Optical characterization of native aerosols from e-cigarettes in localized volumes

Measuring the size distribution of aerosols typically requires processing a sample, specifically to adjust the particle concentration to an adequate level. Unfortunately, this manipulation can significantly alter the native composition of some aerosols, which can lead to unreliable or even unusable measurements. We demonstrate that coherence-gated dynamic light scattering is suitable to measure the size distribution of native aerosols without the need for sample processing. Another novel aspect of the present work is the first demonstration of these type of localized light-scattering-based measurements in aerial media. Measuring the size distribution reliably in optically dense aerosols is possible thanks to the interferometric amplification of single scattering in an optically isolated, picolitre-sized coherence volume. We carried out proof-of-concept experiments in aerosols from electronic cigarettes, which poses a challenge mainly due to their high concentration, volatility, and hygroscopicity. We generated aerosols using two common moisturizers, propylene glycol and glycerol, and measured their particle size distribution as a function of the burning power. The aerosols generated in the presence of glycerol are more polydisperse and have larger particles with increasing burning power. This unique characterization of native aerosols can provide valuable information for dosimetry and hosting sites in the respiratory system.


Introduction
Aerosols are dynamic systems consisting of solid or liquid particles suspended in a gaseous phase and are ubiquitous in nature even in the most hostile and pristine conditions.Their composition i.e., chemical content, concentration, and particle size distribution, constitute a valuable source of information about their material origin, the generating mechanism, and their potential impact [1].Among these properties, the particle size distribution has been crucial in many applications, including environmental, pharmaceutical, and biomedical.For instance, in aerosols of natural origin such as clouds, haze, and fog, the particle size distribution largely determines the atmospheric scattering characteristics [2], which are critical for weather and communications applications.Notably, the size distribution of biological aerosols has been used for the identification of aging biomarkers and aging-related diseases [3] as well as for the construction of metabolic tissue profiles in cancer research [4].
Measuring the particle size distribution of aerosols typically requires a processing stage, which may include filtering, diluting, staining, or labeling the particles, in order to prepare the sample for the equipment, specially to adjust the particle concentration to a suitable level [5].Unfortunately, this manipulation can significantly alter the native composition, which can cause not only inconsistencies among measurements and techniques but, more importantly, to produce unreliable or even unusable outcomes.
In this regard, aerosols of anthropogenic origin such as the smoke from electronic cigarettes (EC), or vapers, are a clear example where measuring the native composition at the emission site is desirable but, at the same time, it poses a significant challenge [6].These aerosols consist of liquid particles, that is, nanoscopic droplets; they are generated with a high particle concentration and their native composition is highly susceptible to change due to their high volatility, coalescence, condensation, and hygroscopicity.Nevertheless, measuring their size distribution is fundamental to understanding their vaporization characteristics [7], transport mechanisms [8], and deposition in the respiratory system [9][10][11].
Among different alternatives, optical characterization techniques, especially those based on light scattering, are preferred mainly due to their potential for noninvasive measurements [5,12].Indeed, dynamic light scattering (DLS) has been used to measure the particle size distribution in relation to aerosols in different settings.For instance, DLS measurements have been performed with commercial equipment on aqueous suspensions that are engineered to be considered optically equivalent to the aerosol of interest, and the results are extrapolated to describe the characteristics of the aerosol [13].Another approach consists of collecting particles from the aerosol under study, drying them, and resuspending them in a suitable liquid medium, at convenient concentrations, to perform DLS measurements with standard equipment [14,15].More realistic DLS measurements have been performed on the actual aerosol, but they involve complicated, custom setups for collecting the aerosol, adjusting its concentration, and working with transparent containers to be able to perform free-space measurements [16][17][18].
In this work, we carried out novel proof-of-concept experiments where we measured the size distribution of native aerosols from EC using coherence-gated dynamic light scattering (CG-DLS), without the need for sample processing.Thanks to the heterodyne amplification of single scattering in localized, pico-litter-sized coherence volumes, this technique has proved useful in a number of scenarios involving optically dense aqueous media, including colloidal suspensions [19,20], hydrogels [21,22], and blood [23,24]; in this regard, another novelty of the present work is that it constitutes the first demonstration of CG-DLS in aerial media.We performed our measurements in freshly collected aerosols, at the native concentrations, requiring only a short stabilization time to ensure Brownian dynamics.We generated aerosols with a commercial-grade EC using two common moisturizers, propylene glycol and glycerol, and measured the dependence of their particle size distribution on the burning power.The results show that the aerosols generated in the presence of glycerol are more polydisperse and have larger particles with increasing burning power.This unique characterization can provide valuable information of dosimetry and potential hosting sites in the respiratory system.

