Fluid Dynamics of Respiratory Infectious Diseases.

The host-to-host transmission of respiratory infectious diseases is fundamentally enabled by the interaction of pathogens with a variety of fluids (gas or liquid) that shape pathogen encapsulation and emission, transport and persistence in the environment, and new host invasion and infection. Deciphering the mechanisms and fluid properties that govern and promote these steps of pathogen transmission will enable better risk assessment and infection control strategies, and may reveal previously underappreciated ways in which the pathogens might actually adapt to or manipulate the physical and chemical characteristics of these carrier fluids to benefit their own transmission. In this article, I review our current understanding of the mechanisms shaping the fluid dynamics of respiratory infectious diseases.


INTRODUCTION
As of June 2021, the number of people who have tested positive for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) surpassed 179 million worldwide, and the global death toll attributed to the associated coronavirus disease 2019 (COVID-19) exceeded 3.8 million (1). Since the first reported human SARS-CoV-2 infections in late 2019, COVID-19 has placed unprecedented pressure on national health care delivery systems and the global economy. Vaccines are among the most successful public health tools in human history; however, they are not always highly effective (2) or immediately available, particularly for new pathogens such as SARS-CoV-2, or available at all, even for well-known ones such as adult tuberculosis (TB). Even when available, eradication or deployment to maximize population protection takes planning and care and is not always straightforward (3). Therefore, COVID-19 continues to affect the daily life of the general public in unprecedented ways, including the need for increased hand hygiene, regular surface decontamination, social (i.e., physical) distancing, self-quarantine/isolation, and wearing of masks in public, as well as complete or partial closure of entire sectors of the economy, education, and care for the most vulnerable-all in an effort to curb the rapid spread of the virus while the world awaits the widespread availability and distribution of an effective vaccine.
Although it took a global pandemic to make such infection control strategies part of the collective consciousness, respiratory infectious disease outbreaks and epidemics have occurred with some regularity. Within the past 20 years alone, the original SARS affected 26 countries (4) Wells pioneered the physics-based dichotomized view of respiratory disease transmission via large versus small droplets (aerosols) based on settling versus evaporation timescales. These ideas still underpin the WHO and CDC classification and risk management guidelines for respiratory infectious diseases.
Chapin, a leading US public health officer, and Langmuir, a prominent figure in the formation of the CDC, fervently opposed the idea of airborne transmission of respiratory diseases, but regretted doing so later in life in the context of measles. Langmuir later admitted the burden of "doctrinal" teaching and called for maintaining the "eternal skepticism of the true scientist" in public health decision-making (Langmuir 1980, p. 40).
Pioneers of modern bacteriology. Pasteur, a chemist, is credited with the discovery of vaccination and experimental validation of germ theory. Koch, a physician, discovered the pathogens causing anthrax, tuberculosis, and cholera and experimentally validated the concept of infectious disease transmission.

Figure 1
Core ideas about germ theory and transmission and their implications for epidemiology and public health, stemming from the legacy of Pasteur, Koch, Snow (not shown), Flügge, and Wells, established modern infection control and epidemiology approaches but were regularly misunderstood, resisted, and dismissed throughout much of history. This is still the case today, as ideas of airborne transmission are often dismissed upon the emergence of new pathogens such as SARS (20) or, more recently, SARS-CoV-2 (21,22). Abbreviations: CDC, Centers for Disease Control and Prevention; SARS, severe acute respiratory syndrome; SARS-CoV-2, SARS coronavirus 2; WHO, World Health Organization.
understanding of the mechanisms shaping the fluid dynamics of respiratory infectious diseases, with a focus on the transmission processes that are at the core of respiratory infectious disease outbreaks, epidemics, and pandemics such as COVID-19.

PENDULUM SWINGS OF TRANSMISSION THEORY: HISTORICAL PERSPECTIVE ON THE ORIGIN OF THE AIRBORNE VERSUS DROPLET ROUTE OF TRANSMISSION DICHOTOMY
The notion that diseases could be contracted through inhalation of so-called bad air dates back millennia and was the foundation of the long-held miasma theory, which posited that epidemics spread through the inhalation of noxious vapors originating from decomposing matter. Not until the mid to late nineteenth century did the germ theory garner scientific support, primarily through a series of definitive experiments by Louis Pasteur (18) and Robert Koch (19), paving the way toward an understanding of the causative agents of disease and to systematic investigations of how contagious diseases spread (Figure 1).
In the context of respiratory infectious diseases, at the turn of the last century much interest centered on understanding the host-to-host transmission of TB. The prevailing opinion among physicians at the time was that TB was transmitted mostly through the inhalation of deposited, dried-up tuberculous sputum that is broken up and resuspended in the air in the form of dry dust (23). Today, the dried-out, solid remnants of exhaled mucosalivary droplets are often called aerosols, an umbrella term that denotes any type of matter carried in the air (24). From a fluid dynamic perspective, they are referred to as droplet nuclei or residues, as they are the solid remnants of liquid droplets following desiccation.
False dichotomy: large versus small droplets Wells (1934Wells ( , 1955 "Small" droplets (now referred to as aerosols) "Large" droplets d ≥ 100 μm The isolated respiratory drop emission paradigm, which remains the foundation of current infection control guidelines: the dichotomy between isolated small-and large-droplet respiratory emissions (or droplets versus aerosols) (33,35). The cutoff of 100 µm introduced by Wells relies on a competition between timescales: the settling timescale [from a 2-m height, assuming Stokes settling speed u s = (gd 2 /18μ)(ρ d − ρ a ), where g is gravity and ρ a and μ are the density and dynamic viscosity of the ambient air phase, respectively] and the evaporation timescale (assuming a quiescent ambient and an isolated droplet evaporation model). The model relies on the assumption that respiratory droplets are emitted in isolation (i.e., as a pure liquid spray).
Carl Flügge, then professor of hygiene at the University of Breslau in Germany, challenged this singular focus on airborne transmission via "resuspended dust" and argued that, in infectious diseases of the respiratory system, transmission through contact with liquid mucosalivary droplets emitted by infected individuals presents a much greater risk of infection (23,(25)(26)(27). Over a decade, Flügge and his assistants conducted a series of systematic laboratory and clinical investigations into the host-to-host transmission of TB via exhaled droplets, recognizing such notions as evaporation, settling time, infectious dose, and the role of background airflows in dispersing exhaled droplets (25,(27)(28)(29)(30)(31). Importantly here, and contrary to today's nomenclature of droplets versus aerosols, Flügge termed all respiratory emissions "droplets," irrespective of their size and final state (liquid droplet or droplet residue/aerosol), as long as they were not yet deposited. In these experiments, collection was done hours after emission, allowing ample settling to occur and not differentiating between liquid and dried droplets. They established that such emissions can easily travel several meters from coughing or sneezing subjects and that it is generally possible to demonstrate the infectivity of the emissions captured at various distances from the patients.
While the extensive work by the Flügge school established a nuanced understanding of respiratory infectious disease transmission in TB, it was soon erroneously reduced to the notion that only ballistic (large) droplet emissions that remain in the liquid state are the major route of respiratory infectious disease transmission. This dramatic reduction changed the view of respiratory infectious disease transmission from the airborne dry resuspended dust route to that of ballistic liquid droplet infection (32,33).
This view remained unchallenged for 30 years, until William Wells started to examine TB transmission with a focus on the fate of isolated droplet emissions (33)(34)(35). On the basis of the droplet diameters, he dichotomized respiratory droplets as large versus small (or droplets versus aerosols), using a 100-µm-diameter cutoff between large and small (Figure 2). In his framework, isolated large droplets settle faster than they evaporate and hence fall to the ground within a few seconds, whereupon they contaminate surfaces, typically close to a patient. Small isolated droplets, in contrast, evaporate faster than they settle and form droplet nuclei. Without ambient flow, high www.annualreviews.org • Fluid Dynamics of Respiratory Infectious Diseases air resistance (drag) prevents isolated small droplets from reaching further than a few centimeters, even if emitted at high speed. Even though it was underpinned by a sound mechanistic model of isolated droplet evolution in the ambient environment, Wells's unified view of TB transmission faced significant opposition and resistance from the medical and epidemiology establishment (36) (Figure 1).
One of Wells's chief opponents, Charles Chapin, was an influential public health officer who forcefully defended the view that communicable respiratory diseases are spread by close contact or through indirect contact via droplet transmission (32). Broader failure to recognize the possibility of airborne transmission of diseases such as measles was later attributed, at least in part, to the failure of the leading epidemiologists of the time to question the prevailing doctrines and to conduct conclusive field studies that may have settled the issue for a variety of infectious diseases (32). Therefore, doctrine prevailed over scientific method. Despite such resistance, however, Wells persisted and eventually prevailed. Aided by supporting evidence from a series of systematic and careful laboratory experiments, Wells's physics-based dichotomy of respiratory infectious disease transmission was eventually accepted and is still at the core of World Health Organization (WHO) and Centers for Disease Control and Prevention (CDC) classification schemes (37,38), albeit with a default focus on the droplet route for most respiratory diseases, even in the absence of confirmatory evidence in the time of emerging pathogens, such as SARS-CoV-2.
Today's classification schemes employ various arbitrary diameter cutoffs, from 5 to 10 µm, to classify host-to-host transmission into droplet versus aerosol routes on the basis of the inhalation potential of such drops. Consequently, infection control strategies have been built around the dichotomy between a large-droplet route, with a focus on droplet precautions and surface contamination, and an aerosol route, with a focus on high-grade respirators. Not surprisingly, in the event of an emerging respiratory infectious disease outbreak, as was the case with SARS-CoV and now SARS-CoV-2, public health officials have become interested in droplet size distributions to determine which route dominates and which precautions should be mandated. However, the problem with this dichotomy, as we now know, is that it is false.

