Direct and indirect effects of viral pathogens and the environment on invasive grass fecundity in Pacific Coast grasslands

Authors


*Correspondence author. E-mail: seabloom@science.oregonstate.edu

Summary

1.  Pathogens can have strong effects on their hosts and can be important determinants of biological invasions. In natural systems, host–pathogen interactions may be mediated by direct environmental effects on pathogen communities and host fitness.

2.  While environmental mediation of host–pathogen interactions has been investigated experimentally and at single sites, there have been few studies tracking pathogen effects on lifetime host fecundity across large naturally occurring environmental gradients.

3.  If environmental factors directly mediate both pathogen transmission and host fecundity, laboratory and local-scale studies may not predict pathogen effects across large spatial scales.

4.  Here we investigate the relationship between host fecundity and infection by a suite of RNA viral pathogens, by surveying two invasive annual grasses at 18 locations along a 1200-km latitudinal gradient on the west coast of North America.

5.  Infected hosts of both species had 28–30% lower fecundity than uninfected hosts in our field surveys. However, the correlation of reduced fecundity to infection arose from indirect effects of the environment on both host fecundity and pathogen prevalence, rather than direct effects of the pathogen on the host. Pathogen prevalence was highest at sites where uninfected hosts had lowest fecundity.

6.Synthesis. In past experimental inoculations, virus infection reduced fecundity of these host species. Against this background, the results of our geographic-scale survey demonstrate the challenges not only of inferring cause from correlation, but also of extrapolating from local studies and experimental inoculations to larger spatial scales. Our results highlight a need for experimentally manipulating infection across environmental gradients. Such an integrated approach would allow quantification of the fitness impacts of infection, even when the environment directly affects both prevalence and host fecundity.

Introduction

Pathogens can have strong effects on the composition of natural communities. Determining the effects of pathogens on hosts is of particular relevance for invasive species, as pathogens can mediate biological invasions (Anderson & May 1986; Dobson & Crawley 1994; Tompkins, White & Boots 2003; Torchin & Mitchell 2004; Mitchell et al. 2006; Borer et al. 2007b). However, pathogen effects on hosts occur within a larger abiotic and biotic environmental context. Environmental gradients in resources, competitors and consumers all directly alter host vital rates, independent of pathogens (Mitchell et al. 2006). In addition, environmental factors can alter pathogen prevalence by changing pathogen transmission (Gregory 1973; Agrios 1978; Fitt, McCartney & Walklate 1989; Madden, Yang & Wilson 1996; Aylor 1999), vector communities (Cumming & Guegan 2006), resource supply rates (Mitchell et al. 2006), host community composition (Power & Mitchell 2004; Keesing, Holt & Ostfeld 2006) and abundance of natural enemies (Packer et al. 2003; Malmstrom et al. 2006). As a result, pathogen and environmental effects on host fitness are potentially confounded (Hassell et al. 1982; Holmes 1982; Thrusfield 2005).

There are at least two types of effects on host fecundity that could be obscured by such environmental confounding. Environmental conditions can modulate pathogen effects on host fecundity. This phenomenon of environmental modulation is well studied using experimental inoculations or natural infections in the laboratory or single field sites (Park 1948; Malmstrom et al. 2006). Conversely, pathogens may mediate environmental effects on host fecundity. This phenomenon occurs when infected and uninfected hosts respond differently to abiotic gradients (Malmstrom et al. 2006), competition (Park 1948; Malmstrom et al. 2006; Borer et al. 2007b) or consumer effects (Packer et al. 2003; Malmstrom et al. 2006).

However, few studies have attempted to separate these effects on large-scale, naturally occurring environmental gradients. This may be because of the difficulty of assembling a data set in which the infection status and lifetime fate and fecundity of individual hosts are known for a spatially extensive set of sites with known environmental conditions. While there are many large-scale studies of pathogen prevalence (Berger et al. 1998; Smith et al. 2002), these rarely measure host fecundity or fitness (but see Pioz et al. 2008). As a result, it remains unclear whether local-scale studies of pathogen effects can be scaled up to mesh with larger-scale observations of disease prevalence. Scaling up can lead to incorrect conclusions when the environmental conditions at a local study site are not representative of environmental conditions across the region. It also remains unclear how large-scale surveys can be used to predict pathogen impacts on local host populations. Scaling down can lead to incorrect conclusions when local sites vary environmentally, and there is either environmental confounding of host fecundity and infection, or environmental modulation of the fitness impacts of infection.

