Studies on early dynamics of COVID-19 (Li et al., 2020a) revealed that the epidemic doubled in size every 6.4 (Wu et al., 2020) to 7.4 (Li et al., 2020a) days, with a reproductive number (R₀) of infectious cases of 2.2 (Li et al., 2020a) to 2.7 (Wu et al., 2020). Later investigations followed disease spreading to Singapore (Pung et al., 2020), Germany (Hoehl et al., 2020), France, and Finland (www.ecdc.europa.eu/en/covid-19-pandemic) (Lipsitch et al., 2020; McMichael et al., 2020; Sun et al., 2020). However, major uncertainties remained on SARS-CoV-2 transmission dynamics (Lipsitch et al., 2020). Considerable effort across major research institutions was invested into modeling SARS-CoV-2 spreading determinants. Models were generated, which took into account, among others, global traveling, population density, demographic characteristics, age distribution, social dynamics, governmental policies, air pollution, virus infectious capacity, and SARS-CoV-2 containment procedures, together with economical and healthcare factors (Baker et al., 2020; Hernandez-Vargas and Velasco-Hernandez, 2020; Kennedy et al., 2020; Kissler et al., 2020; Lipsitch et al., 2020; Pearson et al., 2020) (10.21203/rs.3.rs-82122/v1). However, limited, if any, regional heterogeneity in COVID-19 transmission could be identified using such diffusion models. We reasoned that fundamental variables were missing from current analyses, and we went on to identify such missing factor(s).

SARS-CoV-2 was suggested to be sensitive to temperature and humidity, which may affect diffusion across diverse climate areas (Sundell et al., 2016) (papers.ssrn.com/sol3/papers.cfm?abstract_id=3550308; ssrn.com/abstract=3556998; www.medrxiv.org/content/10.1101/2020.02.22.20025791v1). Accordingly, initial climate-dependent propagation models predicted a limited impact of COVID-19 in the Southern hemisphere, during seasons that were infection-prone in the Northern hemisphere (papers.ssrn.com/sol3/papers.cfm?abstract_id=3550308; ssrn.com/abstract=3556998). However, early foci of infection were detected in Australia and New Zealand (Figure 1). Outbreaks were also revealed in South America and extended to Central America and Mexico. Further infection foci were revealed in Saudi Arabia and Africa and extended to sub-Saharan countries (Supplementary Tables S1, S2), questioning simple models of cold-climate-dependent SARS-CoV-2 transmission.

Viral evolution processes (Eletreby et al., 2020) may mimic regional COVID-19 spreading dynamics (Rausch et al., 2020; Volz et al., 2021; Korber et al., 2020; Li et al., 2020b). SARS-CoV-2 possesses a single-strand RNA genome (Mousavizadeh and Ghasemi, 2020), prone to acquire genomic mutations (nextstrain.org/ncov/; www.gisaid.org/). However, the SARS-CoV-2 RNA polymerase has error-correcting capacity and shows replication error rates >10-fold lower than other RNA viruses (Rausch et al., 2020). Correspondingly, SARS-CoV-2 overall sequence diversity is low (Korber et al., 2020). Spike proteins in particular showed few overall mutations at early time points of the pandemic (Korber et al., 2020). Still, SARS-CoV-2-bearing distinct sets of mutations were isolated in different regions of the world (nextstrain.org/ncov/global), leaving the question open as to whether viral genetic drift, driving distinct disease evolution, could account for heterogeneous disease courses across different geographic areas.

We thus went on to challenge regional versus genetic evolution models of COVID-19 at a population-wide level. The best chances for detecting basic transmission determinants of SARS-CoV-2 were expected before any large-scale defensive approach was implemented (Carleton et al., 2021). Western Europe provided a vast terrain for this approach, because of the large-scale outbreaks of COVID-19 early-on during the pandemic. A further advantage was provided by Europe’s high healthcare management and data collection standards (Bloomberg Global Health Index, 2018, www.bloomberg.com/; WHO, www.who.int/whr/en/; worldpopulationreview.com/countries/best-healthcare-in-the-world/; (Lozano et al., 2020)), which supported a robust detection of basic diffusion parameters of COVID-19.

