1. Epidemiology and Global Health
  2. Microbiology and Infectious Disease
Download icon

Mapping global environmental suitability for Zika virus

  1. Jane P Messina Is a corresponding author
  2. Moritz UG Kraemer
  3. Oliver J Brady
  4. David M Pigott
  5. Freya M Shearer
  6. Daniel J Weiss
  7. Nick Golding
  8. Corrine W Ruktanonchai
  9. Peter W Gething
  10. Emily Cohn
  11. John S Brownstein
  12. Kamran Khan
  13. Andrew J Tatem
  14. Thomas Jaenisch
  15. Christopher JL Murray
  16. Fatima Marinho
  17. Thomas W Scott
  18. Simon I Hay Is a corresponding author
  1. University of Oxford, United Kingdom
  2. University of Washington, United States
  3. University of Melbourne, United Kingdom
  4. University of Southampton, United Kingdom
  5. Harvard Medical School, United Kingdom
  6. University of Toronto, Canada
  7. St Michael's Hospital, Canada
  8. Flowminder Foundation, Sweden
  9. Heidelberg University Hospital, Germany
  10. Heidelberg partner site, Germany
  11. Ministry of Health Brazil, Brazil
  12. University of California Davis, United States
Research Article
Cited
39
Views
29,658
Comments
0
Cite as: eLife 2016;5:e15272 doi: 10.7554/eLife.15272

Abstract

Zika virus was discovered in Uganda in 1947 and is transmitted by Aedes mosquitoes, which also act as vectors for dengue and chikungunya viruses throughout much of the tropical world. In 2007, an outbreak in the Federated States of Micronesia sparked public health concern. In 2013, the virus began to spread across other parts of Oceania and in 2015, a large outbreak in Latin America began in Brazil. Possible associations with microcephaly and Guillain-Barré syndrome observed in this outbreak have raised concerns about continued global spread of Zika virus, prompting its declaration as a Public Health Emergency of International Concern by the World Health Organization. We conducted species distribution modelling to map environmental suitability for Zika. We show a large portion of tropical and sub-tropical regions globally have suitable environmental conditions with over 2.17 billion people inhabiting these areas.

https://doi.org/10.7554/eLife.15272.001

eLife digest

Zika virus is transmitted between humans by mosquitoes. The majority of infections cause mild flu-like symptoms, but neurological complications in adults and infants have been found in recent outbreaks.

Although it was discovered in Uganda in 1947, Zika only caused sporadic infections in humans until 2007, when it caused a large outbreak in the Federated States of Micronesia. The virus later spread across Oceania, was first reported in Brazil in 2015 and has since rapidly spread across Latin America. This has led many people to question how far it will continue to spread. There was therefore a need to define the areas where the virus could be transmitted, including the human populations that might be risk in these areas.

Messina et al. have now mapped the areas that provide conditions that are highly suitable for the spread of the Zika virus. These areas occur in many tropical and sub-tropical regions around the globe. The largest areas of risk in the Americas lie in Brazil, Colombia and Venezuela. Although Zika has yet to be reported in the USA, a large portion of the southeast region from Texas through to Florida is highly suitable for transmission. Much of sub-Saharan Africa (where several sporadic cases have been reported since the 1950s) also presents an environment that is highly suitable for the Zika virus. While no cases have yet been reported in India, a large portion of the subcontinent is also suitable for Zika transmission.

Over 2 billion people live in Zika-suitable areas globally, and in the Americas alone, over 5.4 million births occurred in 2015 within such areas. It is important, however, to recognize that not all individuals living in suitable areas will necessarily be exposed to Zika.

We still lack a great deal of basic epidemiological information about Zika. More needs to be known about the species of mosquito that spreads the disease and how the Zika virus interacts with related viruses such as dengue. As such information becomes available and clinical cases become routinely diagnosed, the global evidence base will be strengthened, which will improve the accuracy of future maps.

https://doi.org/10.7554/eLife.15272.002

Introduction

Zika virus (ZIKV) is an emerging arbovirus carried by mosquitoes of the genus Aedes (Musso et al., 2014). Although discovered in Uganda in 1947 (Dick et al., 1952Dick, 1953) ZIKV was only known to cause sporadic infections in humans in Africa and Asia until 2007 (Lanciotti et al., 2008), when it caused a large outbreak of symptomatic cases on Yap island in the Federated States of Micronesia (FSM), followed by another in French Polynesia in 2013–14 and subsequent spread across Oceania (Musso et al., 2015a). In the 2007 Yap island outbreak, it was estimated that approximately 20% of ZIKV cases were symptomatic. While indigenous transmission of ZIKV to humans was reported for the first time in Latin America in 2015 (Zanluca et al., 2015; World Health Organisation, 2015), recent phylogeographic research estimates that the virus was introduced into the region between May and December 2013 (Faria et al., 2016). This recent rapid spread has led to concern that the virus is following a similar pattern of global expansion to that of dengue and chikungunya (Musso et al., 2015a).

ZIKV has been isolated from 19 different Aedes species (Haddow et al., 2012; Grard et al., 2014), but virus has been most frequently found in Ae. aegypti (Monlun et al., 1992; Marchette et al., 1969; Smithburn, 1954; Pond, 1963; Faye et al., 2008; Foy et al., 2011b; WHO Collaborating Center for Reference and Research on Arboviruses and Hemorrhagic Fever Viruses: Annual Report, 1999). These studies were based upon ancestral African strains of ZIKV, but the current rapid spread of ZIKV in Latin America is indicative of this highly efficient arbovirus vector (Marcondes and Ximenes, 2015). The relatively recent global spread of Ae. albopictus (Benedict et al., 2007; Kraemer et al., 2015c) and the rarity of ZIKV isolations from wild mosquitoes may also partially explain the lower frequency of isolations from Ae. albopictus populations. Whilst virus transmission by Ae. albopictus and other minor vector species has normally resulted in only a small number of cases (Kutsuna et al., 2015; Roiz et al., 2015), these vectors do pose the threat of limited transmission (Grard et al., 2014). The wide geographic distribution of Ae. albopictus combined with the frequent virus introduction via viraemic travellers (McCarthy 2016; Bogoch et al., 2016; Morrison et al., 2008; Scott and Takken, 2012), means the risk for ZIKV infection via this vector must therefore also be considered in ZIKV mapping.

The fact that ZIKV reporting was limited to a few small areas in Africa and Asia until 2007 means that global risk mapping has not, until recently, been a priority (Pigott et al., 2015b). Recent associations with Guillain-Barré syndrome in adults and microcephaly in infants born to ZIKV-infected mothers (World Health Organisation, 2015; Martines et al., 2016) have revealed that ZIKV could lead to more severe complications than the mild rash and flu-like symptoms that characterize the majority of symptomatic cases (Gatherer and Kohl, 2016). Considering these potentially severe complications and the rapid expansion of ZIKV into previously unaffected areas, the global public health community needs information about those areas that are environmentally suitable for transmission of ZIKV to humans. Being a closely related flavivirus to DENV, there is furthermore the potential for antigen-based diagnostic tests to exhibit cross-reactivity when IgM ELISA is used for rapid diagnosis. Although ZIKV-specific serologic assays are being developed by the U.S. Centers for Disease Control, currently the only method of confirming ZIKV infection is by using PCR on acute specimens (Lanciotti et al., 2008, Faye et al., 2008). Awareness of suitability for transmission is essential if proper detection methods are to be employed.

In this paper, we use species distribution modelling techniques that have been useful for mapping other vector-borne diseases such as dengue (Bhatt et al., 2013), Leishmaniasis (Pigott et al., 2014b), and Crimean-Congo Haemorrhagic Fever (Messina et al., 2015b) to map environmental suitability for ZIKV. The environmental niche of a disease can be identified according to a combination of environmental conditions supporting its presence in a particular location, with statistical modelling then allowing this niche to be described quantitatively (Kraemer et al., 2016). Niche modelling uses records of known disease occurrence alongside hypothesized environmental covariates to predict suitability for disease transmission in regions where it has yet to be reported (Elith and Leathwick, 2009). Contemporary high spatial-resolution global data representing a variety of environmental conditions allows for these predictions to be made at a global scale (Hay et al., 2006).

Results

Figure 1A shows the locations of the 323 standardized occurrence records in the final dataset, classified by the following date ranges: (i) up until 2006 (before the outbreak in FSM); (ii) between 2007 (the year of the FSM outbreak) and 2014; and (iii) since 2015, the first reporting of ZIKV in the Americas. This map is accompanied by the graph in Figure 1B, showing the number of reported occurrence locations globally by year. These figures highlight the more sporadic nature of reporting until recent years, with the majority of occurrences in the dataset (63%) coming from the recent 2015–2016 outbreak in Latin America.

Figure 1 with 1 supplement see all
(A) Map showing the distribution of the final set of 323 ZIKV occurrence locations entered into the ensemble Boosted Regression Tree modelling procedure.

