Impact of large wildfires on PM10 levels and human mortality in Portugal

Uncontrolled wildfires have a substantial impact on the environment, the economy and local populations. According to the European Forest Fire Information System (EFFIS), between the years 2000 and 2013 wildfires burnt about 170,000-740,000 ha of land annually on the south of Europe (Portugal, Spain, Italy, Greece and France). Although most southern European countries have been impacted by wildfires in the last decades, Portugal was the most affected, having the highest percentage of 5 burned area comparing to its whole territory. For this reason, it deserves a closer attention. However, there is a lack of knowledge regarding the impacts of the wildfire-related pollutants on the mortality of the population. All wildfires occurring during the fire seasons (June-July-August-September) from 2001 and 2016 were identified and those with a burned area above 1000 ha were considered for the study. To assess the spatial impact of the wildfires, these were correlated with PM10 concentrations measured at nearby background air quality monitoring stations, provided by the Portuguese Environment Agency (APA). Associations 10 between PM10 and all-cause (excluding injuries, poisoning and external causes) and cause-specific mortality (circulatory and respiratory), provided by Statistics Portugal, were studied for the affected populations, using Poisson regression models. During the studied period (2001-2016), more than 2 million ha of forest were burned in mainland Portugal and the 48% of wildfires occurred were large fires. A significant correlation between burned area and PM10 have been found in some NUTS III (regions) on Portugal, as well as a significant correlation between burned area and mortality. North, centre and inland of Portugal are the 15 most affected areas. The high temperatures and long episodes of drought expected on the future will increase the probabilities of extreme events and therefore, the occurrence of wildfires. 1 https://doi.org/10.5194/nhess-2021-38 Preprint. Discussion started: 12 February 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
Wildfires have a considerable impact on the environment and humans worldwide. Climate change has lately been identified as a very important variable in this matter (Gillett et al., 2004) since the future projections suggest an increase of the number of 20 droughts, heat waves and dry spells (Turco et al., 2019). Global warming will produce changes in temperature and precipitation patterns leading to a higher prevalence and severity of wildfires (Settele et al., 2015;Bowman et al., 2017) and consequently impacting future air quality (Schär et al., 2004). In fact, this could not only extend the burnt area in chronically impacted areas (Ciscar et al., 2014), but also affect new ones, like Sweden in the summer of 2018 (Lidskog et al., 2019). According to the 2016 European Forest Fire Information System (EFFIS) report (San-Miguel-Ayanz et al., 2017), the south of Europe 25 (Portugal, Spain, France, Italy and Greece) is the area most affected by wildfires since 1980 until today, considering Europe, Middle East and North Africa. In the last decades, Portugal was by far the country with the largest burned area, almost 50% between the southern European countries (Parente et al., 2018). Although there has been a slight decreasing trend in the burnt area in this region since 2000 after an increasing period in the previous 20 years (European Environment, Agency, https: //www.eea.europa.eu/data-and-maps/indicators/forest-fire-danger-3/assessment), recent extreme events like the 2017 fires in 30 Portugal and the 2018 fires in Greece which resulted in a severe loss of human lives are confirming the worst projections.
Uncontrolled wildfires emit numerous pollutants derived from the incomplete combustion of biomass fuel, which cause damage to human health, particularly the cardiovascular and respiratory systems (World Health Organization, 2010). Examples include particulate matter (PM), carbon monoxide, methane, nitrous oxide, nitrogen oxides, volatile organic compounds (VOCs), and other secondary pollutants (Cascio, 2018) that are released mainly into the atmosphere but can be transported 35 to many other environmental compartments. Moreover, they can affect the physicochemical properties of the atmosphere, as for instance the interaction of PM with solar radiation which can prompt a modification of the temperature depending on the characteristics of the aerosol (Trentmann et al., 2005). Consequently, some of these chemicals are regulated by the European Directive 2008/50/EC of 21 May 2008 of the European Parliament and of the Council on Ambient Air Quality and Cleaner Air for Europe, which establishes threshold values for a safe air quality. But although wildfire emissions are a crucial parameter 40 for the local air quality (Knorr et al., 2016), where in some cases there are already chronically-exposed populations due to the frequency and dimension of the events, they are not contained by political borders and can also affect areas far from the ignition points due to the atmospheric transport of the pollutant plumes. A number of studies (e.g. Lin et al. (2012); Im et al. (2018); ; Augusto et al. (2020), among others) report the influence of natural and anthropogenic emissions on air quality composition across different countries, especially PM and tropospheric O 3 . For wildfires it is also important to take 45 into account some factors which influence the plume dispersion, such as the duration and space evolution of the fire event and the meteorological conditions associated (Lazaridis et al., 2008). An increase of cardiovascular and respiratory morbidity and mortality are some of the impact these contaminants can have on humans (Johnston et al., 2012;Tarín-Carrasco et al., 2019).
For instance, there is a strong evidence of the relationship between PM in general and mortality, especially from cardiovascular diseases, for both long-term and short-term exposure (Anderson et al., 2012). Although some studies corroborate the existence 50 of a link between the exposure to wildfire-related air pollutants and hospital admissions, visits to emergency clinics or even respiratory morbidity (Liu et al., 2015;Reid et al., 2016), the impacts on human health are difficult to quantify and the real effects still poorly known.
Concerning PM, a recent study focusing on 10 southern European cities revealed that cardiovascular and respiratory mortality associated to PM10 (particles with aerodynamic diameter below 10 µm) was higher on days affected by wildfires' smoke 55 than in smoke-free days (Faustini et al., 2015). The authors also found that PM10 from forest fires increased mortality more than PM10 from other sources. So, the estimation of mortality due to exposure to wildfire-generated pollutants is key to manage health resources and the necessary public funds towards prevention and remediation, setting up appropriate policies and protocols (Rappold et al., 2012).
The two main factors to take into account for the wildfire's effects are the location and, most importantly, the size of the fire 60 event (characterised by the respective burnt area). When the wildfire occurs close to a large conurbation, the population exposed is higher. But as Analitis et al. (2012) showed in their study, small fires do not seem to have an effect on mortality, whereas medium and large episodes (with burnt areas >1000 ha) have a significant impact on human health, which increases with the size of the fire. Aiming to enhance the knowledge on the effects of wildfires on human health, this study describes the pattern of wildfires in Portugal for 16 years (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016) and assesses the impact of those events on the country's population mortality 65 during the fire season (June, July, August and September). In this work, the focus is placed on indirect effects of pollutants emitted by wildfires, namely assessing the influence of wildfire-generated PM10 on the Portuguese population mortality. The relationship between the burned area of large wildfires and PM10 and this same pollutant and mortality was studied. The Nomenclature of Territorial Units for Statistics (NUTS) level 3 (NUTS III) geographical division has been used to be able to compare the effects of the fires in different parts of the country. Finally, monthly deaths due to all-cause (excluding injuries, 70 poisoning and external causes) and cause-specific mortality (cardiovascular and respiratory) for all ages for each NUTS III has been studied. These causes have been selected due to their well-known connection with air pollution.

