Environment influences SARS-CoV-2 transmission in the absence of non-pharmaceutical interventions

As COVID-19 continues to spread across the world, it is increasingly important to understand the factors that influence its transmission. Seasonal variation driven by responses to changing environment has been shown to affect the transmission intensity of several coronaviruses. However, the impact of the environment on SARS-CoV-2 remains largely unknown, and thus seasonal variation remains a source of uncertainty in forecasts of SARS-CoV-2 transmission. Here we address this issue by assessing the association of temperature, humidity, UV radiation, and population density with estimates of transmission rate (R). Using data from the United States of America, we explore correlates of transmission across USA states using comparative regression and integrative epidemiological modelling. We find that policy intervention (`lockdown') and reductions in individuals' mobility are the major predictors of SARS-CoV-2 transmission rates, but in their absence lower temperatures and higher population densities are correlated with increased SARS-CoV-2 transmission. Our results show that summer weather cannot be considered a substitute for mitigation policies, but that lower autumn and winter temperatures may lead to an increase in transmission intensity in the absence of policy interventions or behavioural changes. We outline how this information may improve the forecasting of SARS-CoV-2, its future seasonal dynamics, and inform intervention policies.


Introduction
When analysed jointly, the R 0 of all USA states are fairly well predicted by all explanatory 117 variables included in the regression model (i.e. population density, temperature, absolute 118 humidity and UV radiation), with an overall model adjusted r 2 of 58% (supplementary 119 table S1). However, UV radiation is a very weak predictor of R 0 , while temperature 120 and absolute humidity show sufficiently strong correlations with each other (r = 0.85) 121 that we cannot disentangle their contributions to R 0 due to high inflation of variances 122 (supplementary table S1). This is further demonstrated through principal components 123 analysis, where temperature and absolute humidity fall along the same principal compo-124 nent axis (supplementary figure S1). We therefore focused on temperature as the best 125 fitting climate variable (assessed by Pearson's r, supplementary table S2). 126 When regressed against temperature and log 10 -transformed population density only, we 127 find that R 0 significantly increases with population density and decreases with tempera-128 ture ( fig. 1; both p < 0.001, table 1). We see a stark difference, however, when analysing 129 R t during lockdown (defined as the mean R t recorded over the 14 day period following a 130 stay at home order): much less of the variation in R t is explained by the regression model 131 (adjusted r 2 = 18%), vastly lower model coefficients for explanatory variables (i.e., much 132 lesser correlations; supplementary table S3, but note that population density is still a 133 significant predictor), and much lower R t estimates overall (paired t 39 = 21.1; p < 0.001; 134 figure 1b). Additionally, if we regress the combined R 0 and R t estimates against tem-135 perature and population density, using lockdown as a binary interaction term, we find 136 a significant interaction between lockdown and temperature (p < 0.001, supplementary 137 table S5), i.e. lockdown mediates the effects of climate on transmission. 138 The strong correlates of population density and temperature on R 0 across the United 139 States were echoed in our climate-driven Bayesian modelling of daily variation in R t .

140
Posterior medians of the scaled coefficients of (log 10 -transformed) population density and 141 daily temperature were 0.68 and −0.48, respectively. These coefficients were strongly 142 supported (both Bayesian probabilities > 99.9%), and suggest that greater population 143 6 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101 temperature ( Table 1: Population density and temperature are drivers of R 0 at state-level in the USA. Multiple r 2 = 0.6037, adjusted r 2 = 0.5822, F 2,37 = 28.18, p < 0.001. Scaled estimates are coefficients when predictors are scaled to have mean = 0 and SD = 1. Scaling our explanatory variables means our coefficients are measures of the relative importance of each variable. In contrast to our epidemiological modelling, temperature is a greater driver of pre-lockdown R 0 than population density (log 10 -transformed is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101 (a) (b) Figure 1: R 0 is affected by the environment, but the impact of lockdown is greater. (a.) R 0 plotted against temperature (averaged across the two weeks prior to the R 0 estimate) and log 10 -transformed population density (people per km 2 ) for each USA state (grey points). Surface shows the predicted R 0 from the regression model (table  1). Temperature has a negative effect on R 0 at state-level in the USA, whilst population density has a positive effect (table 1). (b.) The mean R t for the two weeks following a state-wide stay-at-home mandate (i.e., during lockdown) plotted against average daily temperature for the same period and log 10 -transformed population density. The effects of temperature and population density are much weaker in the mobility-restricted data and R is reduced overall. The same colour scale, given in the centre of the figure, is used across both sub-plots.  Figure 2: The relative importance of temperature and population density as drivers of pre-lockdown R 0 . (a.) Heatmap of the regression model R 0 predictions, with USA state-level R 0 point estimates overlaid. High population densities and low temperatures drive increases in SARS-CoV-2 R 0 . This is a 2D representation of the regression plane in fig. 1a, using the same colour-scale. (b.) Residuals from a linear regression of R 0 against log 10 -transformed population density ("Corrected R 0 "), plotted against temperature. This illustrates that, when considering population density alone, R 0 is overestimated in cold states and underestimated in warm states. After accounting for population density, there is a significant effect of temperature upon R 0 (see table 1). In both figures, points are highlighted with standard two-letter state codes; MN and FL refer to Minnesota and Florida, respectively, and are referred to in the discussion.
8 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . : Average mobility reductions required to mitigate differences in population density and temperature. This figure shows the percent reduction in average mobility (measuring retail, recreation, grocery, pharmacy, and workplace trips) needed to compensate for a given temperature (brown) or population density (blue) driven increase in R t . These calculations assume a 'background' R 0 of 1 and a baseline 'background' mobility (defined as '0' by Google 36 ). Solid lines represent the median mobility reduction required, dashed and dotted lines the 75% and 90% posterior credibility intervals respectively.

