Journal list menu

Volume 6, Issue 6 art104 p. 1-21
Article
Open Access

Avian malaria in Hawaiian forest birds: infection and population impacts across species and elevations

Michael D. Samuel

Corresponding Author

Michael D. Samuel

U.S. Geological Survey, Wisconsin Cooperative Wildlife Research Unit, University of Wisconsin, Madison, Wisconsin 53706 USA

E-mail:[email protected]Search for more papers by this author
Bethany L. Woodworth

Bethany L. Woodworth

U.S. Geological Survey, Pacific Island Ecosystems Research Center, Hawai‘i National Park, Hawai‘i 96718 USA

Present address: Department of Environmental Studies, University of New England, Biddeford, Maine 04005 USA.

Search for more papers by this author
Carter T. Atkinson

Carter T. Atkinson

U.S. Geological Survey, Pacific Island Ecosystems Research Center, Hawai‘i National Park, Hawai‘i 96718 USA

Search for more papers by this author
Patrick J. Hart

Patrick J. Hart

University of Hawai‘i, Hilo, Hawai‘i 96720 USA

Search for more papers by this author
Dennis A. LaPointe

Dennis A. LaPointe

U.S. Geological Survey, Pacific Island Ecosystems Research Center, Hawai‘i National Park, Hawai‘i 96718 USA

Search for more papers by this author
First published: 30 June 2015
Citations: 49

Corresponding Editor: D. P. C. Peters.

Abstract

Wildlife diseases can present significant threats to ecological systems and biological diversity, as well as domestic animal and human health. However, determining the dynamics of wildlife diseases and understanding the impact on host populations is a significant challenge. In Hawai‘i, there is ample circumstantial evidence that introduced avian malaria (Plasmodium relictum) has played an important role in the decline and extinction of many native forest birds. However, few studies have attempted to estimate disease transmission and mortality, survival, and individual species impacts in this distinctive ecosystem. We combined multi-state capture-recapture (longitudinal) models with cumulative age-prevalence (cross-sectional) models to evaluate these patterns in Apapane, Hawai‘i Amakihi, and Iiwi in low-, mid-, and high-elevation forests on the island of Hawai‘i based on four longitudinal studies of 3–7 years in length. We found species-specific patterns of malaria prevalence, transmission, and mortality rates that varied among elevations, likely in response to ecological factors that drive mosquito abundance. Malaria infection was highest at low elevations, moderate at mid elevations, and limited in high-elevation forests. Infection rates were highest for Iiwi and Apapane, likely contributing to the absence of these species in low-elevation forests. Adult malaria fatality rates were highest for Iiwi, intermediate for Amakihi at mid and high elevations, and lower for Apapane; low-elevation Amakihi had the lowest malaria fatality, providing strong evidence of malaria tolerance in this low-elevation population. Our study indicates that hatch-year birds may have greater malaria infection and/or fatality rates than adults. Our study also found that mosquitoes prefer feeding on Amakihi rather than Apapane, but Apapane are likely a more important reservoir for malaria transmission to mosquitoes. Our approach, based on host abundance and infection rates, may be an effective alternative to mosquito blood meal analysis for determining vector-host contacts when mosquito densities are low and collection of blood-fed mosquitoes is impractical. Our study supports the hypothesis that avian malaria has been a primary factor influencing the elevational distribution and abundance of these three species, and likely limits other native Hawaiian species that are susceptible to malaria.

Introduction

Infectious diseases can have significant impacts on biological conservation and biodiversity (Daszak et al. 2000, Dobson and Foufopoulos 2001). The increasing emergence of wildlife diseases with potential threats to ecological systems, as well as domestic animal and human health, emphasize the importance of understanding disease dynamics and associated risks to biological conservation and human health. As invasive species, human development, and climate change alter host or vector communities and habitats, this information is increasingly essential for long-range conservation planning for species that face significant climate (Harvell et al. 2002) and environmental change, particularly for threatened or endangered species. However, determining wildlife disease dynamics including rates of infection, drivers of transmission, vector feeding preferences, and host mortality presents significant challenges (McCallum et al. 2001, Wobeser 2008). The importance of these epizootiological parameters for understanding host-pathogen dynamics, population effects, and on host-pathogen evolution have long been recognized, but are infrequently addressed (Scott 1988, Oli et al. 2006, Murray et al. 2009, Lachish et al. 2011a). Quantifying epizootiological parameters in wildlife populations is difficult because methods used in human epidemiology seldom apply (McCallum et al. 2001, Caley and Hone 2004, Lachish et al. 2011a); however, recent applications of multi-state mark-recapture models have provided an important tool to assess infection dynamics and population impacts (Faustino et al. 2004, Senar and Conroy 2004, Conn and Cooch 2009, Atkinson and Samuel 2010) while accounting for differential capture heterogeneity (Jennelle et al. 2007).

In addition to having direct wildlife conservation and management implications, diseases can profoundly affect ecological and evolutionary processes. For example, host and parasite diversity can influence disease prevalence, host and parasite abundance, and evolutionary outcomes (Holt and Pickering 1985, Altizer et al. 2003, Keesing et al. 2006). When pathogens impose a significant fitness cost on their hosts, spatial and temporal variation in the risk of infection can generate differential selection pressures that affect host adaptation, extinction risk, and genetic variation (Altizer et al. 2003, Poulin 2007, Wolinska and King 2009, Lachish et al. 2013). In turn ecological heterogeneity can alter disease dynamics and shape the interaction and coevolution between hosts and parasites (Ostfeld et al. 2005, Poulin 2007, Real and Biek 2007, Wolinska and King 2009, Lachish et al. 2013). Characteristically, vector-borne pathogens are generalists that infect multiple hosts; therefore, heterogeneity in host resistance can also alter species-specific risk of infection and community dynamics (Altizer et al. 2003, Keesing et al. 2006). Studies on avian blood parasites, especially Plasmodium species, have shown that host prevalence varies temporally, spatially, and among sympatric species due to climatic, host susceptibility, vector abundance, or vector habitat availability (Merilä et al. 1995, Sol et al. 2000, Wood et al. 2007, Loiseau et al. 2010, Sehgal et al. 2011, Lachish et al. 2013). An improved understanding of how these biotic and abiotic factors influence host-parasite interactions is of particular conservation relevance given rapid changes in the world's climate and increasing habitat fragmentation. Understanding the ecological drivers of vector borne diseases in wildlife also has strong implications for transmission and potential control of similar diseases to humans.