SHIFT IN PARADIGM: FROM RESPIRATORY ISOLATED DROPLETS TO THE TURBULENT MULTIPHASE CLOUD
While physically elegant, Wells's isolated droplet model does not capture a central physical characteristic of exhalations. Exhalations do not eject isolated MS droplets into the environment. Exhalations-ranging from speaking to yelling, coughing, and sneezing-are made up of a turbulent multiphase cloud whose physical properties have significant implications for the range and longevity of the ejected mucosalivary droplets and, hence, for their fate, their range of contamination in space and time, and associated risk-assessment and risk-mitigation strategies (17,21,(39)(40)(41). The turbulent exhalation cloud consists of warm and moist exhaled air laden with mucosalivary droplets ranging in size from O(1 µm) to O(1 mm) (39). The locally warm and moist atmosphere within the exhaled cloud protects the smaller droplets from immediate evaporation, and its momentum carries them forward much farther (by a factor of more than ∼200 for drops below 10 µm) than they would be able to travel were they emitted in isolation (21,39,40). Importantly, the range of all droplets is extended, and contrary to the Wells picture, the smallest droplets tend to travel the farthest (21,39) (Figure 3a). In fact, it is not the size of the droplets at emission that determines their range but rather the characteristics of the warm and moist gas cloud that is emitted and carries them forward. Only after the emitted cloud slows down sufficiently do droplet and residue sizes become important determinants of dispersal in background low-velocity indoor airflows (a few centimeters per second) and penetration of the respiratory system.  (a) Paradigm shift from Wells's isolated droplet picture to that of the high-momentum turbulent (high-Re) multiphase exhalation cloud that carries droplets much further than if they were emitted in isolation (by a factor of at least 200 for drops <10 µm). Moreover, the smaller drops are carried over macroscopically important distances in a few seconds, even in the complete absence of background ventilation flow. The physical picture of respiratory multiphase cloud emission (Figure 4) applies to a range of exhalations, as long as the key assumptions of the point-source emission and sufficiently high Re are satisfied to ensure turbulence of the cloud. (b) These assumptions remain valid for breathing as well as for more violent exhalations such as coughing and sneezing, as was recently confirmed in experiments on speech and high-fidelity numerical simulations of violent exhalations (42,43). Abbreviations: CDC, Centers for Disease Control and Prevention; Re, Reynolds number; WHO, World Health Organization.

