Rapid functionalisation and detection of viruses via a novel Ca2+-mediated virus-DNA interaction

Current virus detection methods often take significant time or can be limited in sensitivity and specificity. The increasing frequency and magnitude of viral outbreaks in recent decades has resulted in an urgent need for diagnostic methods that are facile, sensitive, rapid and inexpensive. Here, we describe and characterise a novel, calcium-mediated interaction of the surface of enveloped viruses with DNA, that can be used for the functionalisation of intact virus particles via chemical groups attached to the DNA. Using DNA modified with fluorophores, we have demonstrated the rapid and sensitive labelling and detection of influenza and other viruses using single-particle tracking and particle-size determination. With this method, we have detected clinical isolates of influenza in just one minute, significantly faster than existing rapid diagnostic tests. This powerful technique is easily extendable to a wide range of other enveloped pathogenic viruses and holds significant promise as a future diagnostic tool.

In this study, we have established a novel method to label and modify enveloped viruses using calcium chloride (CaCl 2 ) and DNA. By adding fluorophores to the DNA, we were able to efficiently image and characterize viruses by their size in a single-particle tracking assay. This approach allowed us to detect viruses more rapidly than the currently available tests (with our results being available within one minute), and worked on all enveloped viruses tested, including viruses present in clinical samples. No amplification or purification steps were required as viruses could be detected directly in complex solutions such as cell culture media, virus transport media and allantoic fluid. We were also able to immobilise modified viruses, directly assess aggregation of fluorescent virus particles, and count labelled viruses using a fluidics-based optical approach. We therefore provide a novel tool for the rapid functionalisation, detection and quantification of enveloped viruses.

Results
Calcium chloride labelling and single-particle tracking as a novel method for virus detection.
Our study began with a serendipitous discovery while investigating methods of fluorescently labelling intact influenza virus particles. We observed that brief incubations of virus particles with short, fluorescently-labelled, non-specific DNAs in the presence of high concentrations of CaCl 2 generated bright fluorescent particles that diffused slowly in solution, and resembled labelled viral particles (Movie S1). The fluorescent particles were extremely bright, pointing to the presence of multiple fluorescent DNA molecules per particle, and diffused much more slowly than free DNAs, which, due to their small size (only a few nm), diffused so rapidly that their motion blurred out and merely contributed to the overall background of the sample.
To establish whether the slowly diffusing particles were indeed labelled viral particles, we studied their mobility and number using single-particle tracking 9 . Fluorescently-labelled particles were observed using a widefield microscope with variable angle epifluorescence microscopy (VAEM) to reduce the background signal 10 (Fig. 1A) and single-particle tracking software was used to track the bright and slow-moving particles (Fig. 1B). By localizing visible peaks and associating nearby localizations in subsequent frames with each other, the two-dimensional paths of the diffusing particles were reconstructed for the duration of their visibility in the focal plane (~700 nm thick). The diffusion coefficients of the tracked particles were then plotted on a histogram and the mean diffusion coefficient (D m ) was used to give an estimate of the average diameter of the diffusing particles (d st ) using the Stokes-Einstein relation ( Fig. 1C and Eq. 5). Thus, the diffusion coefficients and Stokes' diameters obtained from single-particle tracking experiments could be used as an observable for particle detection.
Initially, we fluorescently labelled the H1N1 A/Puerto Rico/8/1934 (PR8) strain of influenza by adding 5.25 × 10 6 PFU/mL PR8 to 0.65 M CaCl 2 and 1 nM Atto647N-labelled single-stranded DNA (DNA 1, see methods), after which the sample was immediately transferred to the well of a glass slide placed on the microscope. Imaging of the sample showed the presence of multiple bright spots in the field-of-view (Fig. 1D, left panel). These spots were tracked (a 1000 frame acquisition at 30 Hz, taking 33 seconds) (Fig. 1D, right panel). Subsequent data analysis, taking less than 20 seconds, allowed diffusion coefficients to be calculated for each track (Fig. 1E). Fitting a Gaussian to the logarithm of the diffusion coefficients (D [µm 2 /s]), gave a mean diffusion coefficient log(D m ) of 1.42, which provided a Stokes' diameter of 134 nm. It should be noted that the broad distribution of diffusion coefficients arises from the higher uncertainty in analysing short tracks and is not necessarily indicative of a broad size distribution in the particles themselves 11 . The overall assay time took approximately 1 minute. Our results were also consistent with the tracking of commercial, fluorescently-labelled microspheres of a similar size (Fig. S1).
