confidence intervals for each group.

Full size image

Variable analysis

The sPLS-DA performed on the training data selected the best components with the more appropriate product ions. Sample and selected product ions are represented in a Clustered Image Map (CIM) that enables the visualization of the sample classification per group as well as samples and variables relationship based on the VOCs signature (Fig. 4).

Figure 4
figure 4

Clustered Image Map from the sPLS-DA analysis performed on collected cell supernatants. (A–E) Clustered image map obtained from the sPLS-DA analysis performed on the cell supernatant collected at 1H (A), 10H (B), 24H (C), 48H (D), and 72H (E) respectively. Each row corresponds to one sample, from the training samples. Each column corresponds to one selected variable from the sPLS-DA first two components. Colors are defined by infection status: blue for the control cells, grey for the H5N8/LP samples, and orange for the H5N8/HP samples.

Full size image

At all-time points, except at 24H, all samples were correctly grouped by infection status. At 24H, one H5N8/LP sample was associated with the H5N8/HP samples. This is in agreement with Fig. 3D, which present a sample closely related to the H5N8/HP 95% confidence ellipse.

Interestingly, the sample dendrogram on the left on the CIM shows that H5N8/LP and control cells sample are closer than H5N8/HP at 1H, 10H, and 72H. Then H5N8/HP and control cells group are closer than H5N8/LP at 24H. At 48H, H5N8/LP and H5N8/HP are closer than control cells.

sPLS-DA theoretical performances

Cross-validation performed during the variable selection allows the evaluation of the theoretical performances of the model. Therefore, based on the selected components and variables, at each time point, it is possible to predict the classification error rates for external samples. The results, presented in Supp data Table 4, show that error rates per group are low, mostly below 10%.

The theoretical performances do not indicate that one group is easier to predict.

Test data prediction

VOCs selected signature was used to predict the infection status of the test samples as external samples. These samples (n = 9) shared the same pre-processing steps. The results, presented in Table 1, show that all samples were correctly assigned to the good infection status with the exception of one H5N8/HP infected sample, which was classified as control cells at 72H.

Table 1 Test data prediction using the select VOCs by sPLS-DA.
Full size table

These results show that at any timepoint, the selected VOCs signature enabled a sensitivity of 100% and a 100% specificity for the 1H to 48H time points and an 83.3% specificity for 72H. At 72H, the general accuracy of the selected signature was calculated to be 88.8%. General accuracy is defined as the probability for an unknown sample to be correctly classified. Additionally, positive and negative prediction values were calculated to be 100% and 75%, respectively.

At any time point, no classification error was made between the two viruses (Table 1). The signature strongly differentiated the pathogenicity of the virus used to infect the cells.

Finally, statistical analysis of all data regardless of the sampling timepoint and by dividing data into two groups, infected cells, and control group, present a 96.67% sensitivity, 100% specificity, general accuracy of 97.78%, a predictive positive value of 100%, and 93.75% predictive negative value.

Conclusion and discussion

In this study, headspace gas analysis of infected and non-infected cells supernatant was analyzed by SIFT-MS coupled with an sPLS-DA data analysis method. The aim was to investigate the ability of VOCs signature for sample discrimination based on their infectious status, the precocity of the possible discrimination, and external sample prediction. Samples discrimination not only aimed to differentiate infected and uninfected samples but also to assess the ability of VOCs to discriminate closely related viruses infection. Therefore, chicken cells were infected by a LPAIV or a HPAIV H5N8. Cells supernatants were collected at 1H, 10H, 24H, 48H, and 72H for both vRNA detection and quantification and SIFT-MS measures. VOCs signatures for sample discrimination and external sample prediction were selected using a sPLS-DA method34 based on the analysis of four independent experiments called training data. Selected VOCs signature at any time points were ultimately used for the prediction of a fifth independent experiment, called test data.

