Virus clearance from hosts

Experimental infections of Hendra, Nipah, and Marburg viruses in bats have suggested that, after a short incubation period, bats have an acute systemic phase of viral replication (Table 2, S1 Table). RNA from Hendra virus was detected in urine and feces for an average of seven days [25]; Hendra virus was isolated from bat tissues 10 days, but not 21 days, after infection [29,30]. Likewise, in three Marburg virus experiments, bats experienced a short period of viremia followed by clearance from most sampled tissues by 10 to 14 days after inoculation [31–33]. There were no signs of morbidity or mortality induced by Hendra, Nipah, Ebola, or Marburg viruses, consistent with previous virus detections in apparently healthy wild bats (Table 2) [6,33–35]. However, all henipavirus experiments and one Marburg virus experiment were conducted with individual wild-caught bats without knowledge of the prior infection and immune histories.

The sequelae to the acute period of viral infection and shedding in bats are less well characterized. Although virus was not isolated from bats at the conclusion of any experiment, RNA was frequently detected in tissues at necropsy (S1 Table). For instance, Hendra virus RNA was detected in the lung, liver, spleen, and kidney of bats at the conclusion of a three-week experiment [25]. Marburg virus RNA was detected in the spleen at the conclusion of one four-week experiment [31] but was cleared from all sampled tissues at the conclusion of another three-week experiment [32]. To understand viral clearance, two important questions must be resolved. First, is the absence of PCR detection sufficient to demonstrate clearance? Measles virus RNA, for example, can be detected in humans for months after active infection has cleared [36]. Second, what experimental duration is adequate for assessing viral persistence?

Virus persistence within hosts

Although there is no direct evidence that the henipaviruses or filoviruses persist within individual bats, it is commonly assumed that bats host persistent infections [37]. This assumption may be derived from evidence of persistent flavivirus and Rio Bravo virus infections during experiments conducted in the 1960s and 1970s [38,39], and a common paradigm that reservoir hosts carry persistent infections (e.g., hantavirus in rodents and simian immunodeficiency virus [SIV] in primates [40,41]). More recently, simultaneous shedding of several viruses (and sometimes bacteria) from many individuals was observed during pulses of excretion [e.g., 42–44]; this observation could be explained by within-host persistence. Some features of the immune systems of bats appear to be different from those of other mammals, and this has fueled speculation that bats may host persistent infections [16]. For example, the set point of interferon responses in bats appears to be relatively high compared to other mammals [45]. The latter is consistent with the possibility that bats rapidly control viral replication, thus avoiding the pathological consequences of disease observed in other species [45]. Moreover, it has been proposed that bats can coexist with pathogens due to adaptations inadvertently acquired during the evolution of flight [46–48]. Despite this growing body of research on bat–pathogen interactions, there is no direct evidence that henipaviruses or filoviruses persistently infect their bat hosts.

If bats do not clear infections, there are at least two mechanisms by which viruses may persist within bat hosts [18]. First, bat hosts may tolerate viruses. In this case, a tempered inflammatory response would minimize immunopathology; viral replication may fluctuate as a function of the immunological competence of the host [18]. Second, persistent latent or low-level infections from which virus can be reactivated may become established in bats [13]. In theory, persistent bat viruses could reactivate in response to processes that affect host immunity, such as stress, pregnancy, or poor nutrition [5,14]. Whether a particular individual clears infection or remains persistently or latently infected may depend on immune status, dose received, coinfections, route of infection, history of infection, and individual variation—all factors that may differ amongst individuals and through time [18].

Latent and recrudescent Hendra and Nipah encephalitis cases were observed in humans, and, therefore, recrudescence was proposed as a mechanism of persistence in bats. However, recrudescent virus in humans was not transmitted and, therefore, did not contribute to epidemiological dynamics [e.g., 49,50–52]. Nevertheless, the different cellular interactions between viruses and reservoir hosts versus incidental hosts may lead to different mechanisms of persistence. For example, the ability of bats to control viral replication may limit the capacity of viruses such as Hendra and Nipah to spread to the central nervous system of bats. In general, characterization of host pathogen dynamics in a spillover host does not provide evidence for the dynamics being similar in reservoir hosts [5,27]. Recrudescent Nipah virus infection was reported in a bat, but the evidence presented was weak and alternative explanations were not ruled out [13].

