key: cord-1028517-24pp67fw
authors: Choi, Y.-W.; Tuel, A.; Eltahir, E. A. B.
title: An environmental determinant of viral respiratory disease
date: 2020-06-07
journal: nan
DOI: 10.1101/2020.06.05.20123349
sha: 52937652033e049ef82368a7c662bcf3c0a1babd
doc_id: 1028517
cord_uid: 24pp67fw

The evident seasonality of influenza suggests a significant role for weather and climate as one of several determinants of viral respiratory disease (VRD), including social determinants which play a major role in shaping these phenomena. Based on the current mechanistic understanding of how VRDs are transmitted by small droplets, we identify an environmental variable, Air Drying Capacity (ADC), as an atmospheric state-variable with significant and direct relevance to the transmission of VRD. ADC dictates the evolution and fate of droplets under given temperature and humidity conditions. The definition of this variable is rooted in the Maxwell theory of droplet evolution via coupled heat and mass transfer between droplets and the surrounding environment. We present the climatology of ADC, and compare its observed distribution in space and time to the observed prevalence of influenza and COVID-19 from extensive global data sets. Globally, large ADC values appear to significantly constrain the observed transmission and spread of VRD, consistent with the significant coherency of the observed seasonal cycles of ADC and influenza. Our results introduce a new environmental determinant, rooted in the mechanism of VRD transmission, with potential implications for explaining seasonality of influenza, and for describing how environmental conditions may impact to some degree the evolution of similar VRDs, such as COVID-19.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. in the mid-latitudes, 1 and in the case of COVID-19, some countries have clearly experienced 43 widespread transmission and an explosive growth in cases, while in others, the outbreak seems 44 much more constrained. 2-4 It is evident that social determinants play a major role in controlling 45 transmission, especially given the success of social distancing policies implemented in response 46 to COVID-19. However, this does not necessarily imply that the environment plays no role in 47 shaping VRD spread, as highlighted by the clear seasonality of influenza in mid-latitude 48 countries. 1 49 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 7, 2020. survival of several VRD pathogens, 5-7 although the effect of temperature seems weak in the case 52 of the SARS-CoV-2 virus responsible for COVID-19 8 (Fig. S1 ). High UV radiation is also 53 believed to suppress viral activity and infectivity in the case of influenza viruses 9 and possibly 54 SARS-CoV-2. 10 Additionally, evidence has emerged that viral shedding in mammals is enhanced 55 at low temperatures, 11 making the case for strong biological controls on VRD prevalence. 56

Yet, because VRDs are primarily transmitted by droplets exhaled by infected subjects, 57 environmental conditions may also play a major role in shaping their spread. 12,13 Previous studies 58 have argued that cold and dry environments were conducive to the survival and transport of 59 VRD-infected droplets, unlike warm and humid environments. 1, 6 This hypothesis seems 60 supported by empirical relationships applied to country-level data, 7,14,15,16 though in the case of 61 COVID-19 initial results suggest that weather and climate conditions may have limited effects 62 on the spread of the disease. 17,18 63 One important limitation of such studies is their focus on temperature or humidity as separate 64 covariates to understand or predict VRD prevalence. Different relationships are developed for 65 tropical and mid-latitude countries 1 although the physics of droplets is the same. Additionally, 66 relationships are sometimes found to be non-monotonic: in the case of COVID-19, the 67 transmission efficiency may first be enhanced as temperature and absolute humidity drop, and 68 then decline beyond a certain threshold. 15 Therefore, while evidence points to some degree of 69 environmental control on VRD spread and prevalence, the lack of a consistent and physically-70 based framework makes it all the more difficult to assess. 71 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 7, 2020. coughing. 24 After emission, droplets can contaminate nearby surfaces, or disperse as aerosols and 85 may infect subjects who inhale them. This idea, first developed by Wells, 25 led to the 86 discrimination of "large" and "small" droplets, and has since then influenced strategies to control 87 the spread of infection according to whether the disease was thought to be transmitted primarily 88 through large or small droplets. 26 Droplet diameter cut-offs usually range between 5 and 10µm, 27 89 and the typical associated distance varies between 1.5 and 2m. 28 More recent studies have 90

