Eric M. Janson, Department of Biological Sciences, Vanderbilt University, VU Station B #35-1634, Nashville, TN 37235-1634, USA. Tel.: 1 615 936 3699; fax: 1 615 343 6707; e-mail: firstname.lastname@example.org
Recent studies have shown that symbionts can be a source of adaptive phenotypic variation for their hosts. It is assumed that co-evolution between hosts and symbionts underlies these ecologically significant phenotypic traits. We tested this assumption in the ectosymbiotic fungal associate of the gall midge Asteromyia carbonifera. Phylogenetic analysis placed the fungal symbiont within a monophyletic clade formed by Botryosphaeria dothidea, a typically free-living (i.e. not associated with an insect host) plant pathogen. Symbiont isolates from four divergent midge lineages demonstrated none of the patterns common to heritable microbial symbioses, including parallel diversification with their hosts, substitution rate acceleration, or A+T nucleotide bias. Amplified fragment length polymorphism genotyping of the symbiont revealed that within-lineage genetic diversity was not clustered along host population lines. Culture-based experiments demonstrated that the symbiont-mediated variation in gall phenotype is not borne out in the absence of the midge. This study shows that symbionts can be important players in phenotypic variation for their hosts, even in the absence of a co-evolutionary association.
In mutualistic symbioses, host organisms frequently co-opt microbial phenotypes to exploit novel resources or to express adaptive traits (Ollerton, 2006; Moran, 2007; Janson et al., 2008; Bronstein, 2009; Oliver et al., 2009). Such is the pattern in hereditary symbioses, or associations in which a symbiotic associate is vertically transmitted from mother to offspring across host generations. In these symbioses, the alignment of reproductive interests enhances the evolution of reciprocally beneficial adaptations, such as nutritional compensation (Gündüz & Douglas, 2009) or defence from natural enemies (Oliver et al., 2005). Symbionts can therefore act as a source of ecologically significant phenotypic variation for the host, even facilitating the invasion of novel ecological niches (Moran, 2007; Janson et al., 2008; Oliver et al., 2009).
Hereditary symbioses between bacteria and insects are common and phylogenetically widespread (Buchner, 1965). However, insects also engage in remarkably complex symbioses with fungi (Buchner, 1965; Vega & Blackwell, 2005; Gibson & Hunter, 2010). Insect–fungal symbioses are diverse, occurring in a numerous taxa and providing a wide variety of benefits to the insects (Vega & Blackwell, 2005). While fungal endosymbiosis in insects may currently be underappreciated, a distinctive feature of most described fungal symbioses is that they are not intimately integrated within the cells or tissues of their hosts, as are endosymbiotic bacteria. Examples include the fungus-farming ants in the Myrmicinae, fungus-farming termites in the Macrotermitinae, ambrosia beetles in the Scolytinae and Platypodinae, wood wasps in the Siciridae and gall midges in the Cecidomyiinae (Morgan, 1968; Bissett & Borkent, 1988; Farrell et al., 2001; Six, 2003; Mueller et al., 2005).
Co-evolutionary processes and patterns may differ in important ways between different types of symbioses. Ectosymbioses can differ from endosymbioses in their mode of transmission (opportunity for horizontal transmission and symbiont replacement), amount of exposure to ‘unprotected’ (external) environments, increased opportunity for gene flow from other populations and the overall intimacy of the interaction – all of which may in turn influence the capacity of symbioses to act as a source of phenotypic variation, the process and pattern of co-adaption of symbiont and host, and the likelihood of co-phylogenesis of host and symbiont lineages (Mikheyev et al., 2010).
We used phylogenetic and population genetic analyses, along with phenotypic assays to characterize patterns of co-evolution in symbiosis between a unique gall midge and its fungal associate. The Cecidomyiidae, to which gall midges belong, are a large and diverse family of flies but remain poorly understood. As their common name implies, many form galls on their host plants and a number of these form a symbiotic association with fungi within galls. Only in recent years have studies begun to emerge indicating that gall-associated fungi contribute to both nutrition and defence (Weis, 1982a,b; Stireman et al., 2008, 2010; Janson et al., 2009) and thus potentially act as sources of ecologically significant phenotypic variation, facilitating the success on and exploitation of plant hosts. However, evolutionary studies of the fungal symbionts of gall midge are lacking.
The goldenrod-galling midge Asteromyia carbonifera (Diptera: Cecidomyiidae) forms blister-like galls on the leaves of Solidago spp. throughout much of North America (Gagné, 1968, 1989) (Fig. 1). Asteromyia carbonifera maintains an intimate association with a fungus (Borkent & Bissett, 1985; Bissett & Borkent, 1988; Gagné, 1989), and their blister galls are composed almost entirely of fungal mycelia. Crego et al. (1990) identified four morphologically distinct gall morphologies caused by A. carbonifera on Solidago altissima. They termed these morphologies ‘flat’, ‘cushion’, ‘crescent’ and ‘irregular’ (Fig. 1; see also Gagné, 1968; Stireman et al., 2008). In addition to differing in external shape, these galls consistently differ in the mean number of chambers per gall, the position of larvae within the gall, the position of the gall on the leaf, the thickness of the gall wall and several other characteristics (Gagné, 1968; Crego et al., 1990; Stireman et al., 2008). Each gall morph can be found sympatrically and syntopically, often co-occurring on the same ramet and occasionally the same leaf (Stireman et al., 2008). Two independent studies have demonstrated that A. carbonifera gall morph–associated populations are at different stages of evolutionarily divergence and originated from a common ancestor (Crego et al., 1990; Stireman et al., 2008). Phenotypic diversity in A. carbonifera gall morphology is also correlated with variation in susceptibility to the various parasitoids that attack the galls (Weis, 1982a; Stireman et al., 2008; unpublished data), suggesting a possible adaptive basis to the gall phenotypic variation (Stone & Schönrogge, 2003).
