Avian vampire fly (Philornis downsi) mortality differs across Darwin’s finch host species

In invasive parasites, generalism is considered advantageous during the initial phase of introduction. Thereafter, fitness costs to parasites, such as host-specific mortality, can drive parasites towards specialism to avoid costly hosts. It is important to determine changes in host specificity of invasive populations to understand host-parasite dynamics and their effects on vulnerable host populations. We examined changes in mortality in the introduced avian vampire fly (Philornis downsi) (Diptera: Muscidae), a generalist myasis-causing ectoparasite, between 2004 and 2020 on Floreana Island (Galápagos). Mortality was measured as the proportion of immature larvae found upon host nest termination. Over the time period, the avian vampire fly was most abundant and had low mortality in nests of the critically endangered medium tree finch (Camarhynchus pauper) and had the highest mortality in nests of hybrid tree finches (Camarhynchus spp.). Low larval mortality was also found in small tree (Camarhynchus parvulus) and small ground finch (Geospiza fuliginosa) nests. Selection could favour avian vampire flies that select medium tree finch nests and/or avoid hybrid nests. Overall, the finding of differences in avian vampire fly survival across host species is parsimonious with the idea that the introduced fly may be evolving towards host specialisation.

Niche breadth is a fundamental concept that underpins key hypotheses in species ecology [1][2][3] . The breadth of a niche is the set of conditions in which a species can persist, and can include dimensions such as habitat diversity and climatic variation 4 . In parasitic organisms, niche breadth is often synonymous with host specificity-i.e., the number of host species a parasite can infect, and parasites range from highly host specific to generalist [5][6][7] . Host specificity is mediated by host-parasite co-evolutionary processes 8 . Hosts and parasites enter an arms race in which they adapt and counter-adapt reciprocally at the expense of the other 9 . Hosts are selected to evade or resist the parasite, whereas parasites evolve to more efficiently exploit their host 6 . Higher virulence (damage to the host) and host specificity can lead to increased exploitation of the host by the parasite 10 . Selection for greater host exploitation may break down when parasite fitness is reduced, whereby high exploitation of hosts leads to premature host mortality, leading to decreased parasite growth and fecundity, or increased parasite mortality 11,12 .
Host specificity presents trade-offs for the parasite. Generalist parasites tend to occur on host species that are phylogenetically closely related 7,[13][14][15] . Nonetheless, they incur the cost of maintaining variation in life history, genetic and behavioural traits that enable exploitation of different host species 16 . This relationship can be further complicated in host hybrid zones, where hybrids can be more or less resilient to parasite populations 17 . Despite high host encounter rates due to wide host ranges, generalist parasite populations exhibit slower geographic expansion rates compared to specialist populations 16 . The occurrence of parasite generalism or specialism is influenced by the costs and benefits inherent to occupying different host ranges, including mortality rates in parasite populations associated with particular host species 18 . For blood feeding parasites, for example, the costs of generalist feeding can push species towards specialisation because of the variation in host blood properties and nutritional value for the parasite 19 . High mortality risk or a lack of nutritional value in specific hosts can drive parasites to specialise on hosts that optimise their fitness 20 . Generalist parasites may have a selective advantage when colonising novel environments given their capacity to switch hosts if a primary host population declines, which can increase their chance of persistence despite a range of establishment challenges 21,22 .