Experimental setup
DLS, also referred to as photon correlation spectroscopy (PCS) and quasi-elastic light scattering (QELS), is an optical technique that allows measuring the dynamic properties of systems composed of particulate constituents suspended in a medium [25].Essentially, the sample is illuminated, and the intensity fluctuations of the light scattered from a spatially limited volume within the sample are detected at a known angle.The Brownian motion of the suspended particles induces random fluctuations in the phase of the scattered waves thus resulting in a stochastic light intensity signal.The intensity fluctuations are characterized typically by means of their intensity autocorrelation function, whose characteristic decay time encodes the diffusion coefficient of the particles and can be used to retrieve their size distribution.
CG-DLS is a variant of DLS where spatial and temporal coherence gates are incorporated to optically isolate a picolitre-sized volume [19].The practical implementation of CG-DLS consists of an optical fiber-based, common-path interferometer operating in reflection, as illustrated in then digitized with a data acquisition card (National Instruments model NI DAQ 9205).
The spectrum of the broadband light source has approximately a Gaussian envelope; therefore, its coherence length can be estimated as   = 2ln(2)   2 0 (∆) [26], which results in   ≈ 40 μm for aerial media ( ≈ 1).Considering the size of the core of the MMFs used (50 μm), the coherence volume from where the scattering signal is extracted, is on the order of   ≈ 80 pL.In such small volume, single scattering amply dominates in most practical scenarios; see details below for the present study.At the same time, high sensitivity is provided by the heterodyne amplification.These unique attributes have permitted, among other things, particle sizing in the range from a couple tens of nanometers up to a couple of microns, over several decades of volume concentrations, which can be as high as 10%vol [19,27].Nevertheless, care should be exerted at low concentrations because the volume observed in CG-DLS is around two orders of magnitude smaller than that in commercial DLS equipment.As illustrated in Fig. 1, the intensity detected is a time-fluctuating signal that results from the interference of a reference field,   , arising from the Fresnel reflection at the end facet of the MMF, and the back-scattered field,   , corresponding to light that was singly scattered by the aerosol particles in the coherence volume and coupled back to the MMF.Because |  | ≫ |  |, the intensity detected can be expressed as () ≈   +2    cos(∆()), where  , is the time-averaged intensity of the reference and scattered field, respectively; ∆() is the random phase difference between the two interfering fields, whose temporal variations are determined by the moving particles.The baseline, or average, intensity,   ≈   , provides optical information about the reflectivity of the fiber-medium interface, as it is largely determined by   .The main outcome of a CG-DLS measurement is the power spectrum of the light intensity fluctuations, (), which is the Fourier counterpart of the intensity autocorrelation function measured in DLS by virtue of the Wiener-Khinchin theorem [25].The spectrum of the broadband light source has approximately a Gaussian envelope; therefore, its coherence length can be estimated as l c = √︁ 2ln(2)/π (λ 2 0 /(n∆λ)) [26], which results in l c ≈ 40 µm for aerial media (n ≈ 1).Considering the size of the core of the MMFs used (D = 50 µm), the coherence volume from where the scattering signal is extracted, is on the order of V c ≈ 80 pL.In such small volume, single scattering amply dominates in most practical scenarios; see details below for the present study.At the same time, high sensitivity is provided by the heterodyne amplification.These unique attributes have permitted, among other things, particle sizing in the range from a couple tens of nanometers up to a couple of microns, over several decades of volume concentrations, which can be as high as 10%vol [19,27].Nevertheless, care should be exerted at low concentrations because the volume observed in CG-DLS is around two orders of magnitude smaller than that in commercial DLS equipment.

Aerosol extraction and collection system
As illustrated in Fig. 1, the intensity detected is a time-fluctuating signal that results from the interference of a reference field, E r , arising from the Fresnel reflection at the end facet of the MMF, and the back-scattered field, E s , corresponding to light that was singly scattered by the aerosol particles in the coherence volume and coupled back to the MMF.Because |E r | ≫ |E s |, the intensity detected can be expressed as , where I r,s is the time-averaged intensity of the reference and scattered field, respectively; ∆ϕ(t) is the random phase difference between the two interfering fields, whose temporal variations are determined by the moving particles.The baseline, or average, intensity, I avg ≈ I r , provides optical information about the reflectivity of the fiber-medium interface, as it is largely determined by I r .The main outcome of a CG-DLS measurement is the power spectrum of the light intensity fluctuations, P(f ), which is the Fourier counterpart of the intensity autocorrelation function measured in DLS by virtue of the Wiener-Khinchin theorem [25].