The disconnect between studies of pathogens in the laboratory or at single sites and large-scale epidemics is particularly relevant to studies of the role of pathogens in biological invasions, because of the need to forecast pathogen impacts at large spatial scales and in novel environments. The invasion of exotic annual grasses into grasslands of western North America presents a unique opportunity to examine pathogen–environment interactions within a system of great relevance to conservation biology. The invasion of these grasslands by exotic annual grasses from the Mediterranean region is one of the most dramatic invasions world-wide, including over 9 million ha in California alone (Heady 1977). Invasion of annual grasses into native perennial grass communities may have been facilitated by a suite of phloem-limited pathogens, collectively referred to as barley and cereal yellow dwarf viruses (B/CYDVs) (Malmstrom et al. 2005b; Borer et al. 2007b).

B/CYDVs in their exotic annual hosts are well suited for large-scale surveys and assessment of lifetime fecundity effects of pathogens in natural systems. Many exotic annual grass species have quite extensive latitudinal distributions. Experimental inoculation by these viruses in laboratory and field trials can significantly reduce annual grass biomass and fecundity (D’Arcy 1995; Malmstrom et al. 2005a). For example, greenhouse inoculations with BYDV-PAV decreased total biomass (above- and below-ground) of two widespread and common annual grass hosts that are the focus of the current study, Avena fatua and Bromus hordeaceus (41% and 39% respectively; J.P. Cronin, M.E. Welsh, M. Dekkers, C.E. Mitchell, unpublished data). We surveyed populations of these two invasive annual grass species at 18 grassland sites spanning a threefold gradient in rainfall (434–1448 mm year−1) and 1200 km of latitude along the west coast of North America. We screened each of the 568 individual host plants for four common RNA viral pathogen species (barley and cereal yellow dwarf viruses, B/CYDVs). In addition, the lifetime fecundity of annual grasses is easily measurable in the field as the seed production at the end of a growing season, so we could directly measure lifetime fecundity and infection status in naturally occurring individuals.

We use these data to investigate the following three questions to clarify the roles of the environment and pathogen infection on host fecundity:

  • 1 What is the relationship between infection status and host fecundity? To answer this question, we first compare the fecundity of infected and uninfected hosts across all sites. We then compare fecundity of infected and uninfected hosts after controlling for among-site variability (i.e. we determine the effect of pathogen nested within a site) to test whether infection status explains any residual variation in host fecundity.
  • 2 What is the direct effect of the environment on host fecundity? To answer this question, we use regression to find environmental determinants of fecundity independent of the pathogen (i.e. fecundity in uninfected hosts across all sites).
  • 3 What is the relationship between environmental quality and pathogen prevalence? To answer this question, we compare our pathogen-independent measure of site quality (fecundity of uninfected hosts) with site-level pathogen prevalence.

Materials and methods

Study system

B/CYDVs are a suite of aphid-vectored viruses in the family Luteoviridae that are known from over 150 grass hosts (Irwin & Thresh 1990; D’Arcy 1995). These viral pathogens cause one of the most economically important viral diseases of cereal crops (barley yellow dwarf) and are some of the most prevalent of all pathogens (Irwin & Thresh 1990). Recent work suggests that the presence of these viruses is a necessary precursor to one of the most widespread and persistent plant invasions world-wide (Malmstrom et al. 2005b; Borer et al. 2007b), the conversion of 25% of the area of California to annual grassland dominated by exotic annual species from the Mediterranean region (Heady 1977; Seabloom et al. 2003).

Infection by a B/CYDV leads to increased mortality, stunting and decreased fecundity in crops and natural systems (Rochow 1970; D’Arcy 1995; Malmstrom et al. 2005a). There is no vertical transmission of the viruses, so seedlings of infected parents are initially uninfected (Rochow 1970). B/CYDVs are transmitted by at least 25 different aphid species (Halbert & Voegtlin 1995). The viruses do not replicate in the aphid vectors and are not transmitted to aphid offspring (Rochow 1970; Agrios 1978). Aphids can acquire the viruses in as short a time as 15 min, and viruliferous aphids can inoculate a plant in 2 h, although efficiency increases with acquisition and inoculation time (Gray et al. 1991; Power & Gray 1995). BYDV and CYDV species belong to distinct genera and the different viral species differ in their virulence and the suite of aphids that serve as efficient vectors (Miller & Rasochova 1997).