In order to verify the size and shape of a typical spherical influenza strain, we stained an A/Aichi/68 (X31) virus sample and imaged it using transmission electron microscopy (Fig. S2A,B). We measured the size (average of major and minor axes) (Fig. S2C) of 119 particles and obtained a broad distribution of virus sizes (Fig. S2D), with the largest number of particles (36) falling into the 120-140 nm diameter range. This result was consistent with the results from our single-particle tracking assay, as well as with previous TEM or cryo-electron microscopy studies of influenza particles in the literature, which show that spherical strains generally consist of small spheres or ovoid particles of approximately 120 nm in diameter [12][13][14][15][16] .
To confirm that the signal in our assay was specific to fluorescently labelled PR8 virus particles, we replaced the virus with water; no signal was observed in the absence of virus (Figs 1F and S3). Substitution of either the CaCl 2 or the fluorescent DNA with water also resulted in no signal, indicating that they were both essential requirements for fluorescence labelling to occur. Interestingly, addition of 0.25 M of the calcium chelator EDTA to the sample well during observation of fluorescently labelled virus particles immediately eliminated the signal, suggesting that the labelling is reversible (Figs 1F and S3). We further defined the minimum concentration of EDTA required to completely eliminate the signal to be approximately 0.048 M (Fig. S4), suggesting that the minimal CaCl 2 concentration required for efficient virus labelling is above 0.6 M. Taken together, we have established that the bright, slow-moving particles we observe after incubation of influenza virus with CaCl 2 and DNA are single, labelled viruses.
Virus labelling requires calcium chloride, a fluorescent nucleic acid and a viral envelope. To investigate the nature of the ionic species required for fluorescence labelling to occur, CaCl 2 was replaced with KCl, NaCl, MgCl 2 or the cationic polyamine spermine, which has an overall ionic charge of +4. No tracks were detected when KCl or NaCl were used, while a very small number of tracks were detected when MgCl 2 or spermine was used (an average of 6 and 25 tracks from a 1000-frame acquisition movie, respectively; Figs 2A and S5A). In contrast, when CaCl 2 was used, a large number of bright trackable particles were observed in the field of view (an average of 589 tracks from a 1000 frame acquisition movie; Figs 2A and S5A). This suggests that labelling of virus particles using calcium chloride and DNA must involve a physical mechanism of interaction specific to CaCl 2 .
We also found that it was possible to label virus particles with either fluorescent single-stranded DNA or RNA, but not with a fluorescently-labelled protein (the Klenow fragment of DNA Polymerase; Figs 2B and S5B), suggesting that efficient fluorescence labelling requires the presence of a nucleic acid. To eliminate the possibility that the nature of the dye was responsible for the labelling process, we attempted virus labelling and detection using unconjugated Atto647N dye not attached to a nucleic acid, which did not result in any fluorescent virus particles (Fig. 2B).
Next, in order to test the minimal length requirement of the nucleic acid required for efficient labelling, we tested fluorescently-labelled single-stranded DNAs of different lengths. We found that using short DNAs of 6, 9 or 12 nucleotides in length did not result in any appreciable virus labelling, while longer DNAs of 20, 45 or 64 nucleotides gave rise to bright fluorescent particles that could easily be tracked (Figs 2C and S5C). Whilst use of the 45mer provided the highest number of tracks overall (Fig. 2C), the signal-to-noise improves with increasing length of DNA (Fig. S5C). Overall, our results suggest that fluorescent labelling via CaCl 2 requires a nucleic acid greater than 12 nucleotides in length.
To study the requirement of a lipid membrane for efficient virus labelling, we tested a sample of non-enveloped adenovirus, which has a similar diameter to influenza, however no labelled particles were observed in the field-of-view and no tracks could be detected (Figs 2D and S5D). Next, we calcium-labelled 200 nm vesicles containing 25% anionic lipids (giving an overall negative charge), and found that we could observe and track bright particles diffusing in solution (Figs 2D and S5D). When cationic or neutral 200 nm vesicles were used in place of the anionic vesicles, no tracks were detected. Taken together, this strongly suggests that a negatively charged lipid bilayer is a necessary requirement for virus labelling to occur. We therefore propose a model for our labelling method that suggests that the Ca 2+ ions derived from calcium chloride facilitate an interaction between the negatively charged polar heads of the viral lipid membrane and the negatively charged phosphates of the nucleic acid (see discussion) (Fig. 2E).