Because gall morphology acts as a strong correlate of both genetic and ecological divergence and galls are composed of fungus, the diversification of A. carbonifera gall morphs may be mediated by its fungal symbiont. We investigated the role of the fungal symbiont in A. carbonifera divergence. First, we characterized the identity, evolutionary history and phylogenetic relationships of the fungal symbiont and asked whether genetically divergent populations of fungus are associated with genetically divergent fly populations. Because of the intimate, widespread interaction between fungi and midges, we expected the fungal symbiont to exhibit reciprocal divergence with host flies, as well as evolutionary patterns common to microbial mutualisms, namely, accelerated rates of molecular evolution and A+T nucleotide bias. Second, we investigated whether the fungal symbiont is directly involved in the observed phenotypic variation in gall structure seen in A. carbonifera (Fig. 1). Because gall polymorphism is such a striking feature of A. carbonifera, we predicted that the fungal symbionts of A. carbonifera will exhibit heritable phenotypic variation in culture-based assays. We generally find that none of these predictions are supported, suggesting that gall morphological diversification is not the result of a history of co-evolution and co-adaptation but likely because of the manipulation of fungal growth and development by the gall midges.
Female A. carbonifera gather fungal conidia (asexual spores) from an unknown source and carry them in specialized invaginations (mycangium) in their terminal abdominal segment (Borkent & Bissett, 1985; Heath & Stireman, in press). Conidia are deposited along with eggs on Solidago leaves during oviposition. Conidia germinate soon after oviposition, and fungal hyphae proliferate through the leaf tissue (Camp, 1981; Borkent & Bissett, 1985; Bissett & Borkent, 1988). Unlike most insect-induced galls, however, Asteromyia galls do not exhibit any evidence of plant cell hyperplasy or hypertrophy (Camp, 1981; Rohfritsch, 2008). Instead, larvae come to lie in a gall chamber, completely surrounded by mycelium with fungus tissue making up the vast majority of the gall structure (Camp, 1981; Crego et al., 1990). In vivo sterol analysis (Janson et al., 2009) and in vitro culturing (Heath & Stireman, in press) indicate that A. carbonifera larvae feed exclusively on fungal tissue. Previously, the fungus was identified as Sclerotinium asteris and later as Macrophoma sp. (Bissett & Borkent, 1988; Roskam, 2005). These identifications were based on the morphology of conidia isolated from the mycangia of field-collected females, and from the morphology of mycelia in culture and on the leaves of their host plants.
During the summers of 2005–2008, leaves containing Asteromyia carbonifera galls were collected from Solidago plants in several locations: local sites around Fredericton, New Brunswick, Canada; southwestern Ohio, USA; Nebraska, USA; southern Illinois, USA, and southern and eastern Georgia, USA. At each location, we attempted to sample as many of the four gall morphs as possible and sample from the most dominant Solidago species (usually Solidago altissima). Briefly, leaves containing galls were placed into plastic bags, placed into a chilled cooler and transported back to the lab where they were refrigerated until use. Photographs were taken of each gall prior to processing so that a record of the gall morph could be maintained. Leaves were rinsed briefly in running deionized water and then were excised from the leaf. Excised galls were surface sterilized by sequential immersion in 95% ethanol (10 s), 10% Clorox solution (2 min) and 70% ethanol (30 s). Galls were then dried under sterile conditions and plated on 2% malt extract agar (MEA). Plates were sealed with parafilm, allowed to incubate at room temperature and checked daily for any hyphal growth for up to 12 weeks. When hyphae were observed, a small agar plug was aseptically removed from the plate and transferred to a fresh MEA plate. Care was made to ensure that each subculture was derived from a single hypha. After one hypha was removed from a gall plate, the gall was discarded. Thus, each gall usually yielded only a single isolate. Gall isolates were regularly checked for culture purity and subcultured when necessary. In total, we obtained 186 isolates (65 cushion, 34 crescent, 61 flat and 26 irregular) from Ohio, 103 isolates from Canada (24 cushion, 16 crescent, 57 flat and 4 irregular), 76 isolates from Illinois (7 cushion, 10 crescent, 59 flat), 25 isolates from Nebraska (13 cushion, 12 flat) and 11 isolates from Georgia (three cushion, two crescent, five flat, two irregular), for a total of 401 isolates.
Because fungi isolated from galls could be either the symbiont of A. cabonifera or incidental endophytes/pathogens, we also cultured field-collected A. carbonifera eggs, which frequently have conidia on their surface (Heath & Stireman, in press). Finally, we cloned ITS PCR amplicons from whole gall genomic DNA extractions. We then matched the sequences of the egg-derived isolates and cloned direct-gall extractions to that of the gall-derived fungi to determine the focal species most likely to the symbiont of A. carbonifera. Pure cultures of gall-isolated fungi were allowed to grow at room temperature on 2% MEA for 4–6 weeks prior to genomic DNA extraction. To obtain tissue for genomic DNA extraction for use in Amplified fragment length polymorphism (AFLPs), mycelium was scraped from MEA plates and placed into 14-mL tubes containing ∼7 mL malt extract broth. Liquid cultures were allowed to grow for 4–7 days prior to genomic DNA extraction.
Genomic DNA extraction, PCR and AFLPs
Genomic DNA for use in sequencing and cloning was extracted using the method of Arnold & Lutzoni (2007). Two loci were targeted for sequencing: the internal transcribed spacer regions (including the full 5.8S subunit) of the nrDNA and partial elongation factor 1 alpha. These loci were chosen for their phylogenetic informativeness at potentially low levels of evolutionary divergence (White et al., 1990; Glass & Donaldson, 1995) and their ability to be included in phylogenies with other publically available sequences. The ITS locus was amplified with the primers ITS4 and ITS5 (White et al., 1990). The primers EF1-728F and EF1-986R (Carbone & Kohn, 1999) were used to amplify part of EF-1α. Genomic DNA was amplified in 15 or 20 μL reactions (10–20 ng genomic DNA, 1.5 mm MgCl2, 0.25 mm each dNTP, 0.5 μm forward and reverse primers). One molar betaine was added to the ITS PCR reactions because of the relatively high GC content of the ITS sequences. PCRs were performed with the following amplification program: 2 min at 96 °C, 30 s at 94 °C, 45 s at 52–54 °C (depending on locus), 1 min at 72 °C, followed by a 10 min final extension at 72 °C and an indefinite 4 °C soak. All samples were checked for successful PCR amplification on a 1.5% agarose gel prior to sequencing.