The results show a remarkable ability of sample discrimination as early as 1H post-infection. At any time points, the SIFT-MS VOCs analysis did not only discriminate infected cells from uninfected cells but successfully differentiated the two viruses, suggesting that even closely related viruses could be discriminated in the early phase of the infection. Even though viral entry happened during the first hour, and intracellular signaling pathways have ultimately been altered from the control group, knowing that some of the pathways are associated with specific VOCs10,18,22,41, questions are raised regarding potential contamination and unspecific virus and/or viral infection based VOCs. Notably, HPAIV and LPAIV were amplified on embryonated eggs prior to the cell infection. Even though allantoic fluids were diluted at least 10,000 times and only 50 μL of the diluted solution was used for the inoculation, VOCs associated with allantoic fluids could have been detected at the earliest time points. Feuerherd and colleagues used multi-capillary column ion mobility spectrometry (MCC-IMS) to investigate cell headspace discrimination of SARS-CoV-2 infection from H1N1 or hCoV-NL63 infections20. By monitoring headspace changes every 12.5 min for 72H, they came to the conclusion that discrimination was possible starting at three days post-infection. This raises questions regarding technology performances or the validity of our results in the early phases of the infection. In the same study, the authors used a decision tree model to predict sample discrimination.

The authors revealed that using all data for sample discrimination decreased the discrimination performances (sensitivity and specificity), probably due to viral, not specific cell apoptosis20. These findings align with our prediction performances that decreased at 72H. In fact, nine external samples per time point were classified based on VOCs signature, and only 1 sample was misclassified at 72H. This could suggest a decrease in virus-based VOCs signature specificity.

These results, which highlight 100% sensitivity and a global accuracy of 97.78%, suggest that the selected VOCs signature are strong. The performances of the prediction are high. These results suggest that SIFT-MS headspace analysis, coupled with sPLS-DA methods, has a strong capacity for infection detection.

Our data tends to validate the utilization of SIFT-MS coupled sPLS-DA analytical workflow for infected cell headspace discrimination. The workflow successfully identified VOCs signature associated with viral infection and, more importantly, this signature was virus-dependent. SIFT-MS enables fast, direct, simple, and cost-effective headspace analysis, which could be an excellent strategy for in-line analysis of viral infection. However, full scan analysis, as used in this study, does not allow for the specific identification of biomarkers in such complex gas mixture. More investigation should be performed to identify biomarkers. In particular, GC–MS technology, despite being less user-friendly and more expensive, is well dedicated to VOCs biomarker identification. VOCs analysis could also be associated with transcriptomic analysis to link VOCs with cell signaling pathways. MixOmics R package34 is designed for multiOmics data analysis and could associate virus-induced transcript to VOCs compounds and product ions. This strategy could better understand the VOCs used by SIFT-MS discriminant analysis.

Already, some investigations showed a relationship between VOCs and the TLR318. VOCs signature of airway cells specifically changed upon infection, and these changes were not observed when cells were treated with heat inactivated virus or TLR3 agonist. These suggest that VOCs changes are not associated with the presence of dsRNA or the specific activation of a cellular pathway.

VOCs analysis could be a valuable strategy for infection monitoring, first-in-line, for virus production, and then in field. In fact, viral production in the pharmaceutical world is essential for vaccine production. Early in-line detection of infection and surveillance of the replication is important to reduce production costs and increase effectiveness. Our results suggest that SIFT-MS could be directly implemented on a bioreactor for the detection of infection.

This work has successfully demonstrated that our SIFT-MS headspace analysis coupled with sPLS-DA approach could select VOCs signature, enabling cell supernatant discrimination between their infection status. The extreme sensitivity of the SIFT-MS and data analysis allows for the discrimination of cells infected with two closely related viruses in terms of sequences and cell pathogenicity. This suggests that VOCs signature can be specifically associated with a single virus. In addition, discrimination happened in the early phases of the infection, even before vRNA could be detected by gold-standard RT-qPCR.

More investigation must be pursued to identify VOCs specific biomarkers, but SIFT-MS headspace analyses offer a simple, fast, reliable VOCs based detection of infected samples. This seems to be first required steps before applying this strategy in field where VOCs contamination is important.