www.nature.com/scientificreports www.nature.com/scientificreports/ Detection of clinical samples of influenza. In order to investigate the minimal virus concentration that we could reliably detect using our calcium-based labelling strategy, we tested increasing concentrations of virus. When virus was excluded from the sample (0 PFU/mL), no fluorescent particles were observed in the field-of-view; however, when PR8 virus was added, the number of fluorescent particles observed increased with increasing virus concentration (Fig. 3A). No tracks were detected in multiple acquisitions when virus was excluded from the sample, providing a value of zero for the 'limit of blank' (LOB); i.e., the highest number of tracks found when replicates of a blank sample containing no virus were tested 17 . A small increase in the number of tracks was observed at low concentrations of virus (below 3.5 × 10 4 PFU/mL), followed by an approximately linear increase for the higher concentrations tested (between 3.5 × 10 4 and 26.2 × 10 4 PFU/mL) (Fig. 3B). The 'limit of detection' (LOD; i.e., the lowest number of tracks likely to be reliably distinguished from the LOB and at which detection is feasible) was calculated as 1.86 tracks 17 , corresponding to a minimal virus titre of 2.2 × 10 4 PFU/mL (Fig. 3C).
Next, we examined clinical isolates of influenza. As we were expecting the virus concentration in clinical samples to be lower than laboratory-grown PR8, we detected viruses by adding two DNA species, labelled with either the red dye Atto647N (DNA 1) or the green dye Cy3B (DNA 2) simultaneously; this ensured that fluorescent viruses were concurrently labelled in both detection channels (that depend on excitation by 532-and 640-nm lasers; Figs 4A and S6A,B). We chose to do use this for the low concentration samples to give an unequivocal signal from actual virus particles (co-localised signals were more likely to be actual virus particles, as opposed to cellular debris etc. that might be labelled in just one colour). Using PR8 as an example, 86% of the detected virus particles were tracked in both channels simultaneously (Fig. S6C). As expected, when virus was excluded from the sample, no particles were observed in the field-of-view, and no tracks were detected (Fig. 3D,E). Tracking of PR8 virus resulted in a high number of tracks (average of 632 from a 1000-frame acquisition; Fig. 3E). Two independent clinical samples of A(H1N1)pdm09 (H1N1(i) and H1N1(ii)) and influenza B (Yamagata lineage) (Flu B) were also tested, and resulted in an average 119, 12 and 150 tracks per 1000 frame acquisition (Fig. 3E). Although www.nature.com/scientificreports www.nature.com/scientificreports/ the number of detected viruses was lower in the clinical samples than in the laboratory-grown PR8 stock, this was still well above the LOD of our assay, clearly establishing that our detection assay could be useful in a clinical diagnostic setting. We were able to carry out an analysis of the diffusion coefficients for two of the samples (H1N1(i) and Flu B), however for the third sample (H1N1(ii)), the low number of tracks meant that no size information could be obtained (Fig. S7). The mean diameters for H1N1(i) and Flu B were larger than expected, possibly due to aggregation (discussed further below).

Detection of multiple virus types and strains.
In addition to the PR8 strain analysed previously ( Fig. 1D-F), we were able to successfully detect influenza strains H1N1 A/WSN/1933 (WSN), H3N2 A/Aichi/68 (X31) and B/Florida/04/2006 (InfB) (Fig. S8). Viruses were detected in minimal essential media harvested from infected cells or allantoic fluid harvested from infected eggs, demonstrating the versatility of our assay for virus detection in different complex fluids. Interestingly, we found that different strains of influenza showed varying degrees of aggregation whilst diffusing in solution (for example, Fig. S8A-C shows several particles (circled) that visually appear brighter and larger than others in the field-of-view). Addition of trypsin reduced the aggregation of the virus particles, but did not eliminate it (Fig. S8). The larger Stokes' diameters obtained for some virus strains suggested that we were observing a heterogeneous population of diffusing species consisting of some monomeric particles and some small aggregates. This theory was supported by simulations characterising the motions of populations of diffusing viruses with different sizes (Fig. S9). Using the X31 virus sample as an example, which had a log(D m ) of 1.06 and mean size of 171 nm, we found that a simulation movie consisting of 30% 100 nm particles, 27% 200 nm particles, 23% 300 nm particles and 20% 400 nm particles gave a mean size of 178 nm when analysed (Fig. S9G). Comparison of the diffusion coefficient plots from the experimental and simulated data show an overall similarity in distribution and minimal differences in mean square deviation, while a comparison with simulated data from 100 nm particles alone did not closely resemble the experimental data (Fig. S9G).