Aerosol extraction and collection system
EC are nicotine delivery devices that have become popular for tobacco control, especially to decrease cigarette consumption and help quit smoking, with potentially reduced health risks [28][29][30].Briefly, their principle of operation relies on the partial burning of the smoking liquid produced by the Joule effect (heat dissipation in a conductor due to a circulating electrical current) that takes place in an electrical coil, which is immersed in the liquid reservoir.In general, the smoking liquid in a EC consists mainly of a moisturizer, typically a mixture of propylene glycol and glycerol, which constitutes 99% of the composition, nicotine and flavoring [31].The aerosol forms when the outgoing vapor condenses rapidly in contact with air, forming small liquid particles that can remain suspended in air.Because EC aerosols originate from a partial burning, their process of production is rather chaotic and results in highly-concentrated, polydisperse aerosols, with concentration on the order of 10 9 particles/cm 3 , whose size distribution is sensitive to the generation conditions [6,31] i.e., liquid composition and burning power, which may determine the ultimate deposition site [10,11].
Aerosols were generated with a commercial EC (Vaporesso model Gen Nano, 3.5 mL, 5-80 mW).Sample collection was done in two steps, extraction and storage, using a disposable syringe (60 mL).The process of aerosol extraction, Fig. 2(a), starts by placing the empty syringe, whose plunger is all the way in, right above the EC mouthpiece.Then, the EC is turned on and the plunger is retracted at a flow rate of 20 mL/s, until the syringe is full.Importantly, these extraction parameters (speed and syringe volume) allow preserving the integrity of the native composition [32].
EC are nicotine delivery devices that have become popular for tobacco control, especially to decrease cigarette consumption and help quit smoking, with potentially reduced health risks [28][29][30].Briefly, their principle of operation relies on the partial burning of the smoking liquid produced by the Joule effect (heat dissipation in a conductor due to a circulating electrical current) that takes place in an electrical coil, which is immersed in the liquid reservoir.In general, the smoking liquid in a EC consists mainly of a moisturizer, typically a mixture of propylene glycol and glycerol, which constitutes 99% of the composition, nicotine and flavoring [31].The aerosol forms when the outgoing vapor condenses rapidly in contact with air, forming small liquid particles that can remain suspended in air.Because EC aerosols originate from a partial burning, their process of production is rather chaotic and results in highly-concentrated, polydisperse aerosols, with concentration on the order of 10 9 particles/c m 3 , whose size distribution is sensitive to the generation conditions [6,31] i.e., liquid composition and burning power, which may determine the ultimate deposition site [10,11].
Aerosols were generated with a commercial EC (Vaporesso model Gen Nano, 3.5mL, 5-80 mW).Sample collection was done in two steps, extraction and storage, using a disposable syringe (60 mL).The process of aerosol extraction, Fig. 2(a), starts by placing the empty syringe, whose plunger is all the way in, right above the EC mouthpiece.Then, the EC is turned on and the plunger is retracted at a flow rate of 20 mL s, until the syringe is full.Importantly, these extraction parameters (speed and syringe volume) allow preserving the integrity of the native composition [32].As soon as the syringe is full of aerosol, the EC is turned off and the syringe is rotated to a horizontal position, as illustrated in Fig. 2(b).At this point, the endoscopic MMF probe of the CG-DLS setup has been previously prepared and is ready to be introduced through the syringe's opening.The fiber holder itself seals the container once the fiber is in place.With this procedure, the native aerosol can be extracted, stored, and measured without the need of sample processing; only a short stabilization time of a couple of minutes is needed to ensure the Brownian diffusion regime of the aerosol particles.As soon as the syringe is full of aerosol, the EC is turned off and the syringe is rotated to a horizontal position, as illustrated in Fig. 2(b).At this point, the endoscopic MMF probe of the CG-DLS setup has been previously prepared and is ready to be introduced through the syringe's opening.The fiber holder itself seals the container once the fiber is in place.With this procedure, the native aerosol can be extracted, stored, and measured without the need of sample processing; only a short stabilization time of a couple of minutes is needed to ensure the Brownian diffusion regime of the aerosol particles.

