Evolution of host range in Coleosporium ipomoeae, a plant pathogen with multiple hosts

Significance

Patterns of host breadth evolution in pathogens that attack multiple host species have seldom been quantified. Previous investigations of pathogens attacking a single host indicate that pathogen genotypes evolve a broad host range across host genotypes. By contrast, this investigation demonstrates that, in a pathogen attacking several host species, pathogen genotypes evolve to be highly host-specific. This difference suggests that the evolutionary dynamics of pathogen host breadth may differ depending on whether the pathogen attacks only one or multiple host species.

Abstract

Plants and their pathogens coevolve locally. Previous investigations of one host–one pathogen systems have demonstrated that natural selection favors pathogen genotypes that are virulent on a broad range of host genotypes. In the present study, we examine a system consisting of one pathogen species that infects three host species in the morning glory genus Ipomoea. We show that many pathogen genotypes can infect two or three of the host species when tested on plants from nonlocal communities. By contrast, pathogen genotypes are highly host-specific, infecting only one host species, when tested on host species from the local community. This pattern indicates that within-community evolution narrows the host breadth of pathogen genotypes. Possible evolutionary mechanisms include direct selection for narrow host breadth due to costs of virulence and evolution of ipomoea resistance in the host species.

Much of plant-pathogen coevolution is mediated by “gene-for-gene” (GFG) interactions. These interactions involve R genes in plants and corresponding virulence/avirulence genes in the pathogen (1). At a given pair of corresponding loci, a host may carry either a resistant (Res) or a susceptible (Sus) allele, or both, with Res typically being dominant. The pathogen may carry either a virulent (Vir) allele or an avirulent (Avr) allele. Infection results, unless at one pair of corresponding loci, the plant R locus has a Res allele and the pathogen has the Avr allele. Models of the evolution of GFG systems generally predict that generalist pathogens (those able to infect multiple host-resistance genotypes) will be favored by natural selection over highly specialized genotypes that can infect only one resistance genotype (2–6). Experimental analyses of pathogen host breadth in natural plant–pathogen systems are consistent with these expectations in that pathogen isolates are generally able to infect multiple host-resistance genotypes, especially in host populations with high levels of resistance (7–10).

With very few exceptions (11, 12), the evolution of pathogen host range has been examined, both theoretically and empirically, for a single pathogen species interacting with a single host species. Many pathogens, however, are capable of infecting multiple host species. Predictions of evolutionary models based on a single evolving host species cannot be clearly extrapolated to this situation. Moreover, there are reasons to believe that, with multiple host species, selection for generalism may not be as prevalent. Maintaining infectivity on multiple hosts requires continued success in the coevolutionary arms race with more than one independently evolving host genome. The conditions under which this maintained infectivity can occur are likely more restrictive than with only one host, although this possibility has not been examined theoretically. In addition, selection to maintain infectivity on a particular host is likely weaker when the pathogen population can successfully reproduce on another host (see ref. 13 for an analogous argument with respect to partial resistance). Finally, costs associated with the ability to infect multiple host species (e.g., ref. 14) are likely greater than costs associated with the ability to infect multiple genotypes within the same host. All of these factors would tend to weaken selection for a broad host range and thus promote the evolution of specialist pathogen genotypes within populations.

One approach to determining whether there is an evolutionary tendency for host breadth to be narrowed within populations is to compare pathogen host breadth in its local native community with host breadth on hosts from outside its native community (e.g., refs. 9 and 13). The latter constitutes an estimate of host breadth on host species with which the pathogen has presumably not recently coevolved and is also an estimate of host breadth for a pathogen strain that has recently immigrated into a new community. If evolutionary processes within local communities act to promote specialization, host breadth should be lower on hosts from the native community. In this report, we demonstrate that this pattern is exhibited for a host–pathogen system consisting of one pathogen and three host species.

Methods

Ethics Statement.

No specific permits were required for the collections used in these experiments. Some collections were made in public road rights-of-way, and where collections were made from private land, permission to do so was granted by landowners. No endangered or protected species were affected by this work.

The Ipomoea–Coleosporium Pathosystem.

Throughout the eastern United States, several morning glory (Ipomoea) species are alternate hosts to a red rust, Coleosporium ipomoeae. Among these species are Ipomoea coccinea L., Ipomoea hederacea Jacq., and Ipomoea purpurea (L.) Roth, each of which is an annual herbaceous flowering plant that occurs commonly in agricultural fields, field margins, and other disturbed habitat. Outcrossing rates in these species vary: I. hederacea is highly selfing [93% (15)], I. purpurea’s selfing rate has been reported to be between 20% and 70% (15–17), and, although selfing in I. coccinea has not been explicitly quantified, the species is self-fertile (18). None of the species are capable of hybridizing.

