Effects of habitat fragmentation on choke disease (Epichloë bromicola) in the grass Bromus erectus

Authors

  • Kathleen Groppe,

    1. Botanisches Institut der Universität Basel, Hebelstrasse 1, CH-4056 Basel,
    Search for more papers by this author
  • Thomas Steinger,

    1. Botanisches Institut der Universität Basel, Hebelstrasse 1, CH-4056 Basel,
    2. Institut für Umweltwissenschaften, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich,
    Search for more papers by this author
  • Bernhard Schmid,

    1. Botanisches Institut der Universität Basel, Hebelstrasse 1, CH-4056 Basel,
    2. Institut für Umweltwissenschaften, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich,
    Search for more papers by this author
  • Bruno Baur,

    1. Botanisches Institut der Universität Basel, Hebelstrasse 1, CH-4056 Basel,
    2. Department of Integrative Biology, Section of Conservation Biology (NLU), University of Basel, St Johanns-Vorstadt 10, CH-4056 Basel, Switzerland
    Search for more papers by this author
  • Thomas Boller

    1. Botanisches Institut der Universität Basel, Hebelstrasse 1, CH-4056 Basel,
    Search for more papers by this author

B. Schmid, Institut für Umweltwissenschaften, Universität Zürich, Winterthurerstrasse 190, CH-8057 Zürich, Switzerland (e-mail:bschmid@uwinst.unizh.ch).

Summary

  • 1 The fungal endophyte Epichloë bromicola causes choke disease on Bromus erectus, thereby suppressing maturation of host inflorescences. We conducted a 3-year field experiment to investigate the effects of small-scale habitat fragmentation on the occurrence of choke disease in three calcareous grasslands in the Swiss Jura Mountains, where, overall, 4.3% of all flowering tillers had diseased inflorescences.
  • 2 The number of diseased B. erectus plants (presumed genets), the number of tillers with diseased inflorescences, and the number of tillers with healthy inflorescences were counted over 3 years, but healthy tillers could not be allocated to particular genets. Each of 12 blocks contained one large (4.5 × 4.5 m), one medium (1.5 × 1.5 m) and two small (0.5 × 0.5 m) fragments and corresponding control plots. The percentage of tillers in the plots that were infected but did not show disease symptoms (asymptomatic plants) was estimated in the final year of the study using a diagnostic polymerase chain reaction. On average 1.7% of all tillers without disease symptoms were infected.
  • 3 There were significantly more diseased plants in fragment than in control plots, particularly in small fragment plots or in the third year of the study, indicating that disease incidence in the host plant increased after fragmentation if assessed at the level of the genet population. This was probably due both to a switch of genets from the asymptomatic to the symptomatic state and to increased horizontal transmission of the disease in fragments.
  • 4 The increase in the number of flowering tillers with diseased inflorescences was outweighed by that in the number of tillers with healthy inflorescences. Disease incidence was therefore decreased by fragmentation if assessed at the level of the host flowering tiller population. The effect on healthy plants was probably due to beneficial abiotic edge effects following fragmentation.
  • 5 Plot size affected fragmentation response, with the largest increases in small, followed by medium and large fragments. Similarly, the strength of fragmentation responses increased through time over the 3-year study period.
  • 6 Significant site to site and year to year variation in the number of diseased plants and in the number of tillers with diseased inflorescences suggested that separate experiments with replication within site and year would have yielded a series of interesting but individually different results. Only by repeating the experiments at several sites and over several years was it possible to obtain the general results reported in the previous points.

Introduction

Habitat fragmentation has significant consequences for animals and plants (Soulé 1986; Saunders et al. 1991). It directly affects species in ecological systems by eliminating or altering their habitat (e.g. Soulé 1986; Quinn & Hastings 1987; Bolger et al. 1991; Harrison 1991; Saunders et al. 1991; Margules et al. 1994; Diffendorfer et al. 1995), first by reducing the total area available and secondly by breaking the remaining habitat into more or less isolated fragments. The decreased area leads to reduced population sizes and thus to a higher risk of disappearance of some species from the small fragments, for instance due to chance events (Frankel & Soulé 1981; Shaffer 1981; Gilpin & Soulé 1986; Simberloff 1994). Moreover, because of isolation of fragments, recolonization following local extinction may be low (Hanski 1994; Gonzalez et al. 1998). A third effect is that population dynamics and species interactions may change in fragments through the increased impact of external factors such as physical disturbances, predation and parasitism (Lovejoy et al. 1986; Wilcove et al. 1986; Lord & Norton 1988; Kruess & Tscharntke 1994; Matthies et al. 1995; Gonzalez et al. 1998). These alterations can be accompanied by less obvious, indirect effects (Kareiva 1987; Robinson et al. 1992). For example, if fragmentation affects the distribution and abundance of insects, the resource base of insectivorous birds may also be altered (Wiens 1994). Thus, habitat fragmentation can influence an entire suite of processes, ranging from individual behaviour through population dynamics to ecosystem fluxes.