Data processing and information retrieval
As mentioned before, DLS operates under the premises of single scattering.This condition is amply fulfilled in our experiments.Previous studies report that aerosols emerge from EC with concentrations in the order of 10 9 particles/cm 3 [6,31].By taking an average particle diameter of d = 500 nm, this concentration is equivalent to a volume fraction of ϕ ≈ 5 × 10 −5 , approximately, such that the interparticle distance is much greater than ten diameters.Moreover, at the wavelength of operation (λ 0 = 670 nm), the refractive index of propylene glycol and glycerol is n ≈ 1.44 and n ≈ 1.47, respectively [33].Following Ref. [34], the photon scattering length, l s = 2d/(3Q s ϕ), where Q s is the scattering efficiency, is estimated to be l s ≈ 2.8 mm and l s ≈ 2.5 mm for the propylene glycol and glycerol aerosols, respectively.Even with this crude estimation, the scattering length is already around two orders of magnitude larger than the coherence length; thus, within the extent of the coherence volume, only single scattering takes place.Furthermore, if the calculation of l s is refined by including the anisotropy factor, g ≈ 0.65 for both aerosols, and by taking the backscattering efficiency instead, Q bs <Q s , the effective scattering length, l * = l s /(1 − g), increases approximately by a factor of fifteen for both aerosols thus making it to be even larger than the coherence length.Therefore, in the conditions of our experiments, single scattering amply dominates in the picolitre-sized coherence volume of CG-DLS; thus, the information retrieval can be done as in traditional DLS as follows.
Data were acquired at a sampling frequency of 20 kHz (spectra resolved up to 10 kHz) and integration time of 1 second for 10 minutes; that is, 600 spectra were recorded per measurement.Each spectrum was decomposed into a collection of discrete Lorentzian components, , where f is the frequency; ν i and a i is the corner frequency and the relative amplitude of each Lorentzian function, respectively, with ∑︁ N i=1 a i = 1.Two Lorentzian components (N = 2) were sufficient to describe the envelope of P(f ) in all cases; one of them carried most of the weight while the second one adjusted only a small portion of P(f ) close to the noise floor.The corner frequency of P(f ) is the inverse of the characteristic autocorrelation time, τ, measured in DLS [25], which can be used to estimate the diffusion coefficient of the diffusing particles, D eff = 2π/(q 2 τ), and its corresponding aerodynamic diameter, d a = k B T/(3πηD eff ), where q is the magnitude of the scattering vector q = 4πn sin(θ/2)/λ 0 , with n being the refractive index of the suspending medium, θ the scattering angle (n ≈ 1 and θ = π rad in our case), and λ 0 is the free-space wavelength of the incident light; k B T is the thermal energy, with k B being Boltzmann's constant and T the absolute temperature; and η is the viscosity of the medium where the particles are suspended.