Genomic DNA for use in AFLPs was extracted separately from the genomic DNA used for sequencing. A subset of isolates was chosen for DNA extraction for use in AFLP genotyping. Fungus tissue was removed from liquid culture, dried under vacuum, flash-frozen in liquid nitrogen, crushed with a mortar and pestle and then extracted with a CTAB extraction buffer containing beta-mercaptoethanol, followed by ethanol precipitation as used by Gibbons et al. (see An et al., 2010).
PCR amplicons were sequenced at the University of Arizona Genomic Analysis and Technology Core (GATC) or the Vanderbilt University DNA Sequencing Core. All PCRs were purified with an automated 96-well PCR purification system and then analysed on an ABI 3730xl DNA Analyzer at the GATC or treated with 1 U each shrimp alkaline phosphatase (SAP)/Exonuclease I (ExoI) if sequenced at Vanderbilt. PCR amplicons were sequenced in the 5′ and 3′ directions with the original amplification primers. ITS PCR amplicons from whole gall genomic extractions were cloned into chemically competent E. coli cells using the TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA, USA). Cloned amplicons were sequenced using M13 vector primer at the University of Arizona GATC.
AFLP genotypes were generated using the methods of Vos et al. (1995), with some modifications. Briefly, approximately 150–300 ng of genomic DNA was digested with the restriction enzymes EcoR I and Mse I (NEB, Ipswitch, MA, USA). AFLP adapters were ligated to the ends of genomic restriction fragments with T4 ligase (Roche, Palo Alto, CA, USA). The samples were diluted five-fold and used as a template for preselective amplification. A preselective amplification was performed using two primers complementary to the AFLP adapters and the restriction site sequences (Eco+A primer and Mse+C primer) (Vos et al., 1995). Amplification conditions were 94 °C for 30 s, 56 °C for 1 min, 72 °C for 1 min, for a total of 20 cycles and then held at 10 °C indefinitely. Samples were again diluted five-fold for selective amplification. One additional base was added to the primers for selective amplification, and the forward primer was fluorescently labelled to visualize the DNA during migration through the gel. The following primer combinations were used in the selective amplification: 6FAM EcoR I+AA and Mse+CG, VIC EcoR I+AC and Mse+CT, PET EcoRI+AA and Mse+CC, and NED EcoRI+AT and Mse+CC. Selective amplification conditions were 94 °C for 2 min, 94 °C for 30 s, 65 °C for 30 s (reduced by 1 °C per cycle), 72 °C for 1 min, repeat ten times excluding 94 °C for 2 min; 94 °C for 30 s, 56 °C for 30 s, 72 °C for 1 min, repeat last cycle 30 times, and follow with 72 °C for 30 min and holding at 10 °C indefinitely. Selective AFLP reactions were poolplexed and run on an ABI 3730 DNA Analyzer (Applied Biosystems, Inc., Foster City, CA, USA) at the University of Arizona GATC (Tucson, AZ, USA).
Investigating the evolutionary history and phylogenetic relationships of the fungal symbiont
Phylogenetic reconstruction and analysis of DNA sequence data
Contigs from each sequenced isolate were assembled and edited from bidirectional trace files in Sequencher 4.5 (Gene Codes Corp., Ann Arbor, MI, USA). Each sequence was examined for miscalled bases and manually edited when necessary. The consensus sequences were then manually aligned in MacClade 4.08 (Maddison & Maddison, 2005) and searched for variable sites across all isolates. Once it was determined that B. dothidea was the most likely symbiont (see results for details), we focused on B. dothidea sequences obtained from our sequencing screen for further phylogenetic analysis. No nucleotide substitutions were detected among all B. dothdiea isolates at both loci used for phylogenetic analysis (ITS: N = 150; EF-1α: N = 96). Therefore, sequences from representative A. carbonifera gall morph isolates and geographical locations were chosen for use in phylogenetic reconstructions to reduce the number of redundant sequences. All sequences used in this study were submitted to GenBank. Representative sequences for both loci were entered in BLASTn searches to approximate the genus-level identity of the symbiont. Congeneric and conspecific sequences were then obtained from GenBank and included in all phylogenetic reconstructions. All sequences were aligned using muscle 3.7 (Edgar, 2004) and manually adjusted in MacClade 4.08 when necessary.
Phylogenetic analysis of nucleotide sequence data was conducted by searching for trees using maximum parsimony (MP), maximum likelihood (ML) and Bayesian methods. Prior to phylogenetic analysis, a partition homogeneity test (PHT) with 500 replicates was performed in paup* 4b10 (Swofford, 2003) to determine whether the two loci could be combined in a single concatenated data set. The PHT was nonsignificant (P =0.848), and so all phylogenetic analyses were performed on the concatenated two-locus data set.
The interspecific parsimony analysis was carried out in paup* 4b10. The parsimony analysis consisted of 100 replicate searches of trees generated by random stepwise addition using tree bisection reconnection (TBR) branch swapping. Alignment gaps were treated differently in two separate searches: the first as missing data and the second as a 5th character state. All characters were equally weighted and unordered. Branches of zero length were collapsed and all multiple, equally parsimonious trees were saved. Because of the relatively large number of taxa and low level of sequence divergence in the intraspecific analysis, TNT 1.1 (Goloboff et al., 2008) was used to reconstruct the most parsimonious intraspecific tree. This parsimony analysis consisted of 100 replicate searches generated by random stepwise addition using TBR branch swapping. Branches of zero length were collapsed and all multiple, equally parsimonious trees were saved. The intraspecific MP search used four advanced MP tree searching algorithms (sectorial searching, parsimony ratchet, tree drifting and tree fusing) with their default parameters. The robustness of the MP trees was evaluated with 1000 bootstrap replicates.
Prior to ML and Bayesian analysis, an appropriate model of nucleotide substitution was determined. Models of nucleotide substitution were selected according to the AIC using the program jModelTest 0.1.1 (Posada, 2008). Maximum likelihood (ML) searches were conducted using PhyML 3.0 (Guindon & Gascuel, 2003). Trees were obtained by 100 replicate ML searches of trees generated by creating an initial distance tree using the BIONJ algorithm. Branch swapping was performed using subtree pruning and regrafting. The interspecific data set employed a TPM1uf+G model of substitution (Kimura, 1981) with a gamma shape parameter of 0.0960 (Ncat = 4). Equilibrium nucleotide frequencies were fixed at A = 0.18982, C = 0.31143, G = 0.24692, T = 0.25183. Maximum likelihood was not performed on the intraspecific data set. The robustness of the reconstructed tree was evaluated with 1000 bootstrap replicates using the aforementioned substitution model parameters.