Habitat fragmentation is also known to precipitate a variety of abiotic changes, which can influence habitat characteristics and ecosystem function (Lovejoy et al. 1986; Saunders et al. 1991; Murcia 1995). Direct changes include alteration of relative humidity, wind exposure, light intensity, rainfall interception, and air temperature, and these can further affect nutrient transport, soil moisture, soil ecology and particulate matter deposition with consequences for the flora and fauna within a fragment (Lovejoy et al. 1986; Saunders et al. 1991; Murcia 1995). Long-term studies show particularly large changes in species composition, population dynamics and carbon fluxes along fragment edges (Pimm 1998).

Despite the large number of studies on the effects of habitat fragmentation, its relationship to host–pathogen interactions in plants has only so far been looked at in theoretical or observational studies (e.g. Holmes 1996). Here we report the first experimental study of such an effect. Species of the fungal genus Epichloë form systemic infections in above-ground tissues of grasses, as for instance E.bromicola in hosts from the genus Bromus. Flowering of infected B. erectus induces intercellular fungal hyphae to emerge and proliferate to form a fruiting body (stroma) that causes abortion of developing inflorescences, a condition known as choke disease. Stroma formation nearly always results in complete sterilization of the host (Groppe et al. 1999) and culminates in the production of infectious ascospores (Chung & Schardl 1997).

The growth rate of the fungus (White et al. 1991; White 1992; White et al. 1993; Bucheli & Leuchtmann 1996) and of the host (Kirby 1961; Emecz & Jones 1970) both determine whether choke disease will form stromata on infected plants. Choke disease is only able to manifest itself if fungal mycelium successfully entraps the expanding inflorescence primordium and surrounding leaves in the stroma. Failure to do so results in normal differentiation and elongation of the affected inflorescence (Emecz & Jones 1970; White et al. 1992). Because differences in the rate of development of the flowering apex (Cooper 1952; Cooper 1958) and of the fungal stroma (Kirby 1961; Emecz & Jones 1970; Sun et al. 1990) are controlled, at least in part, by environment, the occurrence of the symptoms of choke disease in a plant population may be affected by abiotic changes due to habitat fragmentation.

The specific aim of our 3-year field experiment was to determine whether small-scale habitat fragmentation affects stroma formation in E. bromicola-infected B. erectus hosts. In addition, we wanted to know if fragmentation effects are dependent on fragment size. We discuss these effects in relation to the effects of habitat fragmentation on the host plants per se.

Materials and methods

Study sites and experimental design

The response of plant–fungal associations to abiotic change caused by small-scale habitat fragmentation was examined in three species-rich calcareous grasslands belonging to the Teucrio-Mesobrometum type in the north-western Swiss Jura mountains (Zoller et al. 1986; Ellenberg 1988). Experimentally fragmented areas of the grasslands were created in spring 1993 by mowing all vegetation above 2 cm around the experimental plots within each of 12 blocks, distributed across the three sites. An experimental unit (= block) contained one large (4.5 × 4.5 m), one medium (1.5 × 1.5 m) and two small fragments (0.5 × 0.5 m), separated by a 5-m wide strip of frequently mown (6–12 times per year in the period from March to October) vegetation and a corresponding set of control plots, which were mirror-symmetrically arranged (Fig. 1). A total of 48 fragments (12 large, 12 medium and 24 small) and 48 control plots of the corresponding size were therefore established. Within each block the positions of fragment-control plot pairs as well as the control and treatment halves were randomized. Five blocks were situated on a SW-facing slope with an inclination of 19–22° near Nenzlingen (10 km S of Basel; elevation 500 m a.s.l.). Three blocks were situated on a SSE-facing slope with an inclination of 20–22° near Movelier (24 km SW of Basel; elevation 780 m). At the third field site near Vicques (24 km SSW of Basel; elevation 570 m) four blocks were situated on a SE-facing slope with an inclination of 15–27°. Detailed descriptions of the vegetation and invertebrate communities at the three sites are given in Baur et al. (1996).

