Ecological Archives E096-123-A2
Chelsea L. Wood, Julia K. Baum, Sheila M. W. Reddy, Rowan Trebilco, Stuart A. Sandin, Brian J. Zgliczynski, Amy A. Briggs, and Fiorenza Micheli. 2015. Productivity and fishing pressure drive variability in fish parasite assemblages of the Line Islands, equatorial Pacific. Ecology 96:1383–1398. http://dx.doi.org/10.1890/13-2154.1
Appendix B. Details of methodology.
Across-islands and within-island gradients
To assess the effects of productivity and fishing pressure on parasite abundance and diversity, we conducted sampling at two different spatial scales: across six islands of the Line Islands archipelago (Fig. 1a) and within the archipelago's largest island, Kiritimati (Fig. 1b). Our goal was to quantify differences in parasite abundance, species composition, and diversity in these two data sets and to interpret this variability in the context of the natural and anthropogenic gradients that exist across these two spatial scales.
The Line Islands archipelago experiences strong natural and anthropogenic gradients in biotic and abiotic conditions. Three of these islands (Jarvis, Kingman, and Palmyra) have never been permanently inhabited or intensively fished, and are currently protected against fishing as U.S. National Wildlife Refuges (Maragos et al. 2008a) within the Pacific Remote Islands Marine National Monument. As a consequence, these islands represent some of the most intact coral reef ecosystems of the tropical Pacific, with fish assemblages distinguished by high biomass of the top predators that are commonly sought in commercial fisheries (DeMartini et al. 2008; Maragos et al. 2008a; Sandin, et al. 2008). The remaining three islands (Teraina, Tabuaeran, and Kiritimati) are part of the Republic of Kiribati, and have been inhabited and fished since the 1850s (see Supplemental Data SI in Sandin, et al. 2008). Due to intensive artisanal fishing pressure, the fish faunas of most of the reefs on these islands are relatively depauperate, with low biomass of top predators and high abundance of low-trophic level hosts like planktivores (DeMartini et al. 2008; Sandin, et al. 2008). In addition to this strong inter-island variability in anthropogenic impacts on the biota, natural gradients in oceanography and climate also differentiate these islands (Maragos et al. 2008b). Oceanographic variability, including a gradient of increasing nutrient concentration and primary productivity and decreasing sea surface temperature at more southerly islands (Appendix A; Sandin et al. 2008) is driven by differences in upwelling and predominant currents (Maragos et al. 2008b). Equatorial upwelling bathes the southerly islands in cold, nutrient-rich water, which lie in the path of the cold, nutrient-rich Equatorial Undercurrent (Maragos et al. 2008b). In contrast, the northerly islands lie outside the influence of equatorial upwelling, and are bathed by the warmer and more nutrient-poor North Equatorial Countercurrent (Maragos et al. 2008b). These natural gradients in productivity and temperature do not covary with anthropogenic gradients in fishing pressure among the six islands (Appendix A; Fig. 1a). For this reason, we were able to reliably parse the influence of natural from anthropogenic influences on parasite communities in the across-islands portion of our analysis.
In contrast, Kiritimati Island experiences a strong west-to-east gradient in both fishing and productivity (Fig. 1b), because the western, leeward side of the island is exposed to both heavy fishing pressure and high levels of productivity (Walsh 2011). The most populous townships on this island of more than 5,000 inhabitants are located on the western side of the atoll (Kiribati Statistics Office 2007). Due to the island's large size (388 km² in land area, ~150 km in perimeter, ~50 km from the largest town to the lowest-fishing pressure sites we sampled) and the expense of car or boat travel over long distances, most fishing occurs in the vicinity of these population centers (Fig. 1b; DeMartini et al. 2008; Walsh 2011). Simultaneously, the leeward side of the atoll experiences high-nutrient, low-temperature conditions, possibly due to prevailing east-to-west currents that drive island wake upwelling (Fig. 1b; Heywood et al. 1990; Walsh 2011). Primary productivity and fishing pressure are therefore positively correlated (Fig. 1c). To investigate the combined influence of fishing and productivity on parasite abundance, species composition, and diversity on Kiritimati, we included fishing and productivity as a single predictor in statistical models (as recommended by Dormann et al. 2013, see details in Statistical analysis: Parasite abundance, below), and used inferences from the across-islands analysis to interpret patterns detected on Kiritimati. This conservative approach allowed us to avoid spurious conclusions about the relative influence of productivity and fishing, which can arise from interpreting unstable parameter values generated by collinear statistical analyses, while still gaining insight into how these two factors shape parasite abundance, species composition, and diversity. Using both the across-islands and within-island datasets allowed us to assess the effects of productivity and fishing at different spatial scales, providing a rigorous test of our hypotheses.
Host sampling
Species were chosen to span a range of body sizes, trophic levels, and taxonomic groups (Table 2). Fish were collected by scuba divers using three-pronged spears (for fish >10 cm) or handnets (for fish <10 cm). We sampled at least 25 individuals of each species from each island, and exceeded this target for most species–island combinations (Appendix C; details in Wood et al. in press). All sampling was conducted at depths between 5 and 25 meters, with the majority of samples taken between 11 and 18 meters (Appendix B). To account for possible depth-dependent variation in parasite abundance, we included the depth at which the host was collected in all across-islands statistical analyses.
Species were chosen to overlap as much as possible with those sampled for the across-islands analysis (Table 2), but the species lists differed due to limitations on some species' availability on Kiritimati. Where the two lists diverged, we kept species pairs as taxonomically and ecologically similar as possible (i.e., Cephaolpholis argus substituted for congener species Cephalopholis urodeta, Plectroglyphidodon dickii substituted for confamilial species Stegastes aureus). Fish were collected by scuba divers using three-pronged spears (for fish >10 cm) or microspears and barrier nets (for fish <10 cm). All collections were conducted between 7 and 15 m depth.
All fish were frozen immediately after collection, and total length, standard length, fork length, and mass were recorded after freezing. In all analyses, total length was used as a proxy for fish body size. Fish were kept frozen before being thawed for dissection.
For the across-islands analysis, fish counts were conducted by a team of divers at each island to characterize the fish assemblage and estimate the abundance, biomass, and size-structure of all coral reef fish species >3 cm TL. Fish were identified to species and counted in belt transects constrained to depths between 7 and 15 meters in fore-reef habitat. Three belt transects were completed at each station. In each transect, divers tallied all fishes ≥20 cm TL along a 25-m long × 4-m wide swath on the first pass and tallied all fishes <20 cm TL along a 25-m long × 2-m wide swath on the return pass. Surveys were conducted in 2005 on Kingman, Palmyra, Tabuaeran, and Kiritimati (Sandin et al. 2008), in 2005 and 2010 on Teraina (S. Sandin, unpublished data), and biennially between 2002 and 2010 on Jarvis (Williams et al. 2011). The total number of transects was 25 on Kingman, Tabuearan, and Kiritimati, 26 on Palmyra, 9 on Teraina, and 29 on Jarvis. Detailed methods are described in Friedlander et al. (2010).
Parasite sampling
We did not count mobile skin parasites or micropredators, as these are easily lost when the host is captured, and we did not search for myxozoan parasites, but all other metazoans should have been easily detected with our protocol. Briefly, for each dissected fish we examined fins, gills, one eye, one filet, inner and outer surfaces of skin, intestine, spleen, heart, and liver under magnification. For all bilateral organs where we examined only one side, parasite counts were doubled. We identified parasite species to the lowest possible taxonomic level (Appendix E) and photographs of each parasite species (along with detailed images of diagnostic morphological features) and voucher specimens are available for examination by request to the corresponding author. As general guides for identifications we used Kabata (2003) and Yamaguti (1963) for copepods, Schultz (1969) for isopods, Gibson et al.(2002) and Bray et al.(2005, 2008) for trematodes, Skryabin (1991) for nematodes, and Khalil et al.(1994) for cestodes. We supplemented with taxonomic literature from the Indo-Pacific region, where available, to achieve more taxonomically resolved identifications (Rigby and Adamson 1997; Rigby and Font 2001; Rigby and Font 1997; Rigby et al.1999).
Each parasite was classified according to its broad taxonomic group (Subphylum Crustacea, Class Monogenea, Class Trematoda, Phylum Nematoda, Class Cestoda), transmission strategy (direct versus trophic transmission), and host specificity (ranked 1–6 based on Brusca [1981], Sasal et al. [1998], and Jones et al. [2007]), with 1 indicating high specificity). We defined "specialists" as those parasites known to use a narrow range of host species for the stage in the life cycle most likely to parasitize a fished species (e.g., the adult stage of trophically transmitted parasites). Because the natural histories of the parasites we detected are poorly known, we surmised life history traits for each parasite based on its membership in higher-order taxonomic groups (i.e., Phylum, Subphylum, or Class level), based on previous assessments (Brusca 1981; Sasal et al. 1998; Jones et al. 2007). While this is a coarse approach, host specificity is known to be phylogenetically conserved within these higher-order taxonomic groups (Sasal et al. 1998; Mouillot et al. 2006), and until life cycles and species identities are worked out for the parasite fauna of the Northern Line Islands, this approach is a strong approximation for understanding how parasite traits might mediate the direction of parasites' response to anthropogenic environmental change.
Oceanography
Values were extracted from the Ocean Color Radiometry Online Visualization and Analysis Monthly Data tool in NASA Giovanni's Ocean Portal (NASA 2013). We used time-averaged data from MODIS on Aqua at a resolution of 4 km, for the eight years prior to fish sampling: May 2002 to June 2010. To derive a single, island-level estimate of [chl-a] for each island in the across-islands analysis, we placed a buffer zone between shore and the 30-m depth contour. We then averaged the [chl-a] value of all grid cells immediately outside that buffer (Appendices A and F). This approach removes cells that can have inflated [chl-a] values due to reflectance from land and shallow bottom substrate (Gove et al.2013). For the within-island analysis, we assigned each Kiritimati Island sampling site the [chl-a] value associated with the nearest grid cell outside the buffer zone (Appendix G).
Statistical analysis: Parasite abundance
For the across-islands data set, we analyzed the abundance of every host–parasite combination that was observed in at least one host individual on at least n – 3 islands, where n was the total number of islands where that host species was collected, for a total of 45 host–parasite combinations (out of 78 total combinations detected). For each host–parasite combination, we used a generalized linear mixed effects model (GLMM) with negative binomial error structure and correction for zero-inflation to assess the response of parasite abundance to fishing pressure, with fishing status (fished versus unfished) and productivity (measured as mean [chl-a] for each island) as fixed factors and island (Jarvis, Kingman, Palmyra, Teraina, Tabuaeran, Kiritimati) as a random factor. Two covariates with the potential to influence parasite abundance were also included: body size of the host (measured as total length) and depth of collection of the host. These analyses were performed with the glmmadmb function (package glmmADMB) in R (R 2.11.1 GUI 1.34, R Foundation for Statistical Computing, 2010). Since many statistical tests were performed, we applied a correction for multiple comparisons (false discovery rate, or FDR correction) to all p values within this suite of statistical tests (Benjamini and Hochberg 1995).
For the within-island data set, we analyzed the abundance of every host–parasite combination that was observed in more than two host individuals, for a total of 47 host–parasite combinations (out of 58 total combinations detected). Across the 25 collection sites, fishing pressure and productivity were correlated and therefore represented collinear predictors of parasite abundance (R² = 0.400, t23 = 3.91, p = 0.0007; Fig. 1c). The influence of collinear predictors on a response variable cannot be distinguished by statistical means and attempts to run statistical models with collinear predictors can result in unstable parameter estimates, inflated variance for parameter estimates and erroneous outcomes of hypothesis testing (Dormann et al. 2013). To circumvent this constraint, we collapsed the collinear predictors into a single predictor in all statistical models – specifically, we retained the productivity variable, renamed it "productivity–fishing gradient value", and interpreted the response of parasite abundance to this predictor as a joint function of productivity and fishing pressure (as recommended by Dormann et al. 2013). Though previous studies conducted on Kiritimati have circumvented this problem by limiting the analysis to a subset of the collection sites for which productivity and fishing pressure are not collinear (Walsh 2011), this option was unavailable for the present analysis because fewer replicates were collected across the productivity–fishing gradient in this study than in Walsh (2011), and restriction of collection sites would have reduced replication to unacceptable levels.
For the within-island analysis, we used a generalized linear model (GLM) with negative binomial error and zero-inflation to assess the response of parasite abundance to the productivity–fishing pressure gradient for each host–parasite combination. Because productivity and fishing pressure are collinear predictors (Fig. 1c), we used only the ecological variable of primary interest (productivity) as a predictor, excluding fishing pressure (as recommended by Dormann et al. 2013). We renamed this predictor "productivity–fishing gradient value", where increasing values of this predictor indicate increasing values of both productivity and fishing pressure (as recommended by Dormann et al. 2013). We considered in our interpretation of results the fact that significant relationships between the productivity–fishing gradient valueand parasite abundance might be affected by either productivity or fishing pressure. Productivity–fishing gradient value was included in statistical models as a fixed factor. One covariate with the potential to influence parasite abundance was also included: body size of the host (measured as total length). Depth was not included as a covariate, because collection depth was held constant (7–15 m). These analyses were performed with the glmmadmb function in R and we applied the FDR correction for multiple comparisons to all p values within this suite of statistical tests (Benjamini and Hochberg 1995).
To investigate differences in the response to productivity and fishing among parasite taxa detected in the across-islands data set, we performed meta-analyses. For effect size estimates, we used regression coefficients for the effect of productivity on abundance of each parasite in each host, extracted from the GLMMs described above. All analyses were performed with the metafor package in R. We began by calculating a cumulative effect size of productivity across all host–parasite combinations, using a fixed-effects model weighted by the inverse of the variance for each effect size, to test Hypothesis 1. We tested for heterogeneity within these effect sizes with the test statistic QT. Where heterogeneity was detected, we hypothesized that it was due to underlying ecological differences among host–parasite combinations. We tested our hypotheses with several meta-analytic fixed-effects general linear models. One model (Model 1) tested our a priori hypothesis that parasite higher-order taxonomic grouping would influence the mean response of parasites to productivity and another (Model 2) tested a priori hypotheses about the interaction between parasite and host traits. Model 1 included the moderator higher order taxonomic grouping of the parasite (Crustacea, Monogenea, Trematoda, Cestoda, Nematoda), and was designed to test Hypothesis 2a. Because this model explicitly tested for heterogeneity within and between taxonomic groupings, the single acanthocephalan taxon we detected was excluded from this analysis. Model 2 included parasite transmission strategy (direct versus trophic; Hypotheses 2a and 2b), host specificity (ranked 1–6 based on Brusca [1981], Sasal et al. [1998], and Jones et al. [2007], with 1 indicating high specificity; Hypothesis 2c), response of host abundance to productivity (standardized coefficient for the effect of productivity on host density from ANOVA models performed within host species; Hypothesis 2b), and the interaction between host response and parasite transmission strategy (Hypothesis 2b). We also used this procedure to derive mean effect sizes for the effect of fishing on parasite abundance (i.e., substituted as effect sizes the regression coefficients for the effect of fishing on abundance of each parasite in each host) for each higher order taxonomic grouping (Wood et al., in press).
To investigate the degree to which results from the within-island analysis were consistent with results from the across-islands analysis (Hypothesis 3), we compared effect sizes for each of the parasite higher-order taxonomic groupings (Crustacea, Monogenea, Trematoda, Cestoda, Nematoda) between the two data sets. We began by calculating mean effect sizes for the within-island data set by building a fixed-effects general linear model with the moderator higher order taxonomic grouping of the parasite (Crustacea, Monogenea, Trematoda, Cestoda, Nematoda, Acanthocephala; Model 3). Another model (Model 4) included the moderator parasite transmission strategy (direct versus trophic). To test for correspondence between the across-islands and within-island data sets, we ran a linear model predicting the z-score for the effect of the Kiritimati productivity–fishing gradient on abundance of Crustacea, Monogenea, Trematoda, Cestoda, and Nematoda (from Model 3) with two factors: (i) z-score for the effect of productivity on abundance of these parasite taxa from the across-islands analysis (from Model 1) and (ii) z-score for the effect of fishing on abundance of these parasite taxa from the across-islands analysis (from Wood et al., in press). We performed a log+1 transformation on the response variable to satisfy the assumption of normality. The form of the model was (transformed z-score for effect of productivity–fishing gradient in within-island analysis) ~ (z-score for effect of fishing in across-islands analysis) + (z-score for effect of productivity in across-islands analysis). Acanthocephalans were not included because only one acanthocephalan was detected in the across-islands data set. This analysis allowed us to circumvent the problem of collinearity of fishing and productivity in the within-island data set: instead of attempting to disentangle the influence of fishing versus productivity as predictors of parasite abundance, we quantified their joint effect on parasite abundance and then asked whether the response of parasites to the within-island productivity–fishing gradient was consistent with the patterns uncovered in the across-islands analysis: (i) a positive effect of productivity on trophically transmitted parasites, but not on directly transmitted parasites (described above), and (ii) a positive effect of fishing on directly transmitted parasites and a negative effect of fishing on trophically transmitted parasites (Wood et al., in press).
Statistical analysis: Parasite diversity
We also tested for differences in parasite taxonomic diversity as a function of productivity. Because differences in sampling effort can confound the comparison of diversity among communities, we accounted for differences in number of hosts sampled and number of parasites detected by calculating richness estimates. We included in our analysis all parasite taxa detected, including those that were excluded from the abundance analyses described above due to their low abundance. We used the non-parametric jackknife estimator to project parasite taxon richness at the saturation of the species accumulation curve (Zelmer and Esch 1999), calculated using the SPECIES package in R. This analysis was conducted for each host–island combination in the across-islands analysis, and for each host–site-grouping combination (Fig. 1c) for the within-island analysis.
For the across-islands analysis, we ran a mixed-effects general linear model with island-level productivity ([chl-a]), fishing status (fished versus unfished), and their interaction as fixed effects and island and host species as random effects, where replicates were jackknife-estimated parasite taxon diversity for each host–island combination (Hypothesis 1). We included the covariates depth (average depth at which hosts of a given species were collected) and host body size (average host TL within host species) to ensure that these factors were not driving patterns in parasite diversity. Covariates were removed from the model through backwards elimination if they were not significant at α = 0.10. This analysis was conducted using the lmer function (package lme4) in R, and p values were extracted with pvals.fnc (package languageR).
To evaluate how parasite taxon diversity varied among sites within Kiritimati Island, we split sites into three natural groupings in [chl-a]–fishing pressure space (Fig. 1c). We needed to group sites to increase replication (i.e., number of hosts sampled) and thereby allow calculation of jackknife estimates of parasite taxon diversity. We chose to create three roups as a trade-off between (i) maximizing replication within groups and (ii) evaluating parasite taxon diversity across a variety of levels of the joint productivity–fishing pressure gradient. We assigned sites to groups of "low", "intermediate", and "high" productivity–fishing effort by considering their position along the regression line in Fig. 1c> (Appendix D). We excluded three sites with extreme values of either [chl-a] or fishing pressure that prevented categorizing them into one of the three groups. Where a site might logically be included in either of two groups, it was included in the group where its contribution to replication would equalize replication across groups. We then ran mixed-effects general linear models with productivity–fishing effort site-grouping (i.e., group 1, group 2, group 3; Fig. 1C) as a fixed effect and host species as a random effect, where replicates were jackknife-estimated parasite taxon diversity for each host–site-grouping combination (Hypothesis 3). We included the covariate host body size (average host TL within host species) to ensure that this factor was not driving patterns in parasite diversity. This analysis was conducted using the lmer function (package lme4) in R, and p values were extracted with pvals.fnc (package languageR).
Literature cited
Benjamini, Y., and Y. Hochberg (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society B 57:289–300.
Bray, R., D. Gibson, et al. (2005). Keys to the Trematoda, Vol. 2. CABI.
Bray, R., D. Gibson, et al. (2008). Keys to the Trematoda, Vol. 3. CABI.
Brusca, R. C. (1981). A monograph on the Isopoda Cymothoidae (Crustacea) of the eastern Pacific. Zoological Journal of the Linnean Society 73:117–199.
DeMartini, E., A. Friedlander, et al. (2008). Differences in fish-assemblage structure between fished and unfished atolls in the northern Line Islands, central Pacific. Marine Ecology Progress Series 365:199–215.
Dormann, C. F., J. Elith, et al. (2013). Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography 36:27–46.
Friedlander, A. M., S. A. Sandin, et al. (2010). Spatial patterns of the structure of reef fish assemblages at a pristine atoll in the central Pacific. Marine Ecology Progress Series 410:219–231.
Gibson, D., A. Jones, et al. (2002). Keys to the Trematoda, Vol. 1. CABI.
Gove, J. M., G. J. Williams, et al.(2013). Quantifying climatological ranges and anomalies for Pacific coral reef ecosystems. PLoS One 8: e61974.
Heywood, K. J., E. D. Barton, et al. (1990). The effects of flow disturbance by an oceanic island. Journal of Marine Research 48:55–73.
Jones, C. M., L. Nagel, et al. (2007). Host specificity of two species of Gnathia (Isopoda) determined by DNA sequencing blood meals. International Journal for Parasitology 37: 927–935.
Kabata, Z. (2003). Copepods parasitic on Fishes: Keys and Notes for Identification of British Species. Shrewsbury.
Khalil, L. F., A. Jones, et al. (1994). Keys to the Cestode Parasites of Vertebrates. CABI.
Kiribati National Statistics Office. (2007). Kiribati 2005 Census – Volume 2: Analytical Report. Tarawa, Kiribati: Kiribati National Statistics Office.
Maragos, J. E., A. M. Friedlander, et al. (2008a). US coral reefs in the Line and Phoenix Islands, Central Pacific Ocean: Status, threats, and significance. In: Coral Reefs of the USA. Eds: B. M. Riegl and R. E. Dodge. Springer. pp. 643–654.
Maragos, J. E., J. Miller, et al. (2008b). US coral reefs in the Line and Phoenix Islands, Central Pacific Ocean: History, geology, oceanography, and biology. In: Coral Reefs of the USA. Eds: B. M. Riegl and R. E. Dodge. Springer. pp. 595–641.
Mouillot, D., B. R. Krasnov, et al. (2006). Conservatism of host specificity in parasites. Ecography 29:596–602.
NASA. (2013) Giovanni Ocean Color Radiometry Online Visualization and Analysis Monthly Data. Accessed 23 Feb 2013, from http://gdata1.sci.gsfc.nasa.gov/daac-bin/G3/gui.cgi?instance_id=ocean_month.
Rigby, M. C., and M. L. Adamson (1997). Spirocamallanus species of French Polynesian coral reef fishes. Canadian Journal of Zoology 75:1270–1279.
Rigby, M. C., and W. F. Font (1997). Redescription and range extension of Spirocamallanus istiblenni Noble, 1966 (Nematoda: Camallanidae) from coral reef fishes in the Pacific. Journal of the Helminthological Society of Washington 64:227–233.
Rigby, M. C., and W. F. Font (2001). Statistical reanalysis of the distinction between Spirocamallanus istiblenni and S. monotaxis (Nematoda: Camallanidae). Journal of Parasitology 87:1210–1213.
Rigby, M. C., C. M. Lo, T. H. Cribb, L. Euzet, E. Faliex, R. Galzin, J. C. Holmes, and S. Morand (1999). Checklist of the parasites of coral reef fishes from French Polynesia, with considerations on their potential role in these fish communities. Cybium 23:273–284.
Sandin, S. A., J. E. Smith, et al. (2008). Baselines and degradation of coral reefs in the Northern Line Islands. PLoS One 3: e1548.
Sasal, P., Y. Desdevises, et al. (1998). Host-specialization and species diversity in fish parasites: Phylogenetic conservatism? Ecography 21:639-643.
Schultz, G. A. (1969). How to know the marine isopod crustaceans. William C. Brown Company.
Skryabin, K. I. (1991). Keys to Parasitic Nematodes, Volumes 1–4.
Walsh, S. M. (2011). Ecosystem-scale effects of nutrients and fishing on coral reefs. Journal of Marine Biology: Article ID 187248.
Williams, I. D., B. L. Richards, et al. (2011). Differences in reef fish assemblages between populated and remote reefs spanning multiple archipelagos across the western and central Pacific. Journal of Marine Biology: Article ID 826234.
Wood, C. L., S. Sandin, et al. (in press at Ecology). Fishing drives declines in fish parasite diversity and has variable effects on parasite abundance: Evidence from fished and unfished coral atolls of the Line Islands.
Yamaguti, S. (1963). Parasitic Copepoda and Branchiura of Fishes. Interscience Publishers.
Zelmer, D. A., and G. W. Esch (1999). Robust estimation of parasite component community
richness. Journal of Parasitology 85:592–594.