Bayesian analysis was carried out in MrBayes 3.1.2 (Ronquist & Huelsenbeck, 2003). The interspecific data set employed a HKY+G model of substitution with the transition/transversion ratio, equilibrium nucleotide frequencies and gamma shape parameters optimized by MrBayes. The intraspecific data set employed a HKY+I model of substitution with the transition/transversion ratio, equilibrium nucleotide frequencies and proportion of invariant sites parameters optimized by MrBayes. Each analysis was run using four chains (one cold and three hot) for 2–4 million generations, until stationarity was reached. Trees were sampled every 1000 generations, and the first 25% of the sampled trees were discarded as burn-in. In all phylogenetic reconstructions, trees were either rooted with sequences from Guignardia philoprina (interspecific tree) (see Slippers et al., 2004) or Botryosphaeria corticis (intraspecific tree) obtained from GenBank.
Tajima’s relative rate test for three sequences (Tajima, 1993) was performed in mega 4.02 using several different ingroup and outgroup taxa depending on the comparison being made. P-values were Bonferroni corrected for multiple comparisons for this particular analysis. Nucleotide composition was calculated in BioEdit 22.214.171.124 (Hall, 1999) and statistically evaluated with a χ2 test in jmp 8.0.1 (SAS Institute, Cary, NC, USA). Two additional loci were included in these tests that were not used for phylogenetic reconstruction because of their lack phylogenetic informativeness between free-living and Asteromyia symbiotic fungus isolates (data not shown): partial 28S large ribosomal subunit (amplified with primers LR0R and LR16; Moncalvo et al., 1993; Rehner & Samuels, 1994) and partial beta-tubulin (amplified with the primers Bt2a and Bt2b; Glass & Donaldson, 1995).
AFLP genotyping and analysis
Fluorescent AFLP fragment peaks were automatically scored with Peak Scanner 1.0 (Applied Biosystems, Inc.). Peak Scanner was set to filter out all peaks that were less than 100 bp, as fragments < 100 bp tend to exhibit significant size homoplasy (Vekemans et al., 2002). Fragment data obtained in Peak Scanner was standardized and further filtered with the software package RawGeno 1.1–2 (Arrigo et al., 2009) implemented in the R 2.10.0 statistical software package (R Development Core Team, Vienna, Austria) to remove as much noise as possible from the data set. Fragment bin size and peak filtering were performed with the following parameters: tolerance: 0.97, maximum bin width: 2 bp, minimum bin width: 0 bp, close bins: 0.05%, low intensity bins: 50 std RFU, low intensity peaks: 0.05%, low frequency bins: 4 Nblnds (∼5% of the sample size). Phylogenetic analysis of the AFLP data set was conducted in phylip 3.69 (Felsenstein, 2005) by coding each fragment as a dichotomous character (absent: 0/present: 1) and searching for trees using MP with the PARS module. Data was unweighted, and the Wagner parsimony analysis consisted of 100 replicate random sequence additions and thorough searches for the best tree at each replicate. A maximum of 500 000 trees were saved at each replicate. Robustness of reconstructed phylogenetic relationships was evaluated with 1000 bootstrap replicates.
To visualize this uncertainty and determine whether distance-based methods resulted in similar phylogenetic relationships, we created a splits diagram using splitstree 4.10 (Huson & Bryant, 2006), which may provide a more appropriate representation of relationships at the intraspecific level. A splits diagram represents all inferred splits in a network diagram and is composed of parallel edges, rather than a pruned bifurcating tree representing only a consensus of the optimal tree or trees (Huson & Bryant, 2006). We derived a Dice distance matrix from the binary AFLP data. A neighbour-joining algorithm (‘NeighborNet,’Bryant & Moulton, 2004) was then used to construct an unrooted dendrogram, which is a visualization of the equal-angle split transformation we performed on the AFLP distance matrix.
FST was calculated for isolates over all gall morphotypes using AFLP-surv (Vekemans, 2002). The significance of the FST values (i.e. significantly greater than zero) and 95% confidence intervals were calculated via permutation tests employing 1000 resamplings. The relative isolation of populations associated with geography vs. gall morph was assessed using analysis of molecular variance (amova), as implemented in GenAlEx 6.2 (Peakall & Smouse, 2006). The amova was performed on a distance matrix calculated in GenAlEx 6.2 with the genetic distance formula for haploid binary data of Huff et al. (1993). Grouping levels were geographical location (OH vs. GA) and gall morph (crescent, cushion, flat and irregular), and significance was established by 1000 permutations of the data. To determine whether genetic recombination (sexual reproduction) occurs in the fungal symbiont of A. carbonifera, a statistical test on AFLP genotypes called the index of association (IA; Maynard Smith et al., 1993; Taylor et al., 1999) was performed in lian 3.5 (Haubold & Hudson, 2000). This method examines multilocus genotype data for a nonrandom statistical association among alleles at each locus (linkage disequilibrium). Linkage disequilibrium among large numbers of loci is suggestive of low to nonexistent levels of genetic recombination and thus asexual reproduction (Maynard Smith et al., 1993; Burt et al., 1996). Significance testing of the null hypothesis of linkage equilibrium was performed through a Monte Carlo simulation procedure that resampled the data without replacement 1000 times (Haubold & Hudson, 2000). Here, isolates from different gall morphs were combined, because our results demonstrated no evidence of genetic divergence along gall morph lines (see Results).