In addition to being applicable to the detection of a broad range of influenza strains, we tested our labelling method on two other enveloped virus types, the spherical human respiratory syncytial virus (RSV) and the Autographa californica nuclear polyhedrosis virus (AcNPV), a rod-shaped baculovirus of insects. Imaging and tracking of RSV gave a mean diffusion coefficient log(D m ) of 0.94, which provided a mean diameter (d st ) of 195 nm (Figs 4A-C and S10A-D), in good agreement with the size expected from the literature 18,19 . Baculovirus particles could also be efficiently detected (Figs 4D,E and S10E-H). By modelling the viral particles as ellipsoids, with a characteristic rotational diffusion that was significantly smaller than the exposure time of our experiment (30 ms), we assumed that the measured motion was isotropic 20 . In this way, we were able to use the observed trajectories to calculate a diffusion coefficient histogram (Eq. 4, Materials and Methods) to give a mean diffusion coefficient log(D m ) of 0.26 (Fig. 4F). By assuming a mean axial ratio of 5.9 (calculated by dividing a mean value of 325 nm for the long axis of the baculovirus particle by 55 nm for the short axis 20 ), we obtained an approximate size of 91 nm by 536 nm for the baculovirus particles. Similarly to our experiments with influenza, over-estimation of virus size may result from aggregation. Nevertheless, our results have established the applicability of our virus detection methods to other enveloped viruses with a more complex geometry.

Downstream applications of calcium-labelled virus detection.
A simple and rapid method for fluorescent labelling of virus particles has multiple downstream applications. In order to demonstrate the applicability of our technique for virus counting 4,5 , we used CaCl 2 and fluorescent DNA to label and directly quantify single virus particles in a simple fluidics system. Fluorescent viruses were pumped through a microslide channel (100 µm high × 1 mm wide) and illuminated using a confocal microscope with alternating red and green laser excitation (Fig. 5A). For this purpose, the fiber-coupled outputs of a green (532 nm) and red laser (638 nm) were alternated with a modulation frequency of 10 kHz. Both beams were coupled into an inverted confocal microscope, and bursts of fluorescence corresponding to the passage of each virus particle through the confocal volume were split spectrally onto two avalanche photodiodes detecting red and green fluorescence. Fluorescence data recorded by the avalanche photodiodes was acquired using custom-written LabVIEW software. Analysis of the data was performed using custom-written MATLAB code, which implemented a threshold-based sliding window algorithm to detect fluorescent bursts above background (see materials and methods). www.nature.com/scientificreports www.nature.com/scientificreports/ As an initial proof-of-principle, we tested 110 nm fluorescent microspheres, which showed frequent bursts of simultaneous red and green fluorescence as the spheres passed through the illuminated volume (Fig. 5B). Next, we used our fluidics system to detect PR8 virus that was simultaneously labelled with green or red fluorophores www.nature.com/scientificreports www.nature.com/scientificreports/ using CaCl 2 . Only low-level background fluorescence was detected when virus was excluded from the sample (Fig. 5C, top panel), however, when labelled PR8 virus was flowed through the channel, bursts of fluorescence showing simultaneous peaks of green and red fluorescence were observed (>90% of detected bursts showed both red and green labelling) (Fig. 5C second panel and 5D). The frequency of bursts increased linearly with increasing concentrations of virus (Fig. 5C, third and fourth panels and 5E).
Both the microspheres and the virus particles were extremely bright, giving rise to many bursts exceeding 1000 counts (Fig. 5B,C). When we quantified the mean intensity of the detected fluorescent bursts, we found that the intensities of the virus particles and the fluorescent microspheres were within error (standard deviation from multiple acquisitions) (Fig. 5F). By adding a biotin modification to the fluorescent DNA used for virus labelling we were able to surface-immobilise particles and obtain an estimate of the number of fluorescent DNAs per virus particle (Fig. S11A). By integrating the area of each intensity peak we obtained a mean intensity value of 3134 a.u. for the virus particles (Fig. S11B,C). By comparison, single immobilised fluorescent DNAs were significantly less bright than immobilised viruses, with a mean intensity value of 94 a.u. (Fig. S11D-F). This gives an average number of 33 DNA molecules per virus particle. Overall, the brightness of the virus particles arising from our labelling method suggests that it may be useful for viral detection methods based on direct quantification of intact virus particles, such as flow cytometry.