The Respiratory Exhaled Cloud Governs the Range of Air and Surface Contamination
From a fluid dynamic perspective, the respiratory cloud emission is a discrete emission of mass and momentum that undergoes two phases: (a) an initial continuous jet phase that is brief compared with the subsequent evolution of the cloud and (b) a longer-lasting puff phase. Both jets and puffs are momentum-driven turbulent bodies of fluid ejected from a point source (here, the mouth). Along their trajectory s(t), the lateral dimension r(t) of the jet and puff [measured orthogonal to s(t)] grows as r(t) = αs(t), where α is a (constant) entrainment coefficient, due to lateral momentum diffusion and edge eddies entraining quiescent ambient air (17, 39) (Figure 4a,b). (a) Exhaled air with initial volume V 0 and momentum I 0 containing mucosalivary droplets of a given size distribution forms the multiphase cloud of initial density ρ c (0) and initial buoyancy B 0 . (b,c) The turbulent point-source emission evolves according to a self-similar growth with a signature conical shape along its curvilinear trajectory (s, θ ). The mean velocity of entrainment from the edges of the cloud, W, is proportional to the mean forward velocity of the cloud, U(s); in other words, W(s) = αU(s), where α is the entrainment coefficient. The dynamics of the cloud and the fate of its droplet payload are modeled using a continuous settling model, enabling the prediction of the entire trajectory of the cloud, as well as the ranges over which droplets of a given size leave the cloud and how far away they would settle on surfaces or remain suspended. This model has been validated in both (d,e) analog experiments and ( f,g) human exhalations. These findings reveal the importance of accounting for the cloud's multiphase turbulent puff nature, as the characteristic scaling evolution of s ∼ t 1/4 is robustly demonstrated (39). The cloud dramatically extends the distance reached by the droplets, even in the absence of ambient airflow or ventilation, enabling the smaller drops to travel up to 8 m in a few seconds in a concentrated manner rather than being well mixed in indoor spaces. The implications are important for evaluation of risk and contamination and for infectious disease guidelines (e.g., Figure 5) (21,41).
Assuming discrete momentum-dominated emissions, the initial jet-like phase is characterized by conservation of momentum flux, M 0 ∼ ρ c r 2 (ds/dt) 2 , where ρ c is the density of the multiphase cloud, throughout the brief, continuous emission process (150-200 ms for a sneeze and 200-300 ms for a cough). Combining the conservation of momentum flux with the self-similar growth of the cloud's lateral dimension, s ∼ r, leads to the scaling law s ∼ t 1/2 for the initial temporal evolution of the gas cloud (17,39). Subsequent to the emission period, momentum is no longer continuously injected; rather, it is conserved. This is the longer-lasting puff phase. The conservation of initial momentum I 0 ∼ ρ c r 3 (ds/dt), when combined with the self-similar growth of the cloud, leads to the scaling law s ∼ t 1/4 during the puff phase (17,39). These scaling laws are concrete predictions of the turbulent multiphase cloud model that derive from the entrainment closure of the governing equations (39,44), a surprisingly robust characteristic of such pointsource turbulent clouds across applications ranging from geophysical and industrial systems (45) to exhalations (Figure 4e,g). The t 1/2 and t 1/4 scaling-law predictions have been experimentally verified in physical analog experiments, in human sneezes and coughs (21,39) (Figure 4d-g), and, more recently, in high-fidelity numerical simulations of violent exhalations (43) and further human exhalation experiments such as speech (42) and additional violent exhalations (E. Villermaux & L. Bourouiba, original data) (Figure 4g). When exhalations are performed in rapid succession, a continuous emission closer to the limit of a jet with s ∼ t 1/2 is recovered for a time (42).
The motion of an exhalation cloud is well captured by a point-source multiphase turbulent puff physical picture (39). An isolated mass of turbulent gas laden with liquid drops moves through a quiescent environment with an average speed U(s) along its curvilinear trajectory, at first driven by its initial and discrete momentum, with a possible role for buoyancy depending on ambient conditions. Assuming an initial momentum I 0 and initial buoyancy B 0 , a one-phase cloud of density ρ c , suspended droplet number N, cloud volume V, buoyancy B, and momentum I, one can show that the cloud's evolution is governed by where η is a shape parameter that reflects the ellipsoidal shape of the cloud (39,45), v p is the volume of the (here monodisperse) droplets in the cloud, ρ a is the density of the ambient air, θ 0 is the initial angle of the trajectory s(t) measured with respect to the horizontal direction, and g is gravitational acceleration (39). In this validated model, the speed of the cloud decreases as ambient quiescent air is entrained into the cloud. Droplets of size d continuously settle out of the cloud when their speed, U s (d), exceeds the continuously decreasing mean forward speed of the cloud, U(s) (39) (Figure 4c). The resultant model is closed in the sense that it does not contain any free parameters that need to be fitted: It is predictive from initial conditions. The system of Equations 1-6 is valid for exhalations ranging from tidal breathing to coughing and sneezing and can be used to predict contamination as a function of distance and time since emission (L. Cooper, C. Lu & L. Bourouiba, original data). The turbulent cloud model predicts that the smallest droplets go the farthest. Rather than encountering significant air resistance, limiting their motion to a few centimeters in the absence of ambient airflow, droplets are instead carried forward by the momentum of the turbulent cloud (39), extending the range of droplet emissions up to 7-8 m for sneezes (21,39,40). This is in contrast to the view to which Flügge's extensive work had been reduced, namely that ballistic large-droplet emissions set the range of contamination to a maximum of 1-2 m from a cougher/sneezer (Figure 3a) in the absence of background airflows. Interestingly, the long range of such droplet emissions during coughs and sneezes had already been observed in the 1890s by von Weismayr (46) and Laschtschenko (29) despite their relatively limited understanding of the fluid dynamics involved. They rinsed the mouths of healthy volunteers with a saline suspension of Bacillus prodigiosus (Serratia marcescens) and had the subjects speak naturally at various intensity levels, cough, and sneeze. They showed that colonies grew in petri dishes positioned as far as 185 cm from the speakers, 4 m from the coughers, and 9 m from the sneezers (29, 46).

The Cloud and Its Intermittency Change Droplet Evaporation
The warm and moist local cloud environment and the cloud's turbulent intermittency delay droplet evaporation in the moving multiphase cloud, by a factor of up to 1,000 (21), compared with evaporation of isolated droplets, with significant implications for pathogen persistence. Recent high-fidelity numerical simulations of respiratory point-source emissions estimated a delay in evaporation by a factor of at least 30 in standard indoor conditions, due to the cloud's sheltering effect on its droplet payload (43).
Given the success of the validated turbulent cloud model in explaining and predicting the fluid dynamics of respiratory emission events, it has become obvious that the current paradigm of respiratory infectious disease transmission, still based on Wells's dichotomized picture of isolated large versus small/aerosol droplet emissions (Figure 2), needs to be abandoned or reframed to incorporate the underlying physics of air versus surface spatiotemporal dynamics of contamination. Accordingly, the WHO and CDC risk and infection control guidelines need to be updated in favor of a paradigm that incorporates the notion of a continuum of droplet sizes and the more nuanced and accurate physics of fluids governing their dynamics, range, and evaporation, thus determining risks of contamination in space and time, patterns of transmission, and respiratory protection needs in a variety of natural and built environments.

THE MULTIPHASE TURBULENT CLOUD MEETS AMBIENT AIRFLOWS
As the exhaled multiphase turbulent puff cloud advances upon emission (Figure 4), its forward speed decreases as it grows through entrainment of quiescent ambient air. After sufficient time elapses, the forward speed of the cloud becomes comparable to the comparatively low [O(1-10 cm/s)] background air speed naturally found indoors. When this occurs, the droplets and droplet nuclei that remain trapped in the cloud and carried forward over O(10 m) for the most violent exhalations (21,40) may settle if their settling speed is greater than the ambient ventilation airflow. Otherwise, they are advected by the background airflow and disperse, rather than remaining concentrated in a coherent cloud moving forward.
Under well-mixed indoor conditions, average ambient airflow is 25 cm/s in the occupied zone of the room. It is largely determined by the specifics of the indoor ventilation (e.g., mechanical, natural, displacement, or mixing ventilation) and the geometry of the space, though additional factors can also affect indoor airflow. Human occupants and electric devices produce heat, inducing plumes that contribute to the overall airflow, on average and instantaneously, leading to peak velocities of O(10 cm/s) and up to 600 L/min in the vicinity of a person in otherwise-still ambient conditions of average temperature (∼20°C) (47). However, these airflows do not account for the momentum-driven exhalations directing exhaled flow horizontally toward others. Normal tidal breathing alone can generate the turbulent multiphase jets discussed above, with a Reynolds number (Re) of ∼600-800, that can easily reach distances on the order of 1-2 m (Figure 3).
Upon slowdown of the respiratory cloud and the balance between its mean velocity and the background airflow in an indoor space, the remaining suspended droplets and droplet nuclei disperse via background turbulence. Such dispersal can be estimated using classical Richardson's law for relative dispersal in a turbulent flow (48,49): δ Ri (t) = At β/2 , where β = 3 and δ Ri (t) is the root-mean-square separation between two particles (here, droplets or their residues). The constant A is the product of the Richardson constant and the turbulent dissipation rate per unit mass, which itself is a function of the intensity of the ventilation imposed in the space. The root-meansquare distance between droplets then evolves as δ Ri (t) ∼ t β/2 , assuming the flow is statistically steady.
We therefore have two regimes: (a) early respiratory emission governed by the dynamics of the respiratory point-source turbulent puff cloud entrainment described in Section 3, during which the cloud's speed continuously decreases as it entrains ambient air (Figure 4), and (b) cloud dispersal via turbulent background flow. At a transition time t * , the cloud's growth via entrainment,ṙ(t ) ∼ t −3/4 , becomes negligible compared with its dispersal via background turbulence,δ Ri (t ) ∼ t (β−2)/2 . The transition time t * between regimes is obtained by balancing these two growth/dilution rates: The prefactors A and B account for the specifics of the turbulence and sources. Note that β = 3 from Richardson's dispersal theory (48), but it is an upper estimate of the power law of δ Ri (t). Instead, experimental studies support lower values for the δ Ri (t) power law (50). Experimental observations in quiescent ambient suggest a transition time t * from a few dozen seconds to minutes, depending on indoor conditions, type of respiratory emission, and initial source emission momentum I 0 (Figure 4; Section 3).