Investigating the intrinsic ability of fungal symbionts to generate phenotypic variation
We also examined the degree to which the phenotypic (gall morph) differences observed in nature are a stable property of the fungus or a more complex result of the interaction with the midge. We subjected isolates from each of the four divergent midge lineages to three culture-based experiments. Prior to all growth rate tests, fungal isolates were allowed to acclimate to lab conditions by performing at least 12 weeks of continual subculturing. The first culture experiment involved growing isolates on unamended potato dextrose agar (PDA) in complete darkness. The second experiment involved growing isolates on PDA amended with varying amounts of KCl to assess growth under water stress (five treatments total; Kim et al., 2005). In the final experiment, isolates were grown on Czapek-Dox agar (CDA) with reduced concentrations (10% of standard recipe amount) of the nutrients phosphorous, carbon and nitrogen, plus an unaltered control. The standard recipe for the CDA was as follows: 30 g L−1 sucrose, 3 g L−1 sodium nitrate, 0.5 g L−1 potassium chloride, 0.5 g L−1 magnesium sulfate heptahydrate, 0.01 g L−1 iron(II) sulfate heptahydrate, 1 g L−1 dipotassium hydrogen phosphate, 15 g L−1 agar. Three millimetre plugs of advancing mycelium were punched out of 2% MEA with a cork borer and transferred to a growth rate experimental plate. Growth rates were then determined by taking two perpendicular measurements from the edge of the plug to the edge of the advancing mycelium on days two, three and four post-transfer using a digital caliper. The mean of these two measurements was then calculated, and the daily growth rate (mm day−1) was obtained by subtracting the mean growth of the previous day from the mean growth up to the measurement day. The two separate growth rate measurements (from day 2–3 and from day 3–4) were then averaged to obtain an overall mean growth rate for the replicate. The replicate growth rates were then averaged to obtain a mean growth rate for the isolate. Each isolate was replicated three times for each treatment. When necessary, as in the case of slowly growing isolates, measurements were taken from consecutive days beyond day four. Those isolates that grew too quickly were repeated, so that two daily growth rates could be obtained and averaged. All tests were performed at 23 ± 0.5 °C and ambient humidity (∼40%) in a temperature-controlled environmental chamber. Growth rate data were natural log transformed and analysed with manova in jmp 8.0.1. In the event of a significant manova, univariate anovas were performed.
Investigating the evolutionary history and phylogenetic relationships of the fungal symbiont
Fungal identity, phylogenetic reconstruction and other evolutionary metrics
Based on BLASTn searches of ITS sequences, several fungal genera were isolated from plated A. carbonifera galls, including some common, cosmopolitan soil and plant-pathogenic fungi (Table 1). Most of these appeared to be co-occuring pathogens/latent endophytes as the most common non-Botryosphaeria isolates were isolated from ungalled Solidago leaves in addition to occasional isolation from gall tissue (Table 1; see also Adair et al., 2009). B. dothidea was never isolated from Solidago leaf tissue. Botryosphaeria dothidea was isolated from plated A. carbonifera eggs and cloned using ITS PCR product. The specific strain of B. dothidea isolated in these additional experiments (A. carbonifera egg culturing and cloned whole gall extractions) was identical across their ITS and/or EF-1α region to the B. dothidea isolates acquired from gall tissue cultures. Taken together, these results indicate that the fungal associate of A. carbonifera belongs to the genus Botryosphaeria. Therefore, further phylogenetic analysis was performed using Botryosphaeria sequences obtained from GenBank. All DNA sequences used in this study are freely available on GenBank.
Table 1. Fungi isolated from whole Asteromyia carbonifera galls, A. carbonifera eggs, Solidago sp. leaves, and/or cloned from PCR amplifications of genomic DNA isolated from whole A. carbonifera galls.
Numbers in parentheses indicate the number of confirmed (either by culture morphology or PCR amplification and sequencing of ITS) isolations of specific taxa out of the total number of confirmations in that category. If a particular category is absent from a fungal taxon, it indicates that taxon was not detected in that category.
Plated Asteromyia carbonifera gall tissue (117/199) Plated A. carbonifera eggs (32/32) Cloned A. carbonifera whole gall genomic extractions (70/70)
Single phylogenetic lineage/sublineage in all cases (described in this paper)
Plated A. carbonifera gall tissue (43/199) Plated S. altissima leaf tissue (12/25)
Multiple strains/species detected by PCR
Plated A. carbonifera gall tissue (29/199) Plated S. altissima leaf tissue (11/25)
Multiple strains/species detected by PCR
Collectotricum sp./Glomerella sp.
Plated A. carbonifera gall tissue (1/199)
Likely leaf spot misidentification
Plated A. carbonifera gall tissue (1/199)
Plated A. carbonifera gall tissue (8/199) Plated S. altissima leaf tissue (2/25)
The interspecific phylogenetic reconstruction (MP with gaps as 5th character states) included 39 isolates of 19 Botryosphaeria species. The aligned nucleotide data set consisted of a total of 920 characters. Of those characters in the combined data set, 386 were constant, 187 variable characters were parsimony uninformative and 347 were parsimony informative. The MP search recovered six trees of length 1142. One of those most parsimonious trees (MPTs) is displayed in Fig. 2. Trees reconstructed with different methods were highly consistent, only differing in the placement of B. tsugae and B. protearum. The reconstructed trees were also consistent with other phylogenetic reconstructions of the genus Botryosphaeria (e.g. Slippers et al., 2004). All symbiotic isolates were placed within a clade that included the epitype specimen of B. dothidea (strain CMW8000) with consistently high bootstrap and posterior probability support (Fig. 2). This clade also included the species B. populi, which was previously described as a distinct species based on morphology (Phillips, 2000) but was later deemed synonymous with B. dothidea (Phillips et al., 2005).
The intraspecific phylogenetic reconstruction (Fig. 3; MP tree with gaps treated as a 5th character state) on the concatenated ITS and Ef-1a data sets included 92 isolates of B. dothidea. The aligned data set contained 764 characters. In total, 695 characters were constant, 40 variable sites were parsimony uninformative and 25 variable sites were parsimony informative. The MP search recovered a total 76 trees of length 84. One of the MPTs is displayed in Fig. 3. The intraspecific tree contains two core monophyletic clades, one representing two isolates from South America and the other containing the bulk of the B. dothidea diversity. There was little evidence of host plant or geographical structure to the phylogeny, underscoring B. dothidea’s host plant generalist lifestyle and cosmopolitan distribution. All of the A. carbonifera symbiotic B. dothidea isolates (Fig. 3, grey shading) belonged to a single monophyletic clade, with three isolates from southeastern Canada forming a weakly supported subclade. This subclade was reconstructed based on a two nucleotide indel that was not found in other isolates, which resulted in no statistical support for the clade when gaps were not treated as informative (Fig. 3). The A. carbonifera symbiont clade also included a few free-living isolates and the symbionts of several species of Asphondylia gall midges found in South Africa and Australia.