Figure 1.

Layout of fragment and control plots in a block. Fragmented and control half were assigned randomly to left or right side of block. There were 12 blocks such as this one distributed over three sites. Shaded areas: grassland, mown once late in the year; unshaded area: very short vegetation kept at 2 cm by frequent mowing during the growing season.

Study organisms

Bromus erectus s.l. Hudson, or according to newer taxonomic work Bromopsis erecta (Huds.) Fourr., is a perennial grass species characteristic of dry and semi-dry soils (Hubbard 1984; Stace 1991). It is wind pollinated, self-incompatible and is capable of both propagation through seed production and clonal spread through tillers. Tillers are relatively closely packed within genets. Here we use the term ‘plant’ to refer to a cluster of tillers which presumably belong to a single genet. Epichloë bromicola Leuchtm. & Schardl (Clavicipitaceae, Ascomycotina) is a fungal endophyte of grasses of the host genus Bromus (Leuchtmann & Schardl 1998). E. bromicola forms stromata on developing inflorescence primordia in B. erectus hosts. Mixtures of inflorescences and stroma on a single host individual occur very rarely (Groppe et al. 1999) and infected plants are therefore generally either completely sterile or completely fertile. For reasons not well understood E. bromicola is able to infect seeds of other host species such as B. benekenii and B. ramosus, allowing vertical transmission, but this does not occur in B. erectus (Leuchtmann & Schardl 1998; Groppe et al. 1999). E. bromicola invades florets and ovaries in all three host species, but its hyphae disappear during further development of the gynoeceum of infected B. erectus so that seeds are endophyte free (Leuchtmann & Schardl 1998). E. bromicola was shown in preliminary surveys to infect B. erectus at all three field sites chosen for this study. Since the pathogen must spread horizontally, it takes two or more host growing seasons to become symptomatic in B. erectus. In the first season contageous ascospores, probably mostly transmitted via wind and rain, infect a previously uninfected grass plant (Chung & Schardl 1997), which in the second or a subsequent growing season then produces culms in which the inflorescence is sterilized by the fungal stroma. Each of these stroma-carrying flowering tillers which do not show an inflorescence (Fig. 2), or, in healthy plants a flowering tiller carrying a healthy inflorescence, was counted as a separate unit.

Figure 2.

Tillers of a diseased Bromus erectus plant. The stromata of Epichloë bromicola prevent inflorescence development in tillers at an early stage such that the inflorescences cannot be seen. This photograph illustrates that diseased plant genets can be identified because of their distinct characteristic, the stroma-carrying flowering tillers, as long as their frequency in the population is relatively low.

Measurements

All control and fragment plots within each of the three experimental sites (96 plots in total) were surveyed at peak flowering time (mid-June to early-July) in 1993, 1994 and 1995. Each plot was divided into 0.25 × 0.25 m segments using ropes and the number of diseased plants, the number of stroma-carrying flowering tillers, and the number of tillers with healthy inflorescences were counted. The number of diseased plants could be estimated because each typically formed a distinct cluster of choked tillers which was surrounded by healthy tillers. Healthy plants, however, could not be counted because often it could not be determined to which individual clump or genet a healthy tiller belonged. Therefore it was also not possible to determine disease incidence at the level of plants (i.e. proportion of infected genets out of all genets). A map was constructed for each plot showing the location of each diseased plant and the number of stroma-carrying flowering tillers per plant. Stromata of E. bromicola are morphologically distinct and could be readily distinguished from stromata of other Epichloë species present in the field experiment. We counted roughly 45 000 B. erectus tillers with healthy inflorescences and 2000 stroma-carrying flowering tillers per year.