Field survey of prevalence and fecundity

In May 2006, we sampled B/CYDV prevalence in natural grasslands at eight research reserves in California and Oregon (Fig. 1; Table S1 in the Supporting Information). At the larger reserves, we sampled up to three populations per reserve that were separated by more than 500 m for a total of 18 sample sites. Selected sites were in open oak woodland that was not actively grazed, although cattle and sheep grazing occurs at some of the reserves.

Figure 1.

 Sites included in survey of B/CYDV prevalence and invasive grass fecundity conducted on the west coast of the United States. Numbers indicate the number of sites nested within larger reserves that have apparently overlapping points on this large-scale map.

We randomly collected up to 20 B. hordeaceus and 20 A. fatua whole individual plants (568 total hosts assayed) from a site. Bromus hordeaceus was collected at all sites, while we were only able to collect A. fatua at 10 of our sites (Table S1). Hosts were collected at peak biomass while leaf tissue was still green. While this collection time optimizes detection of the virus, seed mass is likely underestimated. However, it is unlikely that this downward bias would change our overall results, because at this point in the season, plants are not producing new flowers, total seed number is fixed, and total seed mass and seed number are tightly correlated in these types of annual grasses (see Results).

Fresh leaf tissue from each host was tested for infection by BYDV-PAV, BYDV-MAV, BYDV-SGV and CYDV-RPV via enzyme-linked immunosorbent assay (ELISA) using antibodies provided by Agdia (Elkhart, IN, USA) (Rochow 1986). In the rare cases where potential infections by two serologically related viruses were associated nearly 1 : 1 within individuals of a host species at a site, we regarded the potential infections of the virus with the weaker assay response (relative to standard controls on each microplate) as cross-reactions to the other virus, rather than as coinfections. The entire seed head of each host with all seeds attached was removed from each host, dried to a constant mass at 60 °C, and weighed to the nearest 0.01 g.

At each site, we quantified plant biomass by clipping, drying and weighing two 10 × 1 m strips to the nearest 0.01 g. We estimated the areal cover of each plant species present in two 0.5 × 1 m quadrats. Cover was estimated independently for each species, so total cover can sum to more than 100%. We collected and air-dried three 2.5 × 10 cm deep soil cores which were analysed for total phosphorous, nitrate, potassium, organic matter, sand, silt, clay and pH by A & L Western Agricultural Laboratories (Modesto, CA, USA).

All statistical analyses were conducted using r version 2.5.1 (R Foundation for Statistical Computing, Vienna, Austria). Prevalence data were analysed with logistic regression using the glm function in r. Among-site variability was treated as a random effect with population nested within reserve using the lme function from the nlme library in r. Note that we also fitted models in which site and site-within-population were nested within state, but the addition of this nesting factor did not improve model fit as there were no consistent differences between the two states.

Assessment of fecundity metrics

As part of a larger experiment (Seabloom et al. 2003), we established a series of experimental plots in a restored grassland at Sedgwick Reserve in Santa Barbara County in 2000. This is the location of one of our southern collection sites (Table S1). These plots ranged in size from 9 to 25 m2 and were planted with exotic annual and/or native perennial grasses. In addition, the plots were subjected either to a single summer burn, nitrogen addition (4 g N m−2 year−1 as NH4NO3) or watering to match the upper 95th percentile of the monthly rainfall (see Seabloom et al. 2003). Here we do not investigate the specific treatments, but use them as a broad range of environmental conditions to examine how changes in the biotic and abiotic environment alter seed production. In each of the 120 plots, we collected the above-ground portion of 10 individual B. hordeaceus plants just prior to seed fall in 2006. We counted all seeds from each plant and weighed the total seed mass and mass of non-reproductive tissue after drying to a constant mass at 60 °C.