Methodology and data
In this study, the effects of short-term exposure to pollutants due to wildfires on human mortality were quantified. The forest fire pollutant emissions were estimated for the period 2001-2016 during the summer months (June-July-August-September) 75 in mainland Portugal (23 NUTS III and more than 10 million people). In Portugal, large forest fires usually occur during the months of June, July, August, and September, which correspond to the time of the year with the highest temperatures and driest conditions. By focusing our study only on these 4 months, we can have enough data to perform a valid statistical treatment, while avoiding a strong influence of PM10 from other sources in colder months (such as home heating or traffic). At the same time, we do not include in the analysis the deaths due to, for instance, cold and flu that could become confounding factors.

80
For the quantification, two steps were followed. First, an assessment of the incidence, patterns and variations of burned area on a large time frame and spatially integrated by NUTS III was done on the levels of air pollutants. PM10 and burned area were correlated through linear regression, while the mortality data and PM10 were correlated with Poisson regression. Data was processed and ordered by NUTS and by month and year. Finally, the correlation between the pollutants emitted by forest fires, the wildfires burned area and the different causes of mortality during the period 2001-2016 for the summer months was studied.

85
The study is focused on PM since it is one of the main pollutants emitted by wildfires, which can increase PM concentrations up to 50% and more (Lazaridis et al., 2008). Moreover, there is a clear relation with several effects on human health (including mortality), in particular with respiratory and circulatory diseases (Kollanus et al., 2016;Reid et al., 2016;. There was not enough PM2.5 data collected from the Portuguese air quality management network to establish a correlation (only 20 stations measure PM2.5 in the mainland). For all these reasons this study focuses on PM10.