Locations are classified by year of occurrence to show those which took place (i) prior to the 2007 outbreak in Federated States of Micronesia; (ii) between 2007–2014; and (iii) during the 2015–2016 outbreak; (B) the total number of locations reporting symptomatic ZIKV occurrence in humans globally over time.

https://doi.org/10.7554/eLife.15272.003

The final map that resulted from the mean of 300 ensemble Boosted Regression Tree (BRT) models is shown in Figure 2A (with greater detail shown for each region in Figures 2B–D). Figure 2—figure supplement 1 shows the distribution of uncertainty based upon the upper and lower prediction quantiles from the 300 models. We restricted our models to make predictions only within areas where i) mosquito vectors (in this case Ae. aegypti) were able to persist and ii) where temperature was sufficient for arboviral replication within the mosquito. The former of these was calculated by taking the Ae. aegypti probability of occurrence (Kraemer et al., 2015c) value that incorporated 90% of all known occurrences (Kraemer et al., 2015b) (giving a threshold value of 0.8 and greater) while the latter was evaluated using a mechanistic mosquito model (Brady et al., 2013; 2014), which identified regions where arboviral transmission could be sustained for at least 355 days (one year minus the human incubation period) in an average year. Figure 3 is a country-level map distinguishing between those countries that are currently reporting ZIKV, those which have reported ZIKV in the past, those which have highly suitable areas for transmission, and those which are unsuitable. Our models predicted high levels of risk for ZIKV in many areas within the tropical and sub-tropical zones. Large portions of the Americas are suitable for transmission, with the largest areas of risk occurring in Brazil, followed by Colombia and Venezuela, all of which have reported high numbers of cases in the 2015–2016 outbreak. In Brazil, where the highest numbers of ZIKV are reported in the ongoing epidemic, the coastal cities in the south as well as large areas of the north are identified to have the highest environmental suitability of ZIKV. The central region of Brazil, on the other hand, has low population densities and smaller mosquito populations, which is reflected in the relatively low suitability for ZIKV transmission seen in the map. Although ZIKV has yet to be reported in the USA, a large portion of the southeast region of the country, including much of Texas through to Florida, is also highly suitable for transmission. Potential risk for ZIKV transmission is high in much of sub-Saharan Africa, with continuous suitability in the Democratic Republic of Congo and surrounding areas and several sporadic case reports in western sub-Saharan countries since the 1950s. Although no symptomatic cases have yet been reported in India, a large portion of this country is at potential risk for ZIKV transmission (over 2 million square kilometres), with environmental suitability extending from its northwest regions through to Bangladesh and Myanmar. The Indochina region, southeast China, and Indonesia all have large areas of environmental suitability as well, extending into Oceania. While only representing less than ten percent of Australia’s total land area, the area shown to be suitable for ZIKV transmission in its northernmost regions is considerable (comprising nearly 250,000 square kilometres).

Figure 2 with 4 supplements see all
Maps of (A) global environmental suitability for ZIKV, ranging from 0 (grey) to 1 (red), showing greater detail for (B) the Americas, (C) Africa, and (D) Asia and Oceania.
https://doi.org/10.7554/eLife.15272.005
Status of ZIKV reporting as of 2016 by country, showing countries that are highly environmentally suitable (having a suitable area of more than 10,000 square kilometres) but which have not yet reported symptomatic cases of ZIKV in humans. 'Currently reporting' countries are those having reported cases since 2015.
https://doi.org/10.7554/eLife.15272.010

Our models showed ZIKV risk to be particularly influenced by annual cumulative precipitation, contributing 65.0% to the variation in the ensemble of models. The next most important predictor in the model was temperature suitability for DENV transmission via Ae. albopictus, contributing 14.6%. These are followed by urban extents (8.3%), temperature suitability for DENV via Ae. aegypti (5.7%), the Enhanced Vegetation Index (EVI; 3.8%), and minimum relative humidity (2.5%). Effect plots for each covariate are provided in Figure 2—figure supplement 2. Validation statistics indicated high predictive performance of the BRT ensemble mean map evaluated in a 10-fold cross-validation procedure, with area under the receiver operating characteristic (AUC) of 0.829 ( ± 0.121 SD). Due to the uncertainty about Ae. albopictus as a competent vector for ZIKV, we also provide results for an ensemble of models which did not include temperature suitability for dengue via this mosquito species in Figure 2—figure supplement 3.

A threshold environmental suitability value of 0.397 in our final map was determined to incorporate 90% of all ZIKV occurrence locations. This was used to classify each 5 km x 5 km pixel on our final map as suitable or unsuitable for ZIKV transmission to humans. Using high-resolution global population estimates (WorldPop, 2015; SEDAC, 2015), we summed the populations living in Zika-suitable areas and have identified 2.17 billion people globally living within areas that are environmentally suitable for ZIKV transmission. Table 1 shows a breakdown of this figure by major world region, also showing the top four contributing countries to the potential population at risk. Asia has the most people living in areas that are suitable for ZIKV transmission at 1.42 billion, accounted for in large part by those living in India. In Africa, roughly 453 million people are living in areas suitable for ZIKV transmission, the largest proportion of which live in Nigeria. In the Americas, more than 298 million people live in ZIKV-suitable transmission zones, with approximately 40 percent of these people living in Brazil. Within the majority of environmentally suitable areas for ZIKV in the Americas, prolonged year-round transmission is possible. Southern Brazil and Argentina, however, are more likely to see transmission interrupted throughout the year, as is the case with the USA should autochthonous ZIKV transmission occur there. Using high-resolution data on births for the year 2015 (WorldPop, 2015), we also estimate that 5.42 million births will occur in the Americas over the next year within areas and times of environmental suitability for ZIKV transmission.

Table 1

Population living in areas suitable for ZIKV transmission within each major world region and top four countries contributing to these populations at risk.

https://doi.org/10.7554/eLife.15272.011
Region/CountryPopulation living in areas suitable for ZIKV transmission (millions)
Africa452.58
Nigeria111.97
Democratic Republic of the Congo68.95
Uganda33.43
United Republic of Tanzania22.70
Americas298.36
Brazil120.65
Mexico32.22
Colombia29.54
Venezuela22.22
Asia1,422.13
India413.19
Indonesia226.04
China213.84
Bangladesh133.29
World2,173.27

Discussion

A large number of viruses (circa 219) are known to be pathogenic (Woolhouse et al., 2012). Of the 53 species of Flavivirus, 19 are reported to have caused illness in humans (ICTV, 2014). Some flaviviruses, such as DENV, YFV, Japanese encephalitis virus, and West Nile virus, are widespread, causing many thousands of infections each year. The remainder, however, have been recognized as being pathogenic to humans for decades, but have highly focal reported distributions and are only minor contributors to mortality and disability globally (Hay et al., 2013; Murray et al., 2015). As a result, many are of relatively low priority when research and policy interest are considered (Pigott et al., 2015b). The recent spread of ZIKV across the globe highlights the need to reassess our consideration of these other flaviviruses, to gain a better understanding of the factors driving their spread and the potential for geographic expansion beyond their currently limited geographical extents.

Environmental suitability for virus transmission in an area does not necessarily mean that it will arrive and/or establish in that location. Arboviral infections in particular are dependent on a variety of non-environmental factors, with their movement having historically been largely attributed to human mobility from travel, trade, and migration, which introduce the viruses to places where mosquito vectors are already present (Murray et al., 2013; Weaver and Reisen, 2010; Nunes et al., 2015; Gubler and Clark, 1995). The identification of locations with permissible environments for transmission of emerging diseases like ZIKV is crucial, as importation could give rise to subsequent autochthonous cases in these locations (Hennessey et al., 2016; Zanluca et al., 2015). In order to identify places potentially receptive for ZIKV, we assembled the first comprehensive spatial dataset for ZIKV occurrence in humans and compiled a comprehensive set of high-resolution environmental covariates. We then used these data to implement a species distribution modelling approach (Elith and Leathwick, 2009) that has proven useful for mapping other vector-borne diseases (Bhatt et al., 2013; Pigott et al., 2014a; Mylne et al., 2015; Messina et al., 2015b), allowing us to make inferences about environmental suitability for ZIKV transmission in areas where it has yet to be reported or where we are less certain about its presence. How the ongoing epidemic unfolds in terms of case numbers (or incidence) will depend on a range of other factors such as local transmission dynamics, herd immunity, patterns of contact among mosquitoes and infectious and susceptible humans (Stoddard et al., 2013), and mosquito-to-human ratios as recently shown for dengue (Kraemer et al., 2015a) and chikungunya (Salje et al., 2016).

Globally, we predict that over 2.17 billion people live in areas that are environmentally suitable for ZIKV transmission. We also estimate the number of births occurring in the Americas only, as it is the region for which the most accurate high-resolution population data on births exists (Tatem et al., 2014; Sorichetta et al., 2015) and because it is the focus of an ongoing outbreak, which is the largest recorded thus far. In the Americas alone, an estimated 5.42 million births occurred in 2015 within areas and at times that are suitable for ZIKV transmission. It is important to recognize that not all individuals will be exposed to ZIKV. Like with other flaviviruses, a ZIKV outbreak may be temporally and spatially sporadic and, even in the most receptive environments, is unlikely that all of the population will be infected. Furthermore, increasing herd immunity of this likely sterilizing infection will rapidly reduce the size of the susceptible population at risk for infection in subsequent years (Dick et al., 1952) and work is ongoing to predict the likely infection dynamics after establishment. Instead, the estimates are intended as indicators of the total number of individuals or births that may require protection during the first wave of the outbreak. Specifically, these populations should be the focus of efforts to increase awareness and provide guidelines for mitigating personal risk of infection. In future analyses, our estimates could be extended to include ZIKV incidence and the virus’ effect on incidence of associated conditions such as Guillain-Barré syndrome and microcephaly. Before appropriately caveated estimates can be generated, however, more information is needed regarding: (i) the background rate of these conditions due to other causes; (ii) how risk may vary throughout the course of a pregnancy; (iii) the proportion of the population exposed during outbreaks; and (iv) whether or not immunity acquired through a mother’s prior exposure is protective.