Broadly diverse climate regions around the CET longitude (15°E) were severely exposed to COVID-19. Spain and Italy were the countries with the highest early incidence of COVID-19 in Europe (Figure 1, Supplementary Figure S1, Supplementary Table S3). The heaviest initial casualties in Italy were suffered by Lombardy and Veneto, i.e., cold and humid areas during wintertime. Markedly warmer and drier climate conditions prevail in Southern regions of the country. A further shift toward warmer/drier conditions occurs in Spain. Scandinavian countries provided a reference for cold winter temperatures, over a Sweden–Finland–Norway axis (Supplementary Table S4). This offered unique opportunities to assess a climate-dependent coronavirus infection model. Such analysis was performed at a whole-population level on 86,498 laboratory-confirmed SARS-CoV-2 infection cases in Italy, 64,095 in Spain, and 17,496 cases in Scandinavia (github.com/pcm-dpc/COVID-19) (Supplementary Appendix). Diffusion data in France (Supplementary Table S5), Germany (Supplementary Table S6), and the United Kingdom (Supplementary Table S7) provided a validation dataset, encompassing 210,239 COVID-19 cases. This model was then merged with that of coronavirus genetic evolution (Figure 2), for detecting signs of positive selection for increased aggressiveness across the analyzed regions. While such epidemiological models limit the analysis to available, correlated information, a whole-population analysis provided the largest possible data collection scale and was expected to average-out distinct demographic and social inhomogeneities.

Large efforts have gone into modelling COVID-19 transmission, according to global and local population dynamics, demographics, governmental policies, and infectious ability of the virus (Lipsitch et al., 2020; Pearson et al., 2020; Hernandez-Vargas and Velasco-Hernandez, 2020; Kennedy et al., 2020; Baker et al., 2020; Kissler et al., 2020; Rausch et al., 2020; Volz et al., 2021, #34796; Korber et al., 2020; Li et al., 2020b; Mwalili et al., 2020). Most models, though, showed an inadequate capacity for predicting the regional/climate-associated diffusion dynamics of the pandemic (Baker et al., 2020).

We speculated that fundamental variables associated with the COVID-19 uneven diffusion remained to be identified and set a search for discovering such factor(s). We went on to perform a population-wide analysis, on 378,328 laboratory-confirmed SARS-CoV-2 infection cases in continental Europe and the United Kingdom. A robust determination required collecting epidemiological data (Cimoli et al., 2004; Ambrogi et al., 2006), before intervention via disease-containment measures. We thus went on to identify landmark dates, as inflection points of disease incidence curves associated with disease-taming procedures, throughout Western Europe. This led us to identify a quasi-universal pattern of linear growth of COVID-19 cases over time in most of the regions analysed. Such a growth pattern was not predicted by most pandemic spreading models (Carleton et al., 2021), supporting the analytical validity of our novel data collection strategy. This also suggested a unique diffusion mode of SARS-CoV-2 (Thurner et al., 2020), which was largely robust to conventional modeling of disease diffusion dynamics. Within such a unique diffusion mode of SARS-CoV-2, distinct COVID-19 transmission rates were identified as associated with different geographic regions.

Still, the accumulation of mutations of SARS-CoV-2 may have led to distinct selection for disease progression over different regions. An indicator of selective pressure for viral evolution has been that of progressively larger prevalence across different geographic locations (Korber et al., 2020), as indicated for the spike D614G mutation. However, D614G was associated with higher upper respiratory tract viral loads, but a much more limited impact was found in early analyses on disease severity (Li et al., 2020b; Korber et al., 2020; Volz et al., 2021). This appeared puzzling (Rausch et al., 2020), as higher viral loads have been associated with worse disease courses (Yu et al., 2020) and with increased mortality (Westblade et al., 2020). The hypothesis of positive selection of spike D614G was further investigated in the United Kingdom using more than 25,000 whole-genome SARS-CoV-2 sequences. This indicated 614G increases in frequency relative to 614D as consistent with a selective advantage, but not in all cases (Volz et al., 2021). This suggested that “a combination of evolutionary selection for G614 and the founder’s effects of being introduced into highly mobile populations may have together contributed in part to its rise” (Korber et al., 2020). These findings were recapitulated “as a slow genetic drift of a highly stable [SARS-CoV-2] genome” (Dearlove et al., 2020; Rausch et al., 2020), during the early timeframes of the pandemic.