People-Air-Surface-Space Integrated Infection Control Management: Early and Late Regimes for Pathogen Dispersal and Persistence
Both the pre-and post-t * regimes are important to determine indoor exposure risk and to plan for an integrated people-air-surface-space (PASS) (41) management strategy of transmission risk mitigation (Figure 5b,c). The early regime is that of the respiratory cloud evolution. Although relatively short, it can evolve rapidly and over long distances and involve exposure to fresh emissions of potentially high pathogen concentration. Additionally, fresh emissions contain pathogens that have a higher probability of being alive and infectious (see Section 5.3 on pathogen persistence). In this cloud evolution regime, the cloud's droplet payload is not diluted homogeneously in the volume of the indoor space but rather remains highly concentrated in the evolving volume of the cloud and is diluted only by entrainment of ambient air, where the cloud volume (∼r 3 ) scales as r 3 ∼ t 3/4 for t < t * (puff phase), and then ∼t 3β/2 , where β 3 for t > t * (the abovementioned Richardson turbulence dispersal). During the puff phase, exposure from inhalation is also governed by the settling of droplets/nuclei over time. The settling of droplets on surfaces leads to a reduction of risk from inhalation and is governed by the competition between the puff cloud's average velocity and the settling speed of the droplets/particles determining their fallout. For this first regime, source control and physical distancing strategies, possibly integrating new designs of local airflows and office spaces, for example, are critical (Figure 5c,d). Masks reduce the range of all exhalations (51) (Figure 5b). However, unless the masks are fitted to be sealed, leakage occurs with an emerging cloud that has a reduced momentum (Figure 5a; Section 4.2).
The late regime is that of dispersal by background airflow, which can last for hours to days, depending on the indoor ventilation and its maintenance. Here, dilution is more important than in the first regime, so exposure is eventually reduced if the well-mixed limit of the space is reached.   However, the well-mixed limit in occupied indoor spaces is rarely reached, and care needs to be taken to ensure that no high-risk zones of accumulation or airflow recirculation exist. Note that airflow patterns and heterogeneity in an indoor space change on the basis of occupancy number and duration, coupled with the type of venting system and its maintenance. In particular, displacement ventilation, in which fresh air is supplied at relatively low velocity close to the floor and extracted near the ceiling, is known to have energetic advantages as well as advantages in terms of its ability to maintain a fresh breathing zone. It does require sufficiently high ceilings to ensure the development of a transition layer that separates the contaminated upper stratum from the lower fresh air region (52), in addition to ensuring that such a transition layer be sufficiently high and stable to perturbations, such as opening a door or window or connecting to another room (52,53). When mechanical ventilation is not optimal, natural ventilation can remarkably increase the dilution level in an indoor space: Open windows can lead to 20 air changes per hour (ACH). For comparison, indoor spaces with optimized mechanical ventilation can typically reach 10 ACH (54). Air changes are important to estimate with respect to occupancy: A crowded space requires more air changes than a nearly empty one. Thus, the minimum injection of clean air per person is an important additional metric-O(7-10 L/s per person) is recommended (54)-although it was established in the context of odor and temperature comfort. In the context of respiratory pathogen transmission, such as Mycobacterium tuberculosis, a minimum of O(15 L/s per person) is needed to reduce the attack rate (56).

P A S S in t e g r a t e d m a n a g e m e n t P A S S in t e g r a t e d m a n a g e m e n t
It is also important to avoid recirculation of untreated air whenever possible. If air recirculation is necessary, its decontamination (mechanical filters, chemical or light-driven decontamination) is critical, particularly in a time of pandemic. Finally, when leveraging natural ventilation to optimize air changes, it is important to not neglect the potential to mitigate risk during the early-time regime of exhalations. Although opening of windows and doors in a classroom can dramatically increase the ACH on average (tackling the late regime of dispersal), it remains important to maintain nonoverlapping breathing zones of the occupants and avoid ending up extending, via directional persistent flow patterns, the breathing zone of one occupant into the inhalation zone of another, even if masks are in use. Rather than focusing only on wearing masks or on engaging in physical distancing, an integrated PASS approach to indoor risk mitigation must take into consideration the occupants' activities (e.g., speaking versus exercise), crowding, ventilation type and capacity per person, and decontamination of both air and surfaces to assess and mitigate relative transmission risk (Figure 5).

Source Control and Inhalation Protection: Face Coverings
While the use of personal protective equipment of various gradations (e.g., surgical masks or N95 or P100 respirators) is common practice, and even mandated in certain clinical environments during certain procedures and amid pandemics, one of the most passionately discussed topics outside of Asia is the use of face masks by the general public. In that context, masks aim to serve two purposes: to trap the pathogen-bearing droplets of an infected individual (source control) (21) and to protect individuals wearing masks by filtering pathogens in the air. Various studies have shown that the use of masks for source control and protection may help limit airborne transmission (57)(58)(59)(60).
Compared with cloth masks or medical/surgical masks, high-grade P100 or N95 respirators typically offer greater protection. A mask's filtration efficiency (FE) is a key metric that quantifies its protective capacity. FE is calculated as FE = 1 − C D /C U , where C U and C D are the upstream and downstream number densities of particles per volume of air, respectively; C D /C U is also referred to as the penetration. FE depends on the tightness of the material's weave, fiber or thread diameter, and, in the case of nonwoven materials, the manufacturing process. FE is also a function of the particle diameter and flow rate used to challenge the tested material, not accounting for the fit of the mask on the face. The fit factor (FF, or fit) is the ratio of concentration into and out of the mask, accounting for both the FE and seal of the mask. In contrast to penetration, FF is measured when a user is wearing the mask (Figure 5a,b). For a mask to be maximally effective, a tight seal is needed.
High FE of a given material is achieved by combining mechanical and electrostatic filtration approaches, typically Brownian motion, inertial impaction, and electrostatic forces for particle interception and capture. Interception via Brownian motion is effective for small particles below 30 nm, while inertial impaction is more effective for larger particles above the micrometer scale, resulting in a nadir in FE for particles between 100 nm and 1 µm (61). Regarding mechanical interception, the increase in filtration is associated with a pressure drop across the mask; therefore, a resistance to breathing and a consequent compromise between FE and breathability are required. Charging the material enhances filtration without further pressure drop. Thus, procedures that degrade the charge should be avoided, raising key questions about the efficacy of the mask after various procedures of decontamination or prolonged usage. Indeed, increased humidity, for example, may affect filter charge. Data collected in various humidities with varying airflows are conflicting (62)(63)(64)(65).
Previous studies reported FE for cotton-based fabrics ranging from 10% to 60% when tested with particles between 20 nm and 1,000 nm in diameter and with an airflow speed of 5.5 cm/s (66). Some materials had slightly decreased FE when the velocity was increased to 16.5 cm/s (66). A more recent study showed that FE for other cotton-based fabrics ranged between 9% and 96% for particles <300 nm and, similarly, between 14% and 99.5% for particles >300 nm with a flow rate of 1.2 feet 3 /min (0.03 m 3 /min) (67). This wide range of reported FE values shows the importance of not substituting physical distancing with masks. Both should be used under a comprehensive PASS integrated infection control strategy (Figure 5b,c; Section 4.1). Additionally, the current fabric testing paradigm addresses only filtration of particulate matter or droplets, ignoring the effects of the high-momentum turbulent gas cloud from violent exhalations. Indeed, the intensity of exhalation combined with filtration and face seal determines the reduction of momentum and load of particles exhaled in the cloud that escape sideways, posing a residual risk to bystanders (Figure 5a). There remains limited information on respirators' extended use, reuse, and decontamination. For decontamination, certain methods are promising, but further research for validation is needed. The methodologies for testing filtration and persistence of biological agents on masks are not standardized, leading to contradictory results; consequently, it remains difficult to compare the efficacy of decontamination approaches. Understanding the properties of the inhaled particles is an important first step, discussed in the next section (L. Cooper & L. Bourouiba, original data).