Avian malaria, a worldwide disease caused by the blood parasite Plasmodium relictum, was likely brought to Hawai‘i in the early 20th Century (Laird and van Riper 1981, van Riper et al. 1986, Atkinson and LaPointe 2009a). Malaria, together with the previously introduced avian pox virus (Avipoxvirus spp.), posed a major new threat to immunologically naïve Hawaiian birds (Warner 1968). Both avian diseases are readily transmitted by the southern house mosquito (Culex quinquefasciatus), which was introduced as early as 1826 (Halford 1954, Hardy 1960). A wave of native bird extinctions during the 1920s and 1930s has been attributed to avian malaria, and native birds below 1,500 m elevation continue to be at risk from malaria (Goff and van Riper 1981, van Riper et al. 1986). Above that elevation mosquitoes are rare, allowing native forest birds to survive. Later studies reported many endemic Hawaiian species, especially Hawaiian honeycreepers, are highly susceptible to avian malaria, effective disease transmitters, and chronically-infected, life-long reservoirs of disease (Atkinson et al. 1995, Atkinson et al. 2000, Atkinson et al. 2001a, b, Atkinson and LaPointe 2009b, Atkinson and Samuel 2010). In contrast, malaria has minimal impact on the survival of non-native birds, which also have a limited period of effective disease transmission. Because of their high susceptibility, prior workers hypothesized that malaria played a key role in historical population declines of native birds (Warner 1968, van Riper et al. 1986, Atkinson et al. 1995). However, data that quantify disease transmission or address the demographic consequences of malaria in native Hawaiian birds are limited. This information is critical for assessing disease risks, developing conservation strategies (Hobbelen et al. 2012), and predicting the future impact of climate on disease and birds (Benning et al. 2002, Atkinson and LaPointe 2009b).

The goal of our study was to investigate species and elevation patterns in malaria infection of native birds along a 1700 m altitudinal gradient on the Island of Hawai‘i. We focused on three species of Hawaiian honeycreepers that differ in their susceptibility to malaria and movement across the landscape. We used serology and change in disease status (susceptible to recovered) of captured and marked birds to identify susceptible and recovered (chronically infected and immune) birds. We combined multi-state capture-recapture (longitudinal) models with age-prevalence (cross-sectional) models to simultaneously estimate disease transmission and mortality, survival, and capture rates (Atkinson and Samuel 2010). This novel approach is advantageous because age-prevalence models can potentially estimate disease transmission from single capture (cross-sectional) data for known age animals, but multi-state models require recapture of marked individuals. Estimation of these epizootiological and demographic parameters was conducted using a Bayesian state-space model of capture and recapture data (Kery and Schaub 2012, King 2012). Unlike past studies, this approach also allowed us to estimate species-specific patterns of malaria infection, mortality, and population impacts across the altitudinal gradient in Hawai‘i.

Methods

Study species and area

We studied three of the most abundant honeycreepers remaining in Hawaiian forests. The Iiwi (Vestiaria coccinea) is highly susceptible to avian malaria (Atkinson et al. 1995) while Hawai‘i Amakihi (Hemignathus virens) and Apapane (Himatione sanguinea sanguinea) are moderately susceptible and chronically infected individuals are believed important reservoirs for this disease (Atkinson et al. 2000, Yorinks and Atkinson 2000, Atkinson et al. 2001a, Atkinson and Samuel 2010, Atkinson et al. 2013). Iiwi and Apapane are highly mobile, traveling across elevations in search of seasonal or ephemeral nectar resources; in contrast, the more generalist Amakihi is relatively sedentary throughout the year (Scott et al. 1986, Ralph and Fancy 1995, Fancy and Ralph 1997, Hart et al. 2011). These species also differ in their ability to exploit high quality food resources (Pimm and Pimm 1982).

Our evaluation involves four longitudinal studies of avian malaria infection on the Island of Hawai‘i. The study area comprises approximately 1100 km2 on the eastern flanks of Mauna Loa volcano in the southeast corner of Hawai‘i (Appendix: Fig. A1). The Biocomplexity study was conducted as part of a collaborative research effort (NSF grant DEB 0083944) on the Biocomplexity of Introduced Avian Diseases in Hawai‘i. For the Biocomplexity study, nine 1-km2 study sites were established along an altitudinal gradient from 25 to 1800 m above sea level and stratified into three major disease “zones” based on elevations identified by van Riper et al. (1986). We had two study sites (SOL and CJR) at high elevation (>1650 m), four (COO, CRA, PUU, WAI) at mid elevation (1000–1300 m), and three (BRY, MAL, NAN) at low elevation (<300 m). The second study focused on Apapane within a 0.5-ha mid-elevation (1200 m) study site at Kilauea Volcano (KV) from 1992 to 1998. These data were originally reported in Atkinson and Samuel (2010), but we conducted a reanalysis using new statistical models (below) and interpret the results in the larger context of the avian malaria on Hawai‘i. The Kulani study was conducted on a 0.5-ha high elevation (1765 m) site from February 1992 to July 1994 concurrently with the KV study. The Ainahou site (AIN) was located at mid-elevation (915 m) in Hawaii Volcanoes National Park. Ainahou was an operating cattle ranch from 1941 until 1971 and is an open understory, mesic ohia forest (Kilpatrick et al. 2006a). All study sites were in mesic-wet forest (840–4200 mm annual rainfall) dominated by ohia (Metrosideros polymorpha), the primary canopy tree and food source for nectarivorous honeycreepers in Hawai‘i. There were broad similarities in substrate age, rainfall, and vegetation for sites within the same altitudinal zone. Intensity of disease transmission varies across this landscape primarily driven by seasonal and altitudinal differences in both temperature and rainfall that affect abundance of Culex quinquefasciatus and the intrinsic incubation of P. relictum (Ahumada et al. 2004, Ahumada et al. 2009, Samuel et al. 2011). Mean monthly temperatures range from 24°C at low elevation to approximately 13°C at high elevation.