Target area
With 89,015 km 2 (9.11 Mha) mainland Portugal accounts for over 96% of the country's area and hosts over 10 million inhabitants in the west Iberian Peninsula (southwestern Europe). With the largest urban areas along the west Atlantic coast, particularly around the capital (Lisbon) -more to the south-and the second largest city (Porto) in the north (see Figure 1,left), the country has most of its mountain ranges in the north, reaching 1,993 metres above sea level in Serra da Estrela. Although

95
showing a Mediterranean climate, this topographic display leads to various climate patterns along the country, with increasing temperature and decreasing rainfall from northwest to southeast (Moreira et al., 2011;Oliveira et al., 2017). In terms of land cover, Figure 1,left shows a predominance of agriculture by 2015 (over 50% and mainly in the south), followed by forests and shrublands, which comprise 43% of the territory (mainly in the north and southwest). This, combined with high temperatures in the summer months, represents a potential fire hazard, which unfortunately has been often proved almost every summer for 100 many years.

NUTS III boundary data
The target domain was divided by NUTS (Nomenclature of Territorial Units for Statistics) level 3 (Figure 1,right) for Portugal mainland. NUTS is a geocode standard for referencing the subdivisions of countries for statistical purposes developed by the 105 European Union. The geocode is divided in three levels (I, II, III) which are established by each EU member country. NUTS III from mainland Portugal (in total, 23) at a 1:60 million scale were retrieved from the Eurostat web page (Eurostat, 2019) and treated with QGIS3 software.

Wildfires data
The wildfire data, collected for the period from 2001 and 2016, was obtained from the Portuguese Institute for Nature Conser-110 vation and Forests (https://www.icnf.pt/). For this study only forest fires were considered, and from them, only those with more than 1,000 ha of total burned area (which we denominated large fires) were selected. In total, there were 323 events under that category (less than 1% of the number of total fires), which were responsible for 46% of the total burned area. Table 1  drought periods (Turco et al., 2019), the data in Table 1 suggests that it is not possible to perceive a yearly pattern of wildfires in the country. For instance, in 2008 no large fires occurred, whereas 2003 accounted for the highest number of occurrences (81), which were responsible for 80% of the total burned area in that year. But the latter contribution was as low as 12% in 2011 and had a mean percentage for the whole period of 34%.
Considering only June, July, August and September 2001-2016 (the months with highest temperatures and drier conditions 120 when more than 86% of the total fires and 311 of the 323 large fires -96%-occurred), this data was divided by month and year and the respective monthly and yearly sums were considered for each NUTS III level region. All NUTS III had at least one large fire during the study period. In terms of burned area 929,766 ha of forest were lost in mainland Portugal from June to September (2001September ( to 2016, with about 53% due to large fires (>1,000 ha). Figure 2 represents the number of large fires and the burned area they were responsible for, by NUTS III.

125
The north and centre of Portugal present the most extensive forest cover in the country (Nunes et al., 2019), particularly abundant in pine and eucalyptus trees, two highly combustible species that have been associated with extreme wildfire events (Maia et al., 2014). Consequently, both areas show the highest number of large fires and respective burned area (being Beiras e Serra da Estrela and Médio Tejo the most affected NUTS III), but with also Alto Alentejo and Algarve (more to the south, see Figure 1) among the NUTS III with a higher incidence. Additionally, dense Mediterranean forests over hard-to-reach 130 mountains can also be found in these areas, which combined enhance the difficulty of the firefighting efforts. Algarve, despite being located in the south coast, also has some mountains with forests, surrounded by a considerably dry and arid terrain,  (Table SM1).

