Pathogen spillover driven by rapid changes in bat ecology

1Zoonotic spillover is the transmission of a pathogen from a non-human vertebrate to a human. Spillovers of viruses from bats have resulted in the emergence and spread of viruses in the human population. For example, SARS-CoV-2, SARS-CoV-1, Nipah and Hendra viruses have caused human mortalities, sometimes after transmission through an intermediate host. Spillover of viruses from wildlife to humans has been correlated with land-use change through studies that associate land use, occurrence of spillover and presence of reservoir hosts, but without data that reveal the mechanisms. In this long-term study, we observed rapid changes in bat behaviour that coincided with the emergence of Hendra virus. We found that bats were responding to environmental change by persistently behaving in ways that were previously observed as temporary responses to climate-driven food shortages. We propose that these behavioural shifts increased spillover risk by increasing contact of bats with domestic horses, the intermediate hosts from which Hendra virus spills over into humans, and by increasing viral shedding from bat populations that have established outside their normal winter range. We developed and applied a Bayesian hierarchical network model to our 25 years of data on reservoir host ecology, behaviour and spillover events (Data Index in ref. ). We identified distal and proximal drivers of links between habitat loss, climate and spillover, and predicted the risk of Hendra virus spillover in space and time. The Hendra virus system illustrates a suite of ecological connections that contribute to land-use-induced spillover of this pathogen: interactions between land-use change and climate altered the behaviour of wildlife reservoir hosts, increasing their proximity to domestic or human recipient hosts. We propose that this phenomenon, coupled with stressors that drive increased pathogen excretion, leads to pathogen spillover.

2Australian flying foxes (Pteropus spp., fruit bats) are reservoir hosts of Hendra virus, a henipavirus in the family Paramyxoviridae that does not cause discernible disease in bats, but has a case fatality rate of 75% in horses (84 fatalities documented) and 57% in humans (four fatalities documented). Among subtropical Pteropodids, Pteropus alecto (black flying fox) is the species most likely to excrete Hendra virus. Infected bats feeding in horse paddocks shed the virus in excreta or spats that horses contact when grazing, leading to infection,. Humans are exposed through infected horses. Hendra virus has probably circulated in bats far longer than Europeans have occupied Australia, yet Hendra spillover was not identified until 1994. After approximately 2006, the frequency of Hendra virus spillovers increased. Forty one of the 63 spillovers documented to January 2021 occurred in subtropical eastern Australia. The majority of spillovers occurred during the Austral winter and were clustered in space and time (Fig. 1 and Supplementary Information Section 1).

3Our research focused on spillover to horses in the Australian subtropics, where Pteropus spp. primarily feed on nectar. Historically, the bats were nomadic, moving among roost sites to track ephemeral pulses of flowering by Eucalyptus spp. over hundreds of kilometres–; continuous occupancy of roosts was uncommon. During summer, many tree species that are abundant and widely distributed provide food for bats. During winter, few tree species provide food for bats, and the naturally limited distribution and abundance of these trees has been further restricted by clearing for urban development and agriculture,. Pulses of mass flowering in remnant forests leads to a high proportion of the bat population becoming concentrated in small areas, typically in coastal lowlands. Loss of these winter food sources can have severe impacts on the bat population,. When trees that provide food in winter or spring do not flower, as occurs every 1 to 4 years owing to variation in temperature and rainfall,, bats experience brief (typically 3 to 12 week) food shortages. They respond to the shortages by roosting in small groups (population fission) close to reliable but often suboptimal food in urban gardens and agricultural areas (for example, fruit from ornamental, commercial or weed species),,–. In the past, these behavioural responses persisted only for the duration of acute food stress, and bats returned to nomadism and nectivory when the food shortage abated,,.

4To investigate the relationships between land-use change, changes in bat behaviour and spillover, we collected empirical data from 1996 to 2020 within an area bounded by the locations of spillovers in the subtropics (Fig. 1a, Extended Data Fig. 1a and Supplementary Information Section 1). These data include the location and timing of Hendra virus spillovers (Dataset A in ref. ), locations and occupancy of the roosts (Datasets B,C),, characteristics of the roosts and foraging areas around the roosts (Datasets B,C,I)–, climate (Dataset D in the Data Index), nectar shortages (Dataset E), measures of bat fitness (Dataset F and G),, flowering pulses during winter (Dataset J) and loss of winter foraging habitat (Datasets K,L and M in the Data Index and Supplementary Sections 1–13).

5From 1996 until approximately 2002, roosting and foraging behaviours were stable, and no Hendra virus spillovers were detected (Figs. 1 and 2a,b). From approximately 2003 until 2020, bat behaviour and the incidence of spillovers changed rapidly: the number of roosts tripled, and 40 spillovers were detected (Supplementary Information Sections 1 and 3 and Figs. 1 and 2). From 2006 to 2020, spillovers were detected in 80% of years, and 75% of spillovers occurred in annual clusters of three or more (Fig. 1). Neither a change in the definition of a Hendra virus case in horses in 2008, nor the availability of a horse vaccine in late 2012, explain the observed pattern of spillover events (Supplementary Information Section 1).