When a generalist parasite colonises a novel environment and suite of potential host species, the differences in fitness due to altered selection creates a window of opportunity to study niche and host specialisation shifts www.nature.com/scientificreports/

Methods
Study system. This study was conducted on Floreana Island, Galápagos Archipelago, and followed longterm field protocols as described below 39 . The field work was conducted during the Darwin's finch breeding season in the highlands (01° 17′ S, 090° 27′ W) between the months of January and April in ten non-consecutive seasons spanning 17 years: 2004, 2005, 2006, 2008, 2010, 2012, 2013, 2014, 2016 and 2020. We collected avian vampire fly specimens from the nests of the small ground finch, small tree finch, medium tree finch, and the hybrid Camarhynchus tree finch [53][54][55] . Host species were first determined morphologically, and hybrid tree finches were retrospectively confirmed via genetic analyses 53,54  Study species. The avian vampire fly is an obligate myiasis-causing parasite of birds that feeds on the blood and tissue of developing nestlings 44 . Non-parasitic adult flies feed upon decaying vegetable matter, ovipositing their eggs in active bird nests [57][58][59] . Upon hatching, first and early second instar larvae move to the naris and ear canals of nestlings to feed on blood and keratin 35 . Late second and third instar larvae feed externally on nestlings at night, residing in the base of the nest during the day 35,60,61 . Reports of development times of larval instars vary between field and lab reared specimens, with pupation occurring after 4-10 days of feeding 39,62 . Upon host fledging or death, third instar larvae pupate in the base of the nest, forming frothy cocoons and emerging as adults within 7-18 days 39,62 .
Nest monitoring and vampire fly collection. We analysed data from 280 Darwin's finch nests on Floreana Island with all avian vampire fly specimens per nest collected and stored in ethanol following well-established field protocols 39 . Small tree finch (n = 64), hybrid tree finch (n = 34), medium tree finch (n = 55) and small ground finch (n = 127) nests were monitored for activity and brood size every 3 days during incubation and every 2 days during the nestling phase. Males of these host species build new display nests at which they sing for each new nesting event 63 . The female either selects a display nest or selects a male and they build a new display nest together 63 . Incubation lasts ~ 14 days and, if successful, nestlings fledge the nest approximately 12-14 days after hatching. Brood size was determined using a borescope to view inside the nest once nestlings had hatched. After nesting activity had finished (i.e., nest termination, either through death of the nestling or fledging), the nest was collected and dismantled within 24 h to count the number of avian vampire fly offspring within the nest. Nestling age at death was known for a subset of nests (n = 105) from hatching date or visual aging of the nestlings via borescope. In all sampling years (except for four nests in 2004 and 2005), nestlings found dead in the nest were soaked in 70% ethanol for 24 h to allow first instar larvae within the nestling nares or ear canals to float and be collected. We generally only collect ~ 8 1st and 2nd instar using this method from ~ 8% of nestlings soaked. The 1st instar larvae reside inside the nares for the first two days post-hatch and nestlings tended to survive until d7 post-hatch during 2004 and 2005 39 . All avian vampire fly larvae, pupae, puparia and adult flies were stored in 70% ethanol within 24 h of collection. Larval specimens of 241 nests were assigned an age class via observation using a dissecting microscope, following instar identification protocols 35,49 . Parasite intensity was calculated as the total number of larvae, pupae, puparia and adult flies within a nest. Mortality in the avian vampire fly larvae was measured as the proportion of immature (first and second instar) larvae in the nest at the time of host resource termination 64 . This measure accounts for the possibility of third instars and pupae fully developing into adult flies following host termination 65 . This measure also provides an estimate of parasite mortality per host nest, given that first and second larval instars are unable to continue development in the absence of nutrition 65,66 . Statistical analysis. All models were fitted using R version 4.0.3 67 with the packages lme4 68 , MASS 69 , and car 70 , and were visualised with lattice 71 , ggplot2 72 and effects 73 . Total number of avian vampire fly offspring per nest (log transformed to fulfil the assumption of normality) was analysed in relation to study year, annual rainfall and the Darwin's finch species and the interaction between year and rainfall as fixed effects with a linear regression model on the full data set (n = 280).