Quantification of host mortality

As with all large-scale studies examining pathogen prevalence, patterns in the data do not account for host mortality prior to the sampling. This pre-sampling mortality could create a bias in the results if pathogen-induced mortality is highly variable among sites and is correlated with potential explanatory factors. We used an outplant study to quantify among-site variability in host mortality during the period of potential pathogen transmission (i.e. the period between when aphids arrive and are active to plant senescence at the end of the growing season). We conducted this study at a subset of five of the sites sampled during our 2006 observational study, two in Oregon (Finley and Baskett Slough) and three in California (Sierra Foothill, Mclaughlin and Hopland; Table S1). As it was logistically infeasible to do this study at all observational sites, we increased the range of environmental conditions of this experiment by subjecting the outplants to a factorial combination of nitrogen addition (control or 10 g N m−2 year−1 added quarterly as calcium nitrate) and phosphorous addition (control or 10 g P m−2 year−1 added quarterly as triple super phosphate).

The experiment was a complete randomized block design with two blocks at each of the five sites for a total of 40 experimental units (5 sites × 2 blocks × 2 levels of N × 2 levels of P).

Experimental units were 40 × 40 m. Locally collected seed of each host species was germinated in flats with 25 × 25 mm soil plugs. After c. 4 weeks, 25 uninfected plants of each host species were planted into each of the 40 × 40 m plots in January of 2008 prior to the first aphid flights. Individual hosts plants were planted within 20 × 50 cm quadrats placed 2 m apart along transects within each 40 × 40 m plot. Plants were surveyed after c. 5 months just prior to the start of senescence at the end of the growing season in late May or early June.

Results

Infection status and host fecundity

Overall infection prevalence at our sites ranged from 0% to 70% for B. hordeaceus (mean = 13.8%) and 0% to 55% for A. fatua (mean = 21%). The prevalences of the individual viruses in B. hordeaceus in order of abundance were BYDV-MAV (8.3%), CYDV-RPV (5.5%), BYDV-PAV (5.5%) and BYDV-SGV (4.3%). The prevalences of the viruses in A. fatua in order of abundance were BYDV-SGV (12.5%), BYDV-PAV (9.8%), BYDV-MAV (5.2%) and CYDV-RPV (4.7%). There were no significant differences among the two host species in total prevalence (i.e. infection by any virus) or any of the four individual viruses (P > 0.05 lme with site as a random factor). The three viruses that share vectors with at least one other virus (BYDV-MAV, BYDV-PAV and CYDV-RPV) were significantly positively correlated at the site scale (P < 0.05 and correlations ranging from 0.68 to 0.85), while the virus with only a single specialist vector, BYDV-SGV, was not significantly correlated with the other viruses (P > 0.05 and correlations ranging from 0.01 to 0.21), as has been reported in other work on this system (Seabloom et al. 2009, in press).

Bromus hordeaceus and A. fatua individuals infected with BYDV-SGV had significantly lower fecundity than did uninfected individuals. Infected B. hordeaceus hosts had 30% lower seed mass than infected hosts. Similarly, infected A. fatua hosts had seed masses that were 28% lower than infected hosts. These effects were predominantly associated with infection by BYDV-SGV that resulted in a 45.2% reduction in B. hordeaceus hosts and a 42.1% reduction in A. fatua hosts (Fig. 2; Table 1).

Figure 2.

 Overall fecundity of two invasive annual grasses ((a) B. hordeaceus and (b) A. fatua) across populations infected by four species of RNA viral pathogens. Error bars show one SE.

Table 1.   Overall test of relationship between infection status and fecundity [total seed mass; log10(mg)] of two annual grasses: Bromus hordeaceus (total d.f. = 344) and Avena fatua (total d.f. = 170)
Hostd.f.SourceEstimateStandard errort-valueP
Bromus hordeaceus1Intercept2.00350.020597.58600.0000
1BYDV-PAV0.02730.15250.17900.8581
1BYDV-MAV−0.16270.1096−1.48400.1386
1BYDV-RPV0.07390.11950.61800.5369
1BYDV-SGV−0.26480.0973−2.72200.0068
Avena fatua1Intercept2.73070.029293.49200.0000
1BYDV-PAV−0.02930.1098−0.26700.7899
1BYDV-MAV−0.15750.1329−1.18500.2377
1BYDV-RPV0.08140.13820.58900.5566
1BYDV-SGV−0.23750.0781−3.04100.0027

The fecundity differences between infected and uninfected hosts of both species were explained by variation in seed mass among sites. There were no significant relationships between infection by any of the viruses and host fecundity in models that accounted for fecundity differences among sites (i.e. reserves and populations nested within reserves; Table 2).