shown, however, that these arbitrary droplet size cut-offs do not reflect the actual trajectories of 91 exhaled droplets. The dynamics of droplet evaporation and evolution are indeed very dependent 92 on the characteristics of the complex multiphase turbulent flow which the droplets exist in 29 as 93 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint 6 well as background environmental conditions. 12 While influenza transmission has been shown to 94 occur through both the large and small droplet route, 21 at this stage, COVID-19 is still believed 95 to be mainly transmitted by the large droplet path, 30 although aerosol transmission may also be 96 possible. 31 In any case, exhaled droplets are still the major infection route, which implies that 97 environmental controls on droplet evaporation and disappearance may play an important role in 98 determining the spread of the disease. isotropic gaseous media were controlled by the equilibrium between heat and mass exchange at 104 their surface. Both mass and heat transfer involve ambient temperature and humidity, and are 105 therefore strongly constrained by environmental conditions. In steady-state, mass and heat 106 transfer exactly compensate, and one finds that the radius ! of a droplet evolves according to 33 : 107

where &' is the ambient relative humidity, -. the ambient temperature, & , the specific gas 109 constant for water vapour, ρ liquid water density, + , the latent heat of vaporisation, 0 the 110 thermal conductivity of air, 4 the water vapour diffusion coefficient, and 2 3 (T) the saturation 111 vapor pressure at temperature T given by the Clausius-Clapeyron equation. We then define the 112 Air Drying Capacity (ADC, in mm 2 /hr) as the rate of decrease of the droplet surface area: 113

CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint 7 ADC is therefore an atmospheric state-variable uniquely related to air temperature and humidity 115 only. For typical ranges of air temperature and humidity, ADC varies between 0 and 15 mm 2 /hr 116 ( Fig. 1-a,b) . It is a linear function of both relative and specific humidity, but a non-linear 117 function of temperature, consistent with the Clausius-Clapeyron law. ADC strongly controls the 118 time it takes for a free-falling droplet to evaporate, and therefore the diameter cut-off between 119 "large" droplets, that reach the ground before evaporating, and "small" droplets, which turn into 120 aerosols ( Fig. 1-c) . At low ADC values (0-1 mm 2 /hr), only droplets larger than about 25µm will 121 be able to contaminate nearby surfaces, while for high ADC (>10 mm 2 /hr) that threshold moves 122 up to 60µm. Additionally, the potential range of such large droplets is also severely reduced as 123 ADC increases, because they can remain in the air for a significantly shorter time ( Fig. 1-c, Fig.  124 S2). Small (<10µm) droplets -a size typically emitted during normal speech -while never able 125 to contaminate surfaces under the typical range of ADC values, can however potentially be 126 inhaled by subjects in the vicinity of the emitter. Their fate is largely controlled by ADC: a 10µm 127 droplet will evaporate in as much as 25s or as less as 0.5s depending on the background ADC. 128

This may be particularly relevant for VRD pathogens whose infectivity declines once in the dry 129 aerosol phase. 130 131 Data 132 6-hourly temperature, dew point temperature and surface pressure data at 0.75° spatial resolution 133 between 1979 and 2018 were obtained from the ERA-Interim reanalysis 34 (available at 134 http://apps.ecmwf.int/datasets/). Since ERA-Interim is only available up to 2019, we used 6-135 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. GHI(t, C) = # positives, country C, week t population, country C × avg # weekly reports, all countries avg # weekly reports, country C 157 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. 

The spatial distribution of annual-average ADC shows a somewhat meridionally symmetric 162 pattern. The lowest values, between 0-2 mm 2 /hr, can be found above 60° latitude in each 163 hemisphere and over land areas around the equator (Fig. 2-a) . The subtropics in each hemisphere 164 exhibit high ADC values, particularly over the large deserts of North Africa and southwest Asia 165

where temperature is high and humidity is low. Australia, India and the Western United States 166 are all characterised by relatively high ADCs. The situation during winter and spring is overall 167 quite similar, though with notable regional differences ( Fig. 2-b,e) . Europe and Eastern North 168

America both show particularly low ADC values during winter, much lower than in China where 169 ADC remains mostly above 2 mm 2 /hr. ADC over south-eastern Brazil is also at its minimum 170 ( Fig. 2-e) . By contrast, most of Africa, and specifically its large population centres of Ethiopia, 171

Egypt and Nigeria, all show high ADCs. The same can be said for India, particularly during 172 spring. However, consistent with the summer monsoon cycle, ADC becomes much higher during 173 and after the monsoon season over Western Africa and the Sahel region, as well as India, as 174 high-ADC bands move northwards with the rains (Fig. 2-c, 

North America, ADC increases during summer, but remains rather low at around 5 mm 2 /hr. A 176 video showing the space-time evolution of ADC is included with Supplementary Information.  177 178 Testing the relevance of ADC for VRD prevalence 179 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint

The spatial and temporal distribution of influenza cases is highly consistent with that of ADC 180 ( Figs. 3-a, 4-a,b ). ADC appears to set a strong upper bound on influenza prevalence that applies 181 to all countries with available data: influenza has very limited prevalence at ADCs of 5 mm 2 /hr 182 or larger, and clearly increases as ADC approaches 0 (Figs. 3-a, 4-a,b) . The annual cycles of 183 ADC and influenza are also highly consistent, with a clear peak in the disease around when ADC 184 is at its lowest (Fig. 3-c) . Africa stands out due to high ADC values and low influenza 185 prevalence, whereas Europe and North America have low ADCs and generally higher numbers 186 of influenza cases ( Fig. 4-a,b) . While socio-economic factors also play a role in modulating the 187 spread of the disease, it is striking that ADC still constrains the upper end of the range of 188 observed prevalence, consistent with its effect on droplets -the vectors of transmission, 189 particularly the rapid increase in the time needed for droplet evaporation as ADC approaches 0 190 (Fig. S2 ). By contrast, air temperature ( Fig. 3-b) and specific humidity ( Fig. S3-a,c) do not show 191 such clear relationships to influenza, although the annual cycle of temperature appears quite 192 consistent with that of influenza prevalence (Fig. 3-d) . Results for relative humidity do show 193 some enhancement of influenza as the air becomes moister (Fig. S3-b) , but its annual cycle 194 seems quite off when compared to that of influenza incidence (Fig. S3-d) . 195

Interestingly, the spatial distribution of ADC during winter and spring also shows some 196 resemblance to the global map of confirmed COVID-19 cases (Fig. 4-c,d, Fig. S4 South America and Australia is lower (10-500 per million), and even less than that in Africa and 201 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint India. Naturally, many other factors come into play here, like connectivity to the rest of the 202 world, population density and localisation within countries, and public policy measures like 203 social distancing or lockdowns. The number of reported cases also suffers from biases, especially 204 undercounting. Still, countries with low (respectively high) ADCs generally seem to correspond 205 to higher (respectively lower) disease prevalence, a tendency that seems robust to considerations 206 of income levels or test numbers performed by different countries (Fig. S5) . 207 208

VRDs are primarily transmitted between humans through droplets exhaled by infected hosts. 210

Environmental determinants that affect the fate of these droplets can therefore influence 211 transmission of these diseases. We introduced here a new variable, ADC, motivated by droplet 212 growth theory first developed by Maxwell 32 . ADC includes the effects of both temperature and 213 humidity on droplet evolution in the atmosphere. Compared to temperature, ADC turns out to set 214 a much more coherent constraint on influenza prevalence (Fig. 3) . The empirical relationship of 215 ADC with the average prevalence of both influenza and COVID-19 for various world regions is 216 consistent with its physical effects on the decay of droplets through which VRDs are transmitted. 217 ADC directly constrains the evaporation of airborne droplets, potentially setting a strong upper 218 bound on VRD spread and prevalence that appears valid regardless of socio-economic factor 219 (Figs. 4, S5) . It is important to note that ADC also indirectly impacts the survival of liquid 220 droplets even once they have landed on surfaces; high-ADC conditions lead to rapid evaporation 221 from a surface. The transmission of the viruses responsible for COVID-19 and influenza is thus 222 likely impacted by ADC. 223 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint Significant relationships between temperature or humidity and influenza dynamics have been 224 suggested in previous studies for individual countries 14 and temperate regions 6 , but it appears 225 that neither variable, unlike ADC, is able to explain the observed global pattern of influenza 226 prevalence (Figs. 3, S3) . In particular, while specific humidity seems to have a strong effect on 227 influenza virus survival, potentially affecting its transmission during the relatively low-specific 228 humidity peak season in mid-latitude countries 6 , peak influenza in different countries occurs at 229 both times of minimum and of maximum specific humidity 1 . The environmental determinant of 230 VRD proposed in this study has important implications for consistently explaining the 231 seasonality of influenza across the globe. Two kinds of favourable environments have been 232 suggested for influenza transmission: "cold-dry" (as in mid-latitude countries) and "humid-233 rainy" (as in tropical countries), 1 in order to reconcile discrepancies in explaining seasonality of 234 influenza at the global scale. 36 However, if specific humidity were the determinant variable 235 impacting transmission, humid countries would hardly experience any influenza outbreaks, 236 especially during their wet season. Two clusters of high influenza prevalence can be found in the 237 WHO dataset, at both very low and very high humidity (Fig. S6 ). What they have in common are 238 low ADC values, and in fact each cluster corresponds to the period of annual minimum ADC in 239 mid-latitude and in tropical countries. The two proposed influenza regimes may therefore be 240 reconciled by considering ADC framework proposed in this paper. While humidity and 241 temperature may mimic influenza dynamics at the scale of individual countries, 6,14 these same 242 relationships seem less valid when assumed for the world as whole and do not explain the large 243 discrepancies in influenza prevalence between countries. At the global scale, it appears that the 244 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. and north-eastern America, where ADC is low, whereas regions with higher ADC have 260 experienced a much slower growth in cases. In particular, Africa and India stand out by high 261 ADC values and low COVID-19 prevalence ( Fig. 4-a, Fig. 5) . A recent study argued for a 262 reduced transmission rate in Africa potentially linked to the environment, consistent with its 263 higher ADC. 4 Admittedly, COVID-19 data is quite limited, and very much impacted by policy 264 measures taken to limit disease spread. In addition, testing has been inconsistent across the 265 world; in many countries, reported cases largely refer to individuals showing visible symptoms 266 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint of the disease, leaving out many asymptomatic cases. Similarly, influenza data is not free from 267 biases (see Methods). This should make us careful in drawing final conclusions. Still, average 268 influenza and COVID-19 prevalence show a similar and consistent relationship to ADC (Fig. 3) . 269