Nucleotide composition varied among the four examined loci for both free-living and symbiotic fungal isolates (Table 2). G+C content was higher than A+T content for all loci in both groups. Total A+T content was not elevated in A. carbonifera-associated isolates relative to free-living isolates (45.0% vs. 45.8%, respectively). Indeed, overall nucleotide composition did not significantly differ between free-living and symbiotic fungal isolates of B. dothidea (χ23 = 0.31; P =0.96). Tajima’s (1993) relative rate test revealed no evidence of nucleotide substitution rate differences for the symbiotic isolates compared to a free-living isolates (Table 3). There was, however, evidence of substitution rate increase in a fungus species that is involved in a lichen symbiosis (Table 3).
Table 2. Mean per cent nucleotide composition for four nuclear loci in free-living (collected from plant tissue and not in association with an insect) and A. carbonifera symbiotic isolates of B. dothidea. Sequences from all free-living isolates were obtained from GenBank and include isolates from several locations throughout the world found on numerous host plant genera. Sample sizes are as follows: free-living isolates, 28S: N = 11; ITS: N = 242; beta-tubulin: N = 38; EF-1α: N = 79. Symbiotic B. dothidea isolates, 28S = 48; ITS: N = 150; beta-tubulin: N = 48; EF-1α: N = 96.
Partial 28S LRSU
Table 3. Results of Tajima’s (1993) relative rate test for three sequences between various free-living and gall midge symbiotic fungus isolates, and Asteromyia carbonifera symbiotic Botryosphaeria dothidea. Significance of the comparisons was determined by a χ2 test with a Bonferroni corrected P-value of 0.004. Sequences for all isolates not associated with Asteromyia carbonifera were retrieved from GenBank.
†‘Free-living’ indicates that these isolates were not found in association with an insect host, although they are plant pathogens/endophytes.
Miii: number of identical sites in all three lineages; Mijk: number of unique sites in all three lineages; mjii: number of sites unique to lineage one; miji: number of site unique to lineage two; miij: number of sites unique to lineage three (outgroup).
Asteromyia carbonifera symbiotic Botryosphaeria dothidea (North America)
Partial translation elongation factor 1α
Asphondylia symbiotic B. dothidea (South Africa)
A. carbonifera symbiotic B. dothidea (North America)
Test cannot be performed: no nucleotide variation
Partial translation elongation factor 1α
Lichenized (symbiotic) Trypethelium sp.
A. carbonifera symbiotic B. dothidea (North America)
Partial 28S large subunit nrDNA
Lichenized (symbiotic) Trypethelium sp. has the faster rate of evolution for both loci
‘Free-living’Botryosphaeria corticis (North America)†
A. carbonifera symbiotic B. dothidea (North America)
Partial 28S large subunit nrDNA
Partial elongation factor 1α
A. carbonifera symbiotic Botryosphaeria dothidea (North America)
Partial translation elongation factor 1α
AFLP population genetics and phylogenetics
Quality filtering of our AFLP data set removed an average of 19.6% of the initial fragments detected, leaving our final data set consisting of 73 isolates and 1010 loci. The parsimony analysis reconstructed two MPTs of length 10461. These MPTs indicated that there were several distinct clusters of isolates, but these were not consistently associated with gall morph or geographical location. Two free-living isolates were nested well within the symbiotic isolates. The bootstrap support for the AFLP parsimony tree was weak across most of the tree, except for some of the tips. The tips of the NeighborNet tree inferred using splitstree is highly consistent with the results from the parsimony analysis and provides a visual summary of the phylogenetic uncertainties (see Supporting Information). Again, the pattern of genetic variation in the AFLP samples did not consistently correspond to geography or the morphology of the gall morph of origin.
The overall FST among gall morph–associated isolates was 0.0077, lower 95% CI: −0.0033, upper 95% CI: 0.0119 (P =0.11), suggesting a lack of genetic structure among isolates from different A. carbonifera gall morphs. Pairwise FST among gall morph–associated isolates ranged from 0.0001 to 0.0150 (Table 4). Nei’s genetic distance (Lynch & Milligan, 1994) among gall morph–associated isolates ranged from 0.000 to 0.0051 (Table 4). In the amova of AFLP data, the gall morph of origin was responsible for approximately 1% of the total genetic variation, whereas population (sampling site) was responsible for an additional 1% (Table 5). These variance components were not significantly greater than expected, as most of the genetic variation was found within gall morphs and within populations (98%).
Table 4. Below the diagonal: Pairwise FST for fungal isolates associated with the four Asteromyia carbonifera gall morph populations. Above the diagonal: pairwise Nei’s genetic distances for fungal isolates associated with the four Asteromyia carbonifera gall morph populations.
Table 5. Summary of analysis of molecular variance (amova) of amplified fragment length polymorphism data based on Euclidian genetic distances.
%: the percentage of variance explained by each sampling level.
Significance of F statistics (molecular analogues of Fisher’s F statistics) is based on 1000 permutations of samples.
Among geographical locales
Among A. carbonifera gall morphs within geographical locales
Results from the index of association (IA) test are summarized in Table 6. This test indicated that the fungal symbiont of A. carbonifera shows lineage disequilibrium that is greater than would be expected in a sexually reproducing population for two geographical divisions of the data and the entire combined data set.
Table 6. Summary of index of association (IA) calculations and statistical testing.
Mean genetic diversity
Monte Carlo simulations
N: sample size; N loci: number of AFLP loci included in the analysis; VD: observed variance of the mismatch distribution; Ve: expected linkage equilibrium variance of the mismatch distribution; IAS: standardized index of association. See Haubold & Hudson (2000) for details.