Banding and blood sampling

Mist-netting was conducted monthly in each of the nine Biocomplexity study sites, from January 2002 through June 2005, using 18–24 mist-nets at a height of 6 m (Woodworth et al. 2005). Nets were operated for approximately 6 hours each day between 06:30 and 14:00 HST, for 3–4 days/month. At KV, we captured native and non-native forest birds by mist-netting at 1–3 month intervals from January 1992 to June 1998 at 16 fixed locations. Eleven net sites were operated during the entire study, with five additional nets added in August 1992 to increase the number of captures. At Kulani, we captured birds at monthly intervals from February 1992 to July 1994 at 13 fixed net sites within the study site. At Ainahou, we mist-netted, banded, and bled birds for 4 days on alternate weeks from December 2001 to December 2004. A varying number of nets were set at each session depending upon number of bird captures, availability of bird banders, and intensity of tree bloom. Depending on weather conditions, nets were operated from 1-2 h after sunrise to mid-afternoon on 3–5 consecutive days during each sampling period. For all studies, captured birds were banded with US Fish and Wildlife Service numbered aluminum leg bands for subsequent identification. Sex of birds was determined by brood patch or cloacal protuberance, plumage characteristics, and measurements of wing chord, culmen length and tarsus length (Pyle 1997; USGS, unpublished data). We determined age as hatch year (HY), second year (SY), or after-second-year (ASY) based on plumage (Fancy et al. 1993); SY and ASY birds were collectively considered as adults (AD). We obtained blood samples (<1% of body weight) by jugular venipuncture with a heparinized 28.5-gauge insulin syringe. Blood smears were prepared and fixed in absolute methanol in the field. Remaining blood was transferred to microhematocrit tubes and centrifuged to separate plasma, which was frozen at −70°C for serological analysis. All captures and blood samples were collected under approved Animal Care and Use Protocols through the U.S. Geological Survey–National Wildlife Health Center, Madison, WI (1992–1998) or the University of Hawaii–Manoa (2000–2006).

Testing for malarial infection

Within 4 days post-infection (PI) susceptible native birds develop detectable parasitemias (pre-patent period), which peak 12–16 days PI (the “crisis”), and begin a rapid decline by 21–28 days as the humoral and cellular immune systems respond (van Riper et al. 1986, Atkinson et al. 1995, Atkinson et al. 2000, Yorinks and Atkinson 2000). Antibodies can develop as early as 7 days PI (Atkinson et al. 2001b). Mortality typically occurs between 19-30 days PI (Atkinson et al. 1995, Atkinson et al. 2000, Yorinks and Atkinson 2000). Native birds that survive are chronically infected, have immunity to rechallenge with P. relictum, and probably remain infectious to mosquitoes throughout their life (Atkinson et al. 2001a), although detection by microscopy is inconsistent (Jarvi et al. 2002).

Therefore, we used a combination of blood smears and serology to determine the malarial infection status of captured birds. Blood smears were stained with buffered 6% Giemsa, pH 7, for one hour, rinsed with tap water, and dried. We scanned 100 microscope fields (approximately 30,000 to 50,000 erythrocytes) with a 40× objective to identify erythrocytic stages of the parasite. For plasma samples collected as part of the Biocomplexity and Ainahou studies, we used an enzyme-linked immunosorbent assay (ELISA) (Graczyk et al. 1993) to detect antibodies to erythrocytic stages of P. relictum and identify strongly positive and negative samples. Samples with ELISA values (mean absorbance of triplicate samples − the mean absorbance of triplicate negative controls)/(mean absorbance of triplicate positive controls − mean absorbance of triplicate negative controls) × 100) between 15 and 65 were considered equivocal and further tested by immunoblot (Atkinson et al. 2001a). Plasma samples from the Kulani and KV studies were tested by immunoblot alone. Sensitivity and specificity of serological tests are comparable to PCR (Jarvi et al. 2002, VanderWerf et al. 2006) and helped distinguish acute from chronic infections where parasitemias are too low to be consistently detected by microscopy or PCR (Atkinson et al. 2001a, Jarvi et al. 2002).

Captured birds were considered susceptible (antibody and parasitemia negative) or recovered/chronically infected (antibody positive, parasitemia positive or negative). Birds with acute infections were antibody negative and parasite positive (Atkinson et al. 2001a), but these made up less than 5% of captures (Appendix: Table A1). Acutely infected birds were unlikely to be captured as clinical signs of infection include reduced activity, acute morbidity, or mortality (Yorinks and Atkinson 2000, Atkinson and Samuel 2010). For analysis, we considered acutely infected birds that were captured to be recovered because active birds are likely malaria survivors.

Epizootiological modeling

We analyzed longitudinal (capture-recapture) data using Bayesian state-space multi-state models (Kery and Schaub 2012, King 2012). We combined longitudinal estimation of disease transmission with a discrete cumulative age-prevalence model (Atkinson and Samuel 2010) to simultaneously estimate time-specific (i) or seasonal (winter = Wi, spring = Sp, summer = Su, and fall = Fa) capture rate (piS, piR), transition probability (ψiSR) from susceptible (S) to recovered (R), malaria fatality (m), AD survival of recovered (siR) and susceptible birds (siS) for each season-time period (i = 11, 13, 16 and 26 for Kulani (KUL), Biocomplexity, Ainahou (AIN), and KV studies, respectively) (Tables 13). Although we used open population models we did not account for emigration from our study sites; therefore, we calculated apparent survival which underestimates true survival (White and Burnham 1999).

Table 1. State-space transitions for malaria infection in adult Hawaiian forest birds from the true state at time t to the true state at time t +1.
table image
Table 2. Capture probabilities for Hawaiian forest birds at time t. Capture rates for hatch-year birds may be different than capture rates for adult birds.
table image
Table 3. State-space transitions for malaria infection in hatch-year (HY) and second-year (SY) Hawaiian forest birds from the true state at time t to the true state at time t + 1 used to estimate age-prevalence infection rates. HY and SY birds are included only for first capture; therefore, they have a conditional non-malaria survival probability of 1.0. We assumed that transition rates for SY birds were identical to AD, but we allowed different transition rates for HY birds. See text for further details and definition of parameters.
table image

