In this appendix, we provide greater detail and the accompanying equations for the proposed end-to-end modeling framework. Following the format of the main body, we will divide this section into Particle Generation, Particle Transport, and Human Response.
Appendix A.1. Particle Generation
The particle generation submodel characterizes the respiratory particles released into the air by an infectious individual via different emission mechanisms and the subsequent evaporation of most of the volatile liquid (water) content from the particles.
Figure A1 summarizes the workflow of this submodel. The first step in developing a model of respiratory particle emission is to physically characterize the particles. Respiratory particles are heterogeneous and vary in number, size, chemical composition, and viral content depending on the generation mechanism, the stage of infection, as well as inter- and intra-subject variability [
32,
36,
69,
70]. Accounting for this heterogeneity is an important step in quantifying the potential uncertainty in particle emission from an infected individual. The physical characterization of emitted respiratory particles will allow for improved modeling of the fate of respiratory particles in the environment and inform estimates of inhalation, deposition, and infectivity within the respiratory tract of susceptible populations.
Figure A1.
Visualization of the particle generation submodel workflow. Due to the rapid time scale at which it occurs (i.e., within the first 50 s), near-field dynamics of the Particle Transport submodel (Evaporation box in green) is displayed alongside the Particle Generation submodel. Red circles indicate parameters selected for sensitivity analysis (see
Section 2). All notation definitions are provided in
Table 1.
Figure A1.
Visualization of the particle generation submodel workflow. Due to the rapid time scale at which it occurs (i.e., within the first 50 s), near-field dynamics of the Particle Transport submodel (Evaporation box in green) is displayed alongside the Particle Generation submodel. Red circles indicate parameters selected for sensitivity analysis (see
Section 2). All notation definitions are provided in
Table 1.
A simplified schematic of a respiratory particle as it is emitted from an infectious person is shown in
Figure A2. The particle can be described as a combination of volatile liquid (such as water), non-volatile and non-soluble materials (an actively infectious component and other material, such as proteins), and non-volatile, soluble materials that are suspended in the liquid portion (such as salts). As the particle is transported, the volatile material evaporates, leaving the non-volatile and soluble portions of the particle.
Figure A2.
Simplified schematic of a particle.
Figure A2.
Simplified schematic of a particle.
A saliva particle, for example, is composed of approximately 3% well-mixed nonvolatile components and 97% water by mass [
71]. The nonvolatile components are varied, as shown in
Table A1. Most of a saliva particle is made up of salts (NaCl and KCl) and mucin; these components will affect the dehydration and rehydration of particles and, for a high-fidelity model, should be tracked. For dehydration, we used data from [
48] and consultation with Walker to determine post-evaporation diameters from pre-evaporation diameters. We settled on a post-evaporation diameter of 29% of the original pre-evaporation saliva particle size.
Table A1.
Components of a saliva particle. The first two columns were calculated in [
48] and the other columns were calculated.
Table A1.
Components of a saliva particle. The first two columns were calculated in [
48] and the other columns were calculated.
Component | Concentration (Mass per Liter of Fluid) (g/L) | Density of Component (g/L) | Volume of Component/Dry Particle Volume | Weighted Density in Dry Particle (g/L) |
---|
MgCl2 | 0.04 | 2320 | 4.52 × 10−3 | 10 |
CaCl2.H20 | 0.013 | 2240 | 1.52 × 10−3 | 3 |
NaHCO3 | 0.42 | 2200 | 5.01 × 10−2 | 110 |
KH2PO4 | 0.21 | 2340 | 2.35 × 10−2 | 55 |
K2HPO4 | 0.43 | 2450 | 4.59 × 10−2 | 112 |
NH4Cl | 0.11 | 1530 | 1.89 × 10−2 | 29 |
KSCN | 0.19 | 1900 | 2.62 × 10−2 | 50 |
(NH2)2CO (urea) | 0.12 | 1340 | 2.35 × 10−2 | 31 |
NaCl | 0.88 | 2165 | 1.07 × 10−1 | 231 |
KCl | 1.04 | 1984 | 1.37 × 10−1 | 273 |
Mucin | 3 | 1400 | 5.62 × 10−1 | 787 |
DMEM | 1 mL per liter of fluid | - | - | 14 |
Alpha-amylase | - | - | - | - |
Deionized water | 979 mL per liter of fluid | - | 0 | 0 |
Sum | - | - | - | 1706 |
Respiratory particles are primarily generated in three locations within the human respiratory tract. Studies have characterized primary emission locations and mechanisms [
36]. First, respiratory particles are generated during normal breathing and these particles are generally in the range of 1–2 µm in diameter [
72]. Second, respiratory particles may also be generated in the larynx during activities such as speaking or coughing; these particles are also in the range of 1–2 µm in diameter when expelled from the respiratory tract. The final primary location of particle generation is the oral cavity; this generation mechanism is primarily associated with coughing and to a lesser extent with speaking, and these particles are much larger in size (approximately 100 µm in diameter) when measured at emission [
36]. Particles generated by these three locations during speaking, coughing, and normal breathing are quantified in terms of the count median diameter (CMD), geometric standard deviation (GSD), and number concentration (
) [
36]. Values for these three parameters for each respiratory activity are provided in
Table A2.