Diagnostic polymerase chain reaction (pcr)

To test the frequency of asymptomatic infections in B. erectus, vegetative tissues were collected from randomly selected plants with healthy inflorescences. A total of four tillers from each medium and large plot, and two from each small plot, were tested (288 samples). Samples were harvested from the field, frozen on liquid nitrogen and powdered. DNA was extracted by the CTAB-SEVAG method and tested for the presence of fungal DNA by PCR using primers specific for a microsatellite-containing fungal target sequence (Groppe et al. 1995) as described previously (Groppe & Boller 1997). The sequences of the forward and reverse primers are 5′ CGC ACA ATA CGT CAG CTA GGA ATG 3′ and 5′ CCT GAA TCA ACT TTG CTA TCA GGC 3′, respectively. The molecular analysis was only possible in 1995 because appropriate methods were not available in previous years.

Statistical analysis

All measurements were expressed on a per-m2 basis, i.e. adjusted for fragment size, and transformed if necessary to obtain normally distributed residuals and homogeneous group variances in statistical analyses. Preliminary statistical analyses were carried out with the statistical package JMP (SAS 1989). After this exploratory stage the data were analysed using analyses of variance as implemented in the GENSTAT statistical language (Payne et al. 1987; McCullagh & Nelder 1989). The speciality of GENSTAT is the distinction of a treatment and an error model in a comprehensive analysis yielding the correct statistical tests according to the experimental design (e.g. Cochran & Cox 1957). This reduces the danger of using error terms that lead to pseudoreplication (Hurlbert 1984).

The treatment model consisted of the overall mean (m), isolation effects (fragmentation yes or no; ii..), fragment size effects (f.j., partitioned into linear contrast and deviation), year effects (y..k, partitioned into linear contrast and deviation), and the corresponding interaction terms, i.e.

yijk = m + ii.. + f.j. + y..k + (i × f).ik + …

The error model consisted of blocking terms, i.e.

yiklm = m + b..l. + (i × b)i.l. + (i × y × b)ikl. + pi.lm + eiklm

where b..l. are the deviations due to random block effects (blocks numbered 1–12, partitioned into site effects and deviations, i.e. block effects within sites), (i × b)i.l. the deviations due to random halves effects within blocks (equivalent to the isolation-by-block interaction), (i × y × b)ikl., the deviations due to random halves effects within blocks within years (equivalent to the isolation-by-year-by-block interaction), pi.lm the deviations due to random plot effects within halves within blocks, and eiklm the residuals.

Fitting the above two models simultaneously yields an anova-table. We used the year as a split-plot factor rather than a repeated-measures factor. This can lead to an overestimation of the significance of the year effect if serial correlations between years occur. However, it does not affect the significance of the linear contrasts of year and of the interactions of this contrast with other terms as they are single degree of freedom tests (e.g. Elashoff 1986).

Results

Effect of fragmentation on the number of diseased plants

Analysis of variance indicated that fragmentation led to a significant increase in the number of diseased plants (Table 1,Fig. 3). The number of diseased plants was 50% higher in fragments than in control plots in 1993 and 1994 and 100% higher in 1995. The largest increases were always in small fragments (significant fragment × size interaction in Table 1).

Table 1.  Summary table compiled from anovas (see text for full model). Shown are the effects of site, year, fragmentation, and the interactions fragmentation × year, fragmentation × fragment size, and fragmentation × year × fragment size on number of diseased plants, stroma-carrying flowering tillers, tillers with healthy inflorescences and the percentage of flowering tillers with disease. All measures were expressed on a per-m2 basis. The main effect of fragment size and its interaction with year are (among others) not shown because ‘fragment’ size in control plots only referred to the area of sampling and should not have affected the per-area measurements. The degree of freedom (d.f.) for the total in the full anova was 208, the d.f. for the lowest-level residual was 59
 d.f.Number of diseased plants FNumber of stroma- carrying flowering tillers FNumber of tillers with healthy inflorescences FPercentage of flowering tillers with disease F
  • 0.1 ≥ P > 0.05,

  • *

    0.05 ≥ P > 0.01,

  • ***

    P ≤ 0.001.

Site26.26*5.38* 4.23*16.54***
Year24.96*4.29*40.71*** 5*
Fragmentation15.47*0.8949.05*** 1.18
Fragmentation × year20.460.73 9.3*** 0.1
Fragmentation × size24.48*1.1120.4*** 2.53
Fragmentation × year × size40.130.71 6.28*** 0.28
Figure 3.

