ENSO-driven climate variability promotes periodic major outbreaks of dengue in Venezuela

Dengue is a mosquito-borne viral disease of global impact. In Venezuela, dengue has emerged as one of the most important public health problems of urban areas with frequent epidemics since 2001. The long-term pattern of this disease has involved not only a general upward trend in cases but also a dramatic increase in the size and frequency of epidemic outbreaks. By assuming that climate variability has a relevant influence on these changes in time, we quantified the periodicity of dengue incidence in time-series of data from two northern regions of Venezuela. Disease cycles of 1 and 3–4 years (p < 0.05) were detected. We determined that dengue cycles corresponded with local climate and the El Niño Southern Oscillation (ENSO) variation at both seasonal and inter-annual scales (every 2–3 years). Dengue incidence peaks were more prevalent during the warmer and dryer years of El Niño confirming that ENSO is a regional climatic driver of such long-term periodicity through local changes in temperature and rainfall. Our findings support the evidence of the effect of climate on dengue dynamics and advocate the incorporation of climate information in the surveillance and prediction of this arboviral disease in Venezuela.

disease periodicity at a finer temporal scale. To calculate monthly and yearly dengue incidence per 100,000 inhabitants, available and projected yearly population data from each state under study was used 22 .
The El Niño Southern Oscillation (ENSO) is the strongest inter-annual climate cycle on Earth 23 . It is an atmosphere-ocean coupled system that produces quasi-periodic short-term climate and sea surface temperature (SST) changes over the Pacific region with an impact on weather patterns in several countries of the Americas, Africa, and Asia 24 . SSTs are used as an index of ENSO. This climate system oscillation occurs approximately every 3-7 years 23,24 , fluctuating between two extremes known as El Niño corresponding to the warm phase (positive SST anomaly), and La Niña, the cooling phase (negative SST anomaly). El Niño, in fact, refers to the unusual warming of SST. ENSO can be characterized using the Niño 3.4 index, the SST anomaly in the Niño 3.4 region of the equatorial Pacific. El Niño event in turn can be classified according to the SST anomaly index into three categories: weak (0.5 to 0.9 SST anomaly), moderate (1.0 to 1.4 SST anomaly) or strong (>1.5 SST anomaly) 25 . The onset of El Niño events occurs during spring in the Northern Hemisphere, encompassing two calendar years 26 . The events are generally characterized by positive anomalies of SSTs that increase during the Northern Hemisphere's spring, summer and fall of the first year (Year 0), with the maximum SST anomalies occurring during the winter months (December-January-February) of the following year (Year + 1) and SST anomalies decreasing during the spring and summer of the year + 1. El Niño is the main cause of the inter-annual variability of local climate in the central and eastern tropical Pacific Ocean including the northern coast of South America 26 and Venezuela 21 . In Venezuela, El Niño is related with negative anomalies of precipitation, soil moisture and river flows, along with positive (warmer) air temperature anomalies 27 while La Niña has the opposite effect, with cooler temperatures and positive anomalies of precipitation. Here, we used the monthly SST of the eastern and central tropical Pacific as an index of ENSO-region 3.4 (Niño 3.4 index). The SST time-series were obtained from the Climate Prediction Center of the National Oceanic and Atmospheric Administration 28 .
Time series analyses of dengue incidence and climatic variables. Because disease and climatic time-series, as well as their associations, can be strongly non-stationary (varying in time), a specialized time series analysis method known as wavelet analyses (WA) was applied to detect the periodic cycles and dominant components (i.e. the most frequently repeated signal) of the time series and how they change over time 29,30 . In addition, wavelet coherency (WC) methodology was used to compare the frequency components of dengue and climate time-series in order to quantify the statistical (linear) association between variables in a time span 29 . WC provides local information on when two non-stationary signals are linearly correlated and at what particular frequency.
Additionally, the linear relationship of monthly dengue incidence with monthly observations of climatic variables were also explored through cross-correlation functions (CCF). Finally, we calculated the standardized anomalies of dengue incidence, precipitation and temperatures of our study period by subtracting from each seasonal observation (e.g., DecJanFeb, MarApMay, JunJulAug, SepOctNov) the long-term (16years) mean value of each particular season and dividing this by the long-term seasonal standard deviation.