Results and discussion
As mentioned before, aerosols from EC consist of liquid nanodroplets and are polydisperse.
In this regard, it should be noted that, although monodisperse samples are desirable for DLS measurements, DLS tolerates some degree of polydispersity.The polydispersity index (PDI) of the particle size distribution, defined as the square of the ratio of the standard deviation to the mean, is a dimensionless metric commonly used in DLS to evaluate its broadness.Based on the guidelines for commercial equipment, PDI<0.05 are typical of a highly monodisperse samples such as size standards; conversely, samples with PDI>0.7 are considered unsuitable for DLS due to the extreme broadness of their size distribution.As it will be shown later, 0.03 ≤ PDI ≤ 0.08 in our experiments; thus, the samples are suitable for DLS measurements but are not strictly monodispersed, similar to previous studies on aerosols from EC [16][17][18].Because of this, we first evaluated the instrument's intrinsic variability and the integrity of the facet of the MMF during the measurement.
To disclose the intrinsic variability of the CG-DLS instrument, we measured the size distribution of a size standard, an aqueous colloidal suspension of monodisperse polystyrene particles (Sigma-Aldrich 90517-5ML-F) with both CG-DLS and a commercial DLS equipment (Malvern Zetasizer model Nano ZS 5600), as shown in Fig. 3(a).The particles have a nominal average size of 100 nm, and the concentration was adjusted to 0.05 vol%, in deionized water (Milli-Q); this concentration was chosen arbitrarily to be able to perform measurements on the same sample in both instruments.In the case of DLS, the size distribution is retrieved by the equipment based on the cumulants analysis of a single, time-averaged intensity autocorrelation function; in the case of CG-DLS, we constructed the size distribution from the collection of hydrodynamic sizes retrieved from the analysis of all the spectra recorded.Specifically, long-term CG-DLS measurements were performed (600 spectra over 10-minute measurement) for the collection of spectra to reflect a good ensemble average of the sample's composition; the aerodynamic size associated with each spectrum was estimated, and then the entire collection of values obtained was approximated to a probability distribution of sizes.This will become more evident later.In both cases, the measurements were done with similar settings (integration time of 1 second for a total measurement time of 10 minutes).In Fig. 3(a), the red curve is the size distribution measured with the DLS commercial equipment, showing an average and standard deviation of 110.5 nm ± 4.9 nm; the blue curve corresponds to the Gaussian fit of the size distribution measured with CG-DLS, whose average and standard deviation is 109.4nm ± 7.1 nm.In both cases, the size distribution retrieved falls well within the manufacturer's tolerance.Based on our measurements, the PDI values obtained for CG-DLS and DLS were 4.0 × 10 −3 and 5.0 × 10 −3 , respectively, indicating the highly monodisperse nature of the sample.The main conclusion from this sanity check, is that CG-DLS has similar performance to a commercial DLS equipment; therefore, it can resolve the size distribution of monodisperse samples but, more important for our purposes, a broad particle size distribution can be reliable regarded to the polydispersity of the sample, and not to instrument's uncertainty.
particles (Sigma-Aldrich 90517-5ML-F) with both CG-DLS and a commercial DLS equipment (Malvern Zetasizer model Nano ZS 5600), as shown in Fig. 3(a).The particles have a nominal average size of 100 nm, and the concentration was adjusted to 0.05 vol%, in deionized water (Milli-Q); this concentration was chosen arbitrarily to be able to perform measurements on the same sample in both instruments.In the case of DLS, the size distribution is retrieved by the equipment based on the cumulants analysis of a single, time-averaged intensity autocorrelation function; in the case of CG-DLS, we constructed the size distribution from the collection of hydrodynamic sizes retrieved from the analysis of all the spectra recorded.Specifically, longterm CG-DLS measurements were performed (600 spectra over 10-minute measurement) for the collection of spectra to reflect a good ensemble average of the sample's composition; the aerodynamic size associated with each spectrum was estimated, and then the entire collection of values obtained was approximated to a probability distribution of sizes.This will become more evident later.In both cases, the measurements were done with similar settings (integration time of 1 second for a total measurement time of 10 minutes).In Fig. 3(a), the red curve is the size distribution measured with the DLS commercial equipment, showing an average and standard deviation of 110.5 nm ± 4.9 nm; the blue curve corresponds to the Gaussian fit of the size distribution measured with CG-DLS, whose average and standard deviation is 109.4nm ± 7.1 nm.In both cases, the size distribution retrieved falls well within the manufacturer's tolerance.Based on our measurements, the PDI values obtained for CG-DLS and DLS were 2.0 × 10 -3 and 5.0 × 10 -3 , respectively, indicating the highly monodisperse nature of the sample.The main conclusion from this sanity check, is that CG-DLS has similar performance to a commercial DLS equipment; therefore, it can resolve the size distribution of monodisperse samples but, more important for our purposes, a broad particle size distribution can be reliable regarded to the polydispersity of the sample, and not to instrument's uncertainty.Next, to make sure that the facet of the MMF is not affected during the measurements, we used of the average intensity as a metric of the reflectivity, , at the end facet of the fiber, i.e.,   =   ∝ .Besides testing the integrity of the fiber's facet, this measurement also verifies that the magnitude of the local oscillator remains invariant during the measurements; in other words, if the Fresnel reflection at the fiber-medium interface is altered e.g., by particles depositing on its surface, the magnitude of the local oscillator changes and the fluctuations Next, to make sure that the facet of the MMF is not affected during the measurements, we used of the average intensity as a metric of the reflectivity, R, at the end facet of the fiber, i.e., I avg = I r ∝ R.Besides testing the integrity of the fiber's facet, this measurement also verifies that the magnitude of the local oscillator remains invariant during the measurements; in other