The effect of fragment size and year on the number of Bromus erectus plants showing the disease caused by the fungal endophyte Epichloë bromicola per square metre. Plant refers to a clump of stroma-carrying flowering tillers presumably belonging to a single genet. Lines at the top of bars indicate ± 1 SE.

Effect of fragmentation on the number of diseased and healthy inflorescences

There was no significant effect of fragmentation on the number of stroma-carrying flowering tillers (Table 1) although the number was higher in fragment than in control plots in the second and third year of the study, particularly in the small fragments (Table 2, Fig. 4). The numbers of stroma-carrying flowering tillers were 30% higher in fragment than in control plots in 1994 and 64% higher in 1995. Increases were probably not significant due to large variation in stromata production among plots.

Table 2.  The mean number of diseased plants, stroma-carrying flowering tillers and tillers with healthy inflorescences in 1993, 1994 and 1995 in control and fragment plots (data are expressed as means ± 1 SE m−2). The percentage of flowering tillers with disease is also shown
YearTreatmentNumber of diseased plantsNumber of stroma-carrying flowering tillersNumber of tillers with healthy inflorescencesPercentage of flowering tillers with disease
1993Control0.96 ± 0.4    2.4 ± 0.8 92.0 ± 9.22.6 ± 0.75
 Fragment1.44 ± 0.4    2.5 ± 0.8 99.2 ± 9.22.4 ± 0.75
1994Control1.92 ± 0.4    4.0 ± 0.8104.4 ± 9.23.6 ± 0.75
 Fragment2.84 ± 0.4    5.2 ± 0.8146.9 ± 9.23.4 ± 0.75
1995Control1.16 ± 0.4    2.9 ± 0.8 90.6 ± 9.23.1 ± 0.75
 Fragment2.36 ± 0.4    4.8 ± 0.8187.9 ± 9.22.5 ± 0.75
Figure 4.

The effect of fragment size and year on the number of stroma-carrying flowering tillers of Bromus erectus per square metre. Stroma-carrying flowering tillers occur in plant genets infected by the fungal endophyte Epichloë bromicola. Lines at the top of bars indicate ± 1 SE.

There were highly significant fragmentation effects on the number of tillers with healthy inflorescences (Table 1). Healthy inflorescences occur almost exclusively in healthy plants and no infected plants became healthy during the observation period. Nor did we observe the establishment of new B. erectus plants from seed and the changes in the number of tillers with healthy inflorescences must therefore have reflected increased size or flowering intensity of healthy plants. The numbers of tillers with healthy inflorescences were 41% higher in fragment than in control plots in 1994 and 107% higher in 1995 (Table 2). The magnitude of these increases was dependent on fragment size (significant treatment × size interaction in Table 1) and time following fragmentation (significant treatment × size × year interaction in Table 1,Fig. 5). The number of tillers with healthy inflorescences per unit area was highest in small, followed by medium and large fragments (Fig. 5).

Figure 5.

The effect of fragment size and year on the number of B. erectus tillers carrying a healthy inflorescence per square metre. Lines at the top of bars indicate ± 1 SE.

Despite the increases in the number of diseased plants in fragment vs. control plots the percentage of flowering tillers with disease tended to decrease (Table 1) due to the large increases in the number of tillers with healthy inflorescences. There was a 7.6% decrease in the percentage of flowering tillers with disease in fragment vs. control plots in 1993, a 5.6% decrease in 1994 and a 20% decrease in 1995.

Variation in choke disease in natural field populations

Analysis of variance (Table 1) indicated a significant effect of site and year on the number of diseased plants and the numbers of stroma-carrying flowering tillers in the B. erectus populations included in this field study (Table 3). The number of diseased plants ranged from 0.36 to 4.36 m−2 at the three field sites over 3 years. The number of diseased plants was highest in Vicques (2.96–4.36 m−2), followed by Nenzlingen (1.40–1.96 m−2), and Movelier (0.36–0.48 m−2). The numbers of stroma-carrying flowering tillers were similar in Vicques and Nenzlingen (between 4.0 and 6.4 m−2) but were much lower in Movelier (between 0.64 and 0.80 m−2). The percentage of stroma-carrying flowering tillers ranged from 2.7 to 4.1% in Nenzlingen, 3.8 to 4.9% in Vicques and 0.6 to 0.7% in Movelier. There were also significant site to site and year to year differences in the number of tillers with healthy inflorescences (Tables 1 and 3) with a range of 102.4–182.0 m−2.