0.2723 ± 0.0047
0.2723 ± 0.0048
0.3452 ± 0.0049
Investigating the intrinsic ability of fungal symbionts to generate phenotypic variation
The general appearance of each isolate in culture was homogenous across isolates from different gall morphologies and consistent with previously published descriptions of Botryosphaeria dothidea. All cultures (on PDA in total darkness) were initially buff and then became olivaceous grey to black, with a sparse to moderately dense, appressed mycelial mat and occasional aerial mycelium reaching the lid of the Petri dish. Outside of isolate-level variation in growth rate, there were no obvious differences in the cultural appearance of the different gall morph–associated isolates. None of the fungal isolates demonstrated evidence of a gall-like appearance in culture (see also Heath & Stireman, in press).
The mean overall growth rate (on PDA in total darkness) for the fungal symbiont of A. carbonifera was 7.4 mm per day ± 0.2 (N = 93). The results of the growth rate tests for each gall morph–associated set of isolates are detailed in Fig. 4a–c. Figure 4b shows that symbiotic B. dothidea generally grew more slowly on nonorganic sources of carbon (CDA). Isolates from all gall morphs tended to grow more quickly when phosphorous limited, but the mechanism behind this is unclear. Figure 4c shows that mycelial growth rate was inversely related to water stress. Overall, there was no significant difference in growth rate among the isolates from each gall morph within each treatment (manova; F3,48 = 0.5137; P =0.675). There was evidence for individual-level phenotypic variation for growth rate in the fungal symbiont of A. carbonifera. Growth rates for isolates were positively correlated within experiments (Table 7), suggesting that a significant amount of the variation in growth rate is intrinsic to individual isolates but not to groups of isolates from particular gall morphs.
Table 7. Pairwise correlations of growth rates for individual isolates across treatments for the water stress (top) and nutrient limitation (bottom) experiments. Below the diagonal are the parametric correlations on log-transformed data, above the diagonal are the nonparametric correlations.
*P <0.005 for water stress data; P <0.0083 for nutrient limitation (Bonferroni corrected P-values); **P <0.001; ***P <0.0001.
However, we found no evidence of gall morph–associated phylogenetic, genetic or phenotypic divergence in the gall midge symbiotic B. dothidea, no evidence of genomic correlates of a symbiotic lifestyle for B. dothidea and essentially no evidence of evolutionary divergence of the symbiotic isolates (midge associated) from free-living (not insect associated) B. dothidea populations. For gall midges, the fungal symbiont is not sequestered from the environment, and the evolutionary histories of midges and their fungal partners are uncoupled. This pattern may be a general one for fungal ectosymbioses (e.g. Six & Paine, 1999; Kiers & van der Heijden, 2006; Mikheyev et al., 2006, 2007, 2010; Little & Currie, 2007). What is noteworthy about the case of Asteromyia is the degree to which symbiont-mediated phenotypes (Fig. 1) evidently lack transmissible symbiont genotypes.
Botryosphaeria dothidea as a source of phenotypic variation and its influence on evolutionary divergence for Asteromyia carbonifera
Our data suggests that the observed variation in gall morphology is not the result of genetically based phenotypic divergence in the fungal symbiont. Evidence in support of this includes: (i) tightly coupled genetic and phenotypic (gall structure) divergence in A. carbonifera (Crego et al., 1990; Stireman et al., 2008), (ii) an association with a single phylogenetic lineage of B. dothidea and a lack of genetic and phenotypic divergence in the fungus along gall phenotype lines (this study), (iii) lack of gall structure formation in the absence of the midge or when the midge is removed (Heath & Stireman, in press) and (iv) suppression of normal development and maturation in the fungal symbiont by the presence of the midge (Bissett & Borkent, 1988; Adair et al., 2009). Thus, A. carbonifera appears to be adapted to manipulate fungal growth and development, facilitating the observed phenotypic divergence in gall structure in association with a genetically homogenous, nonco-diversified lineage of fungus. This is surprisingly consistent with the consensus view that in all gall forming insects, the gall structure is determined primarily by insect genotype (Stone & Schönrogge, 2003). It appears that A. carbonifera may be the only species in the genus that is able to manipulate B. dothidea to such a degree, although further study is needed. If true, however, symbiont-expressed phenotypic variation in gall structure may be a recent evolutionary innovation in Asteromyia.
How this fungus–midge association and concomitant symbiont manipulation evolved remains an open question, specifically in the light of the fact that B. dothidea does not appear to be a vertically transmitted mutualist. Recent evidence strongly suggests that the variation in gall structure is the result of diversifying selection because of a phenotypically diverse assemblage of midge parasitoids (Weis, 1982a; J. O. Stireman, unpublished data). Therefore, phenotypic and evolutionary divergence in the midge may result from a combination of direct, top-down selection from parasitoids and extrinsic selection against interpopulation mating that causes the expression of maladapted gall phenotypes. The symbiotic association may also facilitate host-shifts, and therefore, host-plant-mediated evolutionary divergence (Stireman et al., 2010), because A. carbonifera does not have to directly contend with gall-induction resistance or deleterious variation in primary or secondary chemistry in its host plants. It has been posited that fungal associations may increase the host plant range of ‘ambrosia’ gall midges compared to gall midge species that do not have fungal associations (Roskam, 2005; Janson et al., 2008).
Implications for the evolution of gall midge–fungus associations
Molecular phylogenetic analysis demonstrated that the symbiotic fungal associate of A. carbonifera is the filamentous ascomycete Botryosphaeria dothidea (Botryosphaeriales: Botryosphaeriaceae). B. dothidea is a cosmopolitan, opportunistic parasite (sometimes endophyte) of dozens of plant families throughout the world, including many economically important woody crop plants (Slippers & Wingfield, 2007). Bissett & Borkent (1988) hypothesized that all symbiotic gall midges are associated with an anamorphic (asexual) Botryosphaeria species, which our study supports. Botryosphaeria species (mostly dothidea) have been found in association with a number of other gall midge species, including Asphondylia species in Australia, South Africa (Adair et al., 2009) and North America (J. B. Joy, E. M. Janson, unpublished data; I. Park, persional communication), other Asteromyia species in North America (E. M. Janson, unpublished data) and at least one Lasioptera species in Europe (Rohfritsch, 1997).