Respiratory Droplet Size Distributions and Emission Load: Variability and Need for Standardized Measurements
Motivated by Wells's isolated droplet-versus-aerosol dichotomy (Figure 2), many studies aimed to determine the sizes and numbers of droplets emitted from breathing, talking, singing, coughing, and sneezing. Figure 6a shows such distributions in the form of size probability density functions (PDFs). Droplets range from submicrometer to millimeter sizes. No single instrument can capture this full size range. Moreover, the variation in instrumentation between studies, or even within one study when combining different instruments, requires scrutiny. When normalizing each PDF by its own average diameter (Figure 6b), a more consistent picture of single-mode distributions per instrument type with comparable standard deviations emerges. The few multimode outlier distributions are attributable to the composite instrumentation. Clearly a more fundamental understanding of the composite mechanisms that govern the fragmentation of MS into droplets is required to clarify the bounds of validity variable distributions (Section 5.2).
We separated the reported data according to emission type and instrumentation used (Figure 7). According to the information found in each study, we used or computed the concentration for each emission. First, it is clear that coughs received much more attention than all other types of emission. Second, the variability in reported concentrations is huge, spanning orders of magnitude [O(1) − O(10 5 ) particles/cm 3 ] for coughs, for example. The instrumentation and methods used to measure such particle sizes and concentrations range from interferometric Mie imaging to laser diffraction or single-particle scattering [aerodynamic particle sizing (APS), optical particle counting or sizing (OPC or OPS)] (94-96), electric mobility [scanning mobility particle sizing (SMPS)], inertial or electric low-pressure impaction (cascade and cyclone impactors), and droplet deposition, as well as impaction and manual or automatic counting from slides via imaging.
The instrumentation, calibrations, and tolerance to background noise (background particle load in the air) associated with these approaches vary widely and require a priori assumptions about optical properties of the droplets or particles; their evaporation rates; impact and contact line dynamics (for impaction); hygroscopic properties when condensation is required for counting (e.g., SMPS); and density, shape, sphericity, and compactness (for electric mobility). These properties are particularly difficult to infer for droplets made of complex, heterogeneous biological fluids or compounds (see Section 5.2). Thus, the measurements vary in sensitivity and precision across the size spectrum, and some of these modalities have only a narrow measurement range, are not necessarily calibrated to a common reference, and may only read out data sequentially across size bins, rendering the measured distributions highly dependent on sampling duration (97). These problems are particularly acute when the sampled flow of interest is a transient impulse, such as an exhalation. Additionally, droplet sizes tend to be determined only at a single location (typically away from the source), which poses a basic undersampling problem in space and time. The high variability between the settings of one instrument and between instruments as well as nonstandardized protocols about location of measurement and environmental conditions make it difficult to compare distributions and reconstruct concentrations across studies (Figures 6  and 7). Measurements of droplet sizes are also particularly challenging because of the dynamic nature of the droplets, which change over time following emission; their complex, heterogeneous biological composition, which is typically unknown; and the fact that the multiphase puff cloud, and its payload of droplets, is itself dilute, moves very quickly and transiently, and is spatially and temporally intermittent due to its turbulent flow nature. Thus, it is critical to exercise caution in reducing methodological variability that confounds the measurements and not to overinterpret the features of these distributions.   Figure 7). In some cases, in single studies, instruments are combined to allow a range from a few millimeters to a few micrometers to be captured. (b) When normalized by the mean diameter of each distribution, the graphs mostly collapse on a unified curve shape for the same instrument, but variability remains. For example, one study (92) combined impaction and aerodynamic particle sizing, with a gap in resolution of drop size in the 10-µm transition region (L. Cooper, C. Lu & L. Bourouiba, original data). Abbreviations: APS, aerodynamic particle sizer; DDA, droplet deposition analysis; FRC, functional residual capacity; OPC, optical particle counter; OPS, optical particle spectrometer; RV, residual volume; SMPS, scanning mobility particle sizer; TLC, total lung capacity.

www.annualreviews.org • Fluid Dynamics of Respiratory Infectious Diseases
The choice of instrumentation clearly has a first-order effect on the range of sizes reported, to the point that the differences in reported concentrations and distributions between studies overpower the differences in range and concentration of emission from one respiratory event to another, or in emission from a healthy versus infected patient, within a single study. Therefore, it is necessary to standardize such experiments and have a standard of reporting so as to ensure comparison and integration of the data reported in the field. Indeed, we found that many studies Compilation of results from the literature on quantification of droplet concentrations in a range of respiratory emissions from both infected and healthy subjects, showing a wide range of variation in instrumentation used (color key) (69, 76-79, 83-86, 89-93). The studies that combined instruments are marked with a square in the color corresponding to the other instrument used. Clearly, cough measurements dominated earlier research, while comparatively little attention has been paid to other emissions. However, even for coughs, the concentration of exhaled droplets varies between studies by orders of magnitude. Further analysis of the data from Figure 6 reveals that the variability between studies is greater than the variability between emission events within a study (e.g., studies comparing a cough and a sneeze with the same setups), making it difficult-without further detailed analyses of experimental conditions and parameters of measurements, which, surprisingly, are not always reported in full-to compare studies and draw conclusions regarding the robustness or validity of reported concentrations, amounts, and size ranges of drops emitted for a particular respiratory maneuver (e.g., sneeze versus cough) or between healthy and infected subjects (L. Cooper, C. Lu & L. Bourouiba, original data). Abbreviations: APS, aerodynamic particle sizer; DDA, droplet deposition analysis; FRC, functional residual capacity; OPC, optical particle counter; RV, residual volume; SMPS, scanning mobility particle sizer.
do not provide sufficient information to infer concentration or emission per unit of time. The lack of standardization in publications, in addition to the lack of consistency in protocols or instrumentation settings, hinders the ability to gain insight into the dynamics and variability of respiratory emissions across studies. Examples of missing information that could allow one to compare results in the literature are prevailing humidity and temperature during sampling at various locations, duration of measurement and distance from source, total volume of air sampled, and number of exhalations used for a measurement.
Refinement of measurement techniques and standardization of how such measurements are performed with indirect methods would gain from being complemented by a mechanistic understanding of how droplets form within and outside the RT. Indeed, such a fundamental understanding could enable us to derive physically based expected physiological bounds on the range of mucosalivary droplet sizes and speeds and their distributions. Such a mechanistic understanding could also enable us to establish or rule out hypotheses about droplet formation and allow for advances in diagnostics and risk assessment based on the biomechanics and underlying physics of the complex, multiscale, unsteady fluid fragmentation governing such processes (discussed in the next section).

