Climate Suitable Conditions Index and Autochthonous Dengue Infections: an Early Warning Index

Background: Aedes aegypti is the primary vector of dengue and contributes to most major epidemics of this virus worldwide. Aedes albopictus is also blamed for certain epidemics, as the 2014 Guangdong dengue outbreak. In Guangdong province, Ae. albopictus is thought to be the dominant vector species, with Ae. aegypti absent from most areas. Whether or not primary mosquito vectors are present, optimal climatic conditions for dengue vector species may play a substantial role in epidemics of the virus. We hypothesise that although vector species are required to initiate and establish an outbreak, favourable weather conditions may then grow and perpetuate the outbreak via complex effects on vector sympatry or interactions. Methods: Vector spices-specic suitable conditions index (SCI) and autochthonous dengue case data were tted to negative binomial (NB) regression models. After accounting for potential confounders, we assessed the relationship between SCI and autochthonous dengue cases. We assumed SCI interaction was a proxy for vector species sympatry and SCI difference a proxy for interspecies competition. Finally, we explored the relationship between these assumed conditions and the autochthonous dengue case. Results: Autochthonous dengue cases are associated negatively with SCI for Ae.albopictus and positively with SCI for Ae.aegypti. According to the NB regression models, autochthonous dengue cases increased 4% (Incident Rate Ratio (IRR): 1.04, 95% CI: 1.02, 1.06) for every unit increase in SCI for Ae. aegypti, but decreased by 3% (IRR: 0.97, 95% CI: 0.96, 0.99) for Ae. albopictus SCI. There was also an interaction between two SCIs and a positive effect of the difference in SCIs on autochthonous dengue cases. These ndings support the hypothesis that vector sympatry and interactions may inuence the risk of a dengue outbreak. Conclusions: Our results conrm the hypothesis that the dengue virus is more transmissible in regions with warmer weather conditions (high SCI for Ae. aegypti). SCI of Ae. aegypti would be a valuable index to predict dengue


Background
Dengue has risen rapidly throughout the world in recent decades. According to model estimates, 390 million dengue virus infections occur yearly [1], and 3.9 billion people are at risk of infection [2]. Dengue cases reported to World Health Organization (WHO) have climbed from 505,430 in 2000 to over 2.4 million in 2010 and 5.2 million in 2019 [3]. To date, dengue is the most prevalent human arboviral infection worldwide. Dengue is an arthropod-borne virus (arbovirus) infection caused by four virus serotypes, transmitted by the two Aedes mosquito species: Ae. aegypti and Ae. albopictus [4]. Ae. aegypti and Ae. albopictus, once zoophilic forest species from sub-Saharan Africa and Asia respectively, are now widespread globally due to rapid human population growth and international trade in the last century and climate change more recently [5,6].
Ae. aegypti is considered highly urban-adapted, while Ae. albopictus can thrive in peridomestic and rural environments over a wide range of tropical, subtropical and temperate climates [7]. In China, Ae. aegypti was once only present in the areas of the tropical zone below 22°N, such as Hainan Island, Leizhou peninsula in Guangdong province and coastal area in Beibu Gulf of Guangxi province. However, since the 2000s, Ae. aegypti has also been found in some bordering counties in Yunnan province, located between 22°N and 25°N [8]. In contrast, Ae. albopictus has a wide distribution in both tropical and temperate regions of China from 41°N to the country's southern reaches [9]. Based on historical climatic data from the last decade, most areas where Ae. albopictus became established had an annual mean temperature above 11℃ and annual mean precipitation above 500mm [10].
Ae. aegypti is considered the principal dengue vector and has contributed to a global resurgence of dengue epidemics in the past three decades. All major epidemics of dengue have occurred only in areas where Ae. aegypti was found [11]. Conversely, Ae. albopictus is considered to play a relatively minor role in dengue virus transmission compared to Ae. aegypti, and is not considered an 'e cient' epidemic vector [11,12]. In China, dengue outbreaks in Hainan province in 1980 and 1985-86 and Xishuangbanna of Yunnan province in 2013 were typically attributed to Ae. aegypti [13,14]. However, outbreaks, largely of classical dengue (i.e., few cases of severe disease) in Guangdong, Fujian, and Zhejiang province between 2004 and 2010, were caused solely by Ae. albopictus in the absence of Ae. aegypti [15,16]. In 2014, China's largest dengue outbreak to date (or at least since dengue became a noti able disease in 1989) occurred in Guangdong, where Ae. albopictus appears to be the sole dengue vector [17].
The dynamics of dengue transmission and outbreak are complicated, including hosts (humans), pathogens and mosquito vectors with their ecological interactions. The WHO recommends that many dengue-endemic nations use vector surveillance to provide a quantitative evaluation of dengue vector population uctuations in the number and geographical distribution to forecast outbreaks and assess management options [18]. However, dengue transmission usually depends on a range of factors, including serostatus of the human host population, mosquito density and climate, rather than at a xed entomologic threshold [19]. Additionally, there has been minimal evidence of meaningful correlations between vector biomass indices and dengue transmission, which might be used to forecast epidemics consistently [20]. Furthermore, the geographical variability in the association between dengue vector abundance and transmission or outbreak occurrence is poorly understood, suggesting that locality-speci c vector population indices may be critical in forecasting dengue transmission increases [21], and other data may have better potential as predictors.
Climate is a crucial environmental determinant of vector geographical distribution and vectorial capacity. Climate factors such as temperature and precipitation and their seasonal patterns can fundamentally affect mosquito population dynamics and individual features relevant to vector biologies like development, reproduction and survival, and consequently dengue virus transmission patterns [22][23][24]. A su ciently warm ambient temperature is also critical for virus replication and dissemination to the salivary glands in female mosquitoes. Conversely, cooler temperatures slow this viral ampli cation, and if the development time of the pathogen exceeds the life span of the infected mosquito, transmission cannot occur [7,25] [30]. Based on these identi ed suitable temperature ranges for dengue vectorial capacity, we have previously developed the Suitable Conditions Index (SCI) for Ae. aegypti and Ae. albopictus to de ne their potential vector capacity. SCI ranks geographic locations based on their climatic suitability for each vector species that transmits dengue and theoretically de nes the spatial parameters for transmission [31]. By incorporating SCI into an outbreak epidemiologic study, it is possible to investigate the potential value of SCI data for dengue transmission or outbreak prediction. Here, we use the case of the Guangdong dengue outbreak in 2014 to assess the potential of a climate-based SCI as a local dengue outbreak early warning index.