To demonstrate that our non-specific labelling approach could be combined with specific detection techniques, we used CaCl 2 and DNA labelled with a green dye (Cy3B) to non-specifically detect immobilized virus particles, followed by staining of viruses with red fluorescent antibodies specific to the virus strain (Fig. S12A). Virus particles were visible in both the green (532 nm) and red (647 nm) channels, and the merged localisations showed many examples of co-localisation between the particles (Fig. S12B,C).
We also studied the ability of calcium-labelled influenza virus to bind to mammalian cells. After incubation of fluorescent PR8 virus with MDCK cells to allow adsorption, the cells were fixed, stained with DAPI to show cellular DNA in the cell nucleus, and imaged (Fig. S13). Small punctate spots were observed on the cell surface around the nuclei using 640 nm excitation, suggesting that fluorescent virus particles were able to attach to host cells (Fig. S13A,B). These spots could not be seen when either PR8 (Fig. S13C), CaCl 2 (Fig. S13D), or fluorescent DNA (Fig. S13E) were excluded from the experiment. Although it is unclear whether the labelled viruses enter the cells, their ability to bind to host cells suggests that labelled virus particles still maintain their structural integrity, and that the DNA does not completely obscure the viral haemagglutinin proteins required for host cell receptor binding on the virus surface.

Discussion
The interactions of calcium ions with lipid membranes have been probed by a variety of experimental methods, including fluorescent spectroscopy, dynamic light scattering and zeta potential measurements, in combination with molecular dynamics simulations 21 . It is generally accepted that the presence of calcium rigidifies and orders lipid bilayers, by causing conformational changes in lipid headgroups, ordering of acyl chains, and lipid dehydration 21 . Studies have demonstrated that calcium binds primarily to the phosphate groups of phospholipids, and simulations indicate that calcium is able to cluster phospholipid molecules via ion-bridges 22,23 , binding at least three lipid molecules concurrently 21,23 . It is plausible that in our experiments this stoichiometry is maintained, and one of the lipid molecules is replaced with a negatively charged phosphate group from the DNA/RNA (Fig. 2E). Further, our experiments suggest that the mechanism of calcium binding is not just charge-shielding, otherwise labelling would work just as well with magnesium or other positively charged ions. It is possible that, in addition to charge, the size and electrostatics of the metal ions may play a role in the efficiency of lipid binding and clustering. We cannot exclude the possibility that CaCl 2 treatment also weakens the viral membrane and creates small pores or invaginations 24 that allow internalisation the nucleic acid, however the instantaneous loss of fluorescent signal upon EDTA addition suggests that the labelling is more likely to be external to the virus particle.
Our method for functionalisation of virus particles has several advantages over existing virus detection techniques. Our assay provides a simple and rapid method for quantitating whole virus particles, which is particularly important in the production of inactivated vaccines where virus titre cannot be estimated using traditional infectivity assays. The method also represents a quick way of assessing virus aggregation, and our results suggest that influenza viruses tend to aggregate independently in solution, in a strain-specific manner. We have shown that viruses could be detected directly in complex biological fluids without the need for purification, and that no fluorescent particles were observed in virus-free minimal essential media, saliva from an apparently healthy individual or allantoic fluid from uninfected chicken eggs (Fig. S14). Labelling was instantaneous and the signal from diffusing fluorescent viruses was readily detectable in just 1 minute; this timescale is more than an order of magnitude shorter than the conventionally used fluorescent dyes used to label viruses for flow cytometry. In addition, the technique is simple and cost-effective, and does not require expensive reagents. Only a small sample volume (1-20 µl) is required, and virus can be detected without the need for an amplification step. We were able to detect influenza virus at a concentration of 2.2 × 10 4 PFU/mL, and RSV virus at a concentration of 7.0 × 10 2 PFU/mL, comparable or better than the estimated lower limits of existing detection methods (4.2 × 10 2 PFU/mL for RT-PCR and 10 6 -10 7 PFU/mL for rapid influenza diagnostic tests) 25 . We anticipate that the detection limit of our assay can be lowered further by taking longer acquisitions, with a small trade-off in overall assay time.