Because acutely infected individuals were seldom captured and have relatively high mortality, we estimated the transition probabilities of susceptible individuals to recovered (chronically infected). For age-prevalence modelling, we estimated cumulative age-prevalence parameters (piS, piR and ψiSR) conditioned on the first capture (survival = 1) of known age (HY and SY) birds which are susceptible at birth (hatch). We randomly assigned a birth season for each HY and SY bird based a species- and elevation-specific multinomial distribution of AD birds with brood patches captured during the Biocomplexity study (Appendix: Table A2). SY birds were randomly assigned a season in the year prior to capture. HY birds were assigned a season that preceded their capture season within the same year. We also estimated separate capture, malaria fatality (m), and malaria infection rates for HY birds, but assumed SY and ASY birds were similar (Tables 1 and 3). For known age birds capture rates have no biological meaning because they are based on captures of birds know to be alive. We estimated annual survival and malaria infection rates as the product of the seasonal survival SY = sWi × sSp × sSu × sFa or seasonal infection IY = (1 − (1 − IWi) (1 − ISp) (1 − ISu) (1 − IFa)) (Atkinson and Samuel 2010), respectively. We also determined the annual population mortality rate from malaria M = IY × S/H × m (Atkinson and Samuel 2010) for HY birds (S/H = 1.0; 100% susceptible at hatch) and for AD (S/H = no. of susceptible AD captured in winter/no. of AD captured in winter). We enhanced model parameter estimates using Bayesian priors for the estimated malaria fatality rates (m) for the three native species based on experimental laboratory studies from high and low elevation Amakihi, Apapane, and Iiwi (Atkinson et al. 1995, Atkinson et al. 2000, Yorinks and Atkinson 2000, Samuel et al. 2011, Atkinson et al. 2013). We used a beta distribution (β(survivors + 1, fatalities + 1)) to estimate priors based on the number of survivors and fatalities in laboratory studies for each species: Iiwi β(2, 9), Apapane β(6, 4), Amakihi β(13, 19), and low-elevation Amakihi β(11, 2). Because our primary focus was on disease processes we also used the 95% confidence intervals on annual survival rates from previous studies (Woodworth and Pratt 2009: Table 8.1) to limit model estimates of seasonal non-malaria survival as follows: Iiwi uniform (0.81, 0.91), Apapane uniform (0.83, 0.97), Amakihi uniform (0.90, 0.95). Preliminary analysis showed these annual survival priors had little effect on disease related parameter estimates.

Models were parameterized so malaria incidence was equal to the transition from susceptible to recovered status (Ii = ψiSR), and occurred prior to survival of recovered birds (Kery and Schaub 2012) which allowed us to estimate the malaria fatality rate (m) as part of the transition (Tables 1 and 3). We used OpenBugs (Lunn et al. 2009) to evaluate 15 models (Table 4) for seasonal, time-specific, or constant capture, survival, and infection rates for each native species (Amakihi, Apapane, and Iiwi) and elevation from the Biocomplexity, KV, Ainahou, and Kulani studies. We used the Bayesian Information Criterion (BIC, Schwarz 1978) to compare alternative models and identify parsimonious models for capture, survival, and transition probabilities. Estimated model parameters are reported as posterior mean estimates and 95% Bayesian Credible Intervals (BCI). Because transient (non-residents) animals are common in most avian populations and their presence negatively biases survival rates, we evaluated capture data for the presence of transient AD birds (Pradel et al. 1997) using TEST 3G.SR in program U-CARE (Pradel et al. 2005). When transient birds were important we removed first captures of AD birds for model selection and to estimate model parameters (Pradel et al. 1997). We ran models with 30,000 MCMC replications for burn-in and an additional 20,000 MCMC replications for model convergence and to estimate deviance and calculate BIC = Deviance + Ln(N) × K, where N is the number of birds and K is the number of model parameters (Link and Barker 2010). Final model parameters were estimated using 70,000 Markov Chain Monte Carlo (MCMC) burn-in replications and 30,000 MCMC replications. Model convergence was assessed based on visual convergence of the MCMC replicates, smoothness of the posterior parameter distributions, and the MCMC error for each parameter being <100-fold smaller than the posterior standard deviation. We used generalized χ2 methods for binomial variables (Sauer and Williams 1989) to evaluate species and elevation differences in malaria infection and survival.

Table 4. Alternative models and description for AD Hawaiian forest birds with malaria transition based on all HY and AD birds (see text for details).
table image

Vector-host interaction

Because densities of adult mosquitoes are low and collection of blood fed Culex is difficult in Hawai‘i, especially at mid and high elevations, we estimated vector feeding preference based on the relative annual infection rates in AD birds adjusted for their population density. We estimated the relative vector preference of bird species j over k for each pair of species for mid- and high-elevation Biocomplexity sites: fpj,k = Ij/Pj/Ik/Pk, where I and P are the annual infection rates (IY) and population density for each species, respectively. In this calculation fpj,k is an estimated odds ratio for preference of species j to k. We also used fp to calculate an index of the relative contribution of native species (in pairs) to infecting susceptible mosquitoes IMj,k = fpj,k × IBj/IBk, where IB is the number of infected birds for each species calculated as the proportion of infectious birds × P. Our approach assumes that avian species have equal likelihood of becoming infected by mosquitoes, which is supported by experimental studies showing 98% infection of all three species from a single bite by an infectious mosquito (Samuel et al. 2011: Appendix B; C. T. Atkinson, unpublished data). Both equations were solved using Bayesian analyses which included the species and elevation specific parameters and variances on the right-hand-side of the equations.

Results

Prevalence and intensity of infection

We obtained blood samples from 5,353 Hawai‘i Amakihi, 2,116 Apapane, and 1,046 Iiwi during four studies on Kilauea and Mauna Loa Volcanoes during 1992–1998 and 2001–2005 (Appendix: Table A1). Overall prevalence of malaria based on parasitemia and serology at first capture ranged from 2.6% (27/1,046) in Iiwi to 40% (837/2,116) in Apapane and 39% (2,072/5,353) in Hawai‘i Amakihi. Prevalence was strongly influenced by elevation, with lowest prevalence of infection at high elevation (2.2% for Iiwi, 7.8% for Apapane, and 1.5% for Amakihi) followed by mid elevation (20% for Iiwi, 60% for Apapane, and 17% for Amakihi), and low elevation studies (no captures for Iiwi, 100% for Apapane, and 85% for Amakihi). We classified >97% of the malarial infections as chronic (recovered) and the remaining birds were acutely infected (Appendix: Table A1), but were classified as recovered in our analyses.

Species-specific patterns—Amakihi

Low-elevation forests

Amakihi were the only species with sufficient captures to evaluate epizootiological patterns across all three elevations (Appendix: Table A1). Amakihi in low-elevation forests were also the only population of native birds with regular capture of acutely infected birds (3.7% of captures). In low-elevation forests the capture data was best represented by Model 12 (Table 5) with equal time-specific capture rates for susceptible and recovered birds that differed between AD and HY birds, constant malaria infection that was equal for AD and HY, and constant non-malaria survival that was similar for disease classes. We found significant (Z = 7.07, P < 0.001) evidence of transient birds and removed first captures of AD. Annual population survival (malaria mortality excluded) was 0.73 (95% BCI = 0.63, 0.83). Seasonal (0.48) and annual (0.92) infection rates were high, indicating that few Amakihi (8%) would avoid malaria infection annually (Table 6 and Fig. 1). AD seasonal capture rates were highest in summer and fall (Table 7).

figure image

Annual malaria infection rates (IY in Tables 6 and 8) in low-, mid-, and high-elevation forest in Hawaii for Biocomplexity (unlabeled), Ainahou (AIN), Kilauea Volcano (KV), and Kulani (KUL) study sites. Mean infection rates with 95% Bayesian Credible Intervals for Amakihi (solid rectangles), Apapane (solid diamonds), and Iiwi (solid triangles).