Data collection
Data on locally acquired and imported dengue cases reported from Guangdong in 2014 were collected from the National Noti able Infectious Diseases Reporting Information System. Dengue has been a statutorily noti able communicable disease in China since 1989 [34]. Dengue is diagnosed according to the national surveillance protocol with standardised case de nitions, including clinically diagnosed and laboratory-con rmed dengue cases. An imported case was de ned as one for which the patient had travelled abroad to a dengue-endemic country within 15 days of the onset of illness. In some cases, importation was de ned based on laboratory results showing that the infection dengue virus had a sequence similarity in the premembrane (prM) and envelop (E) glycoproteins as (prM-E) region of the virus genome compared with viruses isolated from the putative source region where the patient had travelled [35]. In the absence of meeting the criteria for an imported case, a dengue case was considered an autochthonous dengue case. SCI has been developed in our previous research [31] to de ne climate suitability for each species of Ae. aegypti and Ae. albopictus. Based on the optimal temperature range for vectorial capacity in the transmission identi ed by Mordecai et al. using robust mechanistic models [30], the SCI was de ned as the product of the number of suitable months (where the monthly minimum and maximum temperature is within the determined temperature range) and the average monthly precipitation (as a weight). The input data for the SCI model were available on an annual interval and at the county scale. Therefore, we produced SCI values for each species of Ae. aegypti and Ae. albopictus in 2014 for 124 counties/districts in Guangdong.
Normalised Difference Vegetation Index (NDVI) data were extracted from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) website (MODIS Web (nasa.gov)). NDVI data are used to characterise and monitor the global distribution of vegetation conditions and are often used to model global and regional climate [36]. We utilised MODIS Vegetation Indices (MOD13A3 product) version 6 data provided monthly at the 1-kilometre spatial resolution, then aggregated annually and at the county scale [37]. In addition, human population data were obtained from the Sixth National Population Census data (2010), which is available from the National Bureau of Statistics of China (http://www.stats.gov.cn/enGliSH/).