Fluid Fragmentation and Mechanisms of Selection of Droplet Sizes and Loads: Coupling of Unsteady Flow and Complex Fluid
Violent exhalations are a vital defense mechanism complementing mucociliary clearance. They can be triggered by an increase in mucus secretion, inflammation, ciliary dysfunction, or subtle mechanisms deriving from infection that also benefit the dissemination of the pathogens involved (98)(99)(100). The size and speed distributions as well as pathogen concentration of the droplets generated and emitted in the exhalation process govern the transport, persistence, and infectivity of the pathogen payload. Respiratory pathogens surely adapted to this obligatory phase of hostto-host transmission so as to maintain the cycle of infection on which their survival depends. In order to clarify the origin of the physiological difference among individuals and their ability to transmit, it is crucial to decipher the physics driving the amount, composition, and pathogen load of the droplets emitted, coupled with the multiphase physics of the cloud and its momentum, in various exhalation events.
What generates the MS droplets (see the sidebar titled Mucosalivary Fluid for composition) and the cloud that transports them also determines the range and footprint of contamination. Such generation mechanisms are inherently a physical process of fluid fragmentation: the breakup of fluid from bulk into spray. Such fragmentation occurs when shearing, stretching, or otherwise induced thinning forces imposed on a bulk fluid overcome surface tension and viscous forces (101,102). The former tend to promote fluid breakup, while the latter tend to dissipate imparted energy and so mitigate fragmentation. Surface tension and viscous forces also tend to minimize a fluid's surface area; therefore, higher surface tension and fluid viscosity generally favor formation of larger droplets, while higher air speeds impacting on or shearing a fluid surface tend to favor generation of smaller droplets. More specifically, the hydrodynamic instabilities governing the creation of droplets range from shear driven to capillary driven, with more complex versions of classical instabilities such as the Kelvin-Helmholtz, Rayleigh-Plateau, and Rayleigh-Taylor instabilities (103). The regimes of such fragmentation processes are described by the Weber number (We), which quantifies the competition between kinetic and surface energy. When We is high, kinetic energy dominates over surface energy, and fragmentation of a bulk fluid into droplets is favored (101,104). Other nondimensional numbers relevant to fragmentation are the Ohnesorge (Oh) and Deborah (De) numbers, which quantify the effects of viscosity and elasticity, respectively, compared with capillary effects. The timescales relevant for these instabilities include the relaxation time (λ), which, for polymeric fluids such as MS under stretching, is the average time needed for polymers to relax to their equilibrium configuration. Th capillary time (τ ) characterizes, at first order, the breakup timescale of Newtonian low viscosity fluids.
The fragmentation of bulk respiratory fluid is a multiscale process, analogous to-though more complex than-the fragmentation processes reported in other industrial or environmental systems (17,101,103). The complexity of respiratory fluid fragmentation derives primarily from the facts that (a) the exhalation flow from breathing to coughing or sneezing is inherently unsteady and transient; (b) the geometry and physiology of the RT (see the sidebar titled Respiratory Tract for dimensions and scales) are multiscale, with branching, curved, and compliant airways giving rise to rich nonlinear biomechanical and biofluidic phenomena (16,106); and (c) the liquid phase is a complex viscous or viscoelastic biofluid. Thus, the product of the respiratory fluid fragmentation process is governed by the time-varying dynamics of destabilization as well as by the fluid properties (e.g., Newtonian versus non-Newtonian) (107,108). The physicochemical and biological attributes of the fluid phase are therefore important, as are the (highly transient) flow profiles: There is a rapid flow rate increase and then a slow decrease in the course of an exhalation, the relative period of each phase varying in timescale and volume depending on the type of exhalation (see, e.g., Figure 8 for coughs). We expect inherent variability in droplet production and exhalation contamination footprint between individuals, due to differences in airflow capacity, rheology of the respiratory liquid lining in various regions of the RT, and health status. Yet these properties have historically been poorly characterized (with recent progress discussed in the next sections).

Unsteady exhalation flow.
As mentioned above, the respiratory fragmentation and corresponding emission processes are brief: Droplet emission from exhalation flows lasts 100-200 ms for sneezes and 200-300 ms for coughs (39,109) and is driven by a rapidly increasing then slowly decreasing transient airflow (see, e.g., Figure 8l for coughs). Such short-lived, unsteady fluid fragmentation processes confer on the droplets time-varying characteristics that are important determinants of risk of contamination and, hence, for controlling the chain of transmission (17,108). In contrast, for steady fragmentation (101,(110)(111)(112) the properties of the generated spray droplets are independent of time or are all formed at the same time. Unsteady fragmentation typically occurs by transforming a fluid bulk into a sheet followed by ligaments and then droplets via a series of surface tension-dominated instabilities (103,108,113). In the RT, these include mechanisms of (a) film burst, such as bubble or clogging-bag burst, leading to secondary droplet formation from film rupture and destabilization into ligaments and droplets (114-117) (e.g., Figure 8a,b,d), or (b) fluid shearing by time-varying airflow profile (e.g., Figure 8l), giving rise to a more complex version of the Kelvin-Helmholtz instability, which, in turn, eventually leads to emission of ligaments and droplets as well (104).
In fact, these processes are also directly observable in respiratory fluid fragmentation, as the fluid breakup process is not confined to the RT. The sequence of images in Figure 8a-c shows how bulk MS emitted at the mouth is stretched, over the course of less than 5 ms, to form films that rupture to form ligaments and droplets. The droplets formed, here outside of the RT, are superposed on those formed at various locations within the RT (109). Figure 8b shows more clearly the sheet formation that eventually ruptures (Figure 8c), leading to a rim formation due to the pull of surface tension. Similar to bag burst in jets and cross-flows (116), bubble formation and bursting are also observed in respiratory emissions (Figure 8d). These mechanisms share the fragmentation mechanics of bubble bursts and complex and contaminated thin-film ruptures (114-116, 118, 119). Key determinants of these mechanisms are the fluid composition, including surfactant levels and impurities, which here would be in the form of cells or inhomogeneous distributions of clumps of polymers. The location in the RT determines in large part the composition and hence the rheological properties (discussed further in Section 5.2.3). dominant local modes of breakup. Droplets can be produced by a range of mechanisms dominated by capillary, inertial, elastic, or viscous and elastocapillary effects depending on the location within the RT. Regarding film formation and burst, the film thickness and lifetime depend on the film's location within the RT and are critically affected by the local fluid composition [e.g., the surfactant and polymeric content determining its rheology (115,118,119)]. Such a film arises from MS shearing into film; thick plug formation followed by a burst in the upper airway; or periodic, gentler capillary film formation followed by high rupture speeds inducing effective fragmentation into numerous smaller droplets. The thinner the film is, the higher its rupture speed will be. The idealized case is the static homogeneous film governed by the Culick rupture retraction speed, v ∼ 1/ √ h, where h is the film thickness (120). Within the RT, such processes can also be facilitated by airway wall compliance and oscillation enabling plug formation and rupture when the lining of the walls interact. Such airway closure or semiclosure may be responsible for signatures or modes of distributions of droplet sizes superposed on the macroscale droplet sizes generated from shear or bag burst from larger upper-airway plug fragmentation. In addition to affecting the droplet sizes and speeds, the location of droplet formation might influence a droplet's pathogen load, given that some pathogens preferentially invade certain regions of the RT. M. tuberculosis, for example, preferentially proliferates and remains dormant in the alveoli, while rhinoviruses preferentially colonize upper RT cells.