Table 2.   Results of linear mixed-effects model relating infection status of four viral pathogens and fecundity [total seed mass; log10(mg)] of two annual grasses: Bromus hordeaceus (total d.f. = 344) and Avena fatua (total d.f. = 170). Site and populations are treated as random effects with populations being nested within site. Bromus hordeaceus had an estimated standard deviation among sites of 0.1810 and among populations within sites of 0.1816. Avena fatua had an estimated standard deviation among sites of 0.2451 and among populations within sites of 0.1438. Note that the addition of state as a random effect did not improve model fit, so the simpler models are presented here
Hostd.f.SourceEstimateStandard errort-valueP
Bromus hordeaceus1Intercept1.97610.081824.16890.0000
1BYDV-PAV0.09880.11280.87590.3817
1BYDV-MAV−0.02750.0919−0.29970.7646
1BYDV-RPV−0.00320.0898−0.03600.9713
1BYDV-SGV0.00010.07530.00110.9991
Avena fatua1Intercept2.74640.134520.41470.0000
1BYDV-PAV−0.02900.0800−0.36280.7172
1BYDV-MAV−0.08740.0990−0.88270.3787
1BYDV-RPV0.07420.10110.73390.4641
1BYDV-SGV−0.02210.0623−0.35430.7236

Direct effect of environment on host fecundity

Our sample sites spanned a large range of abiotic and biotic environmental variation. Soils were highly variable among sites for phosphorous (4–47 ppm), nitrate (1–10 ppm), potassium (55–485 ppm), organic matter (2.5–7%), sand (34–74%), silt (12–48%), clay (14–26%) and pH (5–7.7). Among sites, precipitation during the year of sampling (2006 rainfall year: July 2005–July 2006) ranged from 434 to1448 mm year−1. While precipitation generally increased with latitude (r = 0.81), we specifically included a longitudinal gradient in California to decouple latitude and precipitation, and our wettest site was in central California (Hopland; Table S1). Total above-ground biomass ranged from 88 to 897 g m−2, proportion host cover (grass cover per total cover) ranged from 18% to 85%, annual grass cover ranged from 8% to 130% and perennial grass cover ranged from 0% to 45%. The total richness of all grass hosts ranged from 3 to 6 species 0.5 m−2.

We looked for direct environmental covariates of uninfected host fecundity using backward selection to remove non-significant terms from an initial model that included July 2005–July 2006 rainfall, mean annual precipitation, latitude, host (i.e. grass) species richness, total grass cover, annual grass cover, perennial grass cover and proportional grass cover (grass cover per total cover). Fecundity of both uninfected A. fatua and B. hordeaceus increased with increasing host richness and precipitation, although the relative importance of these factors differed between our two study species. Fecundity of uninfected B. hordeaceus hosts was highest at sites with high host (i.e. grass) species richness (Fig. 3, Table 3), whereas fecundity of uninfected A. fatua hosts was highest at sites with high precipitation during the 2006 growing season (Fig. 3, Table 3). Biomass and soil data were unavailable from a few sites, however these variables were not significant in any statistical model, so we present results here on the wider range of sites including those with missing biomass and soil data.

Figure 3.

 Primary environmental drivers of fecundity of uninfected host species, B. hordeaceus (a and b) and A. fatua (c and d). Note that precipitation and richness are positively correlated (= 0.46), so both terms are not included in the same model after model selection. However, trends are concordant for fecundity of both hosts along richness (b and d) and precipitation gradients (a and c).