Pollution data
The information available on the levels of pollutants was obtained from the Portuguese Environment Agency air quality network (https://qualar.apambiente.pt/qualar/index.php), established to monitor the concentrations of pollutants according to the  Considering all the pollutants measured on the background stations, PM10 was the one with a potentially higher link to 145 forest fires. Although some stations also measured PM2.5, the coverage in this case was insufficient to draw any significant correlations. The main anthropogenic sources of PM10 include road traffic, industrial activities, and home heating. In this study, to minimize the influence of non-wildfire causes for the PM10 concentrations, we selected only background stations (encompassing urban and semi-urban ones, which are located within urban areas but with minimum influence of road traffic; and rural stations). Therefore, urban stations with road traffic influence and stations close to industrial complexes were not 150 selected. The influence of home heating was already minimized by selecting the summer period as our target timeframe.
As done before for the wildfires data, also here the time  Obstructive Pulmonary Disease (COPD, codes J40-J45) and asthma (ICD-10, code J47), were also considered. However, since in many months and NUTS III in the target time period there were no deaths for COPD and asthma, it was not possible to obtain a data series large enough to correlate with PM10 and wildfires series.

Associations between burnt area, PM10 and mortality
The associations of monthly average PM10 levels, and the occurrence of large wildfires (burnt area >1,000 ha), with mean 185 monthly mortalities (all-cause, respiratory and cardiovascular causes) were studied for the months of June, July, August and September for the period between 2001 and 2016. The estimates of the effects were obtained for each NUTS III region using Poisson regression models (Faustini et al., 2015;Islam and Chowdhury, 2017). Poisson coefficients can correlate a count variable (such as the number of deaths) with a continuous variable. The results were expressed as the Relative Risk (RR) of all-cause, cardiovascular and respiratory mortalities with a 95% confidence interval (95% CI). All regression models were 190 performed using IBM SPSS Statistics 25.0 software.

Relationship between burned area and particulate matter
For the correlation between the burned area from large fires and PM10, a significant positive correlation was found for 7 (out of 13 with available data) of the studied NUTS III, represented by the dotted areas in the map of Figure Table 2, for each NUTS III region and all-cause, cardiovascular and respiratory-related deaths. Results show that almost 30% of all-cause and cardiovascular mortality occur during the extended summer (June, July, August and September), as do 26% of the respiratory mortality.

215
Algarve, Alto Minho, Alto Alentejo and A.M. Lisbon are the NUTS with the higher percentage of all-cause mortality for the studied months, but the NUTS with more per capita incidence are Beira Baixa, Alto Alentejo, Baixo Alentejo and Beiras e Serra da Estrela, areas with lower population density and with mean higher age than the rest of the country. With respect to cardiovascular mortality, the NUTS III which present a high incidence are Algarve, Terras de Trás-os-Montes and Beira Baixa, with the latter, Baixo Alentejo and Alto Alentejo having a higher percentage of population affected.

220
Finally, the results obtained for respiratory mortality show that Algarve, Leizíria do Tejo and Alentejo Litoral are the NUTS III which top the ranking in the summer months, whereas Alto Alentejo is the region with most population affected. Alentejo and Algarve suffer from high temperatures in the summer, which may also be an indicator that contribute for a higher mortality in general (Basu and Samet, 2002) but also due to cardiovascular and respiratory diseases (Pinheiro et al., 2014).
In addition, the aforementioned regions suffer from a considerable afflux of tourists that increase their population in the same 225 period, particularly in Algarve. In Alentejo, the combination of high temperatures with an aged population and less health care resources available may be the justification to have the most population affected by mortality (Chen et al., 2019). In fact, this is a tendency that has been becoming stronger since the beginning of the XXI century, as the percentage of population over