Table 3.  The mean number of diseased plants, stroma-carrying flowering tillers and tillers with healthy inflorescences in 1993, 1994 and 1995 at the three field sites (data are expressed as means 1 SE m−2). The percentage of flowering tillers with disease is also shown
SiteYearNumber of diseased plantsNumber of stroma-carrying flowering tillersNumber of tillers with healthy inflorescencesPercentage of flowering tillers with disease
Nenzlingen19931.40 ± 0.32   4.00 ± 1.16102.4 ± 103.7 ± 0.9
 19941.96 ± 0.32   5.32 ± 1.16123.6 ± 104.1 ± 0.9
 19951.60 ± 0.32   5.04 ± 1.16182.4 ± 102.7 ± 0.9
Vicques19932.96 ± 0.6   4.00 ± 1.0 80.8 ± 8.44.9 ± 0.8
 19944.36 ± 0.6   6.40 ± 1.0137.6 ± 8.44.4 ± 0.8
 19952.96 ± 0.6   4.40 ± 1.0112.8 ± 8.43.8 ± 0.8
Movelier19930.44 ± 0.16   0.64 ± 0.24103.2 ± 7.20.6 ± 0.35
 19940.48 ± 0.16   0.64 ± 0.24113.2 ± 7.20.6 ± 0.35
 19950.36 ± 0.16   0.80 ± 0.24102.4 ± 7.20.7 ± 0.35

Asymptomatic infections

Diagnostic PCR indicated that E. bromicola can infect B. erectus plants without causing disease symptoms. Asymptomatic infections were found in 1.7% of all tillers tested in 1995. The rate of asymptomatic infection varied between sites with 0.85% of tillers from Nenzlingen, 1.0% of tillers from Vicques and 4.2% of tillers examined from Movelier. Comparing these rates with the 2000 or 4.3% diseased vs. 45000 or 95.7% healthy inflorescences surveyed over all 3 years in the field, it can be calculated that on average about 7 out of 10 infected plants show the disease symptoms.

Discussion

Habitat fragmentation may occur on different spatial scales (Lord & Norton 1988; Simberloff 1988; Kareiva & Wennergren 1995; Gonzalez et al. 1998), and may range from small breaks in an otherwise homogeneous habitat to widely scattered units of remnant habitat in a surrounding area (e.g. patches of woody vegetation in an agricultural landscape; Wiens 1989). From the point of view of the organisms inhabiting fragments, habitat fragmentation might occur on the scale of individuals or populations (Wiens 1994).

Little, if anything, is known about the effects of habitat fragmentation on plant disease. Conditions suitable for plant and/or pathogen growth may change following fragmentation leading to fluctuations in population sizes (Burdon 1993) or altered plant–pathogen interactions (Schardl 1996). Habitat fragmentation could directly affect survival, growth rate or reproduction of pathogens. Alternatively, pathogens may be affected indirectly via alterations to the physiological and morphological status of host plants.

We present empirical evidence that small-scale habitat fragmentation increases the occurrence of choke disease caused by the fungal pathogen E. bromicula in the grass host B. erectus. The results of our study showed that the number of diseased plants was higher in fragment than in control plots and higher in small than in large fragments, a finding consistent over 3 years and three field sites. There are two potential explanations for this finding. First, habitat fragmentation could have led to an increased expression of the disease by previously infected but asymptomatic plants and second, habitat fragmentation could have increased fungal transmission rates. Increased establishment from infected seeds can be excluded, even if seedlings had been observed, since E. bromicula is not transmitted via B. erectus seeds. The increases in the number of diseased plants seen in the first year cannot be due to new infections because the pathogen cannot develop to the symptomatic stage in only a few months. Increased fungal transmission rates could nevertheless account for a large proportion of the increases in the number of diseased plants found in the second and especially the third year, although effects seen in the first year of the fragmentation treatment must be due to altered expression of the disease.