Data were normalized previous to the analyses. Time series analyses were performed using R (R Foundation for Statistical Computing. Vienna, Austria. Version 3.3.2; 2016), while WA and WC analyses were performed using Matlab 2017a (The MathWorks, Inc., Natick, Massachusetts, United States) and the toolboxes developed by Cazelles et al. 15

Results
Temporal patterns of dengue incidence. Overall, dengue incidence in Venezuela has exhibited a steady average increase of approximately 9.5% annually between 1991 and 2016 with an average of 39.5 cases x 100,000 inhabitants in the early 1990's to a 10-fold higher mean incidence of 368 cases × 100,000 inhabitants in the last 6 years (2010-2016). An average national incidence of 157 cases × 100,000 inhabitants (range 13-438 cases × 100,000 inhabitants) was observed during these 26 years ( Fig. 2a and Supplementary Figure S1a). Moreover, from 2007 onwards, a total of six epidemic years (years 2007, 2009-2010, 2013, 2014, 2015) were recorded nationally with an intensification in the frequency and magnitude of these outbreaks in comparison with only 4 epidemic years in the previous 16 years (years 1995, 1997-1998, 2001). Overall, most of these epidemic years coincided with an El Niño event (Fig. 2a). Time series of monthly dengue cases (2001-2016) from Aragua and Carabobo regions mirrored the temporal pattern of the whole country. Six major outbreaks encompassing eight epidemic years (2001, 2007, 2009-2010, 2012-2013, 2014 and 2015) were observed in Aragua (Fig. 2b), whereas similar events were registered in Carabobo (Fig. 2c).
The average monthly dengue incidence and precipitation in Aragua and Carabobo regions during the period of 2001-2016 are shown in Fig. 3. Dengue shows a clear seasonal pattern coinciding with the rainy season and peaking between August and November. Although Aragua and Carabobo regions are next to each other, the reported average dengue incidence of the past 26 years is 2.25 times as high in Aragua (199 cases × 100,000 inhabitants) as in Carabobo (88 cases × 100,000 inhabitants).   The wavelet power spectrum (WPS) of local rainfall and temperature exhibited strong significant power at a 1-year scale (seasonal) as expected. Moreover, inter-annual cycles at 3-6-years and 6-years were also detected for minimum and maximum temperatures, respectively (Fig. 5). Interestingly, a significant upward trend over time of the minimum and maximum temperatures was observed in both of the regions under study (Supplementary Figure S1b-e). The periodicity of ENSO is depicted across monthly SST records in Fig. 6. This time-series exhibited inter-annual variability at 2-3 and 5-years cycles during the time period of our study.

Relationship between dengue and climatic variables.
We evaluated the correspondence of the WPS of dengue and SSTs through WC analysis. As Fig. 7 shows, there was a significant but non-continuous coherence between both variables at 1-year and 2-to 3-year scales across the two studied regions. In Aragua state, dengue incidence mainly cohered with SST from 2007 to 2012 at 2-to 3-year cycles (Fig. 7a). In this region, the 1-year coupling between both signals was stronger than in Carabobo region where instead, the link between dengue and ENSO showed a longer (2005-2013) and significant result at the 2-to 3-year scale (Fig. 7b). There were significant but transient coherences between dengue incidence and rainfall (Supplementary Figure S2a (Fig. 8b). Positive anomalies of maximum temperature corresponded with anomalies of ENSO. We also found that positive anomalies of dengue cases (epidemic years) coincided with maximum temperature anomalies. The previous is noteworthy, since some epidemics that did not coincide with ENSO, concurred with positive anomalies of maximum temperatures. This correspondence between dengue and maximum temperatures occurred in almost all epidemic years. In Aragua region (Fig. 8a)  Cross Correlation Functions (CCF) were calculated to identify relevant associations between dengue incidence and climatic variables at a finer temporal scale (monthly). There was no difference in time lags between both Aragua and Carabobo regions for all CCF. Significant correlations between dengue and SSTs were found at similar lags of 4-5 months for both regions, with highest correlation values at lag 4 (r = 0.14) and lag 5 (r = 0.25) for Aragua and Carabobo, respectively. Likewise, dengue had a significant and positive association with rainfall taking place two months earlier than dengue in Aragua and Carabobo (r = 0.23 and r = 0.32, respectively). Finally, high frequencies of dengue incidence were positively correlated with high values of maximum temperatures occurring six months previously for Aragua (r = 0.15, p < 0.05) and 6-7 months previously for Carabobo (r = 0.36; p < 0.05), whereas similar associations were found between dengue incidence and the minimum temperature at lags of 3 (range 3-5) and 5 (range 3-6) months (Aragua: r = 0.26 and Carabobo: r = 0.29; p < 0.05).