For all arboviral diseases, public health education about reducing populations and avoiding contact with mosquito vectors is required in at-risk areas. Specific to ZIKV is the risk of microcephaly in newborns, which has led public health agencies to issue warnings for women who are currently or planning on becoming pregnant in areas suspected to have ongoing ZIKV transmission and the declaration of a Public Health Emergency of International Concern (Heymann et al., 2016). Due to the sensitive nature and implications of these warnings, it is important that levels of risk are rigorously estimated, validated, and updated. Transmission of related arboviral diseases still occurs in many areas we defined as at-risk for ZIKV, which highlights the need for improved vector control outcomes, particularly those targeting Ae. aegypti. Predicted levels of risk for ZIKV transmission are potentially helpful for prioritized allocation of vector control resources, as well as for differential diagnosis and, if a vaccine becomes available, delivery efforts. It should be noted that instances of ZIKV sexual transmission have been reported (Patino-Barbosa et al., 2015; Musso et al., 2015b; Foy et al., 2011a). We did not incorporate secondary modes of transmission into the models we described here, but our map can help inform future discussions about the potential impact of this mode of transmission as its relative importance becomes better understood.

A great deal of basic epidemiological information specific to ZIKV is lacking. As a result, information must be leveraged from our knowledge about transmission of related arboviruses. Previous work has focused on mapping other vector borne diseases that share much of the ecology of Zika, such as DENV (Bhatt et al., 2013) and CHIKV, as well as for its primary vectors, Ae. aegypti and Ae. albopictus (Kraemer et al., 2015c). For this reason, temperature suitability for dengue (Brady et al., 2013, 2014) was entered into the models due to the greater number of field and laboratory studies available for parameterising this metric for DENV. Until more studies related to vector competence and temperature constraints on ZIKV transmission to humans are conducted, this is the most accurate indicator of arboviral disease transmission via Aedes mosquitoes currently available. Indeed, all other covariates in our models could equally be applied to mapping DENV and CHIKV, and ZIKV-specific refinements to modelling covariates will be possible as the disease continues to expand to allow for improvements in future iterations of the map. The relatively smaller amount of occurrence data available for ZIKV (especially prior to recent outbreaks) means that this dataset should also be updated with new information as necessary, leading to a stronger global evidence base and improved accuracy of future maps. Better understanding of ZIKV transmission dynamics will eventually allow for further cartographic refinements to be made, such as the differentiation between endemic- and epidemic-prone areas. Still, all covariates included in the current study have been updated and refined since (Bhatt et al., 2013), and when combined with the most extensive occurrence database available for ZIKV, the resulting map we present here is currently the most accurate depiction of the distribution of environmental suitability for ZIKV. A map highlighting differences in predicted suitability for both diseases is provided in Figure 2—figure supplement 4.

Conclusion

In this study, we produced the first global high spatial-resolution map of environmental suitability for ZIKV transmission to humans using an assembly of known records of ZIKV occurrence and environmental covariates in a species distribution modelling framework. While it is clear that much remains to be understood about ZIKV, this first map serves as a baseline for understanding the change in the geographical distribution of this globally emerging arboviral disease. Knowledge of the potential distribution can encourage more vigilant surveillance in both humans and Aedes mosquito populations, as well as help in the allocation of limited resources for disease prevention. Public health awareness campaigns and advice for mitigation of individual risk can also be focused in the areas we have predicted to be highly suitable for ZIKV transmission, particularly during the first wave of infection in a population. The maps we have presented may also inform existing travel advisories for pregnant women and other travellers. The maps and underlying data are freely available online via figshare (http://www.figshare.com).

Materials and methods

To map environmental suitability for ZIKV transmission to humans, we applied a species distribution modelling approach to establish a multivariate empirical relationship between the probability of ZIKV occurrence and the environmental conditions in locations where the disease has been confirmed. We employed an ensemble boosted regression trees (BRT) methodology (De'ath, 2007; Elith et al., 2008), which required the generation of: (i) a comprehensive compendium of known locations of disease occurrence in humans; (ii) a set of background points representing locations where ZIKV has not yet been reported; and (iii) a set of high-resolution globally gridded environmental and socioeconomic covariates hypothesised to affect ZIKV transmission. The resulting model produces a 5 x 5 km spatial-resolution global map of environmental suitability for ZIKV transmission to humans.

Assembly of the geo-referenced ZIKV occurrence dataset

Information about the locations of ZIKV occurrence in humans was extracted from peer-reviewed literature, case reports, and informal online sources following previously established protocols (Kraemer et al., 2015b; Messina et al., 2014; 2015a). To collate the peer-reviewed dataset, literature searches were undertaken using PubMed (http://www.ncbi.nlm.nih.gov/pubmed) and ISI Web of Science (http://www.webofknowledge.com) search engines using the search term 'Zika'. No language restrictions were placed on these searches; however, only those citations with a full title and abstract were retrieved, resulting in the review of 148 references ranging in publication dates between 1951 and 2015. In-house language skills allowed review of all English, French, Portuguese and Spanish articles for useable location information for human ZIKV occurrence. ProMED-mail (http://www.promedmail.org) was also searched using the term 'Zika', resulting in the review of 139 reports between 27 June 2007 and 18 January 2016. Additionally, the most current database of ZIKV case locations in Brazil was obtained directly from the Brazilian Ministry of Health. From all sources, only laboratory confirmation of symptomatic ZIKV infection in humans was entered into the dataset (mention of suspected cases was not entered). Serological evidence from healthy individuals could represent a past infection, with transmission potentially occurring in a different location to that where the individual currently resides (Darwish et al., 1983), or could be an artefact from possible cross-reactivity with a variety of different viruses (Smithburn et al., 1954). As a result, these less reliable diagnoses of ZIKV were excluded.

All available location information was extracted from each peer-reviewed article and ProMED case report. The site name was used together with all contextual information provided about the site to determine its latitudinal and longitudinal coordinates using Google Maps (https://www.maps.google.com). If the study site could be geo-positioned to a specific place, it was recorded as a point location. If the study site could only be identified at an administrative area level (e.g. province or district), it was recorded as a polygon along with an identifier of its administrative unit. If imported cases were reported with information on the site of infection, they were geo-positioned to this site; if imported cases were reported with no information about the site of infection, they were not entered into the dataset. Informal online data sources were collated automatically by the web-based system HealthMap (http://www.healthmap.org) as described elsewhere (Freifeld et al., 2008). Alerts for ZIKV were obtained from HealthMap for the years 2014–2016, and then manually checked for validity. In total, usable location information was extracted from 110 sources. Information was also collected about the status of symptoms in each reported occurrence, distinguishing between those where symptomatic cases were being reported, versus those where only seroprevalence was detected in healthy individuals.

Due to the potential for multiple independent reports referring to the same cases temporal and spatial standardization was required, as we have described previously in detail for dengue mapping efforts (Messina et al., 2014). In brief, an occurrence was defined as a unique location with one or more confirmed cases of ZIKV occurring within one calendar year (the finest temporal resolution available across all records). Point locations were considered to be overlapping if they lay on the same 5 km x 5 km pixel, and polygon locations were identified by a unique administrative unit code. Furthermore, all polygons whose geographic area was greater than one square decimal degree (approximately 111 square kilometers at the equator) were removed from the dataset to avoid averaging covariate values over very large areas, and only those occurrences comprising symptomatic individuals were retained for modelling purposes to ensure an accurate location of infection. In total, the final occurrence dataset contained 323 unique occurrences to be entered into our BRT modelling procedure. A map of the final set of occurrence locations is provided as Figure 1A.

Generation of the background location dataset

Separate maps of the relative probability of occurrence of Ae. aegypti and Ae. albopictus (Kraemer et al., 2015c) were used to compute a combined metric of the relative probability of vector occurrence, by taking the maximum value from the two layers for all 5 km x 5 km gridded cells globally. The inverse of this combined-Aedes occurrence probability layer (higher values indicating greater certainty of absence) was then used to draw a biased sample of 10,000 background locations. As such, a greater number of background points were sampled in areas where we are more certain that Ae. aegypti or Ae. albopictus do not occur, and therefore where ZIKV is less likely to be transmitted to humans. While it has been demonstrated that predictive accuracy from presence-background species distribution models can be improved by biasing background record locations toward areas with greatest reporting probabilities (Phillips et al., 2009), information on possible reporting biases, or proxies of such spatial bias, are currently unavailable for ZIKV. These 10,000 background locations were combined with the standardized occurrence dataset to serve as comparison data locations in the BRT species distribution modelling procedure. The background locations were weighted such that their total sum was equal to the total number of occurrence locations (n=237; pseudo-absence weighting=0.0237), in order to aid in the discrimination capacity of the model (Barbet-Massin et al., 2012).