Throughout the eastern United States, the rust C. ipomoeae attacks several species of morning glory hosts, including I. hederacea, I. purpurea, and I. coccinea. The pathogen is a heteroecious rust, with pines (especially Pinus taeda) in the southeastern United States as its primary host. In early summer (late May through early June), dikaryotic spores (aeciospores) are formed on pines and subsequently infect the Ipomoea secondary hosts. Infections of Ipomoea produce asexual spores (urediniospores) that reinfect the secondary hosts, with up to 15 asexual generations occurring during the summer. In the fall, asexual spores (teliospores) colonize the primary hosts and subsequently undergo diploidization, meiosis, and fusion to form the dikaryotic hyphae that give rise to the aeciospores in the spring.

Crossing experiments have shown that resistance to individual pathogen inocula (spores collected from the same host species at a particular site) is often determined by genotype at a single genetic locus (19, 20). In addition, these studies provided evidence indicating that loci conferring resistance to a particular pathogen inoculum sometimes differ between populations of the same host species. These results suggest that GFG interactions may frequently be involved in the coevolution of C. ipomoeae and its hosts. In addition, infection by C. ipomoeae reduces fitness substantially in I. purpurea (21, 22), suggesting the potential for the pathogen to drive the evolution of resistance in its hosts.

Field Censuses.

Sites containing at least one of the three focal morning glory species in North Carolina were censused to determine the natural distribution of infection during the years 2006–2008 (Fig. 1). Each year, we visited these communities during the early summer after plant germination had occurred and recorded which Ipomoea species were present. During repeated visits over the summer, we assessed the proportion of plants of each species harboring infections. By late August, these proportions fell into two discrete categories: >90% of plants infected or <10% of plants infected.

Fig. 1.

Map of census site locations and sites used in the cross-inoculation experiment. The presence of host species and their infection status, if present, is indicated by colored portions of divided circles. Each circle includes three portions, one for each host species, as indicated in Inset. White portions indicate the absence of the given host species, green portions indicate presence without infection, and orange portions indicate presence and infection. Approximate locations are indicated by black dots or stars; stars indicate the five censused locations chosen for cross-inoculation experimentation.

Experimental Assessment of Compatibility.

To investigate the pattern of host specificity for pathogen genotypes, we performed a series of cross-inoculations in the laboratory between pathogens collected from a single host species at a particular location, and host plants of each species at that and other locations. The set of inoculations performed represented a compromise between complete coverage of all hosts and pathogens at a given location and coverage of as many locations as possible. Individual inocula were collected from multiple plants of the same species at a given site because propagation of single-spore isolates was not possible. They thus may represent multiple pathogen genotypes. There are two advantages to this approach: (i) It reduces the possibility of attributing an observed compatibility reaction to a sample not representative of the majority pathogen genotype associated with a given host population; and (ii) it decreases the probability of not including minority pathogen genotypes in cross-inoculations. One potential disadvantage is that it may conflate the host ranges of different pathogen genotypes. However, this possibility is a problem only if genotypes in an inoculum have different host specificities. As described below, this possibility can be detected because genotypes able to infect a given host yield visible pustules, whereas genotypes unable to infect that host yield visible signs of a hypersensitive response. Trials yielding both responses on the same plant were very rare (see below).

Seeds from mature plants were collected in August through September of the years 2006–2007, haphazardly and with 3 m between collections to avoid repeated collection from single plants. Assessment of compatibility for collected seeds was carried out in the year after collection, such that the inoculum encountered by experimental plants was that which these plants would have encountered in the field after germinating 1 y after seed dispersal.

Two field locations (CRG and LF) were chosen for complete reciprocal cross-inoculation, and three locations (CB, CL, and MO) were chosen for additional cross-inoculations. Because not all host–pathogen combinations from these additional locations could be tested due to space limitations, a subset of combinations was chosen at random. Each cross-inoculation trial represents a combination of plants from one host population (one Ipomoea species at one location) with a spore inoculum collected from one host species either at the same location or a different location. An average of 12 plants per host species–site combinations were planted to be used as replicate hosts for each trial. Each plant was the progeny of a different maternal plant. Several trials were conducted once during one year, and again during a later year, to assess repeatability of results. Plants used in the inoculation trials were grown for 14 d in the Duke University Greenhouse in fertilized soil (14–14–14) and were watered every other day. Experimental plants were randomly placed into blocks of 36 and grown in identical 36-pot cell packs, each in one greenhouse tray. At 14 d, plants were moved to a climate-controlled growth room with a 16-h photoperiod and thermal regimen of 16 h at 32 °C and 8 h at 22 °C. At 21 d, each 36-plant group was administered an inoculum consisting of a collection of urediniospores from one host species at one location.