Table 5. Model parameter estimates with Bayesian standard errors for Hawaiian Apapane and Iiwi by elevation. See text for description of study sites and final models.
table image
Table 6. Alternative models and Δ BIC values for AD Hawaiian forest birds with malaria transition based on all HY and AD birds. See text for details and Table 4 for model descriptions.
table image
Table 7. Model parameter estimates with Bayesian standard errors for Hawai‘i Amakihi by elevation. See text for description of study sites and final models.
table image
Table 8. Adult seasonal capture rates and Bayesian standard errors for Hawai'i Amakihi, Apapane, and Iiwi at Biocomplexity (Low, Mid, High), Ainahou (AIN), Kilauea Volcano (KV), and Kulani (KUL) study sites on Hawai‘i.
table image

Mid-elevation forests

Most captures of Amakihi in mid-elevation forests were from the Ainahou study, and most of the captures were susceptible birds (Appendix: Table A1). For the Biocomplexity sites, Models 14 and 15 (ΔBIC = 3) provided the best fit (Table 5) with time-constant malaria infection rates, and capture and survival rates that were time-constant and similar for susceptible and recovered birds. We found no evidence for transient birds in the data (P = 0.30). AD capture rates were approximately 0.10 and constant throughout the year (Table 7). Annual population survival without malaria mortality was 0.70 (0.66, 0.79). Annual (0.33) infection rates for AD were moderate, but infection rates were higher for HY (0.68) (Table 6 and Fig. 1).

For the Ainahou study, Model 9 provided the best fit to this much larger data set and we found significant evidence (P < 0.001) of transient birds, so first capture of AD birds were removed (Table 5). A simplified model with seasonal capture rates for AD and HY birds and equal survival for susceptible and recovered birds provided an improved fit to the capture data (Δ BIC = 61). AD seasonal capture rates were highest in fall and winter (Table 7). Annual (0.29) infection rates for AD were moderate. Higher malaria infection and fatality rates for HY birds (Table 6 and Fig. 1) indicates that malaria dynamics may be much different than in AD birds. Annual population survival without malaria mortality was 0.70 (0.66, 0.79).

High-elevation forests

Amakihi were frequently captured in high-elevation forests, but few birds (<2%) had acute or chronic malarial infection (Appendix: Table A1). We found strong evidence (P < 0.03) for transient Amakihi in both the Biocomplexity and Kulani studies and removed first captures of AD birds. Model 15 was the best fit for both studies (Table 5). Model 15 with seasonal capture rates for AD and HY birds provided a modest increase in model fit (ΔBIC = 8) for the Kulani data. For the Biocomplexity study, AD capture rates were approximately 0.15 and constant during the year. At Kulani AD capture rates were lowest in spring and highest during fall and winter (Table 7). Annual population survival without malaria mortality was 0.73–0.74 for these two studies. Annual (0.09–0.17) malaria infection rates were low for AD birds compared to other elevations, but high (0.76–0.84) for HY birds (Table 6 and Fig. 1). Mortality from malaria infection was 0.60 (0.43–0.76) in AD birds at the Biocomplexity sites, but higher (0.75; 0.63, 0.87) at Kulani (Fig. 2B). Malaria fatality was higher in HY birds (0.85; 0.77, 0.92) than AD birds at the Biocomplexity sites, but AD and HY malaria mortality was not different at Kulani (Table 6). These patterns show that malaria dynamics and impact are likely more substantial in HY Amakihi.

figure image

Bayesian prior and posterior estimates (solid diamonds) and 95% Bayesian Credible Intervals (solid lines) of malaria fatality rates (m in Tables 1 and 3) for Biocomplexity (Low, Mid, and High), Ainahou (AIN), and Kulani (KUL) studies for (A) low-elevation Amakihi, (B) mid- and high-elevation Amakihi. Average estimates were calculated using mean and variance for each study.

Comparisons among elevations

AD Amakihi populations in low-elevation forests experience greater annual malaria transmission than mid- (χ2 = 122, P < 0.001) or high-elevation forests (χ2 = 177, P < 0.001) and transmission was greater (χ2 = 6.3, P = 0.012) at mid elevation than high elevation (Fig. 1). We found no evidence of different AD malaria fatality between mid- and high- elevation Amakihi (χ2 = 1.08, P = 0.3), with an average rate of 68% (Fig. 2B). Malaria fatality was lowest in low-elevation Amakihi (2.5%) compared to mid (χ2 = 182, P < 0.001), high (χ2 = 86, P < 0.001), or the combined mortality rates of mid- and high-elevation birds (χ2 = 236, P < 0.001). The malaria fatality rate in low-elevation forests was substantially lower than the prior from experimental studies (Fig. 2A). Malaria fatality rates among mid- and high-elevation studies were slightly higher than prior estimates from laboratory experiments; except for the high-elevation Biocomplexity site where fatality was nearly identical (Fig. 2B). Malaria population impacts were least in low-elevation AD Amakihi compared with mid (χ2 = 24, P < 0.001) and high elevations (χ2 = 2.79, P < 0.095), but mid elevation was not different from high elevation (χ2 = 2.02, P < 0.155). Malaria reduced the low-elevation AD population <1% annually, and reduced the HY population by 1–4%. With average malaria fatality rate of 0.68 for AD (Fig. 2) the annual population mortality from avian malaria in mid-elevation forests was 15% (10–21%) for AD and likely much higher (44–62%) for HY birds. Estimated annual population impacts were relatively small for AD (5.3% and 12%) Amakihi populations at high-elevation study sites, respectively. However, malaria impacts were likely higher for high-elevation HY birds (73% and 57%).

Species-specific patterns—Apapane

Few Apapane were captured in low-elevation forests, and all these birds had previously survived malaria infection (Appendix: Table A1), making it impossible to estimate epizootiological parameters for low-elevation Apapane, which are rare in these forests (Spiegel et al. 2006). By comparison, captures of Apapane were frequent in mid- and high-elevation forests where only a few birds were acutely infected with malaria. Capture rates of AD Apapane varied seasonally at all study sites and were typically highest in fall and lowest in spring (Table 7). For the mid-elevation KV study we found moderate evidence for transient birds (Z = 1.84, P = 0.066) and removed first AD captures. For all other studies there was no evidence of transient birds (P > 0.05).