Table A2.
Size distribution of particles emitted via breathing, speaking, and coughing.
Table A2.
Size distribution of particles emitted via breathing, speaking, and coughing.
| Breathing | Speaking | Coughing |
---|
Location | CMD (μm) | GSD | Concentration (#/cm3) | CMD (μm) | GSD | Concentration (#/cm3) | CMD (μm) | GSD | Concentration (#/cm3) |
---|
Bronchiolar | 1.6 | 1.30 | 0.069 | 1.6 | 1.30 | 0.069 | 1.6 | 1.25 | 0.087 |
Laryngeal | N/A | N/A | N/A | 2.5 | 1.66 | 0.086 | 1.7 | 1.68 | 0.130 |
Oral | N/A | N/A | N/A | 145 | 1.80 | 0.001 | 123 | 1.84 | 0.016 |
For our implementation of the model, emission rate varies with particle size and is determined by assuming a lognormal particle size distribution combining the contributions from each source for particle size bins ranging from 0.1 µm in diameter to 100 μm in diameter. A lower and upper size limit should be selected for virus-containing particles. For example, the lower size limit for virus-containing respiratory particles of 0.1 µm can be chosen based on the 100 nm estimated size of SARS-CoV-2 and the estimated initial water content of the respiratory particles being 90% [
38]. The upper size limit for virus-containing respiratory particles of 100 µm was chosen based on an estimated settling time of less than 20 s for particles larger than 100 µm in diameter. Therefore, the submodel assumes particles larger than 100 µm in diameter will deposit to surfaces quickly and are not a long-term airborne exposure risk in most situations.
For a given pre-evaporation particle diameter
, the total particle emission rate,
, depends on the minute ventilation of the infected subject (
), the source of the particles (bronchiolar, laryngeal, or oral), and the emission mechanism (breathing, speaking, or coughing) within the respiratory tract. More specifically,
where
,
, and
denote the particle number concentration emitted from the bronchiolar, laryngeal, and oral sites, respectively, as determined by the corresponding values of
,
, and
. Note that the minute ventilation
is the product of the tidal volume and breathing frequency both of the infectious subject. The above equation for
is equivent to that given by Equations (2) and (3) in
Figure A1.
Polymerase chain reaction (PCR) tests measure the number of RNA copies per volume of liquid sample collected; however, these viral RNA counts are not an indicator of infectiousness as they can include non-infectious viral genetic material that has been neutralized by the immune system and is cleared by the respiratory tract [
34,
73]. For this model framework we are interested in the infectious viral content being emitted and therefore, introduce an infectivity ratio (
) between RNA copies and plaque forming units (PFUs). The complete conversion process (number of particles to RNA copies to TCID
50) to estimate the viral emission rate is summarized by Equations (2)–(4) in
Figure A1. This study makes the assumption that the concentration of RNA copies in each pre-evaporation particle is equal to the viral load measured by PCR tests. As the particle dries after emission from the respiratory tract, this viral concentration will increase with the decrease in particle volume.
Viral load measured by PCR tests was also found to vary considerably from subject to subject. For example, one study found that viral loads for SARS-CoV-2 positive individuals vary from 6.99 × 10
2 to 4.71 × 10
8 RNA copies/mL with a median of 1.46 × 10
5 RNA copies/mL [
74]. Viral load also depends on the stage of infection and the sample collection site [
41,
63,
75,
76]. In addition, a specific variant of SARS-CoV-2 may have an impact on the measured viral load [
77].