Discussion
We identified long-term (inter-annual) and short-term (seasonal) cycles of dengue incidence in two highly-endemic regions of Venezuela. A strong and significant association was found between dengue inter-annual cycles and ENSO, suggesting that El Niño events have been partly responsible for the periodic outbreaks of dengue in northern Venezuela. We propose that the mechanism of ENSO in driving dengue inter-annual cycles is via warmer local temperatures and lower precipitation. In the studied period, these results were evident in relevant epidemic years, including epidemics of great magnitude, such as the one of 2009-2010, the largest dengue epidemic ever recorded in Venezuela.
Our analysis showed that dengue epidemics at a cyclic frequency of 3-4 years were characteristic of the dynamic of the disease besides its expected 1-year seasonal cycle. Previous studies have documented the occurrence of inter-annual fluctuations in dengue incidence. However, the periodicities reported were different, ranging from 2-3 years 8,15,31,32 to 2-5 years 9 .
Inter-annual cycles of dengue have been related to several environmental (extrinsic) and immunological (intrinsic) determinants 15,17,18,33 which affect both vector and virus 7,24 . Here, however, we focused on characterizing the degree of influence of the ENSO regional event and local meteorological conditions on the temporal disease dynamics. We found a strong and significant coherence between the temporal pattern of dengue and that of these variables showing the role that climate plays in driving disease periodicity. Specifically, we identified a relationship between dengue incidence and SSTs at 2-to 3-year cycles across the two studied regions indicating the significant influence of El Niño in shaping the cycle of dengue epidemics in northern Venezuela. A similar association was also detected between dengue and local climate variables such as minimum and maximum temperatures, with a less clear link observed between dengue and precipitation. At a regional scale, ENSO has been the most commonly studied driver of cyclic climate phenomena in human diseases 34 . In tropical South-America, this event is a periodic climatic oscillation with an average occurrence of 3-7 years, as our analyses showed, and with a strong influence on the inter-annual variability of local climate across different geographical areas 25,34,35 . In Venezuela, temperatures tend to increase by an average of 0.5 °C during El Niño, and severe droughts and negative anomalies of soil moisture, river discharges and rainfall are registered during this climatic episode, the warm phase of the ENSO event. Consequently, warmer and dryer years occur compared with no-Niño years 35 . Inversely, La Niña event shows the opposite effect 25,26 . In agreement with similar observations made in Colombia 36 , our results showed that four out of seven dengue epidemics at national level and in the two studied regions coincided with El Niño events, including a long and extreme dengue outbreak from 2009-2010 which coincided with Niño 0 and Niño +1 years.
Although several studies have tried to disentangle the effects of ENSO on dengue and other vector-borne diseases, this topic is still an area of active investigation together with the impact of climate change on infectious diseases 15,33,[36][37][38][39][40] . A plausible mechanism to explain the influence of ENSO on dengue is related to the fluctuations of local climate conditions that are exhibited during an El Niño event 15,33,36 . By analysing this pathway, we found that temperatures and rainfall periodicities corresponded to the interval of observed periodicity in ENSO (2-6 year periods). Likewise dengue inter-annual cycles corresponded roughly to those of temperature (especially maximum temperature) and rainfall within each region. Likewise, dengue epidemics coincided with positive anomalies of maximum temperatures, and negative anomalies of precipitation. Additionally, positive anomalies of maximum temperatures overlapped with an El Niño event, except for those anomalies that were registered between 2011 and early 2014. Our data is in agreement with that of Gagnon et al. 36 who reported that most of the dengue epidemic peaks happening in Colombia between 1981-1988 were associated with El Niño episodes and warmer temperatures. Interestingly, the local temperature records analysed here showed a significant upward trend. This phenomenon could be linked to an urban heat island effect (UHI), which generally occurs in urban settlements 41 and could affect the transmission of mosquito-borne diseases as it has been previously reported 42 . Further studies are needed to explore and unravel the relevance of this UHI effect on the increment of dengue in Venezuela during the last years.