Table 3.   Backwards-selected regression models of environmental effects on fecundity of uninfected hosts of two annual grasses at 18 sites. Note that Bromus hordeaceus was found at all sites while Avena fatua was found at 10 sites. The total model included July 2005–July 2006 rainfall, mean annual precipitation, latitude, host (i.e. grass) species richness, total grass cover, annual grass cover, perennial grass cover and proportional grass cover (grass cover per total cover). Model selection was based on AIC using step procedure in R on the total model described above
Hostd.f.SourceEstimateStandard Errort-valueP
Bromus hordeaceus1Intercept1.3080.1956.6950.000
1Host richness0.1640.0473.5210.003
Avena fatua1Intercept2.1250.13116.2630.000
1Precipitation (mm)0.0010.0004.6620.002

Direct effect of environment on pathogen prevalence

For both hosts, BYDV-SGV prevalence declined with increasing site quality, as measured by the fecundity of uninfected hosts (Figs 4 and 5). BYDV-SGV infection prevalence also declined with precipitation (r = −0.70), host richness (r = −0.41) and latitude (r = −0.46). Similarly, prevalence of BYDV-PAV and BYDV-MAV in B. hordeaceus was lower at high-quality sites (Table 4). CYDV-RPV prevalence was unrelated to environmental quality for either host (Figs 4 and 5).

Figure 4.

 Relationship between environmental quality, as measured by uninfected host fecundity, and site-level prevalence of four viral pathogen species ((a) BYDV-PAV, (b) BYDV-MAV, (c) CYDV-RPV and (d) BYDV-SGV) in Bromus hordeaceus.

Figure 5.

 Relationship between environmental quality, as measured by uninfected host fecundity, and site-level prevalence of four viral pathogen species ((a) BYDV-PAV, (b) BYDV-MAV, (c) CYDV-RPV and (d) BYDV-SGV) in Avena fatua.

Table 4.   Logistic regressions of site-level prevalence of four viral pathogens in two annual grass hosts. Bromus hordeaceus was sampled at 18 sites and Avena fatua was sampled at 10 sites
HostPathogend.f.SourceEstimateStandard errorP
Bromus hordeaceusBYDV-MAV1Intercept2.80241.61020.0818
1log seed mass−2.69230.86760.0019
BYDV-PAV1Intercept2.05201.90200.2807
1log seed mass−2.51901.02500.0140
CYDV-RPV1Intercept−0.27561.76890.8760
1log seed mass−1.28740.92300.1630
BYDV-SGV1Intercept8.49803.13900.0068
1log seed mass−6.34001.81100.0005
Avena fatuaBYDV-MAV1Intercept−2.16523.50500.5370
1log seed mass−0.25731.30010.8430
BYDV-PAV1Intercept−1.22092.60860.6400
1log seed mass−0.34830.96860.7190
CYDV-RPV1Intercept−3.65263.91020.3500
1log seed mass0.20361.43740.8870
BYDV-SGV1Intercept11.92403.65300.0011
1log seed mass−5.40001.45100.0002

It is possible that the fecundity of uninfected hosts is not independent of prevalence. For example, uninfected hosts could experience competitive release at sites with high levels of infection (Borer et al. 2007a). Furthermore, it is possible that the effects of infection may depend on environmental conditions at a site. There was no evidence in our data of these types of interactions between site quality, infection status, and effects of infection on fecundity. Regressions of seed mass of infected hosts on the seed mass of uninfected hosts had slopes indistinguishable from 1.0 for A. fatua (slope = 0.987, SE = 0.184) and B. hordeaceus (slope = 0.885, SE = 0.170). Thus, the fecundity of both infected and uninfected hosts was strictly a function of the conditions at the local site.

Assessment of fecundity metrics

Our metric of fecundity is the total mass of seed produced by each host plant. Total seed mass is tightly correlated with seed number and total plant size for annual grasses. For example, in B. hordeaceus log10(total seed mass per plant) and log10(total number of seeds per plant) have a correlation of 0.932 (P < 0.0001 on 119 d.f.). Similarly, in B. hordeaceus, log10 (total seed mass per plant) and log10(total plant mass) have a correlation of 0.937 (P < 0.0001 on 119 d.f.). Thus our measure of fecundity accurately reflects both total energetic allocation to reproduction (total seed mass), number of offspring (total number of seeds) and host vigour (total mass per plant).