For the corresponding analyses considering different age classes of the parasite, we had a smaller dataset of n = 241. We repeated the analysis of total intensity in relation to year and rainfall on this subset of data to confirm the same pattern across both data sets. We analysed the total number of first and second instar larvae, third instar larvae, total larvae, pupae and puparia in relation to year and annual rainfall similar to the total number of vampire flies, but as count data with Generalized Linear Models (GLMs) with negative binomial distribution and log link function to correct for overdispersion. Throughout we tested for linear and quadratic relationships of year and rainfall and their additive and interactive effects on the total avian vampire fly intensity, first and second instar larvae, third instar larvae, total larvae, pupae and puparia. Based on the principles of parsimony (the largest amount of variance explained with the minimum number of predictors 74 ), we then selected the model structure that best described our measures of parasite loads at different developmental stages.
Avian vampire fly larval mortality was modelled using the column bind ('cbind') function specifically designed to fit proportion data in logistic regression models with the number of larvae in the first and second larval instar as binomial denominator, and a quasibinomial distribution and a logit link function to correct for overdispersion. We fitted the key response variable avian vampire fly larval mortality to two different data sets: (1) considering all Darwin's finch nests on Floreana Island for which we identified larvae to age class (n = 241); and (2) considering nests where nestling age at death was known (n = 106), with nestling age at death as an additional co-variate. and highlights interspecific host differences in vampire fly larval mortality even when accounting for host age at death 39 . We fitted the study year, annual rainfall and the Darwin's finch species and the interactions (year × rainfall) as fixed effects. Initially, we also controlled for brood size, but this additional predictor did not reveal any significant result and was dropped from the final model to improve sample size (from n = 191 to n = 241 in the data set without brood size). We removed non-significant interaction terms from the models to simplify the statistical approach and interpretation of the results and to ensure a valid interpretation of the remaining additive effects. The effect of host genus (Geospiza and Camarhynchus), excluding hybrid tree finches to remove the effect of hybridisation (n = 213), was analysed using the full model of mortality in relation to year and annual rainfall and their interaction as fixed effects. All quantitative variables were scaled (standardized to mean = 0 and standard deviation = 1) to bring the variables to comparable dimensions and to facilitate the correct interpretation of effect sizes for interaction terms 75 . Residual distributions of the models were inspected visually to assess model fit (diagnostic plots produced by the 'plot' function in the 'base' package: residuals versus fitted values and normal Q-Q plot displaying the theoretical quantiles versus standard Pearson residuals). Throughout, we report model effect sizes (estimates ± SE, derived from the summary function); presented χ2 and p-values are based on an ANOVA Table of Deviance using Type III Wald χ2 tests (ANOVA function in 'car' package). No random factors were considered, as there were no repeated measurements in the data. We tested for correlations of fixed effects beforehand but did not find any indication for co-linearity in our data.

Permits. Permission to conduct this study was given by the Galápagos National Park and Charles Darwin
Research Center, permit no. MAE-DNB-CM-2016-0043, and Flinders University, permit no. E480/19.
Age class distribution and abundance. Analysis of avian vampire fly age classes revealed a significant quadratic relationship with year and the number of first and second instar larvae (GLM, 'year' term estimate: − 0.51 ± 0.14, p = 0.002, Table 2b, Figure S1), with numbers of first and second instar larvae peaking in approximately 2013. The number of first and second instar larvae was consistently higher in hybrid tree finches (Table S1). Furthermore, across all species, third instar larvae and the total number of larvae per nest increased until 2013 and decreased thereafter (GLM, third instar: 'year' term estimate: − 0.32 ± 0.10, p = 0.010; total larvae: 'year' term estimate: − 0.38 ± 0.10, p = 0.001, Table 2c,d, Figure S1). Conversely, the number of pupae and puparia per nest showed the opposite pattern, decreasing until 2013 and 2014 onwards (GLM, 'year' term estimate: 0.46 ± 0.10, p < 0.001; 'year' term estimate: 0.78 ± 0.24, p = 0.002, respectively, Table 2e,f, Figure S1). There was no significant effect of rainfall on the total intensity in these nests (LM, 'rainfall' term estimate: − 0.04 ± 0.03, p = 0.110, Table 2a, Figure S1) or on any other age class considered (i.e., rainfall did not feature in any other parsimonious model, neither in the linear nor quadratic relationship).