Urediniospores were collected from the field in the early summers of 2007 and 2008, soon after infections became visible in field populations. They thus represent the first generation of spores produced on the secondary hosts. Pustule-infected leaves were removed from plants and placed in airtight bags for transport. In the laboratory, spores were washed from live pustules with distilled water, and the resulting spore suspension was diluted to a standard 2,000 particles per mL. Controlled inoculation was carried out by first saturating soil and plants with water 8 h before the end of the light stage of the photoperiod. A clear plastic dome 8 inches (20 cm) high was placed over plant trays at that time to elevate relative humidity and simulate natural conditions in the field at dusk. A total of 5 mL of a standardized spore suspension was then applied via a spray bottle to the undersides of experimental plant leaves and the plants were left undisturbed for 7 d before domes were removed. Plants were observed daily from age 28 to 35 d for scoring; plants on which orange uredia appeared were scored as infected. Plants showing the hypersensitive response, as indicated by the appearance of black flecks or spots on leaves, were scored as resistant.

Statistical Analyses.

The primary response variable was the proportion of replicate plants infected for a particular inoculum, which was arcsine square-root transformed before analysis. Inocula were classified by (i) whether or not the host species inoculated was the same as the species from which the pathogen inoculum was collected; and (ii) by whether the host inoculated was or was not from the same site as the site from which the inoculum was collected. A two-way analysis of variance with interaction was performed with these two categories by using the GLM procedure of the SAS system (Version 9.2) (23). Mean and SE for each treatment were calculated on untransformed proportions by using the MEANS procedure of the SAS system.

For purposes of distinguishing among pathogen genotypes, we constructed an “infectivity vector.” Each element of the vector corresponded to a particular host species–location combination. Each element was scored as a 0 or a 1 based on whether proportion of replicate plants infected was >0.5 or <0.5 (see below). Because not all inocula were tested on all host–location combinations, some elements were unscored in each vector. Two pathogen inocula were considered to represent different infectivity genotypes if, for at least one vector element, one inoculum had a 1 and the other had a 0. Similarly, within each host species, resistance genotypes were distinguished by constructing an analogous “resistance vector,” each element of which indicated resistance (0) or susceptibility (1) to a different pathogen inoculum. Resistance genotypes for a particular host species from two different locations were considered different if for at least one vector element, one vector had a 0 and the other had a 1.

Results

Field Surveys of Infection.

Surveys of communities in North and South Carolina where morning glories are present indicate that they typically contain two or three of the Ipomoea species (Fig. 1). Among these communities, there is abundant variation in the distribution of rust infection: Both the number of species and the combinations of which species are infected varies between communities. However, within each community, the pattern of infection was constant during successive summers. Moreover, in most communities with multiple host species present, more than one species was infected.

Experimental Assessment of Compatibility.

Inoculations of individual plants generally resulted in one of two outcomes: Either a plant became infected, as indicated by the presence of sporulating uredia, or it resisted infection, as indicated by lack of uredia and the presence of small regions of necrotic tissue resulting from the hypersensitive response. Only 3 of the 1,137 observed experimental plants became infected and also exhibited the hypersensitive response, suggesting that only very rarely were plants inoculated with a spore collection that included genotypes with two different specificities. Although we have not yet characterized the genetics of these interactions, this all-or-nothing response is typical of GFG interactions (1, 24).

The mean absolute difference in proportion infection between inoculations replicated across years was 0.089 (± 0.040 SE). When the one replicate that was 0.00 in 2007 and 0.67 in 2008, which may represent a change in infectivity or resistance, is dropped, this difference is 0.060 (± 0.029). These differences are small compared with differences in treatment means (Table S1), indicating that results are generally replicable across years.

Table S1.