Explanatory covariates

A set of six covariates hypothesized to influence the global distribution of ZIKV transmission to humans were used in our models to establish an empirical relationship between ZIKV presence or absence and underlying environmental conditions. These six covariates included: (i) an index of temperature suitability for dengue transmission to humans via Ae. aegypti; (ii) temperature suitability for dengue transmission to humans via Ae. albopictus; (iii) minimum relative humidity; (iv) annual cumulative precipitation; (v) an enhanced vegetation index (EVI); and (vi) urban versus rural habitat type. The underlying hypothesis behind each of the covariates is discussed in more detail below, along with a description of data sources and any processing that was undertaken before entering these covariates into our models. Maps of each covariate layer are provided in the supplementary information in Figure 1—figure supplement 1.

Temperature suitability for dengue transmission to humans

via Ae. aegypti or Ae. albopictus: Temperature affects key physiological processes in Aedes mosquitoes, including age- and temperature-dependent adult female survival, as well as the duration of the extrinsic incubation period (EIP) of arboviruses and the length of the gonotrophic cycle (Brady et al., 2013). While these parameters have yet to be measured experimentally for ZIKV, they have been for the closely related DENV. We obtained temperature data from WorldClim v1.03 (http://www.wordclim.org), which uses historic global meteorological station data from 1961–2005 to interpolate global climate surfaces. MARKSIM software (Jones and Thornton, 2000) was then used to apply the coefficients of 17 Global Climate Models (GCMs) to estimate temperature values for the year 2015. This enabled us to incorporate the quantified effects of temperature on DENV transmission into a cohort simulation model that analysed the cumulative effects of both diurnal and inter-seasonal changes in temperature on DENV transmission within an average year, both for Ae. aegypti and Ae. albopictus separately. The models were then applied to the 2015 temperature data for each 5 km x 5 km grid cell globally. This resulted in maps of temperature suitability for DENV transmission by either Aedes species ranging from 0 (no suitable days) to 1 (365 suitable days). These measures were then used as a proxy for temperature suitability for ZIKV transmission to humans.

Annual cumulative precipitation

Presence of static surface water in natural or man-made containers is a pre-requisite for Aedes oviposition and larval and pupal development. While fine-scale spatial and temporal heterogeneities have been observed between precipitation, vector abundance, and incidence of human DENV infections, there is evidence that areas with greater amounts of precipitation are generally associated with higher DENV infection risk (Chandy et al., 2013; Chowell and Sanchez, 2006; Dom et al., 2013; Pinto et al., 2011; Restrepo et al., 2014; Sang et al., 2014; Sankari et al., 2012; Campbell et al., 2015). Although studies that directly connect levels of precipitation to ZIKV transmission have yet to exist, we assumed for Zika a similar association of precipitation as closely related flaviruses. WorldClim v1.03 precipitation data and MARKSIM software were used as described above for temperature, to estimate annual cumulative precipitation for the year 2015 for each 5 km x 5 km grid cell globally.

Minimum relative humidity

Greater relative humidity has been found to promote DENV propagation in Ae. aegypti mosquitoes in several localized settings (Colón-González et al., 2011; Thu et al., 1998), and has also been found to be an important contributor when predicting DENV risk at a global scale (Hales et al., 2002). Therefore, we again assumed a similar association for ZIKV in the absence of any direct studies, and included the minimum annual relative humidity in our models as a potential limiting factor to ZIKV transmission. Relative humidity (RH) was calculated as a percent of saturation humidity, or the amount of water vapour required to saturate the air given a particular temperature, using the temperature data from WorldClim v1.03 described earlier. The saturation, or 'dew', point (Tdew) was calculated using a tabular relationship (Linacre, 1977). RH was then calculated as follows:

RH=V(Tx)V(Tdew)×100

Where V(Tdew) = 611.21×exp(17.502×T240.97+T) and V(Tx) is the humidity at the given temperature. We then extracted the minimum annual RH for each 5 km x 5 km pixel globally for the year 2015.

Enhanced vegetation index (EVI)

A close association has been shown between local moisture supply, vegetation canopy development, and abundance of mosquito reproduction (Linthicum et al., 1999), with previous studies highlighting the importance of moisture-related measures such as relative humidity to DENV occurrence (Hales et al., 2002). Although resistant to desiccation, both Aedes eggs and adults require moisture to survive (Cox et al., 2007; Sota and Mogi, 1992; Reiskind and Lounibos, 2009; Costa et al., 2010; Luz et al., 2008), with low dry season moisture levels substantially affecting Aedes mortality (Russell et al., 2001; Trpis, 1972; Luz et al., 2008). Vegetation canopy cover has previously been associated with higher Aedes larvae density (Fuller et al., 2009; Troyo et al., 2009; Bisset Lazcano et al., 2006; Barrera et al., 2006) by reducing evaporation from containers, decreasing sub-canopy wind speed, and protecting outdoor habitats from direct sunlight. To account for these factors, we included a 5 km x 5 km resolution measure of the EVI derived from NASA’s Moderate Resolution Imaging Spectrometer (MODIS, Boulder, Colorado) imagery (Wan et al., 2002; Lin, 2012), summarized from gap-filled, 8-day, 1 km x 1 km resolution images acquired globally for years 2000 through 2014 (Weiss et al., 2014) to produce a mean annual EVI layer. This mean EVI product is indicative of amount of photosynthesis taking place in the environment over the course of a year, which is positively correlated with the density of vegetation, and is thus a proxy for the level of moisture available given the relationship between precipitation and vegetative growth.

Urban versus rural habitat type

There is a well-established link between urban areas, some vector borne diseases, and their vectors. In particular, Ae. aegypti is found in close proximity to human dwellings often breeding in artificial containers (Brown et al., 2011; Powell and Tabachnick, 2013; Kraemer et al., 2015c). To identify the relationship between urbanisation and ZIKV presence we adapted probabilistic spatial modelling techniques to predict the spatial distribution of global urban extents at a 5 km x 5 km spatial resolution. We used urban growth rates from the United Nations Population Division (Division, 2014), paired with urban extents measured and tested by the Moderate Resolution Imaging Spectroradiometer Collection 5 (MODIS C5) land-cover product for Asia (Schneider et al., 2009; 2010; 2015). A set of spatial covariate datasets hypothesized to influence the spatial patterns of urban expansion was generated, including the time to travel from each 5 km x 5 km pixel to a major city (Nelson, 2008), the proportion of urbanised land within a buffer of 20 km, human population density (Linard and Tatem, 2012; Stevens et al., 2015; Gaughan et al., 2013), slope (Becker et al., 2009), and distance to water (Arino et al., 2008). A BRT modelling approach was then used to predict areas that would become urban in 2015 (Linard et al., 2013). Outputs were tested against a training dataset comprising points from Asia only, and showed good overall predictive performance (AUC=0.82). The output raster is a 5 km x 5 km gridded surface with urban (1) vs. rural (0) pixels.

Ensemble boosted regression trees approach

The boosted regression tree (BRT) modelling procedure combines regression trees with gradient boosting (Friedman, 2001). In this procedure, an initial regression tree is fitted and iteratively improved upon in a forward stagewise manner (boosting) by minimising the variation in the response not explained by the model at each iteration. It has been shown to fit complicated response functions efficiently, while guarding against over-fitting by use of extensive internal cross-validation. As such, this approach has been successfully employed in the past to map dengue and its Aedes mosquito vectors, as well as other vector-borne diseases (Bhatt et al., 2013; Pigott et al., 2014b; Messina et al., 2015b; Kraemer et al., 2015c). To increase the robustness of model predictions and quantify model uncertainty, we fitted an ensemble (Araújo and New, 2007) of 300 BRT models to separate bootstraps of the data. We then evaluated the central tendency as the mean across all 300 BRT models (Bhatt et al., 2013). Each of the 300 individual models was fitted using the gbm.step subroutine in the dismo package in the R statistical programming environment (Elith et al., 2008). All other tuning parameters of the algorithm were held at their default values (tree complexity= 4, learning rate= 0.005, bag fraction= 0.75, step size= 10, cross-validation folds=10). Each of the 300 models predicts environmental suitability on a continuous scale from 0 to 1, with a final prediction map then being generated by calculating the mean prediction across all models for each 5 km x 5 km pixel. Cross-validation was applied to each model, whereby ten subsets of the data comprising 10% of the presence and background observations were assessed based on their ability to predict the distribution of the other 90% of records using the mean area under the curve (AUC) statistic. This AUC value was then averaged across the ten sub-models and finally across all 300 models in the ensemble in order to derive an overall estimate of goodness-of-fit. Additionally, to avoid AUC inflation due to spatial sorting bias, a pairwise distance sampling procedure was used, resulting in a final AUC which is lower than would be returned by standard procedures but which gives a more realistic quantification of the model’s ability to extrapolate predictions to new regions (Wenger and Olden, 2012). We restricted our models to make predictions only within areas where either Ae. aegypti probability of occurrence (Kraemer et al., 2015c) is more than 0.8 or temperature is conducive to transmission for at least 355 days in an average year. A second ensemble of 300 models was executed which did not take into account temperature suitability for dengue transmission via Ae. albopictus, due to the uncertainty of this species as a competent ZIKV vector. The results of this ensemble of models are provided in Figure 2—figure supplement 3.