9
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101 Here, by combining epidemiological models and outputs with spatial climate data, we 161 show that environment (specifically cold, but also the correlated low-humidity conditions) 162 can enhance SARS-CoV-2 transmission across the USA. Critically, however, these envi-163 ronmental impacts are weaker than that of population density which is, itself, a weaker 164 driver than policy intervention (i.e., lockdown). Below, we suggest that the accuracy of 165 forecasts of SARS-CoV-2 transmission, in particular across seasons, could be improved 166 by incorporating temperature, as well as population density, in a robust, reproducible 167 manner as we have done here.

168
The role that environment plays in transmission 169 Across these state-level USA data, we found a significant negative effect of temperature 170 on SARS-CoV-2's R 0 and a significant positive effect of population density. An important 171 caveat to this, however, is the collinearity between temperature, absolute humidity, and 172 to a lesser degree, UV levels. The strong correlations between these environmental drivers 173 mean that we are unable to discern the effects of each in a single model and therefore we 174 focus on temperature as the most reliable environmental predictor. After accounting for 175 the effect of population density on transmission (table 1), temperature's effect is striking 176 (figure 2). We also tested the effects of our predictor variables on R t for times where 177 strict lockdown measures were in place. When these mobility restrictions are in place, we 178 observe no significant effects of temperature on R t , i.e. the effects of lockdown dampen 179 any environmental effects so as to make them inconsequential (figure 1b; supplementary 180 table S3). Furthermore, under lockdown conditions the overall transmission rates are 181 vastly reduced. Through our epidemiological modelling approach we are able to account 182 for these effects (as mobility changes are explicitly incorporated), and find that higher 183 population densities and lower temperatures drive increased R t . Moreover, the formula- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193250 doi: medRxiv preprint and supplementary figure S2).

187
The precise physiological mechanisms for temperature-dependant inactivation in SARS-

188
CoV-2 are still not known, but animal models for influenza have shown that increased 189 viral transmission at lower temperatures can be due to effects on the host 7,8 . In animal 190 models, this is proposed to be due to the combined effects of higher titres of viral particle  There are important methodological caveats to our findings. Dynamics and reporting 203 between USA states are known to be variable 38 , introducing a level of uncertainty to our 204 findings. Furthermore, lockdown measures were (and continue to be) quite heterogeneous 205 across the USA, with different states displaying different levels of response to COVID-206 19 39 . Through our epidemiological modelling approach we are able to account for these 207 different state-level responses using google mobility data. We can also observe other 208 potential confounding factors in these analyses. Across the USA, the north-eastern states 209 in the vicinity of the major transport hub of New York City (e.g., NY, NJ, ME, PA, RI, 210 and CT) tend to have generally higher R 0 than predicted, whilst west-coast states (e.g.,

211
WA, CA, and OR) have lower R 0 than predicted (fig 2a). While this type of effect could be 212 due to preemptive protective measures taken by states prior to COVID-19 outbreaks, we is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193250 doi: medRxiv preprint temperature affects human behaviour, thus making it difficult to disentangle the effects 216 of climate from changes to mobility. We do find a link between the average mobility and 217 temperature coefficients in our Bayesian modelling, suggesting a degree of collinearity, 218 however (perhaps surprisingly), we see no direct correlations between daily temperature Our results comparing SARS-CoV-2 transmission rate before and during lockdown sup-226 port the idea that the major driver of transmission is public health policy 32,40,41 (see figure   227 1). Once stay-at-home measures were implemented across the USA, we can find no mean-228 ingful signal of temperature on transmission. This provides two important, and timely, 229 insights for policy-makers: summer weather is no substitute for mitigation, and policy can 230 prevent transmission in the winter. At the coarse scale of USA states, population den-231 sity is a greater driver of transmission intensity than temperature in our epidemiological 232 modelling (log 10 (population density) ≈ 1.4× larger scaled coefficient than temperature).