Quantification of host mortality

Our outplant study tracked the fate of 985 A. fatua and 1083 B. hordeaceus individuals. At the end of the growing season after 5 months in the field, 86% (0.03 SE among all forty 40 × 40 m plots) of B. hordeaceus and 79% (0.03 SE among all forty 40 × 40 m plots) of the A. fatua were still alive. Thus, mortality due to all causes (including all pathogens, herbivory, competition) was only 14–21% during the period when B/CYDV can be transmitted. Based on these results, direct mortality from B/CYDV in the field did not exceed 14–21% and is certainly much less than this, given that some significant fraction of the mortality was likely due to competition from the existing plants, herbivores, disturbance and other pathogens. More importantly for this work, there is little variability in survival among sites even with the experimental nutrient addition included as indicated by the small standard errors in survival. The relatively high survival of the host plants and lack of variability among sites makes it unlikely that the observed patterns of prevalence are highly skewed by strong differences in pathogen-induced mortality among sample sites.

Discussion

Taken as a whole, we found that invasive annual grass hosts infected by BYDV-SGV had 42–45% lower fecundity than did uninfected hosts. To a lesser degree, infection by other B/CYDV viruses followed a similar pattern (28–30% reduction in host fecundity). However, the fecundity effects in the field surveys were due to site-level variability in environmental conditions rather than direct pathogen effects. There were no residual differences between the fecundity of infected and uninfected hosts after accounting for site-level variation in fecundity. Similarly, a non-manipulative study at a local site also did not find a consistent negative correlation between B/CYDV infection and host fecundity across three host species including B. hordeaceus (Remold 2002). In contrast, laboratory and field experiments conducted at single sites have demonstrated strong effects of B/CYDV inoculation on fecundity of our two focal host species (Griesbach et al. 1990; Malmstrom et al. 2005a). The lack of coupling between experimental manipulations of pathogens at single sites and both local and large-scale surveys demonstrates that results of large-scale pathogen surveys must be considered within an environmental context. If the environment drives prevalence and host fecundity, then population-level impacts in natural epidemics may not be detectable without experimental inoculations.

Individual host fecundity was strongly correlated with both the abiotic and biotic environment. In particular, high-quality sites, where uninfected B. hordeaceus and A. fatua plants had highest fecundity, had high rainfall and high numbers of grass species. This may have arisen from the sites being favourable for grasses generally as indicated by the more diverse grass flora. Given the observational nature of these data, we are not fully able to assign causation to any environmental covariates. For example, host richness and precipitation are positively correlated, as sites with higher rainfall were also located further north and had higher overall grass diversity. Fecundity of our focal species was not correlated with total biomass, soil nutrient levels or abundance of competitive dominant hosts (perennial grasses; Seabloom et al. 2003).

Several limitations of large-scale surveys limit the ability of this study to establish direct causal links between specific environmental factors and the fecundity of infected and uninfected hosts. Ultimately, host fecundity and pathogen prevalence may be driven by factors not included in our model, such as the presence of irrigated agricultural fields (Griesbach et al. 1990; Hewings & Eastman 1995). Furthermore, our data present a single snapshot in time, while B/CYDV prevalence and impacts can vary widely from year to year (Hewings & Eastman 1995; Seabloom et al. 2009). Because of these limitations, we focus the current work on the more limited goal of examining the correlation between infection status, environment and host fecundity along a gradient in site quality, measured by fecundity of uninfected hosts.

Pathogen prevalence, particularly BYDV-SGV, declined strongly with site quality. For both host species, prevalence of BYDV-SGV declined similarly with increasing precipitation, host abundance and fecundity of uninfected hosts. The observed negative relationship between pathogen prevalence and site quality could arise from either host or vector responses. From the perspective of the host, weakened hosts with low fecundity may also be less able to mount successful countermeasures against pathogen infections or vector attacks. This is unlikely in the case of the B/CYDVs as there has been little evidence of general host resistance in grasses (Wang, Abbott & Waterhouse 2000). However, this is a well-known phenomenon in other plant and animal systems (Lochmiller, Vestey & Boren 1993; Saino, Calza & Moller 1997; Klasing 1998; Fargallo et al. 2002; Kidd 2004; Cunningham-Rundles, McNeeley & Moon 2005; Smith, Jones & Smith 2005).