words, if the Fresnel reflection at the fiber-medium interface is altered e.g., by particles depositing on its surface, the magnitude of the local oscillator changes and the fluctuations would be amplified by a different factor, giving the erroneous impression that the scattering signal, I s , is varying.Figure 3(b) shows the average power measured while the fiber enters and exists the container full of aerosol.First, in regions A and E, the fiber probe is static and sits outside the container, in air; the reflected power is similar, indicating that I r did not change.Erratic measurements are obtained when the fiber is moving, either entering (region B) or exiting (region D), the container.Finally, region C shows that stable measurements can be performed when the fiber sits static inside the container, in the aerosol.The inset shows pictures of the facet of the MMF, where a high contrast can be observed between the core and the cladding before and after entering the container.This simple visual inspection confirms that the aerosol's material did not deposit on the optical fiber.The main conclusion from this sanity check, is that the facet of the MMF is not compromised during the measurement in aerosols.
Figure 4 shows a full measurement of the particle size distribution of a native aerosol generated from propylene glycol (Azumex USP CAS 57-55-6) at a burning power of 30W.Each point in the main plot was obtained by processing power spectra recorded during the integration time of one second for ten minutes (see details in the methodology).As observed previously, during the first minutes (region shaded in red) the measurement is unfeasible due to the aerosol is transitioning from the directional transport regime that governs its emission to a diffusion regime inside the container.
exiting (region D), the container.Finally, region C shows that stable measurements can be performed when the fiber sits static inside the container, in the aerosol.The inset shows pictures of the facet of the MMF, where a high contrast can be observed between the core and the cladding before and after entering the container.This simple visual inspection confirms that the aerosol's material did not deposit on the optical fiber.The main conclusion from this sanity check, is that the facet of the MMF is not compromised during the measurement in aerosols.
Fig. 4 shows a full measurement of the particle size distribution of a native aerosol generated from propylene glycol (Azumex USP CAS 57-55-6) at a burning power of 30W.Each point in the main plot was obtained by processing power spectra recorded during the integration time of one second for ten minutes (see details in the methodology).As observed previously, during the first minutes (region shaded in red) the measurement is unfeasible due to the aerosol is transitioning from the directional transport regime that governs its emission to a diffusion regime inside the container.
Once the aerosol has stabilized, the particle size distribution can be measured reliably.In fact, the shape of the envelope of () in CG-DLS serves as an inherent verification that the diffusion regime has been reached.In a DLS measurement, the intensity autocorrelation function follows a negative exponential when the particles diffuse normally [25].In this regime, the decay time, , encodes the effective diffusion coefficient of the population of particles (see details in the Methodology).In the presence of directional motion, the shape of the intensity autocorrelation function deviates from a negative exponential due to the Gaussian contribution of the flow [35].Similarly, a Lorentzian shape of the envelope of () in CG-DLS, which is the Fourier counterpart of the negative exponential measured in the time domain in DLS, indicates that the particles are diffusing in a Brownian regime.In all cases, the envelope of  () was described by a single Lorentzian function without evidence of distortions due to a directional flow.
As a side note, we noticed that, because aerosols from EC are hygroscopic (due to the moisturizer particles absorb water from the ambient), for the calculation of the aerodynamic diameter both the temperature and relative humidity (RH) need to be known in order to associate the correct value of the air viscosity.All our experiments were performed at 25°C and 60%RH; at these conditions, the viscosity of air is  ≈ 0.6 mPa • s [36,37].No sample processing is required, only a couple of minutes for the aerosol particles to stabilize in a diffusion regime (shaded in red).The inset shows the particle size distribution measured when the fiber sits stably in the aerosol (shaded in green) and its corresponding Gaussian fit.
Once the aerosol has stabilized, the particle size distribution can be measured reliably.In fact, the shape of the envelope of P(f ) in CG-DLS serves as an inherent verification that the diffusion regime has been reached.In a DLS measurement, the intensity autocorrelation function follows a negative exponential when the particles diffuse normally [25].In this regime, the decay time, τ, encodes the effective diffusion coefficient of the population of particles (see details in the Methodology).In the presence of directional motion, the shape of the intensity autocorrelation function deviates from a negative exponential due to the Gaussian contribution of the flow [35].Similarly, a Lorentzian shape of the envelope of P(f ) in CG-DLS, which is the Fourier counterpart of the negative exponential measured in the time domain in DLS, indicates that the particles are diffusing in a Brownian regime.In all cases, the envelope of P(f ) was described by a single Lorentzian function without evidence of distortions due to a directional flow.
As a side note, we noticed that, because aerosols from EC are hygroscopic (due to the moisturizer particles absorb water from the ambient), for the calculation of the aerodynamic diameter both the temperature and relative humidity (RH) need to be known in order to associate the correct value of the air viscosity.All our experiments were performed at 25°C and 60%RH; at these conditions, the viscosity of air is η ≈ 0.6 mPa • s [36,37].
The inset of Fig. 4 shows the particle size distribution obtained from the collection of aerodynamic diameters retrieved using all the spectra recorded when the aerosol is stable (region shaded in green), and its corresponding Gaussian fit (continuous blue curve).