Compared with antibody staining of virus particles, the calcium labelling method is significantly faster, doesn't necessitate sample preparation, requires cheaper reagents, and can be combined with alternative detection methods should specific identification of virus particles be required. Thus, we envisage that our assay has potential as a standalone technique, for example, as a rapid way of immobilizing and visualizing virus particles at the same time, by using biotinylated fluorescent DNA to pull down virus particles onto a microscope slide for observation (as in Fig. S11). However, a further major advantage of the method is its compatibility with other techniques, such www.nature.com/scientificreports www.nature.com/scientificreports/ as existing virus purification or quantification assays. The method could feasibly be used as a rapid initial test to confirm the presence of a virus in a sample, which could then be followed up using antibody-specific labelling or RT-PCR if virus was detected.
In this study, we have demonstrated the use of CaCl 2 and DNA as a general biochemical tool for enveloped virus surface functionalisation, and demonstrated the feasibility of the approach in a rapid single-virus detection assay. The methods and analytical techniques developed here are applicable to the development of assays to study and detect many pathogenic viruses, and may be useful for the labelling or surface modification of any anionic lipid vesicles. Media, Gibco) was collected and the viruses were titred by plaque assay. Titres of WSN, PR8, Udorn and InfB were 3.3 × 10 8 plaque forming units (PFU)/mL, 1.05 × 10 8 PFU/mL, 1.0 × 10 7 PFU/mL and 2.1 × 10 8 PFU/ mL respectively. H3N2 A/Aichi/68 (X31) was grown in embryonated chicken eggs and titred by plaque assay (4.5 × 10 8 PFU/mL). The A2 strain of RSV was grown in Hep-2 cells and titred by plaque assay (1.4 × 10 5 PFU/ mL). All influenza and RSV virus stocks were inactivated by shaking with 0.2% formaldehyde for 24 hours before use. Recombinant baculovirus derived from the Autographa californica nuclear polyhedrosis virus was produced using the Multibac system 26 and was of unknown titre. Chimpanzee adenovirus ChAdOx1-GFP was grown in HEK293 cells and titred by plaque assay (1.1 × 10 12 PFU/mL) 27 .

Virus strains and clinical samples. H1N1
For clinical samples, residual material of oropharyngeal specimens from patients with influenza-like illness, which tested positive for influenza virus by RT-PCR, were used anonymously in this study. Typing and subtyping of influenza viruses was also conducted using RT-PCR, as previously described 28,29 . Specimens were kept frozen at −80 °C before shipment. In total, two independent clinical samples of A(H1N1)pdm09 and one of influenza type B (Yamagata lineage) were tested by the method.
The acquisition and preparation of clinical samples, which included obtaining informed consent from patients for use of the residual volume of their clinical samples for research purposes, was carried out by the National Influenza Reference Laboratory of Southern Greece, Hellenic Pasteur Institute. All samples had been previously received at the Laboratory for clinical diagnostic purposes, and they were de-identified for the purpose of this study. All methods were carried out in accordance with relevant guidelines and regulations, as approved by the Hellenic Pasteur Institute licensing committee. Single-stranded oligonucleotides labelled with either Atto647N, Cy3B or Cy3 dyes were purchased from IBA (Germany). Unless specified otherwise the DNA sequence used was a 64mer labelled with Atto647N (DNA 1). For dual-colour labelling a second 64mer DNA labelled with Cy3B (DNA 2) was used. The RNA was a 34mer labelled with Cy3. Short DNAs used in Fig. 2C were single stranded and labelled with either Atto647N or Cy3B. All sequences are provided in the supplementary information. The fluorescently-labelled protein was the Klenow fragment of E. coli DNA polymerase I 30 , site-specifically labelled at position 550 with ATTO647N and 744 with Cy3B.