Vector responses also could control this pattern, if vector movement rates change with host quality (Kilpatrick et al. 2006). In our system, plant chemistry, particularly free amino acid content, is extremely important for foraging aphid preference, host selection and movement among individuals (Powell, Tosh & Hardie 2006). Stressed plants tend to accumulate relatively high levels of free amino acids (Barnett & Naylor 1966; White 1984), the primary source of dietary nitrogen for aphids (Terra 1988). Thus plants at lower quality sites (e.g. drought-stressed) are likely to have phloem with relatively high free amino acid content which could lead to increased preference by foraging aphid vectors (Powell, Tosh & Hardie 2006; Borer et al. 2009), elevated aphid reproductive output, and ultimately higher aphid densities at poor quality sites (Huberty & Denno 2004; Borer et al. 2009). High densities of aphids tend to increase both short and long-distance aphid movement (Mueller, Williams & Hardie 2001), a primary determinant of transmission rates among hosts (Power & Gray 1995; Borer et al. 2009). Thus, abiotic drivers such as precipitation may have indirectly increased pathogen prevalence by increasing host stress, which would also decrease host fecundity.

We acknowledge that there are limitations inherent in the inferences that are possible from observational data on pathogen prevalence. Ultimately, determining the role of vectors would require surveys of aphid density and fecundity on multiple host species at each site. While this can be accomplished at single sites or in a laboratory setting (Malmstrom et al. 2005b; Borer et al. 2009), it is intractable in a survey of this spatial scale given the irruptive nature of aphid populations. Furthermore, our prevalence data do not account for pre-sampling mortality. As with all prevalence data, it is important to consider potential biases created by pre-sampling mortality. While the scale of this work precluded tracking the mortality of each host throughout the season, our outplant study suggests that roughly 80% of hosts survive the period during which B/CYDVs can be transmitted. More importantly, there was relative little variability in survival among sites. The high survival and low among-site variability make it unlikely that pre-sampling mortality created substantial biases in our analyses.

While not the case in the data presented here, it is also possible to find positive correlations between pathogen prevalence and host fecundity. For example, vectors may preferentially select or reproduce more quickly on larger, fitter hosts (Remold 2002). Infection by a mild pathogen may also alter host susceptibility to secondary attack by other enemies, creating a positive correlation between fecundity and primary infection (i.e. systemic acquired resistance Agrios 1978; Durrant & Dong 2004; Apriyanto & Potter 1990; Gibbs 1980). If these processes occurred within each site, this would contribute to masking any negative effects of the virus on host fecundity, perhaps explaining the lack of correlation after controlling for site-level variation.

Environmental mediation of pathogen impacts is a well-known principle in epidemiology and is well studied in crop, human and domestic animal systems (Hassell et al. 1982; Holmes 1982; Thrusfield 2005). While studies of pathogens in natural systems have proliferated, studies rarely measure pathogen impacts on host fecundity across large-scale environmental gradients. Large surveys of pathogen prevalence exist, yet it remains unclear how to relate local-scale studies of pathogen impacts with larger-scale studies of prevalence. Our results demonstrate that environmental effects on host fecundity and pathogen prevalence can confound measures of pathogen impacts across environmental gradients. Thus, using measurements of pathogen impacts from laboratory and single-site studies to predict large-scale impacts requires integration with the study of large-scale drivers of infection rates and fecundity in natural systems. Similarly, observational surveys are improved when coupled with manipulative experiments. Given our growing awareness of the impacts of pathogens on the fate of invasive and imperilled species (Torchin & Mitchell 2004), a clearer understanding of pathogen impacts spanning entire host species’ ranges is increasingly pressing.

Acknowledgements

Support for this project was provided, in part, by NSF/NIH EID 05-25666 to E.T.B. and E.W.S. and NSF/NIH EID 05-25641 to C.E.M. We also thank the UC Natural Reserve System (Hastings, Mclaughlin and Sedgwick), the UC Research and Extension Centers (Hopland and Sierra Foothill) and the US Fish and Wildlife Service Wildlife Refuge System (Finley and Baskett Slough). E. Orling and A. Brandt assisted in the collection of field data. E. Pulley, S. Waring and M. Welsh assisted in conducting the viral assays.

Ancillary