6We used nectar productivity data from commercial apiarists and measures of flying fox fitness to identify nine winter and spring food shortages during the study period (Supplementary Information Section 7 and Fig. 2a,b). During food shortages, the absence of nectar was associated with a sharp increase in the number of bats admitted into wildlife rehabilitation centres (at least 30 animals in a month) and a low percentage of lactating females with pre-weaning young (<79%; Supplementary Information Section 7 (ref. ) and Extended Data Figs. 2 and 3). Food shortages followed all strong El Niño events (Oceanic Niño Index ≥ 0.8), and also occurred independently of El Niño events (Fig. 2a).

7Early in the study period, the number of roosts was stable. Although food shortages occurred, the bat population fission events were brief (weeks), bats quickly returned to nomadism when the shortage abated and the fission events were not captured in the roost counts. Following this, the number of roosts increased in steps that coincided with food shortages (2003, 2007 and 2010) and then continued to increase steadily (Fig. 2b). As the number of roosts increased, bats roosted in smaller groups that were closer together and fed within smaller areas, and new roosts formed in locations that increased access to anthropogenic foods in agricultural and urban areas (Fig. 3 and Extended Data Figs. 4 and 5a–c). We infer that fissioning of bat populations into smaller groups near anthropogenic food reduces the energetic costs of foraging and allows the bats to mitigate effects of nutritional and energetic stress. Whereas in the past, the behavioural response of bats to food shortages (population fission) persisted only for the duration of acute food stress,,, we observed the behavioural response becoming persistent (Extended Data Fig. 4). To assess the mechanisms associated with the rapid change in bat behaviour, we examined loss of winter foraging habitat across far Southeast Queensland (Supplementary Information 12 and Extended Data Fig. 1b). Before European settlement, winter habitat was extensive in this region (Extended Data Fig. 5d). More than 70% of the forest that provided winter habitat was cleared before 1996, and clearing continued at a constant rate (Extended Data Fig. 5e,g,h). In four of the six years from 1996 to approximately 2002, during which bat behaviour was consistent and no spillover occurred, we recorded aggregations of at least 100,000 nomadic bats associated with mass pulses of flowering in winter habitat in this region (Extended Data Fig. 5f). By 2018, nearly a third of the habitat that was present in 1996 had been cleared (Extended Data Fig. 5e); the number of roost sites had increased fivefold (Extended Data Fig. 4), with 87% of new roosts forming in urban areas. By 2020, winter aggregations of at least 100,000 in far Southeast Queensland had been recorded once in 14 years (Extended Data Fig. 5f). Although bats relocated to both urban and agricultural areas, most spillovers (86%) occurred in agricultural areas (Fig. 3b), presumably because horses were present at higher densities in agricultural areas than in urban areas.

8We developed the parsimonious hypothesis that the combination of acute winter/spring food shortages followed by a lack of winter flowering in the following year yields the highest risk of spillover, especially from roosts in agricultural areas. We tested this mechanistic hypothesis with a multiscale Bayesian network model. We included six variables: strong El Niño events (Oceanic Niño Index ≥ 0.8), food shortages, winter-flowering pulses, land cover within the foraging areas of winter colonies of P. alecto, population fissioning (establishment of new roosts) and number of Hendra virus spillovers (Supplementary Information Section 13 and Supplementary Fig. 1). We validated the predictive ability of the network model with leave-one-out cross-validation and expected log pointwise predictive density. We verified this by using a leave-one-out cross-validation for model selection with more than 1,000 simulated datasets (Supplementary Information Section 13). The best model supported that spillover clusters occur following a sequence of events over three successive years. First, it indicated that strong El Niño events preceded food shortages, and food shortages coincided with population fissioning (Fig. 4a, Extended Data Figs. 6–8 and Supplementary Tables 7–9). Second, the greatest likelihood of spillover from a given roost was associated with P. alecto feeding in agricultural areas during a winter that followed an acute food shortage, but in which there was no pulse of flowering (0.014, 0.126 highest posterior density probability interval; Fig. 4b). By contrast, the presence of a pulse of winter flowering that attracted at least 100,000 bats mitigated spillover risk (0.000, 0.026 probability interval; Fig. 2c and Fig. 4b, Extended Data Fig. 9 and Supplementary Table 5).

9We used the model to make probabilistic predictions of whether a cluster of spillovers (defined as three or more spillovers) would occur in the study area from 1996 to 2021. We used all available information up to a given time point, starting with 2013 (Fig. 4c and Extended Data Fig. 10a,b). For example, to predict a cluster of spillovers in 2020 (Fig. 4d and Extended Data Fig. 10c), we used information collected to the end of 2019. In all cases, model predictions were consistent with observations of the presence or absence of clusters of spillovers in the context of winter flowering. We also verified the predictive ability of our model with fivefold cross-validation; our model accurately predicted the presence or absence of a cluster of spillovers in each of the 25 years in the data (Extended Data Fig. 8).