Lipid vesicles of 200 nm in diameter were prepared as described previously 31 , using extrusion with a 200 nm pore size. Anionic lipid vesicles were composed of 75% 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 25% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate (POPA), cationic vesicles were composed of 50% DOPC and 50% Ethyl-phosphocholine, and neutral vesicles were composed of soybean phosphocholine. Vesicles at a concentration of 10 mg/mL were stored in 100 mM KCl (pH 7.5), 50 mM MOPS and 1 mM MgCl 2 , before being diluted, labelled and imaged. instrumentation. Single-particle tracking experiments were performed using wide-field imaging on a commercially available fluorescence Nanoimager microscope (Oxford Nanoimaging, https://www.oxfordni.com/). Briefly, a green (532 nm) and a red (635 nm) laser were combined using a dichroic mirror and coupled into a fibre optic cable. The fibre output was focused into the back focal plane of the objective (100x oil immersion, numerical aperture 1.4) and, for VAEM, displaced perpendicular to the optical axis resulting in a highly oblique subcritical incident angle on the sample to decrease background fluorophore excitation. Fluorescence emission was collected by the objective, separated into two emission channels and imaged onto a sCMOS camera (Orca flash V4, Hamamatsu). www.nature.com/scientificreports www.nature.com/scientificreports/ Data analysis. For single-particle tracking analysis, the NanoImager software suite was first used to localize the fluorescent molecules in each frame by finding intensity peaks that were significantly above background, then fitting the detected spots with a Gaussian function. The NanoImager single-particle tracking feature was then used to map trajectories for the individual virus particles over multiple frames, using a pre-defined maximum step distance (the maximum distance that a particle may travel between frames) and nearest-neighbor exclusion radius. These parameters were used to maximise the track count; the maximum step distance was used to exclude particles that were diffusing much more quickly than the predicted size of virus particles (e.g. free dye) and the exclusion radius was used to exclude tracks which crossed over, which could lead to ambiguity in trajectory assignment. The 110 nm and 46 nm microspheres were used to calibrate the optimum parameters as a function of the expected distance travelled between frames and the spatial density of localisations. The maximum step distance was taken as is the predicted mean squared displacement, and where D pred is the predicted diffusion coefficient of a spherical particle of the expected size in the solution of interest and t exp is the exposure time (typically 30 ms). The exclusion radius was taken as = .
Exclusion Radius N 0 1 1 (3) where N is the average number of particles per unit area, calculated by averaging the number of particles per unit area in each frame of a movie. The channel with the largest number of localisations was used to calculate the exclusion radius. In cases where the exclusion radius was smaller than the maximum distance, in samples with a high density of fluorescent particles for example, both parameters were set to the maximum distance. In the case of multicolour tracking, a particle observed in both channels was considered to be co-localized if simultaneous localisation in the red and green channels was separated by less than 200 nm. The trajectories were then used to calculate a diffusion coefficient for each track using the equation: where D is the diffusion coefficient, <(Δx) 2 > is the mean square displacement, Δt is the change in time and ϵ is a correction factor to account for a non-zero exposure time 32 . A histogram of the natural log diffusion coefficient for each track was then plotted using a custom-written MATLAB script. The histogram was fitted with a Gaussian function and the diffusion coefficient which corresponded to the curve peak was taken to be the sample mean (D m ). The Stokes' diameter (d st ) was then calculated using: where T is the temperature (in Kelvin), D m is the mean diffusion coefficient and η is the viscosity of the solution. The sample temperature was obtained from the microscope sensors and the viscosity of the solution was corrected for the presence of calcium chloride 33 .
Simulations. Movies simulating the unconfined diffusion of viruses were produced using existing software 34 , with a number of modifications. For each virus particle, an initial position and time at which the particle entered the field-of-view was generated randomly, with the bounds of diffusion set to beyond the field-of-view to simulate unconfined diffusion. The position of each particle was then calculated at 100 sub-steps per frame to simulate the effects of motion blur, using track lengths that followed an exponential distribution. Point-spread-functions were calculated for each localisation and added to a matrix that formed the final movie, before random noise was added to simulate fluctuations in the background photon intensity. The simulated movies were then analysed using custom-written MATLAB software. electron microscopy. H3N2 A/Aichi/68 (X31) virus was visualized by transmission electron microscopy at the Bioimaging Facility of the Sir William Dunn School of Pathology, University of Oxford. Virions were prepared as described previously 35 . Briefly, they were fixed with 4% paraformaldehyde, adsorbed onto grids, negatively stained with 2% aqueous uranyl acetate and imaged in a Tecnai 12 transmission electron microscope (FEI, Eindhoven) operated at 120 kV.