Appendix A.2. Particle Transport
The particle transport submodel describes the fate of particles emitted by the infectious individual in the surrounding environment. The near field transport and dispersion of respiratory particles takes place near the infected individual and within the first seconds of emission; it is dominated by effects of the initial jet flow produced by air escaping the mouth or nose, rapid evaporation as the particles enters a dryer environment, and rapid settling of larger particles. The evaporation and sedimentation of respiratory particles can be expressed as a function of time after emission from the infected individual [
48]; these model results can be used to characterize the near field environmental fate of respiratory particles. Using this information for both saliva particles (oral generation) and lung fluid particles (bronchiolar generation), we can determine the size change and sedimentation of particles after the initial expulsion time period. Our implementation of this model did not account for near field dynamics except for estimating a final particle diameter after dehydration. Our model, therefore, assumes that particles quickly dehydrate to a stable particle size and are introduced into the well-mixed indoor environment; this assumption was partly due to our primary mode of modeling a steady-state concentration to model two individuals who are separated by over 2 m. The additional model and computational complexity from adding a complete characterization of near-field phenomena (e.g, jet flow emission) would be unnecessary for our numerical analyses and sensitivity analysis. If we were examining near-field exposures, including these dynamics would be necessary.
The far field transport and dispersion of respiratory particles takes place after most large particles have settled, the jet flow no longer impacts particle trajectory, and particles have reached a stable size after evaporative drying. For SARS-CoV-2, a time period of 50 s was chosen as the cut-off between near and far field transport models due to the fact that most particles have evaporated down to a stable particle size by that time or had already settled to the ground for those particles large enough to still be experiencing evaporation; note that because of the different time scales in which evaporation-induced particle size reduction and exposure occur (i.e., seconds v. hours) near field dynamics are captured in workflow of the particle generation submodel (see
Figure A1 and
are used to distinguish between the pre-evaporation and post-evaporation particle diameters; moreover, the pre-evaporation particle diameter is proportionally reduced by a factor o
f as shown in Equation (1) of
Figure A1.
Indoor transport and dispersion models include several mechanisms that reduce the concentration of airborne respiratory particles and airborne contagions. First, gravitational settling is modeled using Stokes law to determine the particle settling velocity for a given diameter
(see Equation (5) in
Figure A3); this settling velocity is used to determine a loss rate based on a conservative estimated initial particles height of 2 m above the floor of the room.
Figure A3.
Visualization of the aerosol transport submodel workflow. Equations (6) and (7) are equivalent as the latter is the analytical solution of the former. When executing the workflow, either can be utilized, however, Equation (7) is preferred as the complexity of the numerical integration is reduced, in turn minimizing both the numerical errors introduced and the computational time required. Red circles indicate parameters selected for sensitivity analysis (see
Table 2).
Figure A3.
Visualization of the aerosol transport submodel workflow. Equations (6) and (7) are equivalent as the latter is the analytical solution of the former. When executing the workflow, either can be utilized, however, Equation (7) is preferred as the complexity of the numerical integration is reduced, in turn minimizing both the numerical errors introduced and the computational time required. Red circles indicate parameters selected for sensitivity analysis (see
Table 2).
The second mechanism pertains to viral degradation in the surrounding environment. One study developed an equation to estimate a first order viral infectivity reduction rate for SARS-CoV-2 as a function of air temperature, humidity, and UV radiation [
43]:
where
is the first order viral decay rate (in min
−1),
is the air temperature (in °C),
is the relative humidity (in %), and
is the surface UVB irradiance (in W/m
2). Finally, in the indoor setting, room air ventilation is also used to remove airborne particles from the air while replacing with clean air with zero particle concentration.
To account for these three far field mechanisms in the submodel, we adopt the approach of Dols et al., which uses an ordinary differential equation (see Equation (6) in
Figure A3) to describe how the viral concentration changes with time [
62]. The first term of on the right-hand side of Equation (6) is the time-dependent emission rate, which is determined by the viral emission rate defined in the preceding subsection. The summation of
’s captures the effects due to indoor ventilation, whereas the penultimate term accounts for removal of virus via settling. The last term with
describes viral decay due to environmental factors. We note that this ODE has an analytical solution given by Equation (7).
Appendix A.3. Human Response
To determine likelihood of infection from a contaminated exposure environment, the human response submodel first determines the inhaled viral concentration and then the subsequent total deposited dose.
Figure A4 summarizes this workflow. Ultimately, this submodel aims to quantify the impact of infection.
The amount of inhaled virus is computed by multiplying the time dependent viral concentration for a given particle diameter with an inhalable fraction,
. One study presented an inhalability model that accounts for wind speed, minute ventilation and particle diameter [
57]. Adopting the approach of [
57], the particle diameter-dependent inhalability fraction is computed according to Equation (10) in
Figure A4. For the indoor environment, a value of 0 m/s can be assumed for wind speed, resulting in an inhalable fraction of ~0 for 100-micron particles.