Associations between mortality and PM10
Wildfires are an important source of particulate matter and the associations between mortality, PM10, and the occurrence of large wildfires were assessed in this section. As shown in Figure 6,a, three NUTS III (Alto Tâmega, Beiras e Serra da Estrela, and Viseu Dão-Lafões) present associations between PM10 and all-cause mortality during the studied period. None showed a direct significant association with the occurrence of large fires, likely due to the fact that their contribution to the total burned 235 area in each year from 2001 to 2016 (Table 1) was highly variable (from 12.1% in 2011 to 79.5% in 2003). However, the wildfire origin of PM10 is corroborated by the positive significant correlations obtained for these three NUTS between PM10 and burnt area ( Figure 5), with Viseu Dão-Lafões displaying the highest correlations.
Beiras e Serra da Estrela is the NUTS region most affected by large wildfires during the studied period, both in number (50) and respective burned area (>100,000 ha), and Viseu Dão-Lafões is the third in occurrences (28) corresponding to over 240 58,000 ha burned. This involves high levels of PM10 in a short period of time, which might provoke damage in human health, particularly in an aged population (e.g., for Beiras e Serra da Estrela, 23.8 and 28.7% over 65 years-old in 2001 and 2018, respectively; PORDATA, https://www.pordata.pt)).
In terms of types of diseases, for cardiovascular mortality five NUTS presented associations with PM10: Alto Minho, A.M.
Porto, Região de Aveiro, Região de Coimbra, and Algarve (Figure 6,b). Again, no direct significant associations were obtained 245 with the occurrence of large fires. From these five NUTS, only Região de Aveiro showed a significant correlation between PM10 and burned area, revealing the impact of wildfires in the origin of the PM10. For respiratory mortality, only Viseu Dão-Lafões present associations with PM10, for which a strong correlation between PM10 and burned area was found, suggesting again the impact of wildfires on the presence of PM10 (Figure 6,c). Wildfires do not follow a pattern in number of the occurrences or size during the years studied. This evidence was found despite the difficulties that the uneven scattering of the air quality monitoring stations analysing PM10 in Portugal posed.
In fact, the areas where wildfires are usually more frequent (inland) are far from the urban centres (mainly along the coast), 255 and thus, not abundant in air quality data availability due to the shortage (or even lack in some NUTS III) of monitoring stations. These regions also have an aged population, poorer economy and less health care resources, which can lead to an increase in the mortality rates in general. The socio-economic status of the population affected and the health care facilities and measures existing in the communities have to be taken into account (Oliveira et al., 2017), adding to the countless parameters that may affect these estimations which contribute to considerable gaps identified in this type of studies (Black et al., 2017).

260
Unfortunately, the scarce data available and the lack of accuracy in the existing ones prevented us from estimating/including a correction regarding their influence.
Nevertheless, it was possible to find relationships between very relevant parameters. The significant positive correlation between PM10 and burned area found for 7 of the 13 NUTS III with available data is a good indication of where the influence of wildfires on the emission of PM10 is likely to be stronger. Although the location of the air quality monitoring stations may 265 influence these correlations, especially when they are scarcer, for these NUTS it reveals the influence of wildfires on the local levels of PM10.
Large wildfires tend to be active for several days, releasing high amounts of pollutants to the atmosphere. In Portugal, such as in other Mediterranean countries, most of the wildfires are potentiated by strong winds, which may spread the fire smoke over large distances (Turco et al., 2019;Augusto et al., 2020). Thus, air quality monitoring stations located far from the ignition thousands of kilometres from southern Finland (Niemi et al., 2009;Kollanus et al., 2016).
The negative impact particulate matter can bring to human health is well established and can be translated into several types of diseases (Kim et al., 2015). It is an evidence that it can enter the human body, arrive to the bloodstream and damage some organs or even provoke death due to cardiovascular afflictions like stroke or heart attack, among others, representing a clear hazard to public health (Brook et al., 2010;Hamanaka and Mutlu, 2018). Particulate matter also can damage the human 280 respiratory system. The risk depends on the size of the particle, which if very small can even reach the alveolus (Neuberger et al., 2004;Jo et al., 2017).
In our study, the NUTS III where PM10 concentrations were found to be correlated with the burned area from large fires (Cávado, Ave, Tâmega e Sousa, Região de Aveiro, Viseu Dão-Lafões, Alto Tâmega and Beiras e Serra da Estrela) are indeed the ones where it would be expected to find the strongest influence of the wildfire originated PM10 on the population mortality.