Protective immunity

Experimental evidence for protective immunity following viral infection in bats is inconsistent and suggests differences among species and viruses (Table 2). In the 1960s Sulkin et al. [53] described recurrent viremias in bats and susceptibility to reinfection with Japanese B encephalitis virus, establishing a paradigm for bats as hosts of recurrent infections. By contrast, repeat challenge studies with Rabies virus in one bat species confronted our assumptions about the universal lethality of rabies virus in mammals. Although some individual bats experienced acute, lethal infections, other individuals survived with protective immunity and never shed virus [54]. Contemporary work on emerging bat viruses has been less conclusive. Bats challenged with Marburg virus seroconverted and were protected from viral replication when rechallenged 48 days later, suggesting some short-term protection after recovery from acute viremia [32]. In contrast, 50% of bats inoculated with Hendra or Nipah virus seroconverted, and there was no consistent relation between seroconversion and recovery of viral genome [25]. Moreover, in another study, bats excreted Nipah virus in urine while neutralizing antibody was present in the serum [26]. Together, these studies indicated that antibodies may not be the primary driver of henipavirus clearance in bats [55]. Reinfection experiments with Nipah virus were inconclusive; most bats did not respond to the first inoculation with productive infections [25]. Waning prevalence of Hendra virus antibody was reported in a maternity colony of Pteropus scapulatus over six months [9], and waning maternal henipavirus antibodies have been reported in numerous systems [e.g., 9,56,57], but the effective protection conferred by these antibodies was not determined. The association between antibodies and Ebola viral nucleic acids in samples from bats was unclear [58].

Transmission

As is the case with most wildlife diseases, the process by which bat viruses are transmitted in the wild is poorly understood. Transmission events cannot be observed directly, viral infections of reservoir hosts may be asymptomatic, and sampling of individuals through time is difficult. Therefore, inferences must be drawn from multiple sources of indirect evidence. Bat viruses have been detected in urine, feces, saliva, and soiled fruit, indicating the potential for virus transmission via excreta between bats in the wild [5,59]. However, under laboratory conditions, bat-to-bat transmission for most of these viruses is difficult to achieve (Table 2) [e.g., 32]. Following experimental inoculations, viral shedding by bats for both Nipah and Hendra viruses was minimal [25], although continuous exposure to a low-dose viral rain may increase the probability of infection [5]. Natural and experimental vertical transmission of Hendra virus has been documented, but inconsistently and under specific conditions, such as parenteral inoculation (but not oronasal inoculation) [25,29]. A few longitudinal serological and virological surveys and theoretical models have proposed that seasonal patterns in incidence or spillover could be explained by peaks of transmission in response to demographic or ecological drivers such as birth pulses, waning maternal immunity, and migration [12,15,60,61], but these proposed dynamics have not been confirmed.

Field sampling of seroprevalence

Quantification of seroprevalence often has been the first line of investigation for newly discovered bat zoonoses [e.g., 8,9,11]. Serological assays have some limitations [62], but antibodies are generally easier to detect than antigen; sera can be nonlethally sampled, and, because antibodies generally persist after virus is cleared, the probability of detecting seropositive animals is higher than that of detecting infected animals. However, seroprevalence is a poor metric for quantifying epidemic dynamics. If antibodies persist within bats, seroprevalence measures cumulative exposure. If antibody persistence exceeds the period over which hosts move between populations, then antibody prevalence may not reflect the epidemic history of the sampled population, but simply the infection history of the individual [63]. Thus, longitudinal monitoring of seroprevalence may be most useful for studying the dynamics of viruses that produce short-lived antibody responses in relatively sedentary host species (unlike Pteropid bats).

Measuring age-specific serostatus and antibody titers increases the power of serological surveys. If age-specific seroprevalence increases monotonically as age increases in populations sampled over multiple outbreaks, SIR dynamics can be inferred (Table 1). In contrast, SIRS and SILI dynamics could exhibit complex patterns in cross-sectional serosurveys. These patterns would change through time as epidemics wax and wane. In general, inference from serology alone is unlikely to differentiate among our proposed epidemiological scenarios.

Field sampling of infection prevalence

The difficulty of estimating prevalence of emerging bat viruses may reflect true low prevalence, low test sensitivity, or sampling of tissues in which virus does not persist. Pathogens that cause acute infections may circulate at low prevalence or may be heterogeneously distributed in space and time and therefore require large sample sizes to detect [64].

Pathogens that persist within their hosts may be sequestered in tissues that are difficult to sample non-lethally. For example, Ebola virus RNA has only been detected in the liver and spleen of wild bats [58]. Viremia or shedding in excreta may be periodic, and, therefore, estimation of infection status from these samples may yield false negatives.