Multiscale
Finally, the RT serves as a filter in the selection of the droplets that make it out of the host, and also influences the penetration and deposition of pathogen-laden droplets inhaled by an exposed individual. This filtering capacity arises from the inherent multiscale geometry of the RT and the variation of flow regimes within it, as its curved and branching geometry creates well-known patterns of Dean flow, recirculating and vortical flows (15), and possible impaction points and drop trapping on the way up or down the RT. The combination of these mechanisms (shear or stretching, film or plug burst) of fragmentation, in addition to filtration, can lead to various signatures of drop distributions-associated with different compositions as well-superposed on one another. It is this superposition that would be observed when measuring emissions outside the RT at a single point (e.g., as done in Figures 6 and 7). This superposition, coupled with the unsteadiness of the process, makes it more difficult to disentangle the signature of each fragmentation mechanism involved. Indeed, unsteadiness even for simpler configurations of fragmentation introduces various apparent modes into the droplet distributions, even if the process of fragmentation continues to produce unimodal signatures of drops, though with properties (mean, standard deviation) varying with time (17, 108).

Mucosalivary fluid composition and rheology.
The rheology of the fluid lining the RT is important to determine. MS is a dilute polymer solution based on a Newtonian solvent-water-and a range of polymers that provide the fluid with its viscoelastic properties. Figure 8a-c illustrates the ability of such fluids to form persistent filaments and delay capillary breakup when stretched. Indeed, even in minute amounts, the polymers found in MS can yield dramatic differences in the behavior of the fluid under extension, in comparison to water, due to the increased resistance within extended ligaments to flow. The dynamics observed involve nonlinear effects arising from the finite extensibility of the polymer chains within the fluid. Moreover, under certain flow regimes and when capillary, viscous, and inertial effects balance, additional structures can appear, such as so-called beads on a string (121,122) (Figure 8e-h). These can evolve to create merger and gobbling effects (109) (Figure 8f ). The additional dynamics of viscoelastic stretching in a transient turbulent background flow, ligament persistence prior to drop formation, and beads-on-a-string and gobbling/merger behavior all contribute additional features to the final droplet size distribution-broadening it on the large drop size end-of an exhalation when compared with the distribution from a steady fragmentation process of a Newtonian fluid (123). Similarly, the ligaments are observed to persist at finer scales much longer than water filaments do. Despite the obvious importance of MS properties in determining respiratory fragmentation outcome, the rheology of the human respiratory system fluid lining is not well known. Although several recent studies have reported viscosity and elasticity values, they are inconsistent. We attribute such inconsistencies in part to the methodology of sample collection and preservation, sample age at time of measurement, and precision of measurement. These confounding factors are often not reported, making it difficult to interpret the results. In past and ongoing research (N. Bustos & L. Bourouiba, original data), we have characterized the properties of MS secretions in both healthy and influenza-or cold-infected individuals and noted the strong sensitivity of the results to the method of collection and time of measurement of the sample properties (Figure 8i-k). Such measurements are taken via capillary thinning, during which a sample of MS is stretched between two plates to form a liquid bridge-a filament. A laser micrometer or direct optical measurement tracks the midpoint radius of the filament as it thins under the action of capillary forces. Once fluid inertia can be neglected, the filament thinning process is initially governed by a viscocapillary force balance in which viscous extensional stresses oppose the increasing capillary pressure, followed by a later elastocapillary phase in which stresses generated by the stretching of the polymer chains dominate (Figure 8i). From measurements of the time evolution of the filament radius, one can obtain the breakup (i.e., capillary) time of the filament and the relaxation time λ of the fluid via the evolution R(t)/R(0) ∼ e −t/3λ , where R(t) is the diameter of the filament (Figure 8i). Other techniques, such as pendant drop and shear rheometry, provide measurements of surface tension Local rim Reynolds number: where v b and b are the rim velocity and thickness as measured in the local noninertial reference frame of the rim Local rim Deborah number: De = λ/ ρb 3 /σ in the noninertial frame of the rim of size b and shear viscosity. Figure 8k clearly shows shear-thinning behavior for MS, though in the regime relevant for human exhalations, the viscosity is close to (albeit a little higher than) that of water.
Finally, what is clear from our study is that MS properties are delicate to characterize, as rapid degradation of elastic properties due to sedimentation and clumping of the polymeric content makes the samples inhomogeneous and watery over the first hour (Figure 8i). It is also critical to be cautious when determining inter-and intrasubject variability in health versus infection, given the wide range of relaxation time values we have observed on the basis of more than 560 independent measurements in more than 70 healthy and infected individuals (N. Bustos & L. Bourouiba, original data) (Figure 8j). Therefore, if researchers do not either slow down sample degradation or rapidly measure the rheological properties during collection, they will obtain erroneous results, which could in part explain the wide and contradictory range of values reported in the literature. In fact, most measurements of relaxation times of MS in the literature do not include the sample age, making it difficult to infer the physiological relevance of the values reported. Finally, now that we have gained some clarity regarding the unsteadiness, rheology, and complexity of the problem, it is crucial to break it down into tractable parts and derive mechanistic insights to determine how each part of this complex multiscale problem contributes to the final spray and emission characteristics. Next, I discuss how insights obtained from a standardized analog canonical fragmentation system that captures some of the essential physical features of transient unsteady fragmentation processes relevant to respiratory emissions can help shed light on the role of the various components of the problem (Figure 8m-o).

Canonical, axisymmetric unsteady fragmentation.
The axisymmetric unsteady fragmentation process resulting from the impact of a droplet on a target of comparable size (Figure 8m,n) results in rich, coupled, spatially and temporally varying, multiscale, nonlinear processes that can be broken down into sheet and rim evolution and determination of droplet size and speed distributions (17,124). Upon impact, the droplet is transformed into a rapidly expanding two-dimensional fluid sheet whose rim continuously destabilizes, generating fluid ligaments that, in turn, destabilize and shed droplets, mostly one at a time throughout the sheet expansion process. Rim destabilization depends on the impacting droplet's We and fluid properties: Increasing the viscosity and relaxation time of the surrogate fluid reduces the number of ligaments generated and droplets shed (Figure 8m). Manipulation of these fluid properties results in fragmentation that closely resembles that of fresh human saliva (Figure 8n), showing elongated ligaments and beads on a string reminiscent of those observed via direct imaging of respiratory fluid fragmentation from physiological emission (Figure 8a,b,e-h).
This canonical fragmentation problem provides fundamental insights regarding when the roles of viscosity and elasticity must be incorporated explicitly and when they can be neglected for their effect on the spray of interest. Indeed, in the Newtonian inviscid limit, the droplet size and speed distributions can be predicted exactly and are given by a superposition of Gaussian distributions with time-varying mean, with smaller and faster droplets shed early in the unsteady process and larger and slower droplets shed later (17,124). Therefore, the unsteadiness of the fragmentation process inherently causes the observed skewness of the droplet size and speed distributions. The prediction is predicated on the balance of inertial and surface tension forces in the local reference frame of the rim. This criterion holds for complex fluids for which Re 8 andDe 16, whereRe andDe are the Re and De in the local rim reference frame (Figure 8o). Outside of this region, modifications in the computations of droplet sizes and speeds would be required to account for nonnegligible viscosity and elasticity effects of the MS (Figure 8o).