To determine likelihood of infection from an exposure environment, the model first accounts for how many particles are inhaled, the location of deposition, and the particle composition; this is determined by considering particle-size dependent inhalability [
57], particle size-dependent deposition site [
78] (which would also account for hygroscopic growth of a particle as it enters the humid respiratory tract [
20]), and the viral content of particles [
70]. The severity and course of illness of the disease then could depend solely on demographic factors, immune response, deposited dose, or a combination of factors. For SARS-CoV-2, experimenters were able to show clear dose-dependence for symptomatic/asymptomatic presentation of disease [
17] as well as demographic differences in severity for symptomatic diseases [
51,
79].
Figure A4.
Visualization of the human response submodel workflow. Red circles indicate parameters selected for sensitivity analysis (see
Section 2). All notation definitions are provided in
Table 1.
Figure A4.
Visualization of the human response submodel workflow. Red circles indicate parameters selected for sensitivity analysis (see
Section 2). All notation definitions are provided in
Table 1.
An additional factor to consider is that of hygroscopic growth in the respiratory tract. The URT and proximal lung airways rapidly condition air to ~100% humidity and ~98.6 degrees Fahrenheit [
80]; this environment will cause a hygroscopic particle to grow, often resulting in greater deposition in the URT and lower deposition in the LRT [
20,
24] (due to filtration by the URT).
Using a URT model that accounts for hygroscopic growth, we developed look-up-tables that accounted for ambient temperature and humidity, initial particle size, and initial particle composition for calculating URT and LRT deposition. Plots of URT and LRT deposition fractions as a function of initial particle size are shown in
Figure A5.
Figure A5.
Plots of example look-up table data of particle deposition fraction versus initial diameter of a particle at the inlet to the nasal passages of a susceptible person including deposition in (a) the URT (red), LRT (blue), and total URT and LRT (black), and (b) the tracheobronchial region (TB) (red), pulmonary region (PUL) (blue), and total TB and PUL, or LRT (black). At the inlet to the nasal passages, initial properties are a particle composition of 74% well-mixed nonvolatile components (NaCl) and 26% water by mass, air and particle temperatures of 20 °C, and relative humidity of 50%.
Figure A5.
Plots of example look-up table data of particle deposition fraction versus initial diameter of a particle at the inlet to the nasal passages of a susceptible person including deposition in (a) the URT (red), LRT (blue), and total URT and LRT (black), and (b) the tracheobronchial region (TB) (red), pulmonary region (PUL) (blue), and total TB and PUL, or LRT (black). At the inlet to the nasal passages, initial properties are a particle composition of 74% well-mixed nonvolatile components (NaCl) and 26% water by mass, air and particle temperatures of 20 °C, and relative humidity of 50%.
A dose response model is typically used to determine the probability of infection,
. Adopting the standard form of a probit model (see Equation (14) in
Figure A4),
is taken to be a function of the total deposited dose
(in virions, TCID
50, PFUs, etc.), the median infectious dose
(in equivalent units), and probit slope
; note that here,
and
are measured in terms of TCID
50.
A model that can incorporate infectivity parameters (ID
50 and probit slope) for both the URT and LRT is preferred when data exists to support such a model. If data supports site-specific infectivity, an ID
50 and
will be available for both the URT and LRT. Although a study must be conducted to correlate infectivity likelihood with deposition site for SARS-CoV-2, a decreased infectivity from particles deposited in the URT when compared to the LRT has been observed for several biological agents including F. tularensis [
81], Y. pestis [
82] and B. anthracis [
83,
84]. It is possible to develop a joint probit model given the data or to simply assume that the probabilities of infection for the two deposition sites are independent by using:
where
Ptotal is the overall probability of infection,
PURT is the probability of infection from the dose deposited in the URT, and
PLRT is the probability of infection from the dose deposited in the LRT.
There is not sufficient evidence to correlate initial dose to disease severity or lethality in biological agents; however, for SARS-CoV-2 it has been shown that there are different infectivity parameters for asymptomatic and symptomatic presentation of the disease [
85]. For SARS-CoV-2 we developed a non-dose dependent severity model based on documented incidence of severity, but it does not fit in the discussion of an end-to-end framework as it is independent of all other submodels. It is the authors’ belief that a model of viral reproduction and immune response is necessary for incorporating a model of disease severity within this framework.