Habitat fragmentation could alter the growth rates of either the fungus or its host. Since intercellular fungal hyphae must emerge and proliferate to form a stroma before the plant is able to form a functional inflorescence, either an increase in the growth rate of the fungus or a decrease in the growth rate of the host could increase the incidence of choke disease (Kirby 1961; Emecz & Jones 1970; White et al. 1991; White et al. 1992; White et al. 1993; Bucheli & Leuchtmann 1996). Diagnostic PCR indicated that E. bromicola does infect B. erectus plants without causing disease symptoms, supporting the existence of the pool of infected plants needed for such a shift to occur. This is in accordance with the findings of Clay & Brown (1997) and Wennström (1996) who also showed asymptomatic Epichloë associations in natural habitats. The percentage of asymptomatic infections among stroma-free tillers in our study varied between sites. Interestingly, the site with the lowest density of diseased plants had the highest percentage of asymptomatic infection, and vice versa. Furthermore, the site with the highest percentage of asymptomatic infection (Movelier) had the highest percentage increase in the density of diseased plants between fragments and control plots, while the site with the lowest percentage of asymptomatic infection (Nenzlingen) had the lowest percentage increase of diseased plants. All these observations are consistent with habitat fragmentation increasing the probability that previously infected but asymptomatic plants will switch to the symptomatic state.

Despite the increase in the number of diseased plants, the proportion of flowering tillers with diseased inflorescences in the host population overall was not increased by habitat fragmentation. This was presumably due to a beneficial effect of habitat fragmentation on the growth and flowering probability of healthy plants rather than an increase in the number of healthy plants via seedling establishment. Because the number of diseased plants increased in the fragments, we can also exclude the possibility that some previously diseased plants switched from the symptomatic to the asymptomtic state or overcame the infection. Our results demonstrate that it can be dangerous to report disease incidence in a clonal plant species without defining the level of individual organization to which it applies (Schmid 1990). Habitat fragmentation in our experiment apparently increased disease incidence in the host plant at the level of the genet population (counting entire plants) but decreased it at the level of the flowering tiller population (counting inflorescences). Because the fungal endophyte E. bromicula is systemic it affects entire genets, and thus the genet population is the more appropriate level of definition (see Sackville Hamilton et al. 1987). However, it is often difficult to identify genets. In the present study we could only identify genets when they were diseased and had to infer that the entire genet population did not increase because we did not observe any seedlings that grew into established and flowering genets of B. erectus during the 3-year observation period. We therefore did not use the term disease incidence in the results section.

The beneficial effect of habitat fragmentation on the plants that escaped infection by choke disease could have been due to increased light, temperature and soil nutrient availability along the edges of the fragments (Harris 1988; Saunders et al. 1991; Margules et al. 1994; Samuel Zschokke, unpublished data). The stronger increase in the density of flowering tillers of healthy plants in the smaller than in the larger fragments is in agreement with this explanation, because the relative amount of edge increases with decreasing fragment size. The changed abiotic conditions and subsequent changes in plant and fungal growth could have been responsible for the increased probability of infected asymptomatic plants developing the disease, i.e. changing to the symptomatic state. Edge effects could also have been responsible for higher transmission rates because they may have changed patterns of local wind currents and water splash.

We show that small-scale habitat fragmentation affects not only individual species, but also interactions between species. In addition, the large variation among sites, years, blocks and plots points out the importance of proper design and replication in habitat fragmentation experiments. Had we done separate experiments each at only one site or in only one block with replications within it, we would have obtained a series of interesting but individually different results. To see the general trends through the veil of the individual variation it may often be necessary to cover a larger area than has usually been done in ecological experiments (for another example of repeated experiments see Hector et al. 1999). Our study also highlights the complexities faced when interpreting realistic ecological experiments (Robinson et al. 1992; Margules et al. 1994; Diffendorfer et al. 1995). Predicting the effects of habitat fragmentation on populations requires an understanding of the ecology and biology of individual species, as well as interactions between species.

Acknowledgements

We would like to thank the many dedicated field assistants whose efforts contributed to this work. We would especially like to thank Stephanie Kern and René Husi for technical assistance in the laboratory. Bob Holt and an anonymous referee made very helpful comments about an earlier version of this paper. This research was supported by grants from the Priority Programme Environment, IP Biodiversity, of the Swiss National Science Foundation (grants no. 5001–44621 to T.B., 5001–35229 to B.S. and 5001–35241 to B.B.).

Received 3 December 2000 revision accepted 11 October 2000

Ancillary