Proportions of infected plants for different categories of inocula

Source	All pathogens	I. coccinea	I. hederacea	I. purpurea
WS	0.945 (0.036)	1.000 (0.000)	0.881 (0.072)	1.000 (0.000)
WD	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
BS	0.833 (0.056)	0.667 (0.167)	0.960 (0.0222)	0.681 (0.159)
BD	0.523 (0.064)	0.641 (0.113)	0.424 (0.0970)	0.533 (0.133)

Mean proportion of plants in each of four host-pathogen categories that are infected by an inoculum (SE) for all pathogens or for those collected only from only I. coccinea, only I. hederacea, or only I. purpurea. Analysis of variance of proportion of plants infected in cross-inoculation experiments. Location is either W (within; host from same site as pathogen) or B (between, host and pathogen from different sites). Species is either S (same; host is same species as that from which pathogen was collected) or D (different; host is a different species from which pathogen was collected). Degrees of freedom = 1 for each source. Error MS values are 0.335, 0.416, 0.237, and 0.403, respectively, for all pathogens and pathogens from I. coccinea, I. hederacea, and I. purpurea.

The proportion of trials in which all replicate plants exhibited the same outcome was 0.81. In the remaining trials, most plants showed one outcome, whereas a few showed the other, resulting in a strongly bimodal distribution of outcomes (Fig. S1). This pattern allowed us to categorize the host population represented in a trial as either preponderantly susceptible or preponderantly resistant to a particular pathogen isolate, with the former having <50% of plants infected and the latter having >50% of plants infected. These results indicate that, although resistance or susceptibility to particular pathogen isolates is nearly fixed in host populations, minority resistance genotypes may exist.

Fig. S1.

Distribution of resistance exhibited by groups of experimental plants after inoculation. A group of experimental plants consists of individual plants of one species collected from one location, receiving the same inoculum treatment. Typically, nearly all or nearly no plants in an experimental group exhibited resistance, indicating homogeneity of the compatibility reaction between that group of plants and the administered inoculum treatment. Rarely, an intermediate frequency of resistance was observed among experimental plants in one group receiving the same inoculum.

Comparison of patterns of infectivity across all host species × population combinations (infectivity vectors and resistance vectors) for different inocula reveals extensive genetic variation for infection outcome, both within the pathogen and within each host species (Fig. 2 and Figs. S2 and S3). Among the 12 inocula tested, 11 had unique patterns of infectivity that differ from all other inocula, except for CB-H, which could not be distinguished from the patterns for 5 other inocula (Fig. 2A). There are thus minimally 11 distinct patterns of infectivity and, hence, minimally 11 distinct pathogen genotypes, among the 12 inocula. Similarly, within I. hederacea, I. purpurea, and I. coccinea, there were distinct patterns of infectivity across pathogen inocula: For I. hederacea, there were minimally 4 patterns among 5 populations tested; for I. purpurea, there were 2 patterns among 4 populations tested; and for I. coccinea there were 2 patterns among 4 populations tested (Fig. 2B). This variation indicates that populations of hosts and pathogen have diverged genetically among locations with respect to pathogen inoculum-specific resistance.

Fig. 2.

Compatibility matrices highlighting differences between rust isolates and between host populations. (A) Oriented to show differences between pathogen isolates. Rows represent infectivity vectors. (B) Oriented to show differences between host populations. Rows represent resistance vectors. Black brackets connect isolates or populations that are, based on the extent of our data, not able to be differentiated by their profiles of compatibility. In A, the isolate CB:H is indistinguishable from five different isolates; brackets are omitted to avoid confusion. The one compatibility change between 2007 and 2008 is indicated by a red cell.

Fig. S2.

Composite matrix of compatibility reactions from experimental inoculation experiments over 2 y. Each host species is abbreviated to the first letter of its species name. Numbers are proportions of replicate plants that were infected. Empty cells are those for which experimental inoculations were not carried out. Cells in which two numbers appear are those 21 cross-inoculations that were carried out in both years, 2007 and 2008. The first number is the 2007 result; the second number is the 2008 result.

Fig. S3.

Compatibility matrices as in Fig. 2, but showing proportion of replicate plants infected. (A) Oriented to show differences between pathogen isolates. (B) Oriented to show differences between host populations. Black brackets connect isolates or populations that are, based on the extent of our data, not able to be differentiated by their profiles of compatibility. In A, the isolate CB:H is indistinguishable from five different isolates; brackets are omitted to avoid confusion. The one compatibility change between 2007 and 2008 is indicated by a red cell.

The set of 100 cross-inoculations can be broken down into four categories (Tables S1–S3): WS (within location, same host), inoculation of the same host at the same location from which the inoculum was collected; WD (within location, different host), inoculation of a different host at the same location (blue squares); BS (between locations, same host), inoculation of the same host from a different location (tan squares); and BD (between locations, different host), inoculation of a different host from a different location (white squares). Category WS represents a control for the inoculation procedure, because it involves combinations of host species and pathogen strain for which infection has previously occurred. Collectively, categories BS and BD represent the average number of host species infected at other locations, which corresponds to the expected host breadth of the pathogen inoculum if it were introduced into a new community. Finally, categories WS and WD collectively represent the average host breadth of an inoculum after the pathogen has interacted with the local host community for a period. Under the hypothesis that evolution resulting from interactions within a community does not alter pathogen host breadth, the average host breadth ascertained from categories WS and WD should be the same as that ascertained from categories BS and BD.