Population and births at risk

To calculate the number of people located in an area that is at any level of risk for ZIKV transmission, the global ZIKV environmental suitability map was combined with fine-scale global population surfaces (SEDAC, 2015; WorldPop, 2015). Firstly, the continuous ZIKV environmental suitability map (ranging from 0 to 1) was converted into a binary surface indicating whether there is any risk of transmission. To do this, we carried out a protocol previously used in (Pigott et al., 2015a), choosing a threshold environmental suitability value that encompasses 90% of the ZIKV occurrence point locations. This threshold cut-off of 90% was chosen (rather than 100%) to reflect potential errors or inaccurate locations in the occurrence point dataset. Every 5 km x 5 km pixel in the suitability map with a value above this threshold value was considered at risk for ZIKV transmission. Finally, to estimate the population at risk, we multiplied this binary ZIKV risk map by the global population counts (aligned and aggregated to the same 5 x 5 km grid) for the year 2015 and summed across all cells.

We next estimated the maximum number of births potentially affected by ZIKV in Latin America, as this region is the focus of the recent outbreak and the first to point to a possible association with microcephaly in newborn infants to mothers infected with ZIKV. In order to do this, we first identified the proportion of the year that is suitable for ZIKV transmission within areas that are predicted to be suitable in the binary ZIKV risk map. This proportion was derived from existing temperature suitability models (Brady et al., 2013; 2014), which predict the total number of days within an average year that arbovirus transmission can be sustained in Ae. aegypti, assuming there is a local human reservoir of infection. While the intra-mosquito viral dynamics in this model were parameterised for dengue virus, the limited information currently available on other arboviruses suggests that their dynamics are similar (Lambrechts et al., 2011). Using the resulting 5 km x 5 km map showing the proportion of the year suitable for ZIKV transmission to humans, we then multiplied this by a map (also at a 5 km x 5 km resolution) of the number of births in the Americas for the year 2015, updated from (Tatem et al., 2014; UNFPA, 2014). The resulting map indicates the number of births in the Americas potentially at risk for ZIKV (for 2015), assuming ZIKV currently fully occupies its environmental niche and that births are evenly distributed throughout the year.