Since the high seasonality of influenza is coherent with that of ADC (Figs. 3-c) , this suggests 270 that COVID-19 may also follow ADC seasonality, with potential implications for the current 271 disease hotspots of Europe and north-eastern America, where ADC will increase as summer 272 approaches (Fig. 5) . In regions of Asia outside India, where the seasonality of ADC is very 273 limited, environmental determinants will probably not play much of a role in shaping COVID-19 274 dynamics in the months to come. However, the situation may be more worrying in India and 275

Western Africa, two regions where the summer monsoonal systems will bring low ADC 276 conditions offering favourable conditions for the spread of the disease if effective preventive 277 measures are not taken. 278

Nevertheless, our results present some important caveats. First, indoor heating and cooling will 279 substantially move ADC away from its outdoor value, which we considered in our analysis. 280

Transmission can occur indoors where temperature can be very different from outdoor 281 conditions. Typically, in mid-latitudes, wintertime ADC is much higher inside than outside, and 282 vice-versa during summer. Still, in regions where air conditioning and heating are available, 283 conditions indoors should tend to exhibit much less seasonality than outdoors. In addition, the 284 evident seasonality of influenza makes a strong case for the role of outdoor conditions, given that 285 people spend much of their time indoors year-round 36 . The seasonality of VRDs may therefore 286 primarily reflect outdoor ADC. 287 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint Second, biological determinants of virus survival may be strongly correlated to ADC, meaning 288 that part of the ADC-VRD prevalence relationship may be explained by the effect of 289 environmental conditions on the virus itself, and not on the transmission pathway. In particular, 290 temperature is thought to affect the survival of influenza viruses 5 , though we fail to find a 291 coherent signal in global data (Fig. 3-b) . Similarly, in the case of influenza and, possibly, 292 COVID-19, UV radiation is believed to be severely detrimental to viruses. 9 Low ADC is 293 unmistakably associated with low incoming UV, but at higher levels the relationship becomes 294 less clear (Fig. S7) . Therefore, ADC and UV radiation may well interact and strengthen their 295 respective effects. 296

As the COVID-19 pandemic progresses, better data will become available, and it will become 297 possible to test for the robustness of the relationship between its prevalence and ADC values. So 298 far, evidence points to an influenza-like behaviour, with a pronounced seasonality and mid-299 latitude countries most at risk from late fall to early spring. For the latter, environmental 300 conditions will therefore probably be conducive to a second wave in late 2020, while in Western 301

Africa and India, summer 2020 may bring about favourable conditions for efficient spread of the 302 disease. However, as stressed earlier, conducive environmental conditions are not sufficient to 303 cause VRD spread, and significant outbreaks triggered by social behaviour can occur even under 304 relatively unfavourable environmental conditions. 305 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint

Environmental Predictors of Seasonal

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. Correspondence and requests for materials should be addressed to atuel@mit.edu. 415. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. (Table S1) February-April 2020 ADC against concurrent accumulated confirmed COVID-19 cases for 108 435 countries (Table S1 ). (d) Same as (c), but for the 50 US states. Red, green, light blue, yellow, 436 blue and black colors in (a,b,c) respectively indicate North America, South America, Europe, 437 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted June 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted June 7, 2020. . https://doi.org/10.1101/2020.06.05.20123349 doi: medRxiv preprint