Mid-elevation forests

Model 12 provided the best fit for Apapane captures for KV, while Model 15 was the best fit for the Biocomplexity study (Table 5). For the Biocomplexity and KV studies, seasonal capture rates provided a substantial improvement in model fit; ΔBIC = 49 and >200, respectively. Annual (0.67 and 0.71) malaria infection rates for AD and HY birds were similar for both mid-elevation studies (Table 8 and Fig. 1) demonstrating a high likelihood of susceptible Apapane becoming infected annually. Annual population survival without malaria mortality was 0.49–0.65 for these two studies.

High-elevation forests

Apapane were frequently captured in high-elevation forests, but acutely infected birds were rare and the number of recovered birds was limited (Appendix: Table A1). For the Biocomplexity and Kulani studies, differing seasonal capture rates provided a substantial improvement in the fit for Model 15; ΔBIC = 69 and 10, respectively. Annual (0.10 and 0.28) malaria infection rates for AD Apapane at Biocomplexity and Kulani studies indicated a low to moderate risk of malarial infection (Fig. 1). AD malaria fatality (0.42 vs. 0.43) was nearly identical for both studies (Fig. 3A). However, annual infection and fatality rates were higher for HY birds indicating malaria dynamics and impacts where greater than for AD birds. Estimated annual AD survival rate was 0.50–0.51.

figure image

Bayesian prior and posterior estimates (solid diamonds) and 95% Bayesian Credible Intervals (solid lines) of malaria fatality rates (m in Tables 1 and 3) for Biocomplexity (Low, Mid, and High), Kilauea Volcano (KV), and Kulani (KUL) studies for (A) mid- and high-elevation Apapane, and (B) high-elevation Iiwi. Average estimates were calculated using mean and variance for each study.

Comparisons among elevations

AD Apapane had similar rates of annual malaria infection for mid- (χ2 = 0.4, P = 0.52) and high-elevation (χ2 = 2.2, P = 0.14) studies (Fig. 1). However, malaria infection (χ2 = 50, P < 0.001) was lower in high-elevation forests. Except for the mid-elevation Biocomplexity study, malaria fatality was slightly higher than experimental infection. The two high-elevation studies had posterior estimates that were nearly identical to the prior (Fig. 3A), suggesting that capture data were generally insufficient to improve fatality estimates. The average AD malaria fatality across all sites was 47%, suggesting malaria reduced the AD and HY population approximately 10% and 28–35% annually in mid-elevation forests and 5–12% in high-elevation AD populations. Malaria fatality and population impacts were much higher in high-elevation HY Apapane, indicating that HY birds may have higher malaria impacts than AD Apapane at this elevation.

Species-specific patterns—Iiwi

High-elevation forests

Iiwi were not captured in low-elevation forests and only a few were captured in mid-elevations forests (Appendix: Table A1), restricting epizootiological models to high elevations. These patterns correspond with Iiwi abundance measured during bird surveys (Gorresen et al. 2009, Hart et al. 2011). Capture of either acutely and chronically infected birds at high-elevation sites was infrequent (2%). We found strong evidence for transient birds (P < 0.005) for both studies and removed first captures of AD Iiwi for model estimation. For Biocomplexity sites, Models 14 and 15 best fit the capture data (ΔBIC = 1). We found that Model 15 with seasonal capture rates substantially improved model fit (ΔBIC = 78). For the Kulani study, Model 9 best fit our data (Table 5), and it was also improved by using seasonal capture rates and equal survival for susceptible and recovered birds (ΔBIC = 58). Seasonal captures of AD birds for both studies was highest in fall and smaller in spring (Table 7). Seasonal and annual malaria infection rates were similar for both studies and indicated a moderate risk of annual malaria infection (0.23) for AD Iiwi (Fig. 1). Malaria mortality for AD and HY Iiwi averaged 0.93 (0.87, 0.98%); similar to the 95% fatality reported by Atkinson et al. (1995) for experimental studies (Fig. 3B). Malaria fatality in Iiwi was the highest for any native birds (Figs. 2 and 3). Malaria infection rates were much higher for HY Iiwi for these studies (Table 8 and Fig. 1), indicating that malaria may have a more substantial impact on HY Iiwi populations. Based on 93% average fatality, malaria reduced high-elevation AD and HY Iiwi populations by approximately 16–20% and 55–73% annually, respectively. The annual non-malaria survival rate for Iiwi was 0.55–0.60 (Table 8).

Patterns among species

To evaluate species differences we compared annual malaria infection rates for the same elevation (Fig. 1). We were unable to make species comparisons in low-elevation forests because few Apapane and no Iiwi were found at this elevation (Fig. 1). For AD birds in mid-elevation forests, annual malaria infection rates for Apapane from the Biocomplexity (0.71) and KV studies (0.67) were higher (inline image = 35, P < 0.001) than for Amakihi from Biocomplexity (0.33) and Ainahou (0.30). In high-elevation forests, annual malaria infection rates were similar (inline image = 7.6, P = 0.18) among Amakihi, Apapane, and Iiwi. AD malaria fatality (Fig. 2B) for mid- and high-elevation Amakihi (66%) was greater (χ2 = 16, P < 0.001) than Apapane (47%), but mortality for Iiwi (93%) was greater (χ2 = 1247, P < 0.001) than low-elevation Amakihi (3%), mid-elevation and high-elevation Amakihi (χ2 = 48, P < 0.001), or Apapane (χ2 = 92, P < 0.001). Thus, malaria mortality was greatest in Iiwi, least in low-elevation Amakihi, and intermediate in mid- and high-elevation Amakihi and Apapane. Annual malaria impacts on AD birds in mid-elevation forests were greater for Amakihi (15%) than Apapane (6%) (χ2 = 7.1, P = 0.008). In high-elevation forests malaria impacts were similar (χ2 = 0.003, P = 0.96) for AD Amakihi (9%) and Apapane (8%). However, malaria impacts were highest (χ2 = 4.4, P = 0.036) for AD Iiwi (21%) compared to Amakihi and Apapane.