We looked for further evidence of selection for viral evolution, utilizing broader indicators of the predominance of specific mutation(s) versus disease severity. Four major mutation groups/haplotypes were revealed in all examined European countries. The highest number of accumulated mutations was revealed in Wales and Senegal SARS-CoV-2 isolates, suggesting a correlation with genetic drift, at late stages of the disease. The lowest number of accumulated mutations was recorded in Italy, the country that first showed severe disease outbreaks in Europe, Furthermore, similar mutation loads were observed in Spain, the second hardest-hit country in Europe, and Sweden, a country with much less explosive COVID-19 transmission, suggesting a correlation with disease duration, rather than with selection for higher disease severity. The larger mutation loads were revealed in France and Belgium, both late-disease-insurgence countries, further supporting a relationship between mutation acquisition and length of the disease course. Taken together, our findings add evidence to a model of SARS-CoV-2 genetic drifting during the early course of the pandemic.

COVID-19 spreading models based on population demographics and socioeconomic factors all systematically failed to account for regional diffusion heterogeneity during the pandemic. Our findings show a sharp North–South gradient, with the shortest COVID-19 doubling times in Southern Italy and Spain. At the opposite end of the climate spectrum, Scandinavia showed the longest COVID-19 doubling times, over a Sweden–Finland–Norway axis. This climate model was verified in country-wide validation datasets of COVID-19 cases in Germany, France, and the United Kingdom, which included 210,239 laboratory-confirmed SARS-CoV-2 infection cases. This showed pandemic doubling times that were intermediate between Northern and Southern regions and that were in sharp consistency with the climate-area Köppen–Geiger model (Beck et al., 2018). Thus, our findings support a climate dependency of COVID-19 transmission capacity, usefully adding to the set of variables that are involved in modulating SARS-CoV-2 diffusion.

Findings of a more efficient coronavirus spreading in warmer regions are consistent with the resilience of coronaviruses to high-temperature environmental conditions (Chin et al., 2020) [Kampf, 2020 #33396van Doremalen, 2020 #33393]. Of note, the Middle East Respiratory Syndrome (MERS) was first reported in Saudi Arabia (www.cdc.gov/coronavirus/mers). MERS is caused by the MERS-CoV, which is structurally and genetically related to SARS-CoV, indicating that at least some coronavirus strains may better propagate in high-temperature climate conditions (www.cdc.gov/coronavirus/mers/risk.html).

However, climate areas are associated with complex sets of variables, among them indoor versus outdoor temperature profiles, specific/relative/absolute humidity (Sundell et al., 2016; Babin, 2020; Sajadi et al., 2020), UV exposure versus daily time/season/latitude (Carleton et al., 2021; Sajadi et al., 2020), weather structure, and ventilation, together with social behavior, inter-individual distancing, indoor crowding, lifestyle, and outside physical activity (papers.ssrn.com/sol3/papers.cfm?abstract_id=3556998) (Sajadi et al., 2020) versus community structure, socioeconomic and healthcare factors (www.medrxiv.org/content/10.1101/2020.03.23.20040501v4), including traveling or traveling modes. Recent studies have begun to dissect such determinants, indicating the impact of UV exposure on COVID-19 transmission (Carleton et al., 2021), while suggesting no role of temperature and humidity (SSRN 3567840, 2020; papers.ssrn.com) (Briz-Redón and Serrano-Aroca, 2020; Carleton et al., 2021). Our research confirms these findings and the lack of impact of outside temperature on COVID-19 progression over the areas analyzed (unpublished observations). Additional work is expected to bring in further insight into climate area-associated virus diffusion determinants.

Taken together, our findings suggest higher SARS-CoV-2 resilience in warmer regions than previously predicted and caution that high environmental temperatures may not efficiently tame SARS-CoV-2 infectiousness (Kissler et al., 2020). Very cold regions may be better spared by recurrent courses of COVID-19.