References

  1. 1
  2. 2
    GLOBCOVER the most detailed portrait of earth
    1. O Arino
    2. P Bicheron
    3. F Achard
    4. J Latham
    5. R Witt
    6. JL Weber
    (2008)
    ESA Bulletin pp. 24–31.
  3. 3
  4. 4
    Use of the pupal survey technique for measuring Aedes aegypti (Diptera: Culicidae) productivity in Puerto Rico
    1. R Barrera
    2. M Amador
    3. GG Clark
    (2006)
    The American Journal of Tropical Medicine and Hygiene 74:290–302.
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
    Climate-based descriptive models of dengue fever: the 2002 epidemic in Colima, Mexico
    1. G Chowell
    2. F Sanchez
    (2006)
    Journal of Environmental Health 68:40–44.
  16. 16
  17. 17
  18. 18
    Habitat segregation of dengue vectors along an urban environmental gradient
    1. J Cox
    2. ME Grillet
    3. OM Ramos
    4. M Amador
    5. R Barrera
    (2007)
    The American Journal of Tropical Medicine and Hygiene 76:820–826.
  19. 19
  20. 20
  21. 21
    Zika virus. I. isolations and serological specificity
    1. GW Dick
    2. SF Kitchen
    3. AJ Haddow
    (1952)
    Transactions of the Royal Society of Tropical Medicine and Hygiene 46:509–520.
    https://doi.org/10.1016/0035-9203(52)90042-4
  22. 22
    Yellow fever; a problem in epidemiology
    1. Dick GW
    (1953)
    British Medical Bulletin 9:215–235.
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
    Genetic characterization of zika virus strains: Geographic expansion of the asian lineage
    1. AD Haddow
    2. AJ Schuh
    3. CY Yasuda
    4. MR Kasper
    5. V Heang
    6. R Huy
    7. H Guzman
    8. RB Tesh
    9. SC Weaver
    (2012)
    PLoS Neglected Tropical Diseases, 6, 10.1371/journal.pntd.0001477.
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
    Virus taxonomy: 2014 release
    1. ICTV
    (2014)
    Virus taxonomy: 2014 release, 1st Edition.
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
    Large-scale spatial population databases in infectious disease research
    1. C Linard
    2. AJ Tatem
    (2012)
    Int J Health Geogr, 11, 10.1186/1476-072x-11-7.
  56. 56
  57. 57
  58. 58
    Isolation of Zika virus from Aedes aegypti mosquitoes in Malaysia
    1. NJ Marchette
    2. R Garcia
    3. A Rudnick
    (1969)
    The American Journal of Tropical Medicine and Hygiene 18:411–415.
  59. 59
    Zika virus in brazil and the danger of infestation by aedes (stegomyia) mosquitoes
    1. CB Marcondes
    2. MF Ximenes
    (2015)
    Revista Da Sociedade Brasileira De Medicina Tropical, 10.1590/0037-8682-0220-2015.
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
    Surveillance of the circulation of arbovirus of medical interest in the region of eastern Senegal
    1. E Monlun
    2. H Zeller
    3. B Le Guenno
    4. M Traore-Lamizana
    5. J Hervy
    6. F Adam
    7. L Ferrara
    8. D Fontenille
    9. R Sylla
    10. M Mondo
    (1992)
    Bulletin de la Société de pathologie exotique 86:21–28.
  66. 66
  67. 67
    Global, regional, and national disability-adjusted life years (dalys) for 306 diseases and injuries and healthy life expectancy (HALE) for 188 countries, 1990-2013: Quantifying the epidemiological transition
    1. CJ Murray
    2. RM Barber
    3. KJ Foreman
    4. A Abbasoglu Ozgoren
    5. F Abd-Allah
    6. SF Abera
    7. V Aboyans
    8. JP Abraham
    9. I Abubakar
    10. LJ Abu-Raddad
    11. NM Abu-Rmeileh
    12. T Achoki
    13. IN Ackerman
    14. Z Ademi
    15. AK Adou
    16. JC Adsuar
    17. A Afshin
    18. EE Agardh
    19. SS Alam
    20. D Alasfoor
    21. MI Albittar
    22. MA Alegretti
    23. ZA Alemu
    24. R Alfonso-Cristancho
    25. S Alhabib
    26. R Ali
    27. F Alla
    28. P Allebeck
    29. MA Almazroa
    30. U Alsharif
    31. E Alvarez
    32. N Alvis-Guzman
    33. AT Amare
    34. EA Ameh
    35. H Amini
    36. W Ammar
    37. HR Anderson
    38. BO Anderson
    39. CA Antonio
    40. P Anwari
    41. J Arnlöv
    42. VS Arsic Arsenijevic
    43. A Artaman
    44. RJ Asghar
    45. R Assadi
    46. LS Atkins
    47. MA Avila
    48. B Awuah
    49. VF Bachman
    50. A Badawi
    51. MC Bahit
    52. K Balakrishnan
    53. A Banerjee
    54. SL Barker-Collo
    55. S Barquera
    56. L Barregard
    57. LH Barrero
    58. A Basu
    59. S Basu
    60. MO Basulaiman
    61. J Beardsley
    62. N Bedi
    63. E Beghi
    64. T Bekele
    65. ML Bell
    66. C Benjet
    67. DA Bennett
    68. IM Bensenor
    69. H Benzian
    70. E Bernabé
    71. A Bertozzi-Villa
    72. TJ Beyene
    73. N Bhala
    74. A Bhalla
    75. ZA Bhutta
    76. K Bienhoff
    77. B Bikbov
    78. S Biryukov
    79. JD Blore
    80. CD Blosser
    81. FM Blyth
    82. MA Bohensky
    83. IW Bolliger
    84. B Bora Başara
    85. NM Bornstein
    86. D Bose
    87. S Boufous
    88. RR Bourne
    89. LN Boyers
    90. M Brainin
    91. CE Brayne
    92. A Brazinova
    93. NJ Breitborde
    94. H Brenner
    95. AD Briggs
    96. PM Brooks
    97. JC Brown
    98. TS Brugha
    99. R Buchbinder
    100. GC Buckle
    101. CM Budke
    102. A Bulchis
    103. AG Bulloch
    104. IR Campos-Nonato
    105. H Carabin
    106. JR Carapetis
    107. R Cárdenas
    108. DO Carpenter
    109. V Caso
    110. CA Castañeda-Orjuela
    111. RE Castro
    112. F Catalá-López
    113. F Cavalleri
    114. A Çavlin
    115. VK Chadha
    116. JC Chang
    117. FJ Charlson
    118. H Chen
    119. W Chen
    120. PP Chiang
    121. O Chimed-Ochir
    122. R Chowdhury
    123. H Christensen
    124. CA Christophi
    125. M Cirillo
    126. MM Coates
    127. LE Coffeng
    128. MS Coggeshall
    129. V Colistro
    130. SM Colquhoun
    131. GS Cooke
    132. C Cooper
    133. LT Cooper
    134. LM Coppola
    135. M Cortinovis
    136. MH Criqui
    137. JA Crump
    138. L Cuevas-Nasu
    139. H Danawi
    140. L Dandona
    141. R Dandona
    142. E Dansereau
    143. PI Dargan
    144. G Davey
    145. A Davis
    146. DV Davitoiu
    147. A Dayama
    148. D De Leo
    149. L Degenhardt
    150. B Del Pozo-Cruz
    151. RP Dellavalle
    152. K Deribe
    153. S Derrett
    154. DC Des Jarlais
    155. M Dessalegn
    156. SD Dharmaratne
    157. MK Dherani
    158. C Diaz-Torné
    159. D Dicker
    160. EL Ding
    161. K Dokova
    162. ER Dorsey
    163. TR Driscoll
    164. L Duan
    165. HC Duber
    166. BE Ebel
    167. KM Edmond
    168. YM Elshrek
    169. M Endres
    170. SP Ermakov
    171. HE Erskine
    172. B Eshrati
    173. A Esteghamati
    174. K Estep
    175. EJ Faraon
    176. F Farzadfar
    177. DF Fay
    178. VL Feigin
    179. DT Felson
    180. SM Fereshtehnejad
    181. JG Fernandes
    182. AJ Ferrari
    183. C Fitzmaurice
    184. AD Flaxman
    185. TD Fleming
    186. N Foigt
    187. MH Forouzanfar
    188. FG Fowkes
    189. UF Paleo
    190. RC Franklin
    191. T Fürst
    192. B Gabbe
    193. L Gaffikin
    194. FG Gankpé
    195. JM Geleijnse
    196. BD Gessner
    197. P Gething
    198. KB Gibney
    199. M Giroud
    200. G Giussani
    201. H Gomez Dantes
    202. P Gona
    203. D González-Medina
    204. RA Gosselin
    205. CC Gotay
    206. A Goto
    207. HN Gouda
    208. N Graetz
    209. HC Gugnani
    210. R Gupta
    211. R Gupta
    212. RA Gutiérrez
    213. J Haagsma
    214. N Hafezi-Nejad
    215. H Hagan
    216. YA Halasa
    217. RR Hamadeh
    218. H Hamavid
    219. M Hammami
    220. J Hancock
    221. GJ Hankey
    222. GM Hansen
    223. Y Hao
    224. HL Harb
    225. JM Haro
    226. R Havmoeller
    227. SI Hay
    228. RJ Hay
    229. IB Heredia-Pi
    230. KR Heuton
    231. P Heydarpour
    232. H Higashi
    233. M Hijar
    234. HW Hoek
    235. HJ Hoffman
    236. HD Hosgood
    237. M Hossain
    238. PJ Hotez
    239. DG Hoy
    240. M Hsairi
    241. G Hu
    242. C Huang
    243. JJ Huang
    244. A Husseini
    245. C Huynh
    246. ML Iannarone
    247. KM Iburg
    248. K Innos
    249. M Inoue
    250. F Islami
    251. KH Jacobsen
    252. DL Jarvis
    253. SK Jassal
    254. SH Jee
    255. P Jeemon
    256. PN Jensen
    257. V Jha
    258. G Jiang
    259. Y Jiang
    260. JB Jonas
    261. K Juel
    262. H Kan
    263. A Karch
    264. CK Karema
    265. C Karimkhani
    266. G Karthikeyan
    267. NJ Kassebaum
    268. A Kaul
    269. N Kawakami
    270. K Kazanjan
    271. AH Kemp
    272. AP Kengne
    273. A Keren
    274. YS Khader
    275. SE Khalifa
    276. EA Khan
    277. G Khan
    278. YH Khang
    279. C Kieling
    280. D Kim
    281. S Kim
    282. Y Kim
    283. Y Kinfu
    284. JM Kinge
    285. M Kivipelto
    286. LD Knibbs
    287. AK Knudsen
    288. Y Kokubo
    289. S Kosen
    290. S Krishnaswami
    291. B Kuate Defo
    292. B Kucuk Bicer
    293. EJ Kuipers
    294. C Kulkarni
    295. VS Kulkarni
    296. GA Kumar
    297. HH Kyu
    298. T Lai
    299. R Lalloo
    300. T Lallukka
    301. H Lam
    302. Q Lan
    303. VC Lansingh
    304. A Larsson
    305. AE Lawrynowicz
    306. JL Leasher
    307. J Leigh
    308. R Leung
    309. CE Levitz
    310. B Li
    311. Y Li
    312. Y Li
    313. SS Lim
    314. M Lind
    315. SE Lipshultz
    316. S Liu
    317. Y Liu
    318. BK Lloyd
    319. KT Lofgren
    320. G Logroscino
    321. KJ Looker
    322. J Lortet-Tieulent
    323. PA Lotufo
    324. R Lozano
    325. RM Lucas
    326. R Lunevicius
    327. RA Lyons
    328. S Ma
    329. MF Macintyre
    330. MT Mackay
    331. M Majdan
    332. R Malekzadeh
    333. W Marcenes
    334. DJ Margolis
    335. C Margono
    336. MB Marzan
    337. JR Masci
    338. MT Mashal
    339. R Matzopoulos
    340. BM Mayosi
    341. TT Mazorodze
    342. NW Mcgill
    343. JJ Mcgrath
    344. M Mckee
    345. A Mclain
    346. PA Meaney
    347. C Medina
    348. MM Mehndiratta
    349. W Mekonnen
    350. YA Melaku
    351. M Meltzer
    352. ZA Memish
    353. GA Mensah
    354. A Meretoja
    355. FA Mhimbira
    356. R Micha
    357. TR Miller
    358. EJ Mills
    359. PB Mitchell
    360. CN Mock
    361. N Mohamed Ibrahim
    362. KA Mohammad
    363. AH Mokdad
    364. GL Mola
    365. L Monasta
    366. JC Montañez Hernandez
    367. M Montico
    368. TJ Montine
    369. MD Mooney
    370. AR Moore
    371. M Moradi-Lakeh
    372. AE Moran
    373. R Mori
    374. J Moschandreas
    375. WN Moturi
    376. ML Moyer
    377. D Mozaffarian
    378. WT Msemburi
    379. UO Mueller
    380. M Mukaigawara
    381. EC Mullany
    382. ME Murdoch
    383. J Murray
    384. KS Murthy
    385. M Naghavi
    386. A Naheed
    387. KS Naidoo
    388. L Naldi
    389. D Nand
    390. V Nangia
    391. KM Narayan
    392. C Nejjari
    393. SP Neupane
    394. CR Newton
    395. M Ng
    396. FN Ngalesoni
    397. G Nguyen
    398. MI Nisar
    399. S Nolte
    400. OF Norheim
    401. RE Norman
    402. B Norrving
    403. L Nyakarahuka
    404. IH Oh
    405. T Ohkubo
    406. SL Ohno
    407. BO Olusanya
    408. JN Opio
    409. K Ortblad
    410. A Ortiz
    411. AW Pain
    412. JD Pandian
    413. CI Panelo
    414. C Papachristou
    415. EK Park
    416. JH Park
    417. SB Patten
    418. GC Patton
    419. VK Paul
    420. BI Pavlin
    421. N Pearce
    422. DM Pereira
    423. R Perez-Padilla
    424. F Perez-Ruiz
    425. N Perico
    426. A Pervaiz
    427. K Pesudovs
    428. CB Peterson
    429. M Petzold
    430. MR Phillips
    431. BK Phillips
    432. DE Phillips
    433. FB Piel
    434. D Plass
    435. D Poenaru
    436. S Polinder
    437. D Pope
    438. S Popova
    439. RG Poulton
    440. F Pourmalek
    441. D Prabhakaran
    442. NM Prasad
    443. RL Pullan
    444. DM Qato
    445. DA Quistberg
    446. A Rafay
    447. K Rahimi
    448. SU Rahman
    449. M Raju
    450. SM Rana
    451. H Razavi
    452. KS Reddy
    453. A Refaat
    454. G Remuzzi
    455. S Resnikoff
    456. AL Ribeiro
    457. L Richardson
    458. JH Richardus
    459. DA Roberts
    460. D Rojas-Rueda
    461. L Ronfani
    462. GA Roth
    463. D Rothenbacher
    464. DH Rothstein
    465. JT Rowley
    466. N Roy
    467. GM Ruhago
    468. MY Saeedi
    469. S Saha
    470. MA Sahraian
    471. UK Sampson
    472. JR Sanabria
    473. L Sandar
    474. IS Santos
    475. M Satpathy
    476. M Sawhney
    477. P Scarborough
    478. IJ Schneider
    479. B Schöttker
    480. AE Schumacher
    481. DC Schwebel
    482. JG Scott
    483. S Seedat
    484. SG Sepanlou
    485. PT Serina
    486. EE Servan-Mori
    487. KA Shackelford
    488. A Shaheen
    489. S Shahraz
    490. T Shamah Levy
    491. S Shangguan
    492. J She
    493. S Sheikhbahaei
    494. P Shi
    495. K Shibuya
    496. Y Shinohara
    497. R Shiri
    498. K Shishani
    499. I Shiue
    500. MG Shrime
    501. ID Sigfusdottir
    502. DH Silberberg
    503. EP Simard
    504. S Sindi
    505. A Singh
    506. JA Singh
    507. L Singh
    508. V Skirbekk
    509. EL Slepak
    510. K Sliwa
    511. S Soneji
    512. K Søreide
    513. S Soshnikov
    514. LA Sposato
    515. CT Sreeramareddy
    516. JD Stanaway
    517. V Stathopoulou
    518. DJ Stein
    519. MB Stein
    520. C Steiner
    521. TJ Steiner
    522. A Stevens
    523. A Stewart
    524. LJ Stovner
    525. K Stroumpoulis
    526. BF Sunguya
    527. S Swaminathan
    528. M Swaroop
    529. BL Sykes
    530. KM Tabb
    531. K Takahashi
    532. N Tandon
    533. D Tanne
    534. M Tanner
    535. M Tavakkoli
    536. HR Taylor
    537. BJ Te Ao
    538. F Tediosi
    539. AM Temesgen
    540. T Templin
    541. M Ten Have
    542. EY Tenkorang
    543. AS Terkawi
    544. B Thomson
    545. AL Thorne-Lyman
    546. AG Thrift
    547. GD Thurston
    548. T Tillmann
    549. M Tonelli
    550. F Topouzis
    551. H Toyoshima
    552. J Traebert
    553. BX Tran
    554. M Trillini
    555. T Truelsen
    556. M Tsilimbaris
    557. EM Tuzcu
    558. US Uchendu
    559. KN Ukwaja
    560. EA Undurraga
    561. SB Uzun
    562. WH Van Brakel
    563. S Van De Vijver
    564. CH van Gool
    565. J Van Os
    566. TJ Vasankari
    567. N Venketasubramanian
    568. FS Violante
    569. VV Vlassov
    570. SE Vollset
    571. GR Wagner
    572. J Wagner
    573. SG Waller
    574. X Wan
    575. H Wang
    576. J Wang
    577. L Wang
    578. TS Warouw
    579. S Weichenthal
    580. E Weiderpass
    581. RG Weintraub
    582. W Wenzhi
    583. A Werdecker
    584. R Westerman
    585. HA Whiteford
    586. JD Wilkinson
    587. TN Williams
    588. CD Wolfe
    589. TM Wolock
    590. AD Woolf
    591. S Wulf
    592. B Wurtz
    593. G Xu
    594. LL Yan
    595. Y Yano
    596. P Ye
    597. GK Yentür
    598. P Yip
    599. N Yonemoto
    600. SJ Yoon
    601. MZ Younis
    602. C Yu
    603. ME Zaki
    604. Y Zhao
    605. Y Zheng
    606. D Zonies
    607. X Zou
    608. JA Salomon
    609. AD Lopez
    610. T Vos
    611. GBD 2013 DALYs and HALE Collaborators
    (2015)
    Lancet  386:2145–2191.
    https://doi.org/10.1016/S0140-6736(15)61340-X
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
    Travel Time to Major Cities: A Global Map of Accessibility
    1. A Nelson
    (2008)
    Ispra, Italy: GEMU-JRCOTE Commission.
  74. 74
  75. 75
    Zika: Another sexually transmitted infection?
    1. AM Patiño-Barbosa
    2. I Medina
    3. AF Gil-Restrepo
    4. AJ Rodriguez-Morales
    (2015)
    Sexually Transmitted Infections, 91, 10.1136/sextrans-2015-052189.
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
    ARTHROPOD-borne virus antibodies in sera from residents of south-east asia
    1. WL Pond
    (1963)
    Transactions of the Royal Society of Tropical Medicine and Hygiene 57:364–371.
    https://doi.org/10.1016/0035-9203(63)90100-7
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
    Outbreak of dengue virus serotype-2 (DENV-2) of Cambodian origin in Manipur, India - association with meteorological factors
    1. T Sankari
    2. SL Hoti
    3. TB Singh
    4. J Shanmugavel
    (2012)
    The Indian Journal of Medical Research 136:649–655.
  91. 91
  92. 92
  93. 93
  94. 94
  95. 95
    Gridded Population of the World, v4 (GPWv4)
    1. Socio-economic Data and Applications Center
    (2015)
    Gridded Population of the World, v4 (GPWv4).
  96. 96
    Neutralizing antibodies against arthropod-borne viruses in the sera of long-time residents of Malaya and Borneo
    1. K Smithburn
    (1954a)
    American Journal of Hygiene 59:157–163.
  97. 97
    Immunity to certain arthropod-borne viruses among indigenous residents of Egypt
    1. KC Smithburn
    2. RM Taylor
    3. F Rizk
    4. A Kader
    (1954b)
    The American Journal of Tropical Medicine and Hygiene 3:9–18.
  98. 98
    High-resolution gridded population datasets for latin america and the caribbean in 2010, 2015, and 2020
    1. A Sorichetta
    2. GM Hornby
    3. FR Stevens
    4. AE Gaughan
    5. C Linard
    6. AJ Tatem
    (2015)
    Scientific Data, 2, 10.1038/sdata.2015.45.
  99. 99
  100. 100
  101. 101
  102. 102
    Mapping for maternal and newborn health: The distributions of women of childbearing age, pregnancies and births
    1. AJ Tatem
    2. J Campbell
    3. M Guerra-Arias
    4. L de Bernis
    5. A Moran
    6. Z Matthews
    (2014)
    International Journal of Health Geographics, 13, 10.1186/1476-072X-13-2.
  103. 103
    The effect of temperature and humidity on dengue virus propagation in Aedes aegypti mosquitos
    1. HM Thu
    2. KM Aye
    3. S Thein
    (1998)
    The Southeast Asian Journal of Tropical Medicine and Public Health 29:280–284.
  104. 104
  105. 105
    Dry season survival of Aedes aegypti eggs in various breeding sites in the Dar es Salaam area, Tanzania
    1. M Trpis
    (1972)
    Bulletin of the World Health Organization, 47.
  106. 106
    World Urbanization Prospects: The 2014 Revision
    1. United Nations
    (2014)
    New York.
  107. 107
    The State of the World’s Midwifery 2014: A Universal Pathway. A Woman's Right to Health
    1. UNFPA
    (2014)
    UNFPA.
  108. 108
  109. 109
  110. 110
  111. 111
  112. 112
  113. 113
    Human viruses: Discovery and emergence
    1. M Woolhouse
    2. F Scott
    3. Z Hudson
    4. R Howey
    5. M Chase-Topping
    (2012)
    Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 367:2864–2871.
    https://doi.org/10.1098/rstb.2011.0354
  114. 114
  115. 115
    High Resolution Age-Structured Population Distribution Maps. U. O. S. GeoData Institute
    1. WorldPop
    (2015)
    High Resolution Age-Structured Population Distribution Maps. U. O. S. GeoData Institute.
  116. 116