Figure 5 shows the Gaussian-fitted particle size distributions of native aerosols with different composition, generated at different burning powers.Three moisturizers were tested: propylene glycol (Azumex USP CAS 57-55-6), glycerol (CTR Scientific CAS 56-81-5), and a 1:1 mixture of both.The burning power was swept from 30W to 60W in steps of 10W; the range explored was determined by the specifications provided by the manufacturer to ensure a proper operation of the electrical coil in the EC.The table in each panel shows the average particle, the standard deviation, and the PDI obtained for each burning power (R 2 >0.85 in all cases).The particle size distributions measured in our experiments are in the range from 100 nm to 1 µm, centered at around 400-500nm.We note that our measurements lie within the range of different reports in the literature.For instance, using a combination of condensation particle counting and scanning mobility spectrometry, distributions below 100 nm, centered at around 50 nm, were reported [10]; using aerosol mobility spectrometry, particle size distributions centered at around 200 nm, were measured [11]; using low pressure impactors together with gravimetry, size distributions centered a larger values, at around 1 µm, were reported [38].Importantly, in all these studies diluted aerosols were used, which can affect significantly the size distribution measured depending upon the dilution ratio [11,38].Again, in our case, the The first thing to note, is that these aerosols have moderate to high polydispersity according to their PDI values.Propylene glycol aerosols (Fig. 5(a)) exhibit an average particle size in the range from 0.390 µm to 0.402 µm without a well-defined dependence on the burning power; the variations are within 3%, approximately, which allows concluding that the average particle size does not depend on the burning power for these aerosols.The standard deviation increases marginally with burning power, giving rise to slightly higher PDI values.Glycerol aerosols (Fig. 5(b)), on the other hand, exhibit a clear dependence on the burning power: both the average particle size and the standard deviation increase with increasing burning power.The average particle size increases from 0.414 µm to 0.497 µm when the burning power increases from 30W to 60W (20% increase approximately); similarly, the standard deviation increased by around 25%.This composition leads to the most polydisperse aerosols.Aerosols generated with a mixture of propylene glycol and glycerol, Fig. 5(c), exhibit a similar behavior to the case of pure glycerol: the average particle size increases by 19%, approximately, from 0.412 µm to 0.490 µm, when the burning power increases from 30W to 60W; the standard deviation increased by 20%, approximately.Figure 5(d) summarizes the average particle size for all cases.
The particle size distributions measured in our experiments are in the range from 100 nm to 1 µm, centered at around 400-500 nm.We note that our measurements lie within the range of different reports in the literature.For instance, using a combination of condensation particle counting and scanning mobility spectrometry, distributions below 100 nm, centered at around 50 nm, were reported [10]; using aerosol mobility spectrometry, particle size distributions centered at around 200 nm, were measured [11]; using low pressure impactors together with gravimetry, size distributions centered a larger values, at around 1 µm, were reported [38].Importantly, in all these studies diluted aerosols were used, which can affect significantly the size distribution measured depending upon the dilution ratio [11,38].Again, in our case, the measurements were performed on the native aerosols, without sample processing.
The results of our study are consistent with literature reports where it has been observed that the presence of glycerol leads to particle size distributions with larger sizes that also depend positively on the burning power [10,11,[38][39][40].This has been explained based on the higher volatility and vapor pressure of propylene glycol.On one hand, propylene glycol is aerosolized at a lower temperature; therefore, glycerol condensates in the EC, forming larger particles once the aerosolization temperature is reached [10].On the other hand, according to the ChemSpider and NIST public databases, as well as values reported in the literature [41], at room temperature, propylene glycol and glycerol have a vapor pressure of ∼0.12 mmHg (16 Pa) and ∼1.2 × 10 −4 mmHg (0.02 Pa), respectively; the higher vapor pressure of propylene glycol favors the gas phase while in glycerol aerosols the material tends to condensate more on the particles [10,11,42].
Moreover, based on models provided by the International Commission on Radiology Protection (ICRP) [43], we evaluated the particle deposition in different parts of the respiratory system: the head airway (HA) region, including nasal, pharyngeal, and laryngeal sites; the tracheobronchial (TB) region; and the alveolar (AL) region.The evaluation of all deposition fractions as a function of the particle diameter, d a , can be found in the Appendix.Interestingly, in the range of our measurements (100 nm to 1 µm), the ICRP model predicts a minimum particle deposition in all respiratory regions.In fact, in that range, deposition in the TB region is negligible.Nevertheless, an important observation is that at lower burning powers, where all aerosols have an average diameter at around 400 nm, the deposition fraction in the HA and AL regions are comparable (6.5% approximately).However, the 100 nm increase observed in the average size at the higher burning powers for aerosols containing glycerol, results in a higher deposition fraction in the HA region (10%) as compared to the AL region (7.7%); naturally, these difference in the deposition fraction will be more marked is the entire distribution is considered due to the deposition in the HA region dominates at large particle sizes.
Disclosures.The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Like popular DLS commercial equipment, in CG-DLS the measurement is also carried out at a single, fixed angle; backscattering is measured in this case.Light from a broadband source (Superlum model BLM-S-670-G-I-4; central wavelength λ 0 = 670 nm; bandwidth ∆λ = 7 nm) is launched into a 50/50 multimode splitter (Thorlabs model TM50R5F1B, ∅50 µm, 50:50, 0.22 NA) used as a circulator.The multimode optical fibers (MMFs) used are commercially available (Thorlabs M42L05, ∅50 µm, 0.22 NA).The optical signal was measured with a general-purpose photo-receiver (New Focus model 2001-FC) and then digitized with a data acquisition card (National Instruments model NI DAQ 9205).