Larval mortality. For all samples combined, there was an increase in the proportion of avian vampire fly larval mortality over time ('year' term estimate: 0.48 ± 0.19, p = 0.012, Table 3a) and during years with higher annual rainfall ('rainfall' term estimate: 0.40 ± 0.13, p = 0.002, Table 3a). However, these additive effects should be interpreted with caution due to their involvement in a significant interaction term (estimate 0.58 ± 0.18, p = 0.002; Fig. 2a). Earlier in the study period (2004)(2005)(2006)(2007)(2008), when the years were drier (e.g., ~ 360 mm), vampire fly mortality was lower; while later in the study period (2010-2020), when the years were wetter (e.g., 400-650 mm), avian Table 1. Linear model for avian vampire fly (Philornis downsi) intensity in relation to year and host species collected between 2004 and 2020 from Darwin's finch nests on Floreana. 'Rain' did not feature into the most parsimonious model. Avian vampire flies collected from Darwin's finch nests over 10 years across a 17-year period on Floreana Island. a Species 'small tree finch' was used as a reference category. Note the response variable Philornis downsi intensity was log transformed to achieve normality and all quantitative input variables were scaled and centred. Intercept presented in italics. Sign = significance levels: '***' < 0.001. www.nature.com/scientificreports/  www.nature.com/scientificreports/     Table 3. www.nature.com/scientificreports/ vampire fly mortality was higher (Fig. 1a, Fig. 3). Larval mortality did not differ between the small tree finch, medium tree finch or small ground finch, but was significantly higher in hybrid nests (least square means and post-hoc contrasts Table 3b,c, Fig. 2b). Larval mortality did not differ between Camarhynchus and Geospiza host nests when excluding hybrid tree finches (estimate 'genus' term − 0.15 ± 0.24, p = 0.532; Table 4). When analysing nests where the nestling age at death was known, we found a strong effect of nestling age at death on larval mortality. Larval mortality increased as nestling age at death decreased (estimate − 0.34 ± 0.15, p = 0.025, Table 5, Figure S2c). The interaction effect of year and rainfall on mortality was marginally nonsignificant (estimate 0.38 ± 0.20, p = 0.056, Table 5, Figure S2a). Avian vampire fly mortality was significantly higher in hybrid hosts (estimate 1.36 ± 0.41, t = 3.30, p < 0.001, Table 5, Figure S2b). Larval mortality did not differ between the small ground, small tree, or medium tree finch ( Figure S2b).

Discussion
In this study, we tested patterns of larval mortality in avian vampire fly, a generalist myiasis-causing parasite of Darwin's finches, across time and host species. We did not find a significant increase in parasite mortality across time, but there were clear differences in parasite mortality across host species. Parasite mortality was lowest in nests of the medium tree finch, and highest in hybrid finch nests, even when accounting for chick age at death 39 . If host-specific selection pressures on larval mortality continue or increase, the avian vampire fly may be selected to oviposit in optimal host nests, which may result in host specialisation.
Our results provide some support for the idea that Camarhynchus hybridisation may be an adaptive host response to thwart a novel parasite, in line with previous findings 42 . The Red Queen hypothesis is a powerful  www.nature.com/scientificreports/ theoretical framework to predict host-parasite coevolutionary dynamics, and one expects that host-impacting change caused by the parasite is countered by the host, and vice versa 9 . The newly evolving Darwin's finch and avian vampire fly system is consistent with the idea of oscillating evolutionary dynamics in the wild but requires additional research into genetic and behavioural mechanisms to more fully understand these patterns. Previous research has shown that: (1) during the first part of the decade from 2004 to 2013 39 , the average number of avian vampire flies per host nest increased and then stabilised; (2) one host species, the medium tree finch, consistently has the most avian vampire flies in the nest compared with other host species 40 ; (3) the proportion of hybrid birds increased from 12% in 1998 to between 27 and 55% in later years, and hybrid hosts have the fewest avian vampire flies compared with other host species 52 ; and here we show that (4) avian vampire fly mortality was highest in hybrid finch nests and lowest in the nests of the other host species (small ground, small tree and medium tree finches), even when accounting for nestling age at death. From the perspective of the parasite, it should avoid hybrid finch nests. The mechanisms that may drive host-seeking versus host-avoidance behaviours by the parasite are unknown. However, this study uncovers two concurrent scenarios whereby both parasite intensity and parasite mortality across hosts differed, especially between medium tree finch and hybrid tree finch nests during the early co-evolutionary stages of a host-parasite interaction.