Decision letter

  1. Mark Jit
    Reviewing Editor; London School of Hygiene & Tropical Medicine, and Public Health England, United Kingdom

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Mapping global environmental suitability for Zika virus" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom, Mark Jit, is a member of our Board of Reviewing Editors, and another is Jean-Paul Chretien. The evaluation has been overseen by Prabhat Jha as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The reviewers all felt that this was a timely analysis that provides important first estimates that are needed in light of the ongoing spread of Zika virus into new areas via infected travelers. They also commended the manuscript for increasing the epidemiological relevance by translating global temperatures into model-based estimates of suitability for dengue transmission as a surrogate for Zika virus. These are imperfect and based on static, climatic summaries of conditions that vary seasonally, but are a reasonable choice given the gaps in our present knowledge base. The reviewers also appreciated the open sharing of the datasets used.

However, the reviewers had two major concerns about the manuscript:

a) The definition of limits where "either Ae. aegypti probability of occurrence is more than 0.8 or temperature is conducive to transmission via this mosquito for at least 355 days in an average year" seems quite subjective and presumably helped to keep the mapped values from straying too far from known distributional limits. The reviewers would like you to indicate the objective criteria used to define these limits. Since any criteria are unlikely to be completely indisputable, they suggest that you consider providing some measure of uncertainty for the predicted suitability scores e.g. perhaps the distribution of suitability scores across ensembles.

b) Many (and possibly all) of the covariates in the models also apply to dengue – indeed two of them look at suitability for dengue transmission explicitly (due to lack of current understanding about physiological processes governing Zika transmission). Hence it remains unclear to the reviewers why the expected niche of Zika should be fundamentally different from that of dengue.

The reviewers did identify some differences between the Zika map and your corresponding map for dengue (Bhatt et al.), particularly in Africa. These are presumably driven by the dataset used to parameterise the model (based on Zika case reports in the literature). However, the very small (and potentially biased) body of literature on Zika virus prior to recent outbreaks may limit the accuracy of the estimates presented in the paper.

This needs to be carefully discussed: one solution would be to map the difference between your previous map for dengue (Bhatt et al.) and the map for Zika, with a discussion of the drivers for the difference and whether these seem to be justified (e.g. have at least some face validity).

In addition to the major comments, the reviewers also had some minor comments about presentational issues which we recommend you address, but which do not have to be explicitly responded to in your response letter:

a) On the whole the writing is excellent, but the Introduction is rather long. The focus should be sharpened to address the rationale for the present study. Much of the material presented is relevant, such as the connection to precedents of dengue and chikungunya, or the potential roles of various Aedes species in transmission, but it feels like a wandering review and the language should more directly point the reader to the relevance of each section.

b) The paragraph on cross-reactivity between dengue and Zika and newly emerging symptoms of Zika virus seems like a digression with little connection made to the broader purposes of the paper.

c) For vector competence, it would be useful to cite also very recent work emerging on Ae. aegypti and Ae. albopictus, and Cx. quinquefasciatus. Also to point out that all but the most recent vector competence work was done with ancestral African strains of Zika virus, not those circulating in the Americas.

d) Ae. hensilli should be mentioned as a critical vector on Yap.

e) When you state "wild isolations", it should be "isolations from wild mosquitoes" or similar.

f) The final statement in the Results about expected numbers of births needs additional context to understand how it was determined without the need to consult the detailed methods.

https://doi.org/10.7554/eLife.15272.016

Author response

The reviewers all felt that this was a timely analysis that provides important first estimates that are needed in light of the ongoing spread of Zika virus into new areas via infected travelers. They also commended the manuscript for increasing the epidemiological relevance by translating global temperatures into model-based estimates of suitability for dengue transmission as a surrogate for Zika virus. These are imperfect and based on static, climatic summaries of conditions that vary seasonally, but are a reasonable choice given the gaps in our present knowledge base. The reviewers also appreciated the open sharing of the datasets used.