Persistence, Exposure, and Infection
In addition to the droplet sizes, it is important to determine whether pathogens are present and infective in the droplets as a function of droplet size and history of emission from the source. This determination depends on the droplet volume sampled, handling procedures, and amplification approaches (126,127). The persistence of pathogens refers to their ability to remain alive and multiply, successfully penetrate and hijack a target cell in the case of viruses, and multiply within the cell with various degrees of virulence or effectiveness. For most pathogens, we have only a rudimentary understanding of their persistence and the mechanisms shaping it. We know that SARS-CoV-2 remains infective even after 24 h of deposition on cardboard and 2-3 days on stainless steel (128). For comparison, respiratory syncytial virus or influenza can survive on nonporous surfaces for at least 6 h (129). In contrast, rhinoviruses and adenoviruses have been reported to survive for days on nonporous surfaces (130). When airborne, the measles virus was found to maintain its infectious state for at least 1 h (131, 132), while aerosolized M. tuberculosis has a half-life of approximately 6 h under controlled (rotating drum) conditions (133). Similarly, aerosolized H3N2 influenza virus remains infective for at least 2 h (134) and aerosolized SARS-CoV-1 for 1 h (128). Recent studies of aerosolized SARS-CoV-2 found viable virus up to 3 h postaerosolization (128,135) and up to 16 h in suspended aerosols (136).
Despite these reports, major questions remain unresolved regarding the persistence of pathogens in the environment, whether deposited on surfaces or suspended in the air in a dried or semidried state. The composition of the fluid phase, environmental conditions (temperature and humidity), macro flows, microphysics, and fluid dynamics involved at the small or intermediate scale all play a role in determining pathogen survival. In the context of SARS-CoV-2, lower temperatures appear to promote survival (137)(138)(139), while the pH of the carrier fluid has a negligible effect (137,140); however, the influence of humidity on SARS-CoV-2 survival seems to depend on the characteristics of the carrier fluid (135). Clearly, these are still preliminary results. The fluid dynamic, biological, and environmental factors are extremely entangled, and current experimental results and data do not yet allow us to decipher the key underlying mechanisms governing the wide range of results observed. The interplay between the physics and biology of pathogenbearing droplets in relation to pathogen persistence is very rich, and disentangling these coupled relationships will remain a fascinating problem.

DISCUSSION AND PERSPECTIVE
In the context of respiratory diseases, the generation of mucosalivary droplets and their emission as part of a multiphase turbulent gas cloud, coupled to their evolution in the environment through multiphase droplet and phase-change physics of evaporation and condensation, ultimately determine the nature, composition, and size of the droplets or the droplet nuclei/residue inhaled by a potential host. The size and composition of the inhaled particles determine not only the range and evolution (adhesion or resuspension for example) of the nuclei in the environment but also their potential for penetration deep into the RT and, thus, access to preferential target tissue for the pathogens carried. Inoculation of animals suggests that for the same viral load, the inhalation of an atomized solution results in higher infection and death rates in comparison to dry intranasal inoculation (33,141).
There remain important areas of research to examine, such as deposition and penetration in the context of droplet residues, as a function of the history of the complex biological particles carrying the pathogens of interest. In addition, rich questions regarding the interplay among transmission physics, complex biochemistry, and host immunology remain open. Of special interest regarding this interplay might be the mucous barrier of the upper airways, as it is the first line of defense against respiratory infectious agents (98). Here, the history of the pathogen-laden droplet (evaporation, desiccation, rehydration) determines, in part, the success of host-cell infiltration and infection. Finally, these questions of persistence are at the crux of transmission in indoor spaces, including nosocomial transmission occurring in (a) health care facilities, known to have been critical in the spread of coronaviruses such as MERS (142), and (b) mostly households and indoor environments, for SARS-CoV-2.
At the clinical and practical levels, important questions remain pertaining to generation of contaminated droplets from exhalations as well as from procedures such as intubation or high-flow nasal cannula treatment with oxygen and humid air (143). These procedures are also important in mitigating potential transmission to health care workers and patients due to life-saving procedures. Other questions concerning the environment involve the need for a better understanding of the mixing and dispersal of contaminants, for example, from the motion of people in a confined space or from natural or mechanical ventilation, as well as the need to better characterize and innovate in low-cost ventilation technologies able to produce stable stratified environments (e.g., displacement ventilation) with breathing zones that are free of contaminants.
In summary, we need a mechanistic understanding of the interfacial physics, including the rheology of respiratory secretions in health and infection, and the associated complex multiscale and varied mechanisms of fluid fragmentation in a range of configurations, such as unsteady processes typical of respiratory events (17). Such an understanding will be critical for building the foundation required to capture quantitatively, and thus predict, the complex fluid processes governing spray generation in a range of configurations and dynamic regimes during exhalations from healthy and infected individuals. Similarly, understanding such processes will be crucial for deciphering the role of interventions, such as masks and the environment, including ambient moisture and temperature. With regard to moisture, the large-scale epidemic data sets examined to date seem to suggest reduced transmission in the 40-60% range of relative humidity (C. Verheyen & L. Bourouiba, original analysis), though it has also been argued that lower humidity favors rapid droplet evaporation and hence increases air contamination. It has also been argued that SARS-CoV-2 is comparatively fragile and that viruses that dry faster die faster. Yet, laboratory-based controlled aerosolization experiments with SARS-CoV-2 have resulted in somewhat conflicting results. Moreover, the turbulent cloud of human exhalations shields droplets from immediate evaporation. All this suggests that the role of humidity and temperature and their significance in the chain of respiratory disease transmission are still not mechanistically understood; thus, caution in interpreting claims about their effect is needed. Finally, an understanding of the effect of airflow in a health care environment versus an office can help pave the way toward innovations in infection control that can leverage such processes to reduce transmission in common shared spaces.
COVID-19 has brought to the fore the urgent need to understand and mitigate host-to-host transmission during the ongoing global pandemic. However, the need to base our understanding of this obligatory phase of a respiratory pathogen's life cycle on a solid mechanistic understanding has always been urgent and applies equally to the transmission of seasonal influenza, TB, and a wide range of emerging and reemerging respiratory infectious diseases. Such a mechanistic understanding provides the opportunity to predict and control the risk of transmission. It aids in early detection and allows for new pathways for drug discovery tackling the transmission phase. It also allows for mitigation of epidemics in the critical early phase, prior to the availability of effective vaccines or drugs. In the context of respiratory infectious diseases, the fluid dynamics and biophysics of pathogen extraction, emission, persistence, and host invasion and infection (17) give rise to a large set of biomedical, physiological, and complex and multiscale fluid dynamics questions that deserve to be addressed through rigorous and mechanistic approaches combining novel tools for experimental quantification as well as theoretical and numerical modeling. Finally, it will be crucial to integrate such knowledge in the multifaceted context of testing, tracking, treatment, and vaccination strategies, as well as in creating predictive epidemic models over the months and years to come in order to ensure greater resilience to the inevitable recurring epidemics of respiratory pathogens.

DISCLOSURE STATEMENT
The author is not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.