Negative binomial regression (NB) models
We tted a xed-effects NB model to assess the relationship between autochthonous dengue cases and potential predictors, including SCI for each vector species, NDVI, human population density, and the number of imported dengue cases. An NB model was used to account for the over-dispersed count of the dengue case (Table 1). A log-scale population was used as an offset to control for variation in human population size. Univariable and multivariable analyses were implemented successively. In univariable analyses, we consider each potential predictor in an NB regression model to identify signi cant predictors. In a multivariable analysis, we examined the potential effect of signi cant predictors on dengue case occurrence (presented as incident rate ratio, IRR). Model goodness of t was assessed by Bayesian information criterion (BIC), where smaller values of BIC indicate a better t. We also examined the potential interaction between SCI for Ae. albopictus and SCI for Ae. aegypti (SCI-Ae.aegypti × SCI-Ae. albopictus). We assumed the SCI was a proxy for vector density; a high level of SCI represented high vector density and vice versa. The interaction between the two vectors was examined by assessing how the effect of the SCI of a speci c vector on autochthonous dengue case occurrence changed when the SCI of another vector changed. The SCI difference (de ned as SCI-Ae.aegypti -SCI-Ae. albopictus) was also tted to the NB model. As Ae. albopictus has a different niche to Ae. aegypti in terms of feeding preferences and, critically for the SCI, thermal range, it was reasonable to assume that more considerable difference may indicate a reduced probability of ecological constraints between the two mosquito species. We use the SCI difference to explore the correlation between the probability of ecological constraints and dengue transmission risk.  2/100,000), all of which are located in Guangzhou, the capital city of Guangdong (Fig. 1a).
Guangdong has the most suitable climatic conditions for dengue transmission for each vector (Fig. 1b, c). SCI for Ae. aegypti was highest in the Leizhou peninsula below 22°N and western coastal areas between 22-23°N, with gradually reducing suitability moving further north, and the lowest was observed in the furthest northern part (Fig. 1b). SCI for Ae. albopictus was less likely to be distributed evenly through low to high latitude (Fig. 1c). Regions with the highest SCI for both vectors overlapped and located in the western coastal areas below 23°N (Fig. 1b,c).
The relationship between autochthonous dengue incidence, latitude, imported case and SCI for the two vector species is presented in Fig. 2. The correlation pattern between autochthonous dengue case and latitude was consistent between the two vector species; incidence and SCI were high in 23°N ( Fig. 2a and b). The relationship between imported dengue cases, vectors and autochthonous dengue cases indicated a strong relationship between autochthonous and imported dengue cases, which may remain independent over various levels of SCI for each vector species (Fig. 2c and d).
The relationship between autochthonous dengue incidence and SCI for Ae. aegypti was examined when SCI for Ae. albopictus was held constant at four levels from low to high (34.9 to 58.1, 58.2 to 63.6, 63.8 to 84.5 and 85.5 to 141.8), with results presented in Fig. 2e. There was a moderate relationship between dengue incidence and SCI for Ae. aegypti when SCI of Ae. albopictus was held constant at 34.9 to 58.1 (e.g., minimum to mean -one SD) (Fig. 2e).

Spearman correlation
Spearman correlation coe cients between any two variables considered in this study are presented in Table 2. There was a moderate relationship between autochthonous dengue incidence and SCI for Ae. aegypti (r s = 0.41, p=0.000), while there was a weak correlation between incidence and SCI for Ae. albopictus (r s = -0.18, p=0.000). There was a moderate correlation between dengue incidence and difference in SCI (r s = 0.53, p=0.000). There was a moderate correlation between autochthonous dengue incidence and the other variables, including imported cases, NDVI and human population density. There was a moderate relationship between the two SCIs ((r s =0.43, p=0.000).  (Table 3 Model IV). To avoid possible collinearity between NDVI and population density (r s =-0.73, p=0.000, Table   2), more importantly, the population density was not a signi cant predictor in multivariable models), the population density was excluded from NB regression analyses.