Vector and host interaction

Culex mosquitoes appeared to prefer feeding on Amakihi over Apapane in mid-elevation forests (fp = 3.15; 1.31, 5.79) where malaria transmission was moderate. In high-elevation forests, mosquitoes preferred feeding on Iiwi over Apapane (fp = 10.3; 2.57, 29.7). In these forests Iiwi were somewhat favored over Amakihi (fp = 3.96; 0.79, 13.3), but Amakihi and Apapane had similar feeding preference (fp = 3.76; 0.58, 12.0); these preferences were not statistically significant, likely because malaria infection rates were low and poorly estimated. Apapane were more likely to produce newly infected mosquitoes than Amakihi at mid (IM = 7.23; 3.94, 17.8) and high (IM = 3.29; 0.95, 25.1) elevations. There was no difference between high-elevation Iiwi and Amakihi (5.95; 0.84, 22.4) or Apapane (IM = 1.15; 0.22, 3.7) in their relative contribution to mosquito infections, perhaps partly due to low infection rates.

Discussion

Over the past three decades evidence has accumulated about the prevalence and pathogenicity of avian malaria in native Hawaiian species. However, we previously knew little about malaria transmission patterns and disease impacts on survivorship for different species of honeycreepers, and across the elevation gradients found on Hawai‘i. We used a Bayesian state-space model (Kery and Schaub 2012, King 2012) to combine longitudinal and cross-sectional capture-recapture models (Atkinson and Samuel 2010) to estimate malaria infection and mortality rates, and population impacts of three species of Hawaiian honeycreeper. This approach allowed us to enhance traditional multi-state models by incorporating single captures of known age birds using an age-prevalence method (Caley and Hone 2004, Heisey et al. 2006). Further, the Bayesian framework allowed us to separate infection and disease mortality; therefore, directly estimate disease impacts on each avian species. It also allowed us to incorporate prior information on malaria mortality and annual survival rates. An important assumption in combining multi-state and age-prevalence data is similarity in rates of disease infection and fatality. Our experience showed that including age-prevalence data was most helpful when disease infection and fatality rates for known age birds (primarily HY birds in our case) were similar to those for AD birds. We found this situation for low-elevation Amakihi and mid-elevation Apapane which had improved estimates of malaria fatality (Figs. 2 and 3) and infection. However, when one or both of these rates differ for HY birds there was limited or no benefit from age-prevalence data. In this situation disease and capture parameters for HY birds may be highly correlated and estimation of infection or fatality rates for HY birds may be confounded. As a result, it is only appropriate to conclude that the disease process is operating differently in the known age individuals (a.k.a., young birds). Further, incorporating age-prevalence data with multi-state models may also be beneficial when infection rates are high (so that prevalence changes rapidly with age) or when capture rates are low because many individuals will have insufficient recaptures to estimate state transition rates.

Our study demonstrates that transmission of avian malaria, susceptibility, and population impact depends on the combination of both elevation and species in Hawai‘i. Our results show that malaria infection is lowest in high-elevation forests where climate is less favourable for mosquito populations and malaria parasite development (Ahumada et al. 2004, Samuel et al. 2011). As a result, these high-elevation forests currently provide a relatively disease-free refuge for susceptible honeycreepers (Ahumada et al. 2004, Atkinson and LaPointe 2009b). However, daily and seasonal movements of birds to lower-elevation forests in search of nectar resources (Ralph and Fancy 1995) and/or upslope movement of infected mosquitoes (Freed and Cann 2013) likely contribute to the malaria infection we observed in these high-elevation populations. In contrast, annual malaria infection was 2–4 fold greater in mid-elevation forests and 1.5–2 times greater in Apapane than Amakihi. In low-elevation Amakihi, malaria infection was > 90% which is more than three-fold greater than mid-elevation and more than eight-fold greater than high-elevation forests. We found that malaria fatality was highest in Iiwi, followed by Amakihi and Apapane in mid and high elevations, and lowest in low-elevation Amakihi. We also found no differences in survival rates between susceptible and chronically infected birds during any studies, in contrast to results from Kilpatrick et al. (2006a) for Amakihi at the Ainahou site. The combination of high malaria transmission and fatality for Iiwi (and mosquito feeding preference), but also for Apapane and mid-elevation Amakihi, likely explains why these honeycreepers are absent from low-elevation forests (Samuel et al. 2011).

Our results confirm laboratory studies (Atkinson et al. 2013) that malaria mortality in low-elevation Amakihi is lower (3%) than in mid- and high-elevation Amakihi (42–68%), Apapane (16–43%), or Iiwi (93%). Apparent adaptation through tolerance to malaria provides a demographic advantage that allows low-elevation Amakihi to increase in abundance, despite high levels of malaria infection (Woodworth et al. 2005, Samuel et al. 2011, Atkinson et al. 2013). Our estimated malaria fatality and annual survival rates generally agree with previous laboratory and field studies on Apapane, Iiwi, and Amakihi from mid- and high-elevation forests (Atkinson et al. 1995, Ralph and Fancy 1995, Yorinks and Atkinson 2000, Kilpatrick et al. 2006a, Woodworth and Pratt 2009). We note that our malaria fatality and non-malaria survival estimates were influenced by prior experimental data or constrained to correspond with previous results, respectively. However, we found lower malaria fatality for low-elevation Amakihi (3% vs. 17%; Atkinson et al. 2014), but similar fatality for Iiwi (93% vs. 95%) compared to previous laboratory experiments (Atkinson et al. 1995). A surprising pattern in our results was the apparent higher level of malaria infection and/or malaria fatality we found in young birds. As indicated above, specific disease parameter estimates in this situation may be unreliable; however, these findings indicate that malaria may have a more substantial impact on young birds than on adults. This finding is supportive of previous studies that have suggested that juvenile birds are more susceptible to malarial infection (van Riper et al. 1994) and requires further investigation in the Hawaiian ecosystem. Although we found that malaria fatality was difficult to estimate, in part because we had generally low capture rates, it provides a key parameter in understanding host fitness and evolutionary response (Lachish et al. 2011b), and predicting demographic impacts of malaria infection in Hawaiian birds (Samuel et al. 2011).