However, the reviewers had two major concerns about the manuscript:

a) The definition of limits where "either Ae. aegypti probability of occurrence is more than 0.8 or temperature is conducive to transmission via this mosquito for at least 355 days in an average year" seems quite subjective and presumably helped to keep the mapped values from straying too far from known distributional limits. The reviewers would like you to indicate the objective criteria used to define these limits. Since any criteria are unlikely to be completely indisputable, they suggest that you consider providing some measure of uncertainty for the predicted suitability scores e.g. perhaps the distribution of suitability scores across ensembles.

We thank the reviewers for this suggestion, and agree that our criteria should have been better explained. The final paragraph of the Results section now details the formulation of these criteria and refers to the appropriate sources. Additionally, we have added a map of uncertainty in the suitability scores to the supplementary information and made reference to it within this same paragraph.

b) Many (and possibly all) of the covariates in the models also apply to dengue – indeed two of them look at suitability for dengue transmission explicitly (due to lack of current understanding about physiological processes governing Zika transmission). Hence it remains unclear to the reviewers why the expected niche of Zika should be fundamentally different from that of dengue. The reviewers did identify some differences between the Zika map and your corresponding map for dengue (Bhatt et al.), particularly in Africa. These are presumably driven by the dataset used to parameterise the model (based on Zika case reports in the literature). However, the very small (and potentially biased) body of literature on Zika virus prior to recent outbreaks may limit the accuracy of the estimates presented in the paper. This needs to be carefully discussed: one solution would be to map the difference between your previous map for dengue (Bhatt et al.) and the map for Zika, with a discussion of the drivers for the difference and whether these seem to be justified (e.g. have at least some face validity).

We understand the reviewers’ concerns about the overlap in covariates used for modelling dengue and Zika. We have now addressed this in greater detail in the paragraph in the Discussion. While the covariates are indeed similar and would be well suited for a new iteration of the dengue map, we do not feel it is appropriate to make direct comparisons with the covariates from Bhatt et al. (2013) due to the fact that many refinements have been made to the covariates since that study. Regarding a difference map, while he Bhatt et al. (2013) dengue map is fundamentally different in that it used an evidence consensus layer to constrain predictions to areas with greater certainty of dengue transmission, the current ZIKV map shows environmental suitability for transmission, even in places where we know the disease has yet to be reported. Still, we felt it was useful to provide a difference map as the reviewers recommended, so in order to do so, we subtracted the values of the ZIKV map from those of the DENV map only in places where each had a non-zero prediction. This map is now found in the supporting information (Figure 2 —figure supplement 4), and reference is made to it and the patterns shown in the paper.

https://doi.org/10.7554/eLife.15272.017

Article and author information

Author details

  1. Jane P Messina

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    JPM, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    jane.messina@zoo.ox.ac.uk
    Competing interests
    No competing interests declared.
    ORCID icon 0000-0001-7829-1272
  2. Moritz UG Kraemer

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    MUGK, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
    ORCID icon 0000-0001-8838-7147
  3. Oliver J Brady

    Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, United Kingdom
    Contribution
    OJB, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  4. David M Pigott

    1. Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, United Kingdom
    2. Institute for Health Metrics and Evaluation, University of Washington, Seattle, United States
    Contribution
    DMP, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  5. Freya M Shearer

    Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, United Kingdom
    Contribution
    FMS, Conception and design, Acquisition of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  6. Daniel J Weiss

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    DJW, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  7. Nick Golding

    Department of BioSciences, University of Melbourne, Parkville, United Kingdom
    Contribution
    NG, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  8. Corrine W Ruktanonchai

    WorldPop project, Department of Geography and Environment, University of Southampton, Southampton, United Kingdom
    Contribution
    CWR, Acquisition of data, Analysis and interpretation of data
    Competing interests
    No competing interests declared.
  9. Peter W Gething

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    PWG, Conception and design, Acquisition of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  10. Emily Cohn

    Boston Children's Hospital, Harvard Medical School, Boston, United Kingdom
    Contribution
    EC, Acquisition of data, Analysis and interpretation of data
    Competing interests
    No competing interests declared.
  11. John S Brownstein

    Boston Children's Hospital, Harvard Medical School, Boston, United Kingdom
    Contribution
    JSB, Acquisition of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  12. Kamran Khan

    1. Department of Medicine, Division of Infectious Diseases, University of Toronto, Toronto, Canada
    2. Li Ka Shing Knowledge Institute, St Michael's Hospital, Toronto, Canada
    Contribution
    KK, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  13. Andrew J Tatem

    1. WorldPop project, Department of Geography and Environment, University of Southampton, Southampton, United Kingdom
    2. Flowminder Foundation, Stockholm, Sweden
    Contribution
    AJT, Acquisition of data, Analysis and interpretation of data
    Competing interests
    No competing interests declared.
  14. Thomas Jaenisch

    1. Section Clinical Tropical Medicine, Department for Infectious Diseases, Heidelberg University Hospital, Heidelberg, Germany
    2. German Centre for Infection Research (DZIF), Heidelberg partner site, Heidelberg, Germany
    Contribution
    TJ, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  15. Christopher JL Murray

    Institute for Health Metrics and Evaluation, University of Washington, Seattle, United States
    Contribution
    CJLM, Acquisition of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  16. Fatima Marinho

    Secretariat of Health Surveillance, Ministry of Health Brazil, Brasilia, Brazil
    Contribution
    FM, Acquisition of data, Contributed unpublished essential data or reagents
    Competing interests
    No competing interests declared.
  17. Thomas W Scott

    Department of Entomology and Nematology, University of California Davis, Davis, United States
    Contribution
    TWS, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    No competing interests declared.
  18. Simon I Hay

    1. Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, United Kingdom
    2. Institute for Health Metrics and Evaluation, University of Washington, Seattle, United States
    Contribution
    SIH, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    sihay@uw.edu
    Competing interests
    SIH: Reviewing editor, eLife
    ORCID icon 0000-0002-0611-7272

Funding

European Commission (21893)

  • Jane P Messina
  • Moritz UG Kraemer
  • Thomas Jaenisch

Bill and Melinda Gates Foundation (OPP1053338; OPP1119467; OPP1106023; OPP1093011; OPP1081737; OPP1068048)

  • Oliver J Brady
  • David M Pigott
  • Daniel J Weiss
  • Thomas W Scott
  • Simon I Hay

Wellcome Trust (095066)

  • Simon I Hay

University of Southampton (Economic and Social Research Council's Doctoral Training Centre)

  • Corrine W Ruktanonchai

National Institutes of Health (P01AI098670)

  • Thomas W Scott

University of Melbourne (McKenzie fellowship)

  • Nick Golding

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the Secretariat of Health Surveillance, Ministry of Health of Brazil for providing access to the geographical coordinates of occurrence. JPM, MUGK and TJ receive, and OJB and SIH acknowledge funding from the International research Consortium on Dengue Risk Assessment Management and Surveillance (IDAMS; European Commission 7th Framework Programme (21893)). OJB and SIH are supported by the Bill & Melinda Gates Foundation (OPP1053338). SIH is also funded by a Senior Research Fellowship from the Wellcome Trust (095066), and grants from the Bill & Melinda Gates Foundation (OPP1119467, OPP1106023 and OPP1093011). DMP is also funded by the Bill & Melinda Gates Foundation (OPP1093011). DJW and PWG receive support from the Bill and Melinda Gates Foundation (OPP1068048, OPP1106023). NG is supported by a University of Melbourne McKenzie fellowship. CWR is funded through the University of Southampton’s Economic and Social Research Council’s Doctoral Training Centre. TWS is supported by grants from the National Institutes of Health (P01AI098670) and the Bill and Melinda Gates Foundation (OPP1081737). EC and JSB are supported by the National Library of Medicine of the National Institutes of Health (R01LM010812).

Reviewing Editor

  1. Mark Jit, Reviewing Editor, London School of Hygiene & Tropical Medicine, and Public Health England, United Kingdom

Publication history

  1. Received: February 15, 2016
  2. Accepted: April 10, 2016
  3. Accepted Manuscript published: April 19, 2016 (version 1)
  4. Version of Record published: June 1, 2016 (version 2)

Copyright

© 2016, Messina et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 29,658
    Page views
  • 7,735
    Downloads
  • 39
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Epidemiology and Global Health
    2. Microbiology and Infectious Disease
    Curated by Prabhat Jha et al.
    Collection Updated

    eLife has published papers on many tropical diseases, including malaria, Ebola, leishmaniases, Dengue and African sleeping sickness.

    1. Epidemiology and Global Health
    Ursula Dalrymple et al.
    Research Article