How the El Niño event and related temperature and rainfall patterns affect disease transmission is a matter of conjecture. It is known that the relationship between local climate variables (e.g. precipitation and temperature) and dengue transmission can be complex, affecting both the dengue vector and the virus 11 . Cazelles et al. 15 suggested that the relationship between climate and dengue might be clearer with temperature than with precipitation. Higher temperatures affect the growth rate of Aedes mosquitoes, accelerate the rate of viral replication within the vector and decrease the length of the reproductive cycle, resulting in more infected female mosquitoes over a shorter period of time 11 with the risk of major inter-annual dengue outbreaks. Particularly, it has been documented that higher temperatures (i.e. >32 °C) decrease the length of Ae. aegypti oviposition cycles and egg laying; as a result, the female gonotrophic cycle is shorter whereas blood-feeding frequency is greater [43][44][45] . Other studies have shown that increased temperatures have the effect of diminishing the time of the extrinsic incubation period (EIP) of DENV in Ae. aegypti [46][47][48] . As an example, Watts et al. 46 showed in experimental infections of Ae. aegypti with DENV-2 that the EIP shortened from 12 days to 7 days when the temperature increased from 30 °C to 32-34 °C.
The complexity of the relationship between dengue dynamics and a local climatic factor such as rainfall is exemplified in our study by the positive correlations of dengue with rains in the seasonal period and the negative correlations in the 2-3-y periodic cycle. Differently, precipitation has a crucial influence on mosquito development, because it provides a suitable habitat for the stages of the mosquito life cycle that are water-dependent 11,40,49 . Here, we found a significant coherence between dengue and precipitation at 1-year periods confirming that precipitation is an important driver of dengue occurrence at a seasonal scale as it has been extensively reported 31,50,51 . Indeed, the appearance/increase of dengue cases commonly coincides with both an intensification of rain as observed in our study area and with vector abundance as reported elsewhere 31,[52][53][54] . In Venezuela, the occurrence of household-related potential breeding sites (used car tires, litter outdoors) are a risk factor for dengue transmission during the rainy season. Moreover, the lack of reliable piped water supply has prompted an ever more common behaviour of water storage both indoors and outdoors 52,55,56 especially during dry conditions 57,58 . As a consequence, areas with public service deficiencies or during dry years (i.e. derived from El Niño events or during warming trends) are likely to exhibit a more perennial dengue transmission.
Among the various approaches developed to study non-stationary data, WA is one of the most efficient tools to detect periodicity in epidemiological time series 29 . This technique is more powerful when applied to longer time-series than the one used in our study. Although our findings should be interpreted with some caution in the light of the previous statement, we are confident of the robustness of our overall results. As previously stated, other factors need to be taken into account when explaining inter-annual cycles of dengue, such as population immunity and the (re)introduction of new DENV serotypes 17 . Epidemic outbreaks of dengue fever were first recorded in Venezuela in 1964 and were partly attributed to the (re-)introduction of previously non-circulating DENV serotypes/strains coinciding with increased spread and densities of Aedes aegypti 59,60 . However, from 2000 onwards, the 4 DENV serotypes co-circulate in Venezuela 61 and although the epidemic peak registered in 2001 can be related to the introduction of serotype 3 62 , this peak also coincided with La Niña event. Moreover, after 2000, at least 3 major epidemics occurred with the co-circulation of all DENV serotypes in the same period. Future research should aim to investigate how climate forcing interacts with the effects of cross-immunity and cross-enhancement between serotypes to determine the population dynamics of this arboviral disease in Venezuela. Finally, the epidemics of chikungunya (2014) and Zika (2015-2016) viruses were important epidemiological events that affected a large number of people in Venezuela. During 2014, concomitant chikungunya and dengue epidemics were recorded which may have resulted in a certain degree of misdiagnosis between the two diseases during the first two weeks of the chikungunya epidemic when health personnel were not yet acquainted with this new disease. However, the remarkable clinical picture of chikungunya was quickly recognized and properly diagnosed in the great majority of cases. Regarding the possible epidemic effect of the introduction of Zika virus in 2015, the data showed a steady increase of dengue cases since August 2015, while the first confirmed autochthonous case of Zika was reported by the end of November. Zika epidemic reached its peak in January-February 2016 lingering on during the rest of this year. Therefore, we discard any potential effect of Zika virus epidemic on the dengue time series analysed.

Conclusions
We present evidence suggesting that ENSO and its related local climatic changes (above-normal temperatures and below-normal rainfall) have been important drivers of the biennial and triennial cycles of dengue in the northern part of Venezuela during the last 16-years. The significant upward trend observed in dengue incidence in Venezuela raises the question of its potential relationship with the concomitant increase in local temperatures in the main urban settings. Thus, future research should investigate how regional and local climate interact with viral transmission and with behavioural and other socio-economic factors to determine the population dynamics of dengue in Venezuela. Finally, our findings provide significant evidence of the relevant effect of climate on dengue dynamics and suggest that the local and regional climatic factors here studied should be included in an early-warning system for dengue and other Ae. aegypti-borne viral surveillance and control in Venezuela.