285
And although not for all, indeed associations between all-cause mortality and PM10 were found for three of these NUTS The mortality increase associated with PM10 is consistent with the estimates reported in other European studies, such as APHEA2 (Katsouyanni et al., 2001), APHENA (Samoli et al., 2008), EpiAir (Faustini et al., 2011) and MED-PARTICLES 295 (Faustini et al., 2015), which also reported higher PM10 effects on all-cause, cardiovascular, and respiratory mortalities.
However, some studies present uneven conclusions. Johnston et al. (2011) reported the highest effects on cardiovascular mortality, but Morgan et al. (2010) did not find any consistent effect with cardiovascular deaths in Australia, and Analitis et al.
(2012) registered the highest effects on respiratory mortality in Greece. This high variability may be related to several factors, notably: i) different PM composition or varying gaseous emissions (CO, VOCs, NOx or SO 2 ) from wildfires, which may have 300 different degrees of toxicity on cardiovascular and respiratory systems; or ii) increasing temperature during wildfires, which is known to enhance the effects of PM on more susceptible individuals (e.g., cardiac patients) (Qian et al., 2008). Therefore, the effects we found on all-cause, cardiovascular, and respiratory mortalities during the wildfire seasons may be due to different PM compositions or increasing temperature.
Region-specific associations between PM10 concentrations and mortality were also observed. These may have been influ-305 enced by the factors described above (different PM composition and increasing temperature), but also by the magnitude and duration of the exposure to PM from a given fire; the underlying health status of the population; and the size of the population. The age of the exposed individuals can also be important. In some studies, larger effect estimates in groups of 65 years and older have been reported. Analitis et al. (2012) mentioned that the effect of respiratory mortality in Greece was higher in adults of ages 75 and above during large fires, whereas Haikerwal et al. (2015) observed an increase in risk of cardiac arrests, 310 especially in older adults in Australia, although not all resulted in death. In Brazil, Nunes et al. (2013) reported that older adults had the strongest association between exposure to biomass burning and circulatory disease mortality. In Portugal, the regions traditionally impacted by wildfires coincide with a larger percentage of aged population, which can help explaining the obtained associations.
In our analysis, to study the relationship between the burned area and PM10 concentrations, averaged monthly data were 315 used, as the minimum temporal scale available for the burned area was one month. Other studies relating wildfire-originated PM and mortality are usually based on daily PM concentrations and daily death counts, since they do not account for the burned area as a measure of the wildfire size. Therefore, the monthly approach obviously reduced the number of data available and the possibility of finding more significant correlations, which may have diluted the effects of some wildfires on the population mortality. Moreover, some health effects may not have been detected because wildfires are episodic and local events. Never-320 theless, the results provide an overall context, highlighting the strongest associations between wildfire generated PM10 and all-cause, cardiovascular, and respiratory mortalities. Being able to achieve them with this uneven distribution of available data is an indication that the approach can be very useful to at least uncover tendencies and, in regions with stronger monitoring capabilities and coverage, a way to find stronger and more accurate correlations. This will help legislators and other government bodies to propose ways to protect the population chronically exposed to wildfires or more susceptible to acute reactions 325 to wildfire smoke.

15
Portugal is a country that suffers constantly from serious wildfire incidents, which are bound to pose a risk not only to chronically affected populations but also from acute impacts of the pollutants released in such events. In this work, analysing the summer months (June to September) on a lengthy timeframe (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016), it was possible to find relevant associations between 330 PM10 (associated with large wildfires) and mortality in some NUTS III regions of mainland Portugal (mainly inland and in the north), as well as a significant correlation between burned area and PM10.
In particular, it was found that large fires (in this study considered above 1,000 ha of burned area) have an impact on the health of the population in some areas due to the emission of particulate matter. The lack of data or possible confounding factors likely prevented a higher number of NUTS III with significant correlations. Moreover, in such severe events, the population 335 exposed to a high concentration of pollutants in a short period of time should be considered as a risk modifier of the impacts of air pollution exposure (Desikan, 2017;Rappold et al., 2017).
These episodes occurred during the summer months (June-July-August-September), when high temperatures and long episodes of drought increase the probabilities of undergo one of these extreme events. On a future ruled by climate changes, the high temperatures and long periods of drought that usually fuel big fires are expected to increase, thus leading the way for 340 more extreme and intense events to occur, even outside the typically affected regions. Thus, more population will be exposed more frequently to high pollutant levels, affecting their general health, and increasing chronic diseases and mortality. Hence, restrictive policies and protocols to improve the effectiveness of preventive and mitigation actions must be enforced to face this environmental and societal issue.
Data availability. Data is publicly available through the websites mentioned in the text:
All the compiled data is available upon contacting the corresponding author (pedro.jimenezguerrero@um.es) Author contributions. PT-C wrote the manuscript, with contributions from SA and NR. The manuscript was finally revised by PJ-G. P-TC

355
and SA designed the experiments and led the statistical analysis, with the support of LP-P, NR and PJ-G.