In the context of the symbiosis, association with this particular fungus would be beneficial to gall midges because of B. dothidea’s low host plant specificity, weak pathogenicity and cosmopolitan distribution. In addition, it appears that A. carbonifera is associated with a single lineage within B. dothidea. Given the apparent lack of vertical transmission, the association between A. carbonifera and its symbiont exhibits a level of specificity that might not be expected of an ectosymbiotic association, especially because several nascent phylogenetic lineages comprise B. dothidea (e.g. this study, Adair et al., 2009). This is because nonendosymbiotic symbioses are believed to be particularly susceptible to invasion and symbiont replacement (Buchner, 1965). How A. carbonifera (or any gall midge) maintains this broad level of symbiotic specificity is unclear. As hypothesized for all ambrosia gall midges, it appears that females actively collect conidia (asexual reproductive spores) from the environment sometime after eclosing as adults (Borkent & Bissett, 1985; Bissett & Borkent, 1988; Rohfritsch, 2008). It also appears that they do not collect conidia from the fungus in their natal galls, because reproductive structures are rarely observed at the time of adult emergence (Bissett & Borkent, 1988; Rohfritsch, 2008; Adair et al., 2009) and recently eclosed adult females lack conidia in their mycangia (Bissett & Borkent, 1988; Heath & Stireman, in press). Therefore, females may collect conidia from fruiting fungus found on old galls in the leaf litter (Bissett & Borkent, 1988), from galls where resident midges/parasitoids have died (Adair et al., 2009), or from other parts of the host plant (Rohfritsch, 1997). Broad specificity could be maintained if females only collect conidia that are found in association with Asteromyia galls (e.g. pseudo-vertical transmission; Wilkinson & Sherratt, 2001). On the other hand, females could collect conidia from B. dothidea that is not in association with Asteromyia galls, although how A. carbonifera would be able to distinguish the symbiotic lineage of B. dothidea from nonsymbiotic lineages is unknown.
The association of several insect lineages with a single, nonco-diversified (but not necessarily genetically homogenous) species of ectosymbiont does bear some resemblance to another well-studied fungal ectosymbiosis, the attine ants (Silva-Pinhati et al., 2004; Mikheyev et al., 2006; 2007). Several ant species share nonco-diversified fungal associates, and the symbionts are shared among populations that are geographically isolated. In the attine ants, regular genetic recombination within the symbiotic associate and long distance dispersal appears to be important in maintaining symbiont homogeneity (Mikheyev et al., 2006). For A. carbonifera, genetic homogeneity across divergent midge populations is likely maintained by regular horizontal transmission and long distance dispersal of an asexual symbiotic fungal lineage. Lack of strict vertical transmission and long distance dispersal of asexual genotypes leads to the observed lack of structure and apparent lineage sharing compared to the significant genetic structure observed in the midge. How general these patterns are, is difficult to say because of the lack of studies on the population genetic patterns of truly cosmopolitan, host-plant generalist fungi. Botryosphaeria fungal symbionts of Asphondylia species from Australia and South Africa are similar to the A. carbonifera fungal associate, except there is a well-supported phylogenetic split between the South African and Australian isolates (Adair et al., 2009). Thus, intracontinental long distance dispersal and horizontal transmission may be common for fungi associated with gall midges, while intercontinental dispersal may be less so. Whatever the cause of phylogenetic divergence in Botryosphaeria, the Asteromyia fungal symbiont appears to move between free-living (not insect associated) and symbiotic (insect associated) states and is shared by deeply divergent and geographically isolated genera of gall midges that gall phylogenetically distant plant taxa.
The interaction between gall midges and fungi represent an extreme case where intimacy with and dependency on microbial partners does not necessarily imply a reduction in gene flow from free-living microbial populations or coupling of evolutionary histories. A consistently striking result from other insect–microbial symbioses is that symbiotic microorganisms frequently experience evolutionary forces that are distinct from their free-living hetero- and conspecifics (Moran et al., 1995; Lutzoni & Pagel, 1997; Clark et al., 2000; Wernegreen, 2002; Hosokawa et al., 2006; Kikuchi et al., 2009). Why the fungi associated with A. carbonifera (and other gall midge species) do not appear to experience unique demographic or inter-generational transmission processes remains an open question. There may be little selection on symbiotic traits per se for B. dothidea or such selection may be overwhelmed by selection external to the interaction, because the symbiosis does not appear to be protected or sequestered in any way. Perhaps symbionts are recurrently replaced from free-living stocks, and the A. carbonifera symbiosis has not had ample time to exhibit the signatures of co-evolution and specialization. However, because there are independent instances of fungal symbiosis in gall midges (e.g. Adair et al., 2009), the acquisition of a contemporary B. dothidea would have had to occur convergently in several gall midge lineages around the world. Another possibility is that A. carbonifera and B. dothidea are not involved in a mutualism at all, but instead A. carbonifera (and other ambrosia gall midges) are parasitizing a normally free-living lineage of B. dothidea. The fungal symbiont serves as a source of nutrition (Janson et al., 2009) and defence from natural enemies (Weis, 1982b; Stireman et al., 2008), and thus the benefits conferred to A. carbonifera from B. dothidea are clear. What is less clear is whether B. dothidea receives similar benefits. Such an explanation may underlie the apparent evolutionary misalignment of midges and their fungal symbionts.
We thank Glen Stanosz for providing free-living B. dothidea cultures. Tim Carr provided preliminary information in the fungal symbiont of A. carbonifera. A. Elizabeth Arnold provided technical advice and general mycological expertise. Jeremy Heath collected B. dothidea samples from A. carbonifera eggs and provided unpublished data on aspects of the association. Jeff Joy shared unpublished data about the fungal associate of creosote-galling Asphondylia species. Il-Ju Park shared unpublished data on the fungal symbiont of mesquite-galling Asphondylia. Steve Heard collected the Canadian A. carbonifera gall samples. Vikram Chhatre, Rob Brucker, Charles Hassett and Jessica Mezzanotte provided assistance in the lab. The thoughtful comments of D.J. Funk, D.E. McCauley and two anonymous reviewers greatly improved this manuscript. This work was supported by National Science Foundation grant DEB 0614433 awarded to P.A. and J.O.S.