In parasites that use multiple host species for different life stages, host generalism is the optimal strategy 76 . The avian vampire fly lives its parasitic life stages in a single host environment, and in this case, specialist offspring are predicted to be optimal to maximise arithmetic mean fitness 76 . The observation that different Darwin's finch host species have different average numbers of avian vampire flies per nest, even immediately after host hatching, is in line with the idea of differentiated oviposition in certain hosts 42,46 . Despite specialisation, some specialist lineages hedge their bets by ovipositing in suboptimal hosts 76,77 . In the case of the avian vampire fly, genetic evidence has shown oviposition by multiple females in one host nest; also, females frequently lay fewer eggs than they are able to oviposit at a time, which the supports the idea of bet hedging by ovipositing in multiple nests 78 . However, it is currently unknown if females oviposit preferentially in specific host nests or whether there could be host-specific lineages of avian vampire fly.
Given that high rainfall is associated with more avian vampire flies in host nests 40,45,51 , we would expect to see an increase in competition between larvae during high rainfall years as more parasites compete for the same amount of resources 79,80 . Increased competition may lead to increased parasite mortality. However, in this study, we found lowest parasite mortality in nests of the host species with the most parasites, the critically endangered medium tree finch. The extreme fluctuations in rainfall within parasite lifetimes and across generations on the Galápagos Islands may favour environmental generalists that maintain optimal fitness levels with rainfall fluctuation 76 . Selection pressures from introduced pathogens can lead to swamping of local environmental adaptation in favour of immune response loci 81 . Therefore, there may be a trade-off between achieving optimal parasite fitness across multiple host species and the parasite's capacity to tolerate environmental variation. In the avian vampire fly, such relationships are yet to be explored.
Host specialisation may ease the burden of parasitism in some host species yet may heighten the threat for other neighbouring species, particularly in host-limited, geographically restricted habitats, as occurring on Floreana Island 82 . Smaller, endangered populations, such as the medium tree finch, are more likely to have low genetic diversity with a reduced capacity to evolve in response to parasites 83 . The threat posed by the parasite is further exacerbated by the high intensity of avian vampire fly larvae found in medium tree finch nests. In comparison, the hybrid tree finch that is the result of recombination between the small and the medium tree finch may have increased genetic variation, which may offer novel genes on which selection can act to evolve resistance to parasitism 84,85 . Given the observation that female medium tree finch frequently pair with male small or hybrid tree finches rather than medium tree finch, and the potential for increased hybrid resilience 42,53 , the medium tree finch population may continue to decline, eventually resulting in only a hybrid swarm 17 . Hybrid recruitment, as measured by the proportion of yearling birds in the population, has remained stable across years since 2005, Table 5. Generalized linear model for avian vampire fly (Philornis downsi) in-nest mortality in relation to year and rainfall (interaction term) and species, including the co-variate 'nestling age at death, ranging from 1 to 14 days). Including this co-variate reduces our data set to n = 106. Avian vampire flies collected from Darwin's finch nests in 10 sampling years over a 17-year period (2000-2020) on Floreana Island. Significant estimates indicated in bold. a Species 'small tree finch' was used as a reference category. × indicates an interaction term. Note all quantitative input variables were scaled and centred. Dispersion Parameter for quasibinomial family taken to be 12.273. Intercept presented in italics. Sign = significance levels: '***' < 0.001; '*' < 0.05. www.nature.com/scientificreports/ whereas medium tree finch recruitment rates declined across the same period, suggesting hybrid nestlings and/ or fledglings may have a selective advantage over medium tree finch offspring 52 . Understanding host-specific parasite fitness in this system highlights the need for directed conservation efforts to more exploited