10These data document increasing rates of transmission of a fatal zoonotic virus in a rapidly changing anthropogenic system. Our work suggests that previously transient fissioning of bat populations in response to periods of food shortage has become persistent. This behavioural response leads to the increased use of agricultural and urban areas by bats in winter (Fig. 3). Food shortages occurred throughout the 25-year study period (Fig. 2a,b), probably driven by the lagged effects of climate on Eucalyptus flowering. However, spillovers were not associated with food shortages until bats persistently overwintered in agricultural areas in proximity to horses, and winter-flowering pulses became less frequent (Figs. 2c,3 and 4a). We propose that the loss of winter-flowering habitat and consequent decline in the abundance of winter nectar contributes to the persistence of bats in agricultural and urban areas (Supplementary Information Section 12, Fig. 2c and Extended Data Fig. 5d–h). The consequences of more bats in areas with human settlements include not only increased risk of viral spillover from bats to horses to humans, but also increased conflict with humans.

11The contemporary association between food shortages, population fission and spillover may be explained by both increased proximity of bats and horses, and physiological stress in novel winter habitats. Bats in roosts that recently formed outside of the historic winter range of the species excreted more Hendra virus in winter, especially after food shortages, than bats in roosts within the historic winter range of the species. Bats in urban and agricultural areas rely on suboptimal food, which may lead to nutritional stress that facilitates viral shedding–. Nutritional stress has been linked to infection and shedding of Hendra virus,, coronaviruses and other viruses in reservoir host systems; moreover, Hendra virus spillover events in the subtropics have been associated with small roosts with limited access to native food. The timing of Hendra virus spillover clusters in winter, months after the food shortages in the previous year, may be owing to the cumulative effects of nutritional stress overlaying high energy requirements in winter (thermoregulation and pregnancy) and scarce resources within suboptimal habitats. Although the time-lag between food shortages and spillover requires further investigation, we suggest that the processes driving spillover of Hendra virus are general phenomena that link land-use change and pathogen emergence: behavioural responses to loss of habitat increase contact with recipient hosts, and physiological and immunological responses to food limitation increase pathogen excretion.

12Our data suggest that increasingly rare winter-flowering pulses reduce the risk of spillover. Bats reverted to nomadism and left agricultural and urban areas during pulses of winter flowering in remnant native forest, and spillovers did not occur during these flowering pulses (Fig. 2c). We propose that these pulses of flowering may mitigate zoonotic risk by drawing large numbers of bats (Supplementary Information Section 11 (ref. )) away from feeding in agricultural areas and, therefore, decreasing contact between bats and horses. This nomadism may also reduce competition for food among the bats that remain in agricultural areas. Understanding these mechanisms requires further work. Nevertheless, the loss of native forest that supports large aggregations of nomadic bats appears to be fundamental to the cascade of events that lead to spillover. An extensive programme of ecological protection and restoration of winter-flowering forests (ecological countermeasures) could be a sustainable, long-term strategy to reduce spillover and protect the health of livestock and humans.

13The consistent temporal association of Oceanic Niño Index (ONI) thresholds with food shortages and spillover events allows for the prediction of spillover clusters up to two years in advance through surveillance of climate, or alternatively, one year in advance through surveillance of bat reproduction and bat admissions to wildlife rehabilitation centres (Fig. 4c). However, we cannot as yet predict winter-flowering pulses, and these must be monitored in real time. For example, the conditions leading into winter 2020 suggested that multiple spillover events would occur (Fig. 4d). However, we were unable to predict the pulse of winter flowering that attracted more than 200,000 bats that appears to have prevented winter spillovers. With retrospective information on flowering, our Bayesian network model correctly predicted the absence of a cluster of spillover events in 2020.

14Prevention of spillovers requires characterization of the interactions of reservoir and recipient species with their environment in the context of rapid land-use change and climate variability. Monitoring of ecological processes that occur at different temporal extents, including those longer than natural climatic cycles such as El Niño, is essential to understanding the drivers of spillover. Yet long-term data on reservoir hosts, especially bats, are sparse. Many bat species that are the reservoir hosts of zoonotic pathogens depend on ephemeral resources, and occupy ecosystems in which loss of native vegetation, high livestock density and human populations coincide–. Moreover, some bat species can adapt to human-modified landscapes to mitigate the effects of loss of their native habitat. These species may present greater risks of spillover, as has been observed in other wildlife taxa. We identified key processes connecting land-use change to spillover through behavioural responses of bats to altered food availability. We suggest that behavioural and physiological responses to rapid, human-induced environmental change increased contact between reservoir hosts and recipient hosts, and increased shedding of pathogen within proximity of recipient hosts. This study, therefore, suggests a general framework for examining causes and potential ways to mitigate bat virus spillover in regions without long-term data on reservoir hosts.

15Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

16Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-022-05506-2.