Virus counting. A custom-built confocal microscope was used for single-particle virus counting experiments [36][37][38] . Fluorescent microspheres with a diameter of 110 nm or fluorescently-labelled A/Puerto Rico/8/1934 virus was flowed through a microslide channel (100 µm high x 1 mm wide; Ibidi GmbH) using a syringe infusion pump (Harvard Apparatus, Pump 11 Elite) and illuminated using alternating red (647 nm) and green (532 nm) laser excitation at a modulation frequency of 10 kHz. Bursts of fluorescence corresponding to the passage of each virus particle through the confocal volume were split spectrally onto two avalanche photodiodes detecting red and green fluorescence. Fluorescence data recorded by the avalanche photodiodes was acquired using a custom-written LabVIEW virtual instrument into the single molecule (.sm) format, and converted to plain text using a second virtual instrument. Analysis of the raw time series data was performed using custom-written MATLAB code, which implemented a threshold-based sliding window algorithm to detect events, combined with a Gaussian peak fitting process to identify the height, width, and location of individual peaks. Using a trigger level, re-arming level, minimum event duration, maximum event intensity, and baseline input by the user, the procedure first subtracted the baseline. The trace was then scanned point-by-point, building a map of time points for which the fluorescence intensity exceeds the trigger level without re-crossing the re-arming threshold.
To minimise errors in the Gaussian peak fitting process for short-lived events, data were automatically extracted from either side of these events to ensure that a minimum of 10 data points were available to the peak fitting algorithm. Data from each event were fitted using a Gaussian peak fitting procedure, outputting the peak intensity, duration, and position within the trace of each fluorescent event. In rare instances, where the expansion of the event window caused overlap with neighbouring events, fitting errors may occur. To address this issue, the procedure identified overlapping regions, which were then merged. The peak fitting process then identified only the largest peak, and discarded the smaller peaks. Error checking was then performed, removing events that exceeded the maximum expected event duration, and intensity. Descriptive statistics for detected peaks were output for further analysis and presentation in Origin.
Intensity profile analysis. A/Puerto Rico/8/1934 (H1N1) virus particles were fluorescently labelled by incubation with 1 nM biotin-conjugated DNA labelled with Atto647N, before being immobilised on the surface of a pegylated glass slide treated with neutravadin. The slide surface was imaged using TIRF (total internal reflection fluorescence) with the laser illumination focused at a typical angle of 53° with respect to the normal. Intensity profiles from 20 pixel slices of the fields of view were taken using the built-in 'plot profile' feature on ImageJ. The intensity peaks were fitted using the 'fitpeaks' function in MATLAB, the local background was subtracted, and the peaks were integrated to give an estimate of the intensity of the peak. Multiple particles were analysed and the mean or median peak area of the resulting histogram was taken as the average particle intensity, which was compared between virus particles and free DNA.
Antibody and cell staining. Virus was incubated with 0.65 M CaCl 2 and 1 nM biotinylated Cy3B-labelled DNA before being immobilized via neutravidin on the surface of a pegylated glass slide. Viruses were fixed in 4% paraformaldehyde and permeabilised with 0.5% Triton-X-100, before being incubated with primary and secondary antibodies. A/Udorn/72 virus was stained with an anti-NA primary antibody and Alexa647-labelled secondary antibody and A/WSN/33 virus was stained with an anti-NP primary antibody and Alexa647-labelled secondary antibody. Localisations in each channel were identified using the NanoImager software suite. MDCK cells were grown on 13 mm coverslips in 24-well plates before being infected with A/Puerto Rico/8/1934 virus at a multiplicity of infection of ~20 × 10 6 PFU/mL. The virus was incubated with 0.65 M CaCl 2 and 1 nM Atto647N-labelled DNA before being added to the cells. The cells were incubated for 1 hour at 37 °C to allow viruses to adhere before being fixed for 15 minutes in 4% paraformaldehyde and mounted onto glass microscope slides with Mowiol containing DAPI. Cells were viewed using an Olympus Fluoview FV1200 microscope and images processed with ImageJ. Statistical analysis. The Limit of Blank (LOB) and Limit of Detection (LOD) were calculated as previously described 17 . The LOB was defined as the highest concentration found when replicates of a blank sample containing no virus were tested, and was calculated using: = + . * LOB mean(blank) 1 645 Standard Deviation(blank) (6) The LOD was defined as the lowest virus concentration reliably distinguished from the LOB and at which detection was feasible, determined by utilizing both the measured LOB and test replicates of a sample known to contain a low concentration of virus (0.6 × 10 4 PFU/mL).
(low concentration sample)