233
It should be considered whether thresholds for adaptive and/or intermittent lockdown 234 might be more precautionary (i.e., lower) in colder, more densely-populated regions. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193250 doi: medRxiv preprint differences through time or across space. Regardless, our analysis is too spatially coarse 243 to address such variation. Even quite large variations in climate are more straightforward 244 to mitigate than population density differences (figure 3), and so we suggest that regions 245 with higher population density should continue to be monitored carefully. Finally, we 246 emphasise that population density and temperature are well-known to be strongly cor-  Further, our results support a role for daily temperature changes in transmission, but, we 260 emphasise, do not conflict with other studies suggesting that seasonal forecasting plays 261 a secondary role to mitigation and/or number of susceptible individuals. Such studies 18 262 assumed SARS-CoV-2 responds to climate to broadly similar extents that we find here.

263
What our results do suggest, however, is that future forecasting work should consider the 264 use of the environment to enhance predictions of disease spread. In countries such as the 265 USA with continental climates that swing between extremes of heat and cold, we suggest 266 policy-makers should assume that transmission will increase in winter (and potentially 267 autumn/fall). The timing of the seasons are broadly predictable, and so this is an area 268 in which policy could be proactive, not reactive. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10. 1101 There is no single cause of, or solution to, the current COVID-19 pandemic, and all drivers 271 must be placed in perspective. Here we suggest that both environment, and daily weather, is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020. 09.12.20193250 doi: medRxiv preprint We explored the association between environmental covariates and SARS-CoV-2 trans-283 mission intensity using two approaches. First, we took existing state-level estimates of 284 R 0 and during-lockdown R t for the USA 35 , and regressed them against environmental 285 data in order to test for potential pre-and during-lockdown patterns. In the second 286 approach, we modified and fitted the existing semi-mechanistic epidemiological model 287 used to generate the R0 and Rt estimates above, and fitted it to the observed death is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193250 doi: medRxiv preprint Where AH (g m −3 ) is the absolute humidity, T (K) the temperature in a given cell, RH

322
In these data, R 0 is estimated as R t=0 where t = 0 occurs 30 days prior to the first 10 323 cumulative deaths recorded for each state 32,35 . The date upon which R 0 is estimated 324 therefore differs between states. To account for these temporal differences, we took the is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101 the environmental predictors averaged across the same time period. We used 14 days 333 again here for consistency with our environmental comparison to R 0 . Although mobility 334 restrictions may differ in magnitude between states, these effects are incorporated into 335 the estimates for the R t parameter. In 7 states (Arkansas, Iowa, North Dakota, Ne-336 braska, Oklahoma, South Dakota, and Wyoming) no state-wide stay-at-home mandate 337 was declared. In a further 4 states (Alaska, Hawaii, Montana, and Utah), t = 0 occurred 338 after non-pharmaceutical interventions had already been instated. These 11 states were 339 therefore excluded from the independent validation analyses.
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10. 1101 µ ∼ N ormal(3.28, 0.5) For µ, this is the same as the prior used in the original (non-climate) model 35 (but see and workplace trips (X t,m,1 ), in 'residential' areas (X t,m,2 ), and using public 'transit'

362
(X t,m,3 ). We focus on the vector α k , whose three entries assess the impact of mobility so opted for a model simpler (and so more conservative) in its novel components. In 371 this model formulation, temperature and population density essentially contribute to a 372 latent transmission rate, which is then mediated by the mobility terms to produce the 373 realised R t . Although an interaction between mobility and environment (as found in our 374 regression modelling, see Results) is not explicitly modelled, this formulation produces 375 results analogous that finding, i.e. when mobility reductions are high ("lockdown"), 376 environment has little effect on the realised R t (see supplementary figure S2).

377
We emphasise that the model presented here differs from the original model by fitting is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101 µ that was hierarchically drawn from a common parameter (itself termed µ in Unwin is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. Yin. An initial investigation of the association between the SARS outbreak and is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10.1101/2020.09.12.20193250 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 14, 2020. . https://doi.org/10. 1101