Discussion
We investigated the association between SCI and autochthonous dengue cases in this study and the value of SCI as a predictor of dengue outbreak risk. We found SCI was associated with autochthonous dengue cases, implying SCI might be bene cial in forecasting the probability of local dengue transmission or outbreak occurrence. We also discovered a statistically signi cant interaction between SCI for Ae. albopictus and SCI for Ae. aegypti, which invites the hypothesis that vector species interactions may also in uence dengue virus transmission.
We found opposing relationships between species-speci c SCI and autochthonous dengue occurrence in Guangdong: SCI for Ae. aegypti was positively associated with disease cases while SCI for Ae. albopictus was negatively related to local cases. This nding is not surprising because there are still several signi cant gaps in understanding correlations between climate suitability and virus transmission probability and intensity. Although temperature-dependent transmission models prove valuable for predicting vector spread and dengue transmission, most of these modelling studies nd a very broad correlation of transmission occurring within a temperature range [22,39]. Few can su ciently capture transmission dynamics to provide reliable predictions.
Human mobility and behaviour, urban development and microclimates, aquatic habitat supply, vector control techniques may all impact dengue transmission [23,[40][41][42][43][44][45]. These characteristics may be as important as climate suitability in determining dengue transmission risk. Moreover, transmission dynamics and bionomic responses differ across vector species, mosquito populations and dengue virus serotypes [20,46,47]. While some potential confounders have been adjusted in our study, other elements may contribute to dengue transmission that we have not considered. Finally, whether primary mosquito vectors are present, optimal climatic conditions for dengue vector species may play a substantial role in the outbreak. This suggests the possibility that vector species are a prerequisite for initiating and establishing a dengue outbreak; suitable conditions may expand and maintain the outbreak through complicated effects on vector sympatry or interactions.
Although SCI was not equivalent to mosquito species absence/presence or densities revealed only by entomological surveillance data, speci c vector mosquitoes should only occur at a suitable habitat with favourable meteorological conditions. Our ndings indicate the possibility that SCI might be indirectly associated with autochthonous dengue cases through the variation in dengue vector population, which means that SCI might be bene cial in dengue transmission or outbreaks prediction in the absence of entomological data. This might be meaningful in some dengue-endemic countries where vector absence/presence or density data are usually unavailable and, more importantly, there has been no reliable indication of any consistent association between entomological indices and dengue cases [20].
Our results suggested a signi cant interaction between SCI for Ae. aegypti (main effect) and SCI for Ae. albopictus, which means the expected effect of SCI for Ae. aegypti on autochthonic dengue cases may change over the various levels of SCI for Ae. albopictus. Speci cally, a one-unit increase in SCI for Ae. aegypti may indicate a higher risk of dengue in a region with a low SCI of Ae. albopictus than in a region with a high SCI of Ae. albopictus. Given that regions with high SCI for each Aedes vector have some overlaps in Guangdong, the coexistence of these two vectors is theoretically possible. In areas sympatric for the two species, declining population and displacement of Ae. aegypti have been reported as a result of Ae. albopictus superiority in resource competition at the larval stage and asymmetric sterilisation at the adult stage following interspeci c mating [48]. It has also been supposed that Ae. aegypti prefers to reproduce in areas devoid of Ae. albopictus [49]. Although there is no evidence of vector competitiveness from a eld study in Guangdong, recent evidence from a laboratory experiment suggests competitive displacement is theoretically possible. Larvae of the two species were mixed, allowed to emerge and then cycled through six generations, with the results that Ae. aegypti from the Leizhou peninsula were suppressed (a combination of generation time and abundance was used to judge competitiveness) by Ae. albopictus from Guangzhou but not by Ae. albopictus from other cities [50]. This nding warrants further exploration because the sympatric habit for the two species is expected to expand to additional areas in China [51].
We also observed that the difference in SCI was positively associated with dengue transmission. The two Aedes species are usually distinct in terms of ecological environment, feeding preferences and, critically for the SCI, thermal range. Therefore, high SCI for Ae. aegypti correlating with low SCI for Ae. albopictus is meaningful, indicating that a larger difference in SCI may indicate less competition between the two mosquito species. In other words, areas that are optimally suitable for both species (i.e. low difference in SCI) may, somewhat counterintuitively, produce a decreased dengue transmission risk potentially through the competitive displacement of one species by the other.
There are several limitations to our study. SCI is estimated using annual data and may not accurately capture seasonal variation and time-lag effects in dengue case occurrence. A ner scale of SCI is thus needed in any subsequent spatiotemporal analysis in this geographic region. Since SCI is not equivalent to dengue vector density, SCI should be validated in future studies against actual vector surveillance or entomology survey data. However, as a rst step, our SCI provides valuable insight into the possible relationship between optimal transmission environments and broad disease predictions. Further parameters, such as environmental variation (e.g., oviposition habitat availability, seasonal and daily temperature uctuation) and socioeconomic variables, should be included in the SCI to improve prediction e ciency. Enhanced SCI may be bene cial for extrapolating the possible geographic range of transmission beyond the existing environmental context (e.g., under climate change and for newly invading pathogens).

Conclusions
This study has shown that climate-based SCI is bene cial for evaluating dengue outbreak risk and the potential effect of possible vector competition on dengue transmission. We suggest that the dengue virus in Ae. albopictus vector may be more transmissible in a region with warmer weather conditions (high SCI of Ae. aegypti). SCI of Ae. aegypti could be used to predict dengue transmission even in locations lacking Ae. aegypti but with Ae. albopictus present. Moreover, the SCI can be used to investigate the relative contributions of Ae. albopictus and Ae. aegypti to dengue virus outbreaks in the absence of entomology data in future research.
Abbreviations Aedes: Ae.; SCI: Suitable conditions index; NB: negative binomial; NDVI: Normalized Difference Vegetation Index; BIC: Bayesian information criterion Declarations Figure 1 Geographical distribution of dengue (a) and suitable condition index (SCI) for Ae. aegypti (b) and SCI for Ae. albopictus (c) at a county level in Guangdong province, China, 2014

Figure 2
The relationship between autochthonous dengue incidence, latitude, imported dengue cases and suitable conditions index (SCI) in Guangdong, 2014. The relationship between autochthonous dengue incidence, latitude and SCI for Ae. aegypti (a); the relationship between autochthonous dengue incidence, latitude and SCI for Ae. albopictus (b); the relationship between autochthonous dengue incidence and imported dengue cases and SCI for Ae. aegypti (c); the relationship between autochthonous dengue incidence and imported dengue cases and SCI for Ae. albopictus (d) and the relationship between autochthonous dengue incidence and SCI for Ae. aegypti over various levels of SCI for Ae. albopictus ("scial" for short) (e).