Prevalence data collected across space and time may allow identification of geographic patterns, such as travelling waves, that are likely to be caused by host-to-host transmission (SIR or SIRS dynamics) (Table 1). Age-structured prevalence data with an age-specific incidence skewed towards younger individuals suggests an endemic disease with SIR dynamics. If the duration of maternal-derived immunity is known, the mean age of infection is a direct reflection of the basic reproductive ratio of the disease in the population. Although prevalence data provides many insights into the dynamics of disease, there are few situations in which prevalence data alone can distinguish among our proposed epidemiological scenarios.

Longitudinal sampling of individuals

Longitudinal sampling of the infection status of individuals can help distinguish between scenarios if four conditions are met: (1) Prevalence of the infection must be high enough to detect pathogen within a small number of sampled individuals but must not approach 100% so that variations can be observed. (2) The period over which individuals are resampled must exceed that of the infectious period (and therefore the infectious period must be known). (3) The host’s lifetime must far exceed the infectious period. (4) Pathogen must be detectable without lethal sampling. It is our contention that, if these conditions are met, it should be possible to distinguish among scenarios. In a population with waning immunity and cyclical reinfection, the probability that a given individual becomes infected is expected to be equal to the prevalence of infection among other individuals of similar age and the same sex. Therefore, past infection status is not expected to predict present infection status. In contrast, if individuals are persistently infected but shed episodically, the prevalence of infection among previously positive individuals will be higher than among the population, and past infection increases the likelihood of positive infection status (Table 1). However, large sample sizes and intermediate levels of prevalence probably are required to distinguish cyclical reinfection from persistent infections, and alternative explanations for patterns of prevalence may be difficult to reject. For example, the individuals most likely to be exposed may appear to be persistently infected. In this case, complementing longitudinal sampling with sequence data may improve inferences.

A caveat to this method is that longitudinal, systematic sampling of wild animals, particularly highly mobile or migratory species, requires major effort. For example, large populations of Eidolon helvum (the reservoir hosts of henipaviruses and Lagos bat virus) migrate across continents [22]. Even in resident populations, such as urban populations of Pteropus alecto (the reservoir host of Hendra virus), recapture of individuals within populations that include tens of thousands of individuals is difficult if not impossible. In contrast to studies in migratory canopy-dwelling species, capture-mark-recapture methods have been implemented in cave-dwelling species. For example, Frick et al. [65] reported recapture rates of 0.10 to 0.35 during longitudinal studies of Myotis lucifugusi, and Smith et al. [66] reported recapture rates of up to 0.81 during longitudinal studies of Myotis macropus.

If recapture is possible, generating a consistent time-series of recaptures may be challenging, affecting the ability to make inferences on the temporal resolution of the underlying processes. Nevertheless, some mechanisms may be ruled out even with incomplete time-series data. For example, SIR dynamics are unlikely in an animal in which virus is detected, not detected, and then detected again. However, genomics still is necessary to differentiate SIRS from SILI dynamics.

Viral sequencing

We hypothesize that pulses of infection arising from reactivation of persistent viruses or reintroduction of acute viruses may leave distinguishable evolutionary signatures in viral genomes at the population and individual levels (Fig 3). Viral reintroduction followed by extinction could result in pulses with low levels of viral diversity (Fig 3A). In contrast, pulses driven by reactivation may contain higher viral diversity given the longer evolutionary history of the virus in the population (Fig 3B). The opposite pattern is expected in comparisons of viral diversity across multiple pulses. Each pulse of reintroduced viruses is more likely to represent a different viral lineage, but reactivated viruses may be similar among pulses. Therefore, reintroduced pulses are expected to create low lineage diversity within an outbreak, but higher lineage diversity among outbreaks. The opposite could be expected for pulses driven by reactivation. For example, diverse lineages of Marburg virus circulated simultaneously within and among bat colonies in Uganda across years [67], a pattern that is more consistent with persistence within hosts and reactivation than pulses of reintroduction, which would require simultaneous extinction-recolonization of multiple strains. Information on viral diversity should ideally be coupled with knowledge of the landscape-level and host community-level distribution of viral lineages, enabling inference to the geographic or species origins of viral reintroductions to focal populations [68].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Different within- and between-host mechanisms are hypothesized to produce different evolutionary patterns of viral diversity at the level of populations and individuals.