Table S3.

Proportions of infected plants for different categories of inocula, showing pairwise comparison of treatment differences using the Tukey–Kramer adjustment

Source	All pathogens			I. coccinea			I. hederacea			I. purpurea
	WD	BS	BD	WD	BS	BD	WD	BS	BD	WD	BS	BD
WS	<0.001	0.7950	0.0010	0.0018	0.5370	0.3990	<0.0001	0.9020	0.0089	0.0025	0.6240	0.1930
WD		<0.0001	<0.0001		0.0120	0.0042		<0.0001	0.0106		0.0163	0.0351
BS			0.0008			0.9990			<0.0001			0.8240

Significant effects have P value in bold. Location is either W (within; host from same site as pathogen) or B (between, host and pathogen from different sites). Species is either S (same; host is same species as that from which pathogen was collected) or D (different; host is a different species from which pathogen was collected).

Mean infectivity in the control trials (WS) was 0.945 (Table S1 and Fig. 3), indicating that our infection procedure was highly successful. Infection frequencies were higher for inoculation of the same host species than for inoculation of different host species, regardless of whether they were from the same or different locations (WS and BS) (Table S1). In addition, although there was a substantial probability (0.523) of infection for inoculation of different hosts at different locations (BD), no infections occurred for different hosts at the same location from which the inoculum was collected (WD) (Table S1 and Fig. 3).

Fig. 3.

Means and SEs of infectivity (proportion of replicate plants infected) for four different treatments. WS, same host, same location; WD, different host, different location; BS: same host, different location; BD, different host, different location. All pairs of treatments are significantly different (P < 0.001), except WS-BS in post hoc Tukey–Kramer comparisons.

An analysis of variance indicated that the effects of both host species (same or different) and location (same or different) were highly significant, as was the interaction (Table S2). Post hoc comparisons indicated that all pairs of treatments, except for WS-BS, are statistically significant (P < 0.008; Table S3). The lack of significance for WS-BS indicates that there is no evidence that inocula tested on the host from which they were collected differ in activity on the local and nonlocal hosts. By contrast, the high significance of the WD-BD comparison indicates that in infection of hosts that differ from the collection host, infectivity is much higher when the inoculated host is nonlocal than if it local. Thus, although inocula from each host species are able to infect other host species from other locations, they are unable to infect other host species from the native location. On average, inocula tested on hosts from nonnative communities are able to infect an average of 1.83 host species per community (including the species from which the inoculum was collected); by contrast, when tested on hosts from their native community, inocula are able to infect only 1 host species.

Table S2.

Proportions of infected plants for different categories of inocula, showing analyses of variance of proportions

	All pathogens			I. coccinea			I. hederacea			I. purpurea
Source	MS	F	P	MS	F	P	MS	F	P	MS	F	P
Location	2.72	8.12	0.0051	0.429	1.03	0.3160	1.749	7.38	0.0089	0.210	0.52	0.4770
Species	24.45	73.08	<0.0001	4.799	11.54	0.0017	13.205	55.71	<0.0001	5.507	13.65	0.0009
L x S	6.13	18.33	<0.0001	4.318	10.39	0.0027	0.695	2.93	0.0925	2.893	7.17	0.0121

Significant effects have P value in bold. MS, mean square.

This pattern also appears to hold for when inocula collected from different host species are analyzed separately (Tables S1–S3). In each case, WS and BS exhibited the highest mean infectivities and do not differ from each other. In addition, infectivity was significantly lower for WD compared with BD, with infectivity on a different, nonlocal host ranging from 0.424 to 0.641, whereas infectivity on a different local host was uniformly 0.

Discussion

The primary result of our cross-inoculation experiment is that pathogen host breadth is narrower when tested on hosts from the pathogen’s native community than when tested on hosts from nonnative communities. In particular, although a pathogen inoculum can infect only the species it was collected from in its native community, it can typically infect both that species and others from nonnative communities. This pattern implies that evolution within local communities narrows pathogen host breadth. It also conflicts with both theoretical expectations for, and empirical observations on, plant pathosystems consisting of one host and one pathogen species: Models predict the evolution of broad host range across host genotypes (2–6), and empirical studies are generally consistent with this prediction (7–10).

Several possible mechanisms could explain the narrowