Mlynarek, J.J., W. Knee, and M.R. Forbes. 2014

Host phenology, geographic range size and regional occurrence explain interspeciﬁ c variation in damselﬂ y – water mite associations

Julia J. Mlynarek , W. Knee and Mark R. Forbes

J. J. Mlynarek ([email protected]) and M. R. Forbes, Dept of Biology, Carleton Univ., Nesbitt Building, 1125 Colonel By Drive, Ottawa, ON K1S 5B6, Canada. – W. Knee, Canadian National Collection of Insects, Arachnids and Nematodes, Agriculture and Agri-Food Canada, 960 Carling Avenue, K.W. Neatby Building, Ottawa, ON K1A 0C6, Canada.

In this study, we tested which host species ’ characteristics explain the nature and level of parasitism for host damselfl y (Coenagrionidae) – water mite (Arrenuridae) parasite associations. Prevalence and intensity of mite parasites, and mite species richness were examined in relation to geographic range size, regional occurrence, relative local abundance, phenology and body size of host damselfl y species. A total of 7107 damselfl y individuals were collected representing 16 species from 13 sites in southeastern Ontario and southwestern Quebec, Canada. Using comparative methods, diff erences in prevalence and intensity of parasitism could be predicted by a host species ’ geographic range and phenology. Barcoding based on Cytochrome Oxidase I revealed 15 operational taxonomic units (OTUs) for mite species. Th e number of mite OTUs known to infest a given host species was explained by a host species ’ regional occurrence. Our fi ndings demonstrate the need to measure factors at several ecological scales in order to understand the breadth of evolutionary interactions with host – parasite associations and the selective ‘ milieu ’ for particular species of both hosts and parasites.

Evolutionary history of the association between two species Th us, not only the size of a species’ geographic range, but is expected to have an eff ect on present-day interactions, but also actual representation of host species across the region so are ecological factors (Th ompson 2005). In most cases, of interest could be important. It is expected that a region- that explanatory power of ecological factors depends on ally well-represented host species will come into contact with the host – parasite system under study, perhaps because of more parasite species than host species less well represented certain life history traits (Poulin 2007). Several ecological regionally (cf. Durrer and Schmid-Hempel 1995). Species factors have been reported to explain variation in parasitism are not only unevenly distributed across space, but there are between host species, such as size or landscape confi guration also diff erences in the representation of a species through of habitat (Tella et al. 1999); size, age, reproductive stage time (phenology). Th ese phenological diff erences between or phenology of the host species (Tripet and Richner 1997, species are expected to infl uence their interactions, especially Arneberg et al. 1998, Krasnov et al. 2006, Munoz et al. between groups that are active during diff erent times of the 2007), and size, virulence and life history of the parasite year (Altizer et al. 2006, Locklin and Vodopich 2010). One (Krasnov et al. 2004). Although there are many predictors to expectation is that density of infective stages of parasites explain parasitism, host body size, geographic distribution, would follow density of potential hosts throughout the and population density, seem to be strong predictors in most season of activity (Forbes et al. 2012). More specifi cally, host – parasite associations (Kamiya et al. 2014). the density of infective stages of parasites and measures of Host species ’ characteristics such as geographic range parasitism itself (e.g. prevalence and intensity) should be size, regional occurrence, local abundance, and phenology highest when potential host individuals of one or more spe- have been off ered as explanations for interspecifi c variation cies are most abundant. Local abundance of a host species in parasitism (Morand and Krasnov 2010). As a host spe- might infl uence its interactions with co-occurring species. cies range increases, so does the number of species it interacts Th e intensity of interactions increases with increased host with (Krasnov et al. 2004, Ilvonen et al. 2011) because as a abundance because the probability of encounters with other host species range increases, so too does the probability that species increases (Poulin 1991, Arneberg et al. 1998, Krasnov the host species will come into contact and be parasitized by et al. 2002, Patterson et al. 2008). As a host species becomes more parasites (Price et al. 1988). However within its entire more abundant in a habitat, it is thus expected to be under range, the host species is not evenly distributed in either higher pressure from the parasites (Arneberg et al. 1998). space or time (Th ompson 2005), because there are habitats Our understanding of parasite associations in insect where individuals cannot survive and reproduce well, if at all. hosts is not as well documented as that of vertebrate hosts.

Early View (EV): 1-EV In mites associated with ants, mite prevalence was explained parasitism. We tested the generality of regional occurrence, by the degree of sociality, the size and the generic identity local abundance and body size in explaining interspecifi c of the host species (Campbell et al. 2013). Durrer and variation in parasitism, because they have been shown to Schimd-Hempel (1995) did a comprehensive study of be important factors in vertebrate– parasite systems (Poulin bumblebee parasites and found that parasite load was posi- 2007). tively correlated with average colony size, local abundance and geographic distribution of the host. Th e geographic distribution of damselfl y hosts explained in part the preva- Methods lence and intensity of gregarine endoparasites (Mlynarek et al. 2012). In contrast, host geographic distribution was a Sites poor predictor of Arrenurus (Arrenuridae) water mite prevalence and intensity when comparing pairs of closely related Th irteen sites were chosen and were visited on a weekly damselfl y host species from single sites (Mlynarek et al. basis for eight weeks in June and July 2011 (Fig. 1). Sites 2013). were spread along a 400 km longitudinal transect from the Th e association between damselfl ies and Arrenurus water Queen ’ s Univ. Biological Station (QUBS), 60 km northwest mites is nonetheless a good model system for host – parasite of Kingston, Ontario to Johnville Bog 20 km east of the city studies because of the nature of the obligate interaction char- of Sherbrooke, Quebec to ensure a high number of species acterized by a direct life cycle, the similarity in trophic level interactions occurring (Supplementary material Appendix 2, of the host species, and the known diff erences in distribution Table A1). Th e most western sites (QUBS) were collected at and expected diff erences in phenology and regional occur- the beginning of the week with the eastern sites collected at rence of diff erent host species (Forbes and Robb 2008). the end of any given week. Larval Arrenurus mites contact their future damselfl y hosts Some sites were closer together than others, so there in the water column. As the formerly larval damselfl y ecloses, could have been a spatial bias and non-independence of the the mites move from the exuviae onto the teneral adult and sites. Before running our main analysis, we ran spatial auto- embed their gnathosome through the unsclerotized cuticle correlation on overall parasite prevalence and intensity with into the hemolymph (Smith et al. 2010). Th e mites then each host species run separately and determined that all the form a blind ended feeding tube known as a stylostome and sites were independent from one another. In addition, we start feeding (Smith et al. 2010). Damselfl ies can resist mites interspersed bog and marsh sites across the transect in order by grooming loosely attached mites, or through melanisa- to sample as many host species as possible. Th is likely had tion once a mite has formed a stylostome (Forbes et al. 1999, the observed eff ect of reducing potential autocorrelation in Leung et al. 1999, Mlynarek et al. 2014). species abundance across the transect. In this study, we tested the extent to which geographic range, regional occurrence, local abundance, phenology and body size of host species are good predictors of prevalence Host sampling and intensity of water mite parasitism or of proxies of species richness of mite parasites. Damselfl ies were collected weekly using aerial sweep nets Our study was restricted to damselfl ies of the family by two collectors. At each site, one collector assessed host Coenagrionidae, but we expand upon Mlynarek et al. (2013) species assemblages by collecting individuals, while walking by examining more host species across many sites and by along the length of the shoreline for 20 min. Collecting all using parasite species diversity as an additional measure of damselfl ies within this allotted time also allowed us to index

Figure 1. Map of study sites included in this study in southeastern Ontario and southwestern Quebec.

2-EV relative local abundance of each damselfl ies species. After were sent to the Barcode of Life Database, BOLD; Ͻ www. the initial 20 min collecting period, sampling continued for boldsystems.orgϾ (Ratnasingham and Hebert 2007) for up to three hours or until 30 individuals of each species at barcoding using Cytochrome Oxidase I (COI). Randomly each site were collected. All the damselfl ies were stored in chosen mites were sent to the Canadian Centre for DNA 95% ethanol in individual vials. In the lab, damselfl ies were Barcoding (CCDB; Ͻ www.ccdb.ca Ͼ) at the Biodiversity re-identifi ed and examined for water mites. Inst. of Ontario to perform DNA barcoding using stan- dard, high throughput methods (Ivanova et al. 2006). DNA extractions from 285 larval mites were performed. Th ese Parasite sampling and measures of parasitism larval water mites were selected randomly, based principally on host species and site of host collection. Th is random selec- Damselfl ies were examined for water mites using a ZEISS tion was as follows: if present, ten mites (fi ve from the thorax SteREO Discovery V8 dissecting microscope. All mites and fi ve from the abdomen because thoracic and abdominal were quantifi ed, collected and stored in 95% ethanol. mites are often diff erent species (B. P. Smith pers. comm.)), We used (R ó zsa et al. 2000) defi nitions of prevalence and from every host species at every site were collected. Each mean intensity. Prevalence was defi ned as the proportion of mite was haphazardly chosen from a diff erent host individual host individuals of a particular species infected by at least if there were enough infected individuals. Twelve of sixteen one Arrenurus individual over the total number of hosts col- host species had thoracic mites, and eight of sixteen species lected. Intensity was the number of Arrenurus individuals also had abdominal mites. In those latter species, both tho- on an infected host. We measured the number of Arrenurus racic and abdominal water mites were collected, which is ‘ species ’ hereafter Operational Taxonomic Units (OTUs) why sample size was initially 285. per host species, using a molecular dataset. For most of the Chromatograms were edited and contiguous sequences parameters, we pooled the data from all the sites because ϭ were assembled using Sequencher ver. 4.7 (Gene Codes, there were no signifi cant diff erences in prevalence (F 0.95; Ann Arbor, MI, USA). COI sequences were aligned manu- DF ϭ 12, 80; p ϭ 0.50) or intensity (F ϭ 0.97; DF ϭ 12, 80; ϭ ally in Mesquite ver. 2.74 (Maddison and Maddison 2011) p 0.48) across sites. Analyses were performed in QP3.0 according to the translated amino acid sequence. Sequences (R ó zsa et al. 2000). have been submitted to GenBank (KJ709142-KJ709343). Homologous sequences from four Arrenurus sp. individuals collected from Nehalennia gracilis and N. irene damselfl ies Composite host phylogeny (Mlynarek et al. 2014) and four Enallagma ebrium (Mlynarek To examine potential importance of a host species ’ evolution- et al. 2013) were included in the COI alignment, and seven ary history to its likelihood of being parasitized by specifi c water mite species were selected from GenBank to serve as OTUs of water mites, we created a hypothesized phylogeny outgroups (AB530314, JX838402, JX836526, JX835088, of Coenagrionidae collected in the marshes and bogs of JN018105, JN018109, AB530311). southeastern Ontario and southwestern Quebec, by com- Pairwise distances were calculated using neighbour- * piling recent phylogenies (Chippindale et al. 1999, Brown joining analysis with the uncorrected ( ‘ p ’ ) model in PAUP et al. 2000, May 2002, O’Grady and May 2003, Dumont ver. 4.0b10 (Swoff ord 2003). Phylogenetic analysis of the et al. 2010). Th is approach allowed us to control for evolu- COI dataset was performed using Bayesian inference (BI) tion of the characters among host species. Because molecular in MrBayes ver. 3.1.2 (Huelsenbeck and Ronquist 2001, Ronquist and Huelsenbeck 2003). Th e best-fi t model of phylogenetic information is lacking for many of the damsel- ϩ ϩ fl y species collected, taxonomy is a reasonable substitute to molecular evolution was determined to be GTR I phylogeny (Crozier et al. 2005). We assumed that genera are G, using MrModeltest ver. 2.3 (Nylander 2004). Bayesian monophyletic. We performed analyses of phylogenetic relat- analysis was performed in MrBayes with a Markov Chain edness within the entire species assemblage by assigning the Monte Carlo (MCMC) method, two independent runs, with nucmodel ϭ 4by4, Nst ϭ 6, rates ϭ invgamma, sam- age of the node splits using BLADJ (branch length adjuster) ϭ ϭ module of Phylocom software package (Webb 2000). BLADJ plefreq 1000, four chains one cold and three heated, places internal nodes where age estimates from phylogenetic 10 million generations. Excluding a 10% burn-in, the remain- trees were known. Age estimates were determined through ing trees were used to generate a majority-rule consensus tree. the literature and discussion with experts. BLADJ also sets the remaining branch lengths by spacing the undated nodes in the tree evenly between those dated nodes. Th ere are Host parameters assumptions made, such as the true age of the splits, but this approach reduces variance in branch lengths and the nodal We chose fi ve host species’ characteristics that have been rea- distances are based on evolutionary time and are independent sonably strong explanatory variables of other host– parasite of species richness and levels of resolution within clades. associations (Krasnov et al. 2004, Poulin 2007). Geographic range size (Range) was assessed using the Odonata Central database (Abbott 2007). Th e records in Phylogeny of water mites this database are verifi ed and managed regularly by experts. To determine the size of a host ’ s geographic range, a 2° To address whether host species of sibling species pairs might latitude – longitude grid was superimposed on the geographic share mite OTUs, and to have a preliminary account of host data. Presence– absence was then observed for each cell. Th e species range of those mite OTUs, a subset of water mites area of the species distribution was determined by adding the

3-EV area of each cell where that species was present (see Hassall analysis, 30– 50 individuals of a host species provided appro- 2012 for details). priate measures of parasitism for statistical comparison. Regional occurrence (RO) of a host species was determined simply as the number of sites a host species was collected at within our sampling season. Since we collected at Results 13 sites, the maximum regional occurrence a species can have is 13. Stated otherwise, if the regional occurrence of a Measures of general Arrenurus parasitism host species is 13, that host was collected at all the sites (if 12, at 12 of 13 potential sites, etc.). A total of 7107 damselfl ies were collected (Table 1). Host We assessed relative local abundance by collecting abundance per species per site per week ranged from 61 damselfl ies at a site for 20 min. From this, we could gauge damselfl y individuals collected in Osprey Marsh to 219 how rich the host species assemblage was at a given site and host individuals collected in Stony Swamp. Th ere was varia- the relative abundance of each host species there. Local tion in the cumulative date of emergence of the damselfl y abundance for our study is the proportion of a particular species across the eight weeks (Supplementary material host species over the number of host species collected within Appendix 1, Fig. A1). Based on our analysis, early species 20 min. Because we collected at each site eight times, the included Coenagrion interrogatum , Coenagrion resolutum and relative local abundance is the average local abundance at Enallagma cyathigerum . We collected at least 50% of the each site for eight weeks. Our fi nal measure used for analysis individuals of these species in the fi rst week of collecting. is the mean local abundance (MLA). Th is is the grand mean Th ere are two species, Enallagma aspersum and Enallagma of all the relative local abundances from all the sites. geminatum that emerged late in our collecting. No individ- We assessed phenology of a host species using species uals of these species were collected until the sixth week of accumulation curves (Supplementry material Appendix 1, collecting (Supplementary material Appendix 1, Fig. A1). Fig. A1). We observed the week where we collected 50% of Th ere was also a lot of variation of Arrenurus prevalence and the total individuals of that species. Once the peak of activity intensity between the host species. Prevalence of parasitism for all the species was assessed, we determined peak of dam- varied between 0% in Amphiagrion saucium , Chromagrion selfl y activity by calculating the mean of all the weeks where conditum , Coenagrion interrogatum and Enallagma 50% of the population of a species. We then took, for each geminatum to 37.4% in Enallagma ebrium (Table 1). For the host species, the absolute value of the diff erence between the infected species, mean intensity varied between 1.3 Arrenurus 50% week and the mean of all the weeks. Th rough this, we water mites in Enallagma carunculatum and 13.3 in Ischnura could assess whether the peak fl ight season of a host species verticalis (Table 1) . was near the peak of damselfl y activity across all species. Using digital calipers, we measured wing length from the base to the tip of the right forewing, as a proxy for host body Host phylogeny size. In this case, our measure of wing length is the mean of We collected 16 damselfl y species across all sites which were wing length for each damselfl y species from all individuals of included in our phylogeny of the Coenagrionidae in marshes that species across sites. and bogs of southeastern Ontario and southwestern Quebec (Fig. 2), of which four species (25%) were collected only at one site and two species (12.5%) occurred at all sites Statistics (Table 1). All subsequent statistical analyses were run using this composite phylogeny to control for a potential We evaluated the data using phylogenetically generalized evolutionary signal. least squares (PGLS), which was implemented using the package caper in R (Orme et al. 2012). Fixed eff ect cova- riates included size of host distribution (Range), regional Parasite phylogeny occurrence (RO), phenology (Phenology), grand mean local abundance (MLA) and wing length (WingL). Model selec- A 683bp fragment of COI was amplifi ed from 217 Arrenurus tion was based on Akaike’ s information criterion corrected specimens, a total of 327 characters were parsimony infor- for sample size (AICc ) (Burnham and Anderson 2002). mative and 356 were uninformative. Regression coeffi cients (i.e. β-values on a logit scale) were Bayesian inference was performed on the COI dataset for estimated by model averaging and statistical signifi cance 10 million generations. After burn-in there were 18002 trees, was judged on the basis of sign (positive or negative rela- which were summarized in a 50% majority rule consensus tionship between the predictor and response variables) and tree (TL ϭ 1370, CI ϭ 0.396, RI ϭ 0.927). Th is tree was well the precision of estimates (wherein values with 95% con- supported, with most nodes having moderate to high poste- fi dence intervals that did not overlap with zero were con- rior probabilities (Fig. 3). Th e ingroup was divided into 15 sidered important predictors). Our candidate set of models well-supported clades, OTUs. Average interclade divergence included all additive combinations of the predictor variables. was 10.7% Ϯ 5.8, and average intraclade divergence was To avoid over-parameterization, we did not evaluate interac- 0.7% Ϯ 0.5. Average divergence between OTUs 1 and 2 was tions between variables. 16.6% Ϯ 0.6, divergence between OTUs 3, 4, 5 and 6 was For our analyses of prevalence and intensity, we excluded 7.9% Ϯ 3.2, between OTUs 8, 9, and 10 was 7.5% Ϯ 1.6, the species with sample sizes less than 50 individuals and OTUs 11, 12, 13, 14 and 15 show 9.0% Ϯ 3.2 diver- because of potential sampling bias. Based on power gence. Considering the high level of divergence between

4-EV Table 1. Host parameters of the 16 damselﬂ y species that were collected during the reproductive season of 2011 from 13 sites in southeastern Ontario and southwestern Quebec. N ϭ sample size of each species; RLA ϭ relative local abundance; RO ϭ regional occurrence; OTUs ϭ number of operational taxonomic units; WingL ϭ wing length.

N Prevalence Phenology Range Host species (N inf.) (%) Intensity |d| (10 6 km 2 ) RLA RO WingL OTUs Amphiagrion saucium 8 (0) 0.0 (0 – 36.9) 0.0 (NA) 0.44 1.81 0.050 1 16.03 0 Chromagrion conditum 133 (0) 0.0 (0 – 2.7) 0.0 (NA) 1.44 2.21 0.098 2 21.62 0 Coenagrion interrogatum 52 (0) 0.0 (0 – 6.8) 0.0 (NA) 2.44 2.35 0.058 1 18.90 0 Coenagrion resolutum 542 (15) 2.6 (1.4 – 4.3) 2.1 (1.4 – 2.9) 2.44 6.13 0.046 11 18.21 1 Enallagma aspersum 11 (4) 36.4 (10.9 – 69.2) 3.5 (2.3 – 5.3) 1.56 2.58 0.057 1 19.87 3 Enallagma boreale 242 (32) 12.8 (8.9 – 17.7) 5.9 (4.5 – 7.7) 2.44 6.63 0.046 6 20.00 3 Enallagma carunculatum 123 (3) 2.4 (0.5 – 7.0) 1.3 (1 – 1.7) 2.56 5.20 0.102 3 20.56 ? Enallagma cyathigerum 128 (2) 1.6 (0.2 – 5.5) 3.0 (1 – 4) 3.44 6.50 0.044 6 19.81 3 Enallagma ebrium 1000 (374) 37.4 (34.3 – 40.5) 10.5 (9.2 – 11.9) 0.56 4.35 0.141 12 18.09 7 Enallagma geminatum 3 (0) 0.0 (0 – 70.8) 0.0 (NA) 2.56 2.98 0.047 1 18.48 0 Enallagma hageni 257 (51) 19.8 (15.1 – 25.3) 5.8 (4.3 – 8.3) 0.56 4.25 0.059 8 17.81 4 Enallagma signatum 27 (5) 18.5 (6.3 – 38.1) 5.2 (2.8 – 7.2) 2.56 3.23 0.079 2 20.25 2 Ischnura posita 336 (18) 5.4 (3.2 – 8.3) 1.7 (1.3 – 2.5) 1.56 3.13 0.051 9 15.92 3 Ischnura verticalis 349 (85) 24.4 (19.9 – 29.2) 13.3 (9.9 – 18.4) 0.44 4.36 0.063 13 16.27 6 Nehalennia gracilis 820 (64) 7.8 (6.1 – 9.9) 2.2 (1.8 – 2.8) 0.56 1.51 0.257 5 14.88 2 Nehalennia irene 3076 (395) 12.8 (11.7 – 14.5) 3.3 (3.0 – 3.6) 0.56 4.03 0.486 13 15.60 4

OTUs

4, 5, 8, 9, 10, 11, 15

8, 9, 10, 15

7, 8, 10

7, 8, 9

N/A

9, 10, 14

4, 12

N/A

8, 13, 15

5, 8, 11, 13, 14, 15

1, 2

1, 3, 4, 6

N/A

Figure 2. Composite phylogeny of the Coenagrionidae found at the thirteen sites of southeastern Ontario and southwestern Quebec including the number of Arrenurus OTUs collected from each host species.

5-EV Figure 3. Majority rule consensus tree of 18002 trees generated by Bayesian MCMC analysis (10 million generations) of 683 bp fragment of COI from 224 water mites, 217 ingroup and 7 outgroup specimens (TL ϭ 1370, CI ϭ 0.396, RI ϭ 0.927).

6-EV Table 2. Model selection results for PGLSs evaluating water mite prevalence in relation to host distribution attributes. Table includes the global model and all models with AICc Ͻ 4. Calculations are described in the footnote1 .

Model Log(L) K AIC c Δ i Model likelihood Model prob. R2 Range ϩ PhenologyϪ 38.31 3 84.61 0.00 1.000 0.365 0.66 Range ϩ MLA ϩ PhenologyϪ 37.35 4 86.34 1.73 0.421 0.154 0.67 RO ϩ PhenologyϪ 39.46 3 86.91 2.29 0.317 0.116 0.67 Phenology Ϫ 41.15 2 87.22 2.61 0.271 0.099 0.60 Range ϩ Phenology ϩ WingLϪ 38.05 4 87.74 3.13 0.209 0.076 0.63 RO ϩ Range ϩ PhenologyϪ 38.28 4 88.19 3.58 0.167 0.061 0.61 RO ϩ Phenology ϩ WingLϪ 38.36 4 88.35 3.74 0.154 0.056 0.61 RO ϩ Range ϩ MLA ϩ Phenology ϩ WingLϪ 37.22 6 103.24 17.62 Ͻ 0.001 Ͻ 0.001 0.57

1 K is the number of estimable parameters; AICc ϭ – 2log(L ) ϩ 2K ϩ (2k(K ϩ 1))/(n – K – 1) ; Δ i ϭ AICci – AICcmin; the model likelihood given ϫ ω ϭ R 2 the data is exp (– 0.5 Δ AIC c ); model probability estimated as i (exp( – Δ i))/ Sr exp( – Δ i); R estimated from PGLS.

OTUs and the low divergence within each clade, as well as Additionally, other models were supported by the data the strong support for each clade in the phylogenetic recon- (AIC c Ͻ 4; Table 3). Model averaged regression coeffi cients struction, it appears that each OTU may be distinct. We are (Supplementary material Appendix 2, Table A3) indicated β ϭϪ not implying they are new species and the possibility still a negative association with phenology ( Phenology 2.46, exists that there are many more species present if speciation 95% CI: – 3.81 to – 1.11; Fig. 5a) and a unimodal associa- β ϭ (and adaptive radiation) has occurred recently. tion with range ( Range 1.05, 95% CI: 0.39 – 1.71; Fig. 5b). Host body size and local abundance had little predictive power (Supplementary material Appendix 2, Table A3). Host species characteristics as predictors Our PGLS examining OTU richness in host species in relation to host species characteristics indicated that Our PGLS examining variation in the prevalence of water regional occurrence as the only signifi cant predictor vari- mite parasitism in relation to host species characteris- ϭ ϭ ϭ ω ϭ able (Log(L ) – 27.29, K 2, AIC c 59.71, i 0.29); tics indicated that Range and Phenology were important several other models also were supported by the data ϭϪ ϭ ϭ predictor variables (Log(L ) 38.31, K 3, AIC c 84.61, ( Δ AICc Ͻ 4 ; Table 4). Model averaged regression coeffi cients ω ϭ i 0.37); several other models also were supported by (Supplementary material Appendix 2, Table A4) indicated a Ͻ the data ( Δ AICc 4; Table 2). Model averaged regression positive association between regional occurrence and OTU coeffi cients (Supplementary material Appendix 2, Table β ϭ richness ( RO 0.362, 95% CI: 0.19 to 0.53; Fig. 6). Host A2) showed a negative association between water mite body size and local abundance had little predictive power β ϭϪ prevalence and phenology ( Phenology 10.10, 95% CI: (Supplementary material Appendix 2, Table A4). – 14.91 to Ϫ5.29; Fig. 4a) and a unimodal association with β ϭ range ( Range 2.75, 95% CI: 0.82– 4.68; Fig. 4b). Host body size and local abundance had little predictive power Discussion (Supplementary material Appendix 2, Table A2). Th e best fi tting model in our candidate set of PGLS Th e goal of this study was to determine what host species ’ investigating water mite intensity indicated that again characteristics predict the levels of parasitism (prevalence and range and phenology were signifi cant predictor vari- intensity) and the diversity of parasite interactions. Based on ϭ ϭ ϭ ω ϭ ables (Log( L ) – 27.35, K 3, AIC c 63.71, i 0.31). our results, it is clear that host distribution characteristics

Figure 4. Summary of the important predicting factors explaining Arrenurus spp. prevalence. (a) Phenology and (b) range in relation to Arrenurus spp. prevalence. Cc ϭ Chromagrion conditum ; Ci ϭ Coenagrion interrogatum ; C r ϭ C. resolutum ; E b ϭ Enallgma boreale ; Ec ϭ E. carunculatum ; Ecy ϭ E. cyathigerum ; E e ϭ E. ebrium ; E h ϭ E. hageni ; I p ϭ Ischnura posita ; I v ϭ I. verticalis ; Ng ϭ Nehalennia gracilis ; N i ϭ N. irene.

7-EV Table 3. Model selection results for PGLSs evaluating water mite mean intensity in relation to host distribution attributes. Table includes the global model and all models with AICc Ͻ 4. Calculations are described in the footnote of Supplementary material Appendix 2, Table A1.

Model Log(L) K AIC c Δ i Model likelihood Model prob. R 2 Range ϩ PhenologyϪ 27.35 3 63.71 0.000 1.000 0.307 0.56 Range ϩ MLA ϩ PhenologyϪ 25.21 4 64.14 0.436 0.804 0.247 0.66 RO Ϫ 29.90 2 65.13 1.417 0.492 0.151 0.40 RO ϩ MLAϪ 28.89 3 66.78 3.068 0.216 0.066 0.44 RO ϩ PhenologyϪ 29.30 3 67.60 3.889 0.143 0.044 0.45 RO ϩ Range ϩ MLA ϩ Phenology ϩ WingLϪ 25.07 6 78.93 15.23 Ͻ 0.001 Ͻ 0.001 0.55 can predict damselfl y– water mite associations. Diff erences in Th is trend is mostly driven by two species, I. verticalis and prevalence and intensity of Arrenurus were explained mostly E. ebrium , these are species that are very common at all sites by two host characteristics, range and phenology, as mea- and have the highest numbers of parasite species. sured in this study. Th e diff erences in number of OTUs can Geographic range alone was not able to predict the be predicted by a host species ’ regional occurrence. prevalence and intensity of parasitism. Host phenology was Host geographic range and phenology are important in a strong predictor of prevalence and intensity in Arrenurus predicting prevalence and intensity. A positive relationship spp. parasitism. In systems where hosts demonstrate cyclical between size of host geographic range and the prevalence life history traits, such as the yearly life cycle of many insects, and intensity of parasitism has previous been reported. Price infection levels are expected to vary with time within that et al. (1988) proposed the geographic range hypothesis with cycle (Forbes et al. 2012). Th is has been often demonstrated the prediction that more widespread hosts will be under in infectious diseases (Altizer et al. 2006), bees infected by higher parasite pressure. Tella et al. (1999) supported this internal Crithidia parasites (Gillespie 2010), insects infected hypothesis in a study of diurnal birds of prey parasitized by by gregarines (Zuk 1987, Locklin and Vodopich 2010). hematozoan blood parasites where they proposed that the As previous studies (Gillespie 2010), a seasonal unimodal pattern was potentially due to the evolutionary history of pattern of parasitism was demonstrated in infection levels. specifi c host – parasite associations. Similarly, the prediction Th is seasonal pattern could be linked to host density where of the geographic range hypothesis was supported in a study host density may be either the size of the population of one comparing two species of Calopteryx in Finland (Ilvonen host species (Forbes et al. 2012) or the aggregate density of et al. 2011). In contrast, the geographic range hypothesis was all hosts in the species assemblage (Gillespie 2010). Th ere is not supported in two studies comparing gregarine and water the expectation that parasites time their life cycle to coincide mite parasitism in several pairs of closely related coenagri- with that of their hosts to increase their chances of fi nding a onid damselfl ies where the less widespread species of the spe- host; in this case, the generalist parasite will track when there cies pairs had higher parasitism levels (Mlynarek et al. 2012, is the highest density of potential hosts or the peak of the 2013). In the current study based on an increased number host fl ight season. of non-phylogenetically paired host species, we fi nd some Even if generalist parasites can survive on many host support for the geographic range hypothesis. In our results, species they demonstrate preferences in host use (Tripet and there is a trend towards a unimodal pattern where the species Richner 1997). In this study, generalist parasite species rich- with the smallest and largest geographic distributions have ness or number of OTUs infecting each damselfl y species, the lowest prevalence and intensity of Arrenurus parasitism. was not predicted by host range size or phenology but only by

Figure 5. Summary of the important predicting factors explaining Arrenurus spp. mean intensity. (a) Phenology and (b) range in relation to Arrenurus spp. mean intensity. Cc ϭ Chromagrion conditum ; Ci ϭ Coenagrion interrogatum ; C r ϭ C. resolutum ; E b ϭ Enallgma boreale ; Ec ϭ E. carunculatum ; Ecy ϭ E. cyathigerum ; E e ϭ E. ebrium ; E h ϭ E. hageni ; I p ϭ Ischnura posita ; I v ϭ I. verticalis ; Ng ϭ Nehalennia gracilis ; N i ϭ N. irene.

8-EV Table 4. Model selection results for PGLSs evaluating water mite species richness (number of OTUs per host species) in relation to host distribution attributes. Table includes the global model and all models with AICc Ͻ 4. Calculations are described in the footnote of Supplementary material Appendix 2, Table A1.

Model Log(L) K AICc Δ i Model likelihood Model prob. R 2 RO Ϫ 27.39 2 59.71 0.000 1.000 0.291 0.55 RO ϩ Phenology Ϫ 26.19 3 60.38 0.671 0.715 0.208 0.61 RO ϩ Range Ϫ 26.60 3 61.19 1.482 0.477 0.139 0.59 RO ϩ Phenology ϩ WingL Ϫ 25.23 4 62.09 2.378 0.304 0.089 0.62 RO ϩ WingL Ϫ 27.23 3 62.46 2.755 0.252 0.073 0.52 RO ϩ MLA Ϫ 27.30 3 62.61 2.897 0.235 0.068 0.52 RO ϩ Range ϩ WingL Ϫ 25.61 4 62.85 3.138 0.208 0.061 0.63 RO ϩ Range ϩ MLA ϩ Phenology ϩ WingL Ϫ 24.60 6 70.54 10.831 0.004 0.001 0.60 host regional occurrence. Host species present at more sites size to be a good predictor variable. Studies in ectoparasites were infected by more OTUs. Durrer and Schmid-Hempel of rodents (Krasnov et al. 2004), parasites of fi ssiped carni- (1995) found similar results when comparing bumblebee vores (Lindenfors et al. 2007), and bat fl ies on bats (Patterson species infected by all their parasite taxa. One explanation et al. 2008) clearly demonstrated that host size mattered. could be the fact that if a species is present at more sites it Even in the invertebrate-parasite literature, host size has been will encounter more parasite species. We propose that the shown to be an important predictor in intraspecifi c studies infl uence of host regional occurrence may be explained by (Forbes and Baker 1991). Th ere are other studies that show the fact that widespread hosts have not adapted species- the opposite; there was a lack of support for a relationship specifi c recognition at particular sites and resistance or eva- between ectoparasites infecting fl ying squirrels and host size sion does not occur. If a host species is cosmopolitan there (Perez-Orella and Schulte-Hostedde 2005). Similarly in our may be more gene fl ow between its open populations, which study, host size did not predict parasitism levels. Th e other could maintain their susceptibility to parasitism (Forbes and factor that had little eff ect was host local abundance. Th is Mlynarek 2014). Mlynarek et al. (2014) has documented was also unexpected because a relationship between local this type of pattern in Nehalennia , where N. gracilis, a habi- abundance and parasite levels has been previously reported tat specialist and a closed population species, resists all of an (Arneberg et al. 1998, Krasnov et al. 2002). We would expect Arrenurus sp. whereas N. irene, a ubiquitous species in open that more locally abundant species would be stronger targets populations, does not show any resistance to the same para- of selection by the parasites. Local adaptation is expected to site at the same two sites. evolve in the parasites of locally abundant hosts (Arneberg Th e other two predictor variables tested had a minor eff ect et al. 1998). Even though host abundance and host size were on explaining any of the measures of parasitism observed in expected to predict host parasitism, in this system, the better this study. Based on the literature, we expected host body predictors are ones that are measured at larger scales. Even though hosts were collected at multiple sites within a region, we superimposed the parasitism over the entire range considering it a species’ characteristic, which is common in evolutionary ecology studies (Heff erman et al. 2014). Th e importance of scale has long been understood to be critical in ecology (Wiens 1989, Levin 1992) but it has been tested only infrequently in empirical studies of parasitism (Tack et al. 2014). Diff erent measures of species ’ characteristics and species ’ interactions might depend on diff erent geographic or temporal scales (Heff erman et al. 2014) thus underscor- ing the need of measuring host species ’ characteristics of diff erent scales for a better understanding of the relationship between the host and parasite species assemblages. In this study, we primarily looked at Coenagrionidae – Arrenuridae host – parasite associations from the perspective of host species to explain three measures of parasitism; prevalence, intensity and diversity. As in other host – parasite associations, host characteristics do matter (Poulin 2007). Figure 6. Summary of the important predicting factor explaining Th is study is one of the few in invertebrate host systems that number of OTUs in each damselfl y species. Host regional occur- ϭ elucidate the reasons for the strength of host – parasite asso- rence in relation to Arrenurus OTUs. As Amphiagrion saucium ; ciations, based on host characteristics measured at multiple Cc ϭ Chromagrion conditum ; Ci ϭ Coenagrion interrogatum ; Cr ϭ C. resolutum ; Ea ϭ Enallgma aspersum ; E b ϭ E. boreale ; E c ϭ E. scales. Further research into host – parasite associations in the carunculatum ; Ecy ϭ E. cyathigerum ; E e ϭ E. ebrium ; E g ϭ Enal- this context including understanding turnover of principal lagma geminatum ; E h ϭ E. hageni ; Es ϭ Enallagma signatum ; host species and host ranges across space and time would Ip ϭ Ischnura posita ; I v ϭ I. verticalis ; Ng ϭ Nehalennia gracilis ; be logical next steps in understanding the evolution of the Ni ϭ N. irene. associations between hosts and their parasites.

9-EV Acknowledgements – We thank L. Nagel, A. Morrill and S. Iverson Ivanova, N. V. et al. 2006. An inexpensive, automation-friendly for discussions on this and related projects, V. Levesque-Beaudin protocol for recovering high-quality DNA. – Mol. Ecol. Not. for aid in the barcoding process and M. McPeek and G. Bechly for 6: 998 – 1002. helping with the reconstruction of the Coenagrionidae Phyloge- Kamiya, T. et al. 2014. What determines species richness of para- netic tree. Th is study was funded by a NSERC CGS awarded to sitic organisms? A meta-analysis across animal, plant and fungal JJM and a NSERC Discovery Grant to MRF. JJM and MRF hosts. – Biol. Rev. 89: 123 – 134. designed the study. JJM collected and analysed the statistical data. Krasnov, B. et al. 2002. Th e eff ect of host density on ectoparasite JJM and WK analysed the molecular data. JJM wrote the paper. distribution: an example of a rodent parasitized by fl eas. – Ecology JJM, WK and MRF edited the paper. 83: 164 – 175. Krasnov, B. et al. 2004. Flea species richness and parameters of host body, host geography and host ‘ milieu ’ . – J. Anim. Ecol. 73: References 1121 – 1128. Krasnov, B. et al. 2006. Age-dependent fl ea (Siphonaptera) parasit- Abbott, J. C. 2007. OdonataCentral: an online resource for the ism in rodents: a host’s life history matters. – J. Parasitol. 92: distribution and identifi cation of odonata. – Ͻ www. 242 – 248. odonatacentral.orgϾ . Leung, B. et al. 1999. Grooming decisions by damselfl ies, age- Altizer, S. et al. 2006. Seasonality and the dynamics of infectious specifi c colonisation by water mites, and the probability of diseases. – Ecol. Lett. 9: 467 – 484. successful parasitism. – Int. J. Parasitol. 29: 397 – 402. Arneberg, P. et al. 1998. Host densities as determinants of abundance Levin, S. 1992. Th e problem of pattern and scale in ecology. in parasite communities. – Proc. R. Soc. B 265: 1283– 1289. – Ecology 73: 1943 – 1967. Brown, J. et al. 2000. A phylogenetic perspective on habitat shifts Lindenfors, P. et al. 2007. Parasite species richness in carnivores: and diversity in the north American Enallagma damselfl ies. eff ects of host body mass, latitude, geographical range and – Syst. Biol. 49: 697 – 712. population density. – Global Ecol. Biogeogr. 16: 496 – 509. Burnham, K. P. and Anderson, D. R. 2002. Model selection and Locklin, J. L. and Vodopich, D. S. 2010. Patterns of gregarine multimodel inference a practical information-theoretic parasitism in dragonfl ies: host, habitat, and seasonality. approach. – Springer. – Parasitol. Res. 107: 75 – 87. Campbell, K. U. et al. 2013. Th e diversity and host specifi city of Maddison, W. P. and Maddison, D. R. 2011. Mesquite: a modular mites associated with ants: the roles of ecological and life his- system for evolutionary analysis ver. 2.75. – Ͻ http://mesquite- tory traits of ant hosts. – Insectes Soc. 60: 31 – 41. project.orgϾ . Chippindale, P. et al. 1999. Phylogenetic relationships of May, M. 2002. Phylogeny and taxonomy of the damselfl y genus North American damselfl ies of the genus Ischnura (Odonata: Enallagma and related taxa (Odonata: Zygoptera: Coenagrio- Zygoptera: Coenagrionidae) based on sequences of three mito- nidae). – Syst. Entomol. 27: 387 – 408. chondrial genes. – Mol. Phylogenet. Evol. 11: 110 – 121. Mlynarek, J. J. et al. 2012. Higher gregarine parasitism often in Crozier, R. H. et al. 2005. Phylogenetic biodiversity assessment based sibling species of host damselfl ies with smaller geographical on systematic nomenclature. – Evol. Bioinform. 1: 11 – 36. distributions. – Ecol. Entomol. 37: 419 – 425. Dumont, H. J. et al. 2010. A molecular phylogeny of the Odonata Mlynarek, J. J. et al. 2013. Relative geographic range of sibling (Insecta). – Syst. Entomol. 35: 6 – 18. species of host damselfl ies does not reliably predict diff erential Durrer, S. and Schmid-Hempel, P. 1995. Parasites and the regional parasitism by water mites. – BMC Ecol. 13: 50. distribution of bumblebee species. – Ecography 18: 114– 122. Mlynarek, J. J. et al. 2014. Explaining susceptibility and resistance Forbes, M. R. and Baker, R. 1991. Condition and fecundity of the to a multi-host parasite. – Evol. Biol. 41: 115 – 122. damselfl y, Enallagma ebrium (Hagen) – the importance of Morand, S. and Krasnov, B. R. 2010. Th e biogeography of ectoparasites. – Oecologia 86: 335 – 341. host – parasite interactions. – Oxford Univ. Press. Forbes, M. R. and Robb, T. 2008. Testing hypotheses about para- Munoz, G. et al. 2007. Structure of the parasite communities of a site-mediated selection using odonate hosts. – In: C ó rdoba- coral reef fi sh assemblage (Labridae): testing ecological and Aguilar, A. (ed.), Dragonfl ies and damselfl ies: model organisms phylogenetic host factors. – J. Parasitol. 93: 17 – 30. for ecological and evolutionary research. Oxford Univ. Press, Nylander, J. A. A. 2004. MrModeltest v2. – Program distributed pp. 175 –188. by the author, Evolutionary Biology Centre, Uppsala Univ. Forbes, M. R. and Mlynarek, J. J. 2014. A hypothesis to explain O’Grady, E. and May, M. 2003. A phylogenetic reassessment host species diff erences in resistance to multi-host parasites. of the subfamilies of Coenagrionidae (Odonata: Zygoptera). – Ideas Ecol. Evol. 7: 17 – 24. – J. Nat. Hist. 37: 2807 – 2834. Forbes, M. et al. 1999. Parasitism of Sympetrum dragonfl ies by Orme, D. et al. 2012. caper: comparative analyses of phylogenetics Arrenurus planus mites: maintenance of resistance particular to and evolution in R. – R package ver. 0.5. one species. – Int. J. Parasitol. 29: 991 – 999. Patterson, B. D. et al. 2008. Parasitism by bat fl ies (Diptera: Forbes, M. R. et al. 2012. Seasonality of gregarine parasitism Streblidae) on neotropical bats: eff ects of host body size, in the damselfl y, Nehalennia irene : understanding unimodal distribution, and abundance. – Parasitol. Res. 103: patterns. – Parasitol. Res. 110: 245 – 250. 1091 – 1100. Gillespie, S. 2010. Factors aff ecting parasite prevalence among wild Perez-Orella, C. and Schulte-Hostedde, A. 2005. Eff ects of bumblebees. – Ecol. Entomol. 35: 737 – 747. sex and body size on ectoparasite loads in the northern Hassall, C. 2012. Predicting the distributions of under-recorded fl ying squirrel (Glaucomys sabrinus ). – Can. J. Zool. 83: Odonata using species distribution models. – Insect Conserv. 1381 – 1385. Divers. 5: 192 – 201. Poulin, R. 1991. Group-living and infestation by ectoparasites in H e ff erman, J. et al. 2014. Macrosystems ecology: understanding Passerines. – Condor 93: 418 – 423. ecological patterns and processes at continental scales. – Front. Poulin, R. 2007. Evolutionary ecology of parasites. – Princeton Ecol. Evol. 12: 5 – 14. Univ. Press. Huelsenbeck, J. P. and Ronquist, F. 2001. MRBAYES: Bayesian Price, P. et al. 1988. Parasite-mediated competition – some predic- inference of phylogenetic trees. – Bioinformatics 17: 754 – 755. tions and tests. – Am. Nat. 131: 544 – 555. Ilvonen, S. et al. 2011. Can infection by eugregarine parasites Ratnasingham, S. and Hebert, P. D. N. 2007. BOLD: the barcode mediate species coexistence in Calopteryx damselfl ies? – Ecol. of life data system (www.barcodinglife.org). – Mol. Ecol. Not. Entomol. 36: 582 – 587. 7: 355 – 364.

10-EV Ronquist, F. and Huelsenbeck, J. P. 2003. MrBayes 3: Bayesian hematozoa at small spatial and phylogenetic scales. – Proc. Natl phylogenetic inference under mixed models. – Bioinformatics Acad. Sci. USA 96: 1785 – 1789. 19: 1572 – 1574. Th ompson, J. N. 2005. Th e geographic mosaic of coevolution. R ó zsa, L. et al. 2000. Quantifying parasites in samples of hosts. – J. – Univ. of Chicago Press. Parasitol. 86: 228 – 232. Tripet, F. and Richner, H. 1997. Th e coevolutionary potential of a Smith, I. et al. 2010. Water mites (Hydrachnidiae) and other arach- ‘ generalist ’ parasite, the hen fl ea Ceratophyllus gallinae . – Para- nids. – In: Th orp, J. and Covich, A. (eds), Ecology and clas- sitology 115: 419 – 427. sifi cation of North American freshwater invertebrates, 3rd ed. Webb, C. 2000. Exploring the phylogenetic structure of ecological Academic Press, pp. 485 – 587. communities: an example for rain forest trees. – Am. Nat. 156: S w o ff ord, D. L. 2003. PAUP* . Phylogenetic analysis using parsimony 145 – 155. (* and other methods), ver. 4 beta 10. – Sinauer Associates. Wiens, J. 1989. Spatial scaling in ecology. – Funct. Ecol. 3: Tack, A. J. M. et al. 2014. Th e impact of spatial scale and habitat 385 – 397. confi guration on patterns of trait variation and local adaptation Zuk, M. 1987. Seasonal and Individual variation in in a wild plant parasite. – Evolution 68: 176 – 189. gregarine parasite levels in the fi eld crickets Gryllus veletis Tella, J. et al. 1999. Habitat, world geographic range, and embry- and Gryllus pennsylvanicus . – Ecol. Entomol. 12: onic development of hosts explain the prevalence of avian 341– 348.

Supplementary material (Appendix ECOG-00997 at Ͻ www. ecography.org/readers/appendixϾ ). Appendix 1 – 2.

11-EV