Previous studies on avian malaria have shown spatial differences in Plasmodium prevalence (Sol et al. 2000, Wood et al. 2007, Loiseau et al. 2010, Sehgal et al. 2011, Lachish et al. 2013). Our study builds on these results by demonstrating that pathogen prevalence is driven by both elevation differences in transmission pressure combined with species-specific disease susceptibility. Together, these two key factors affect the distribution and abundance of endemic Hawaiian honeycreepers, restricting most malaria susceptible species to high-elevation forests on Kaua‘i, Maui, and Hawai‘i (Scott et al. 1986, van Riper et al. 1986, Atkinson and LaPointe 2009a). These spatial differences in parasite-mediated selection pressure coupled with spatial host genetic structure and gene flow can significantly influence host adaptation or maintaining genetic diversity (Poulin 2007, Wolinska and King 2009). In Hawaii, the spatial gradient in selection pressure has apparently produced two opposing outcomes—local adaptation of low-elevation Amakihi for malaria-tolerance (Woodworth et al. 2005, Foster et al. 2007, Atkinson et al. 2013) or local extinction of highly susceptible species like Iiwi from low and mid-elevation forests (van Riper et al. 1986). As climate warms and malaria infection risk increases in Hawai‘i (Benning et al. 2002, Atkinson and LaPointe 2009b, Atkinson et al. 2014) it is uncertain whether other endemic honeycreepers will be able to evolve tolerance to malaria. Further research is needed to determine the suite of biotic and abiotic factors that facilitate the coexistence of pathogens and hosts (via either resistance or tolerance) and how these factors interact with host demography and movement across a landscape of heterogeneous selection pressures. The factors driving these outcomes have long been a focus of ecological study (Anderson and May 1991), but are not well understood. In Hawai‘i these processes likely involve host susceptibility and demographic recovery, vector feeding preferences, heterogeneous selection pressure, host genetic diversity, and host gene flow. Understanding the importance of these factors is particularly critical for endemic Hawaiian species that continue to undergo steep population declines and range restrictions as climate warms. Like human malaria, spatial patterns of avian malaria are driven by exogenous temperatures, altitude, rainfall, and suitable habitat for larval mosquitoes to complete their life cycle (Balls et al. 2004, Pascual et al. 2008, Grillet et al. 2010, Grillet et al. 2014). As a result avian malaria in Hawaii may provide a seminal model to understand the environmental and ecological drivers of human malaria and to evaluate alternative control strategies (Samuel et al. 2011, LaPointe et al. 2012).

Heterogeneous encounter rates between hosts and parasites are important determinants of parasite prevalence across host species and disease transmission dynamics, and have significant consequences on host fitness (Kilpatrick et al. 2006b, Medeiros et al. 2013). Previous studies have relied on vector blood meals to evaluate mosquito feeding preferences (Kilpatrick et al. 2006b, Hamer et al. 2009, Simpson et al. 2012, Medeiros et al. 2013). However, in Hawai‘i the low abundance of mosquitos in mid- and high-elevation forests (LaPointe 2000) and their low rate of malarial infection (LaPointe 2000, Samuel et al. 2011) make capturing blood-fed mosquitoes challenging. As an alternative, we used relative host infection rates and relative host abundance to assess host-vector contact rates. In mid- and high-elevation forests, we found that mosquitoes preferred feeding on Iiwi, then Amakihi, followed by Apapane. Despite the preference for feeding on Amakihi, Apapane were much more likely to produce newly infected mosquitoes and therefore are a more important reservoir species, likely because they are less susceptible to malaria mortality, more abundant, and have higher malarial infection rates than other native species. These results suggest that Apapane may serve as a significant reservoir host that enables higher rates of disease transmission to more vulnerable species such as Iiwi, and thereby facilitate apparent competition where one host indirectly competes with others via disease transmission (Holt and Pickering 1985, McCallum and Dobson 1995). Our study identifies the need for research, especially on other Hawaiian islands, to better understand the factors that influence host-vector encounters (e.g., mosquito feeding preference, host defensive behavior, roosting locations) and are potential drivers of differential malaria transmission among species and how these factors influence mosquito infection, the relative impacts of malaria on native species, and composition of the Hawaiian forest bird community.

Overall, we found patterns of malaria transmission across elevations that provide further support that climate, based on an elevational gradient, drives avian malaria through mosquito population dynamics and malaria parasite development rates within the host mosquito (Ahumada et al. 2004, Ahumada et al. 2009, Atkinson and Samuel 2010, LaPointe et al. 2010). As a result, increased temperatures from global warming (Benning et al. 2002, Harvell et al. 2002, Freed et al. 2005, Atkinson and LaPointe 2009b) are likely to increase the severity and frequency of malaria transmission at higher elevations in Hawai‘i. Increasing temperatures may eliminate high-elevation refugia that currently protect many Hawaiian bird populations, and increase malaria transmission and epizootics in mid-elevation forests (LaPointe et al. 2012). There is recent evidence this may already be taking place on Kaua‘i (Atkinson et al. 2014). This combination of factors will likely produce further reductions and extinctions of native Hawaiian birds, particularly for threatened species with small, fragmented populations in high-elevation forests and for species with high susceptibility to malaria such as the Iiwi. In addition, higher transmission of malaria in mid-elevation forests may mean substantial declines in honeycreepers with moderate levels of susceptibility to avian malaria, such as the Hawai‘i Amakihi. The greatest challenge will be development of disease control strategies that conserve the native Hawaiian species while considering the future threats that may occur as climate warms.

Acknowledgments

This research was funded through the U.S. Geological Survey's Wildlife, Invasive Species, and Natural Resource Protection Programs and a Biocomplexity grant from the National Science Foundation (DEB0083944). We also wish to thank our technical staff, postdoctoral researchers, and numerous research interns whose hard work and dedication made this research possible. The use of trade names or products does not constitute endorsement by the U.S. Government. This study was performed under the animal care and use protocols approved at the University of Hawai‘i, Manoa. E. Paxton and V. Henaux provided many valuable comments that improved the paper. The Department of Forest and Wildlife Ecology at the University of Wisconsin-Madison provide assistance with publication costs.

    Supplemental Material

    Appendix

    Table A1. Capture of susceptible (S), acutely infected (I), recovered (R), and total adult (AD) and hatch-year (HY) native Hawaiian forest birds at Biocomplexity (2002–2004), KV (1992–1998), AIN (2001–2004), and KUL (1992–1994) study sites, species, age, and elevation.
    table image
    Table A2. Proportion of adult female native Hawaiian birds with active brood patches by species and season captured in low-, mid-, and high-elevation forests during 2001–2003 at nine Biocomplexity sites.
    table image
    figure image

    Map of Biocomplexity (solid diamonds), Kilauea Volcano (KV solid circle), Ainahou (AIN solid star), and Kulani (KUL solid triangle) study sites located on the eastern slope of Mauna Loa and Kilauea Volcanoes on the Island of Hawai‘i. Biocomplexity study sites are Bryson's (BRY), Malama Ki (MAL), Nanawale (NAN) at low elevation, Crater (CRA), Cooper (COO), Pu'u (PUU), and Waiakea (WAI) at mid elevation, and C.J. Ralph (CJR) and Solomon (SOL) at high elevation. See text and Appendix tables for additional information on study sites.