hosts or those less likely to evolve parasite resistance mechanisms. Our results suggest that such mechanisms may be evolving in the Camarhynchus hybrid group, but at a cost to the medium tree finch population. The effects of host hybridisation on both host and parasite fitness have mainly been documented in plant-parasite systems, as hybridisation is common in plant species 86 . These effects vary between systems. Host hybridisation can, for example, increase susceptibility to parasites, resulting in increased numbers of parasites and decreased hybrid fitness 17,87,88 . In other cases, host hybridisation increases host resistance and tolerance, decreasing parasite loads and increasing host fitness 17,89 . We see this latter pattern in the Darwin's finch system, where hybrid tree finches tend to have fewer parasites per nest than their parent species 42 . In this study, using a sub-sample of nests for which we have accurate data on parasite age class, we also found a pattern of fewer parasites per nest in hybrid nests, though the difference in number of parasites across host species was not statistically significant. There is not much available data on parasite fitness in hybrid versus non-hybrid hosts, which is a research gap that requires attention. In a study on fungal pathogens infecting hybrid plant hosts, pathogens had a fitness advantage in hybrid hosts that was contingent on pathogen hybridisation 90 . In addition, the role of host hybrid fitness is expected to affect parasite fitness 91 . If host resistance and tolerance to parasites increases as the consequence of hybridisation, then parasite fitness could be higher in hybrid hosts able to sustain the parasite, or conversely, parasite fitness could be lower in hybrid hosts that deter parasites from ovipositing. More research is needed to explore different host-parasite evolutionary pathways under conditions of genetic introgression in host and/or parasite.
We don't know why avian vampire fly larval mortality differed across hosts species in this study, but it is known that blood properties of host species can vary in nutritional gain for the parasite 19,28 . Mortality in second and third instar avian vampire fly larvae reared on chicken blood did not differ between formulated diets of varying nutrition, however development time to pupation was fastest on the diet with the highest nutritional value 62 . Decreased developmental time is advantageous when resources can be terminated quickly, such as when Darwin's finch nestlings die young 39 , allowing more larvae to reach pupation faster and hence survive to adulthood in nutritionally optimal hosts. High mortality was found in first instar larvae reared on artificial diets and the possible contamination of the blood with pathogenic bacteria such as Serratia may be driving this high mortality 62 . Serratia, a genus with pathogenic species that affects myiasis-causing and muscid flies, was found to be uniquely associated with avian vampire flies parasitising warbler finches in a microbiome analysis of the fly 92 . Warbler finches in recent years had fewer avian vampire flies and lower host mortality compared to tree finches 46 . Research has further shown that the avian vampire fly microbiome differs significantly across Darwin's finch host species, which is suspected to be associated with differences in finch diets within and across habitats 55,[92][93][94][95] . Overall, the findings of this and previous research suggest that larval mortality may be driven by multiple factors, including host nutritional quality, habitat, and microbiome.
We found high parasite mortality in hybrid avian hosts, which we document in a generalist and recently introduced parasite to the Galapagos archipelago. The parasite did best in nests of the Floreana Island endemic, the medium tree finch. Theory predicts that the vampire fly should be selected to oviposit preferentially in medium tree finch nests, given that it has the highest pupation success in medium tree finch nests, and avoid hybrid finch nests where most of its offspring fail to pupate. Understanding the mechanisms by which the avian vampire fly avoids or selects host nests, invests in generalist or specialist offspring, or alters its strategy to survive in prevailing environmental conditions are at the forefront of research into this rapidly evolving host-parasite interaction system on the Galápagos Islands. Our study provides evidence for differential fitness of an invasive parasite in nests of different host species.