Fig. 1 .
Fig. 1.Schematic of the CG-DLS setup: an optical fiber-based, common path interferometer where the Fresnel reflection at the distal end of the optical fiber serves as the local oscillator for the amplification of singly scattered light in a picolitter-sized coherence volume localized at the tip of the fiber.

Fig. 1 .
Fig. 1.Schematic of the CG-DLS setup: an optical fiber-based, common path interferometer where the Fresnel reflection at the distal end of the optical fiber serves as the local oscillator for the amplification of singly scattered light in a picolitter-sized coherence volume localized at the tip of the fiber.

Fig. 2 .
Fig. 2. Schematic of the aerosol collection process: (a) extraction and (b) storage.The extraction speed and syringe volume allow preserving the composition of the native aerosol.Once stored, the endoscopic optical fiber probe of CG-DLS enters the syringe to perform measurements of the particle size distribution in native aerosols, without the need for sample processing.

Fig. 2 .
Fig. 2. Schematic of the aerosol collection process: (a) extraction and (b) storage.The extraction speed and syringe volume allow preserving the composition of the native aerosol.Once stored, the endoscopic optical fiber probe of CG-DLS enters the syringe to perform measurements of the particle size distribution in native aerosols, without the need for sample processing.

Fig. 3 .
Fig. 3. (a) Particle size distribution of an aqueous colloidal suspension of monodisperse polystyrene particles, measured with CG-DLS and commercial DLS equipment.The table summarizes the more relevant outcomes, including the average particle size, the standard deviation, and the polydispersity index; see text for details.(b) Reflected power measured as the fiber probe of CG-DLS sits static outside the container, in air (regions A and E), and inside the container, in the aerosol (region C), as well as when the fiber is in motion, entering (region B) and exiting (region D) the container, respectively.The inset shows pictures of the facet of the MMF before and after entering the container; see text for details.

Fig. 3 .
Fig. 3. (a) Particle size distribution of an aqueous colloidal suspension of monodisperse polystyrene particles, measured with CG-DLS and commercial DLS equipment.The table summarizes the more relevant outcomes, including the average particle size, the standard deviation, and the polydispersity index; see text for details.(b) Reflected power measured as the fiber probe of CG-DLS sits static outside the container, in air (regions A and E), and inside the container, in the aerosol (region C), as well as when the fiber is in motion, entering (region B) and exiting (region D) the container, respectively.The inset shows pictures of the facet of the MMF before and after entering the container; see text for details.

Fig. 4 .
Fig. 4. Particle size distribution of the native aerosol from an EC.The smoking liquid consists only of propylene glycol; the aerosol was generated at a burning power of 30W.The main plot shows the aerodynamic size retrieved from the collection of spectra (recorded every second), as a function of time, starting from the moment when the fiber enters the container.No sample processing is required, only a couple of minutes for the aerosol particles to stabilize in a diffusion regime (shaded in red).The inset shows the particle size distribution measured when the fiber sits stably in the aerosol (shaded in green) and its corresponding Gaussian fit.

Fig. 5 .
Fig. 5. Particle size distribution of native aerosols from an EC generated from different moisturizers and various burning powers.Panels (a)-(c) show the Gaussian fits of the particle size distributions as a function of burning power: (a) propylene glycol, (b) glycerol, (c) 1:1 mixture of propylene glycol and glycerol.The insets show the average and standard deviation of the particle size distribution.(d) Comparison of the average particle size for the different aerosols generated.

Fig. 5 .
Fig. 5. Particle size distribution of native aerosols from an EC generated from different moisturizers and various burning powers.Panels (a)-(c) show the Gaussian fits of the particle size distributions as a function of burning power: (a) propylene glycol, (b) glycerol, (c) 1:1 mixture of propylene glycol and glycerol.The insets show the average and standard deviation of the particle size distribution.(d) Comparison of the average particle size for the different aerosols generated.

Fig. 6 .
Fig. 6.Inhalable and deposition fractions as a function of the particle diameter according to the International Commission on Radiology Protection model.The shaded region corresponds to the range of particle size in our experiments.See text for details.