Different shapes represent different viral strains within a population and different colors within a shape reflect variation within strains that have a recent common ancestor. (A) In the case of SIR/SIRS dynamics, acute infections are reintroduced and then cleared at the population level between each pulse. At each point in time, the pathogen within individuals in the population either has the same genotype (e.g., blue circles at pulse 1, red squares at pulse 2) or has closely related genotypes with a common ancestor (matching shape but different color). (B) In the case of SILI dynamics, individuals remain infected over time. Genetic diversity is determined in part by within-host viral evolution. Therefore, genotypes are likely to differ among individuals (many unique shape and color combinations, with some consistency over time). (C, D) Illustrative phylogenies of the virus populations across pulses. (C) In a scenario of viral extinction and reintroduction, all strains at a given pulse are closely related and have a recent common ancestor. (D) Divergent strains (different symbols) may be detectable within pulses, and distinctive strains are maintained across pulses. The hexagon and pentagon represent unsampled viral diversity present in the population.

https://doi.org/10.1371/journal.pntd.0004796.g003

In the absence of comprehensive, landscape-level data, other features of viral evolution may help distinguish reactivating from recolonizing viruses. Because viral evolutionary rates can be estimated for many species or can be approximated from better-studied related viruses [69], internal branches of the phylogenetic tree connecting pulses of infection in the same bat population may be too long to be plausible under a model of reactivation. Moreover, because neutral evolutionary rates are likely to be slow while viruses are latent, even less evolutionary distance would be expected between pulses driven by reactivation compared to viruses that circulate continuously over the same timespan [70,71]. Predictions of coalescent theory may also help identify recolonizing viruses because viral effective population size is expected to be driven by transmission between individuals [72]. Therefore, effective population sizes are expected to increase exponentially during pulses driven by reintroduction and between-host transmission but may appear stable in pulses driven by reactivation [73]. However, this effect may dissipate if reactivated infections of a few individuals drive a more extensive outbreak with between-individual transmission (i.e., pulses that are driven by multiple mechanisms). Also, as with any such analysis, reconstructions of effective population size depend on a number of epidemiological and evolutionary factors (e.g., selection) as well as sampling effort, so they must be considered with some caution [72].

Repeated sampling of individuals, although challenging and sometimes impossible, may also provide powerful insights. In pulses of reintroduced viruses, the individual viruses sampled at different points in time might be paraphyletic (i.e., recent samples are not necessarily derived from older samples in a phylogenetic tree), whereas reactivated viruses are expected to be monophyletic, with differences among pulses attributable to within-host evolution (Fig 3). Combining deep-sequencing methods that can characterize mixed-strain infections with repeat sampling could further indicate whether lineages are lost and reintroduced or maintained in individuals over time [74].

Viral sequence data may therefore help to distinguish the mechanisms driving pulses of viral infection in bats; however, several aspects of viral evolution present challenges. First, patterns of evolution within and between hosts are just beginning to be understood and different processes can generate similar patterns [73,75]. For example, strong within-host selection and viral reintroduction could create temporal structure in the virus phylogeny [76]. In this regard, sequencing both neutrally evolving and selected parts of the viral genome would be advantageous and is increasingly practical with advances in whole genome sequencing [74]. These methods will be most powerful in rapidly evolving RNA viruses that are temporally and spatially sampled, but this requires major investment in field sampling of highly mobile species like bats, particularly when the prevalence of detection of viral genomes can be low. Even when abundant molecular data are available, experiments and modeling should be designed to complement field sampling and sequencing of RNA viruses over space and time, particularly in more complex epidemiological scenarios in which our hypothesized expectations may not be realized or detectable.

Numerical exploration of epidemiological scenarios

In the face of incomplete data, modeling is regularly applied to explore the dynamics that would be expected given different scenarios for within- and between-host processes. For example, Blackwood et al. [77] compared the strength of evidence for four epidemiological models of rabies in vampire bats. They concluded that abortive immunizing infections and metapopulation dynamics contribute to population-level maintenance. Assuming an SIR scenario for Filoviruses in Rousettus aegyptiacus bats, Hayman [60] investigated the relative importance of parameters controlling bat life history and infection dynamics, highlighting the key role played by the biannual birth pulses that are characteristic of this species. Plowright et al. [15] used models to demonstrate that Hendra virus prevalence and seroprevalence studies were inadequate to distinguish between SIR or SILI dynamics in fruit bats. Although models cannot prove a certain epidemiological scenario exists, they can identify patterns that may help distinguish between scenarios, and therefore parameters, that are important to measure in the natural systems. It is important to reiterate that, to remain useful, models should be revised throughout the course of research programs to include the latest updates from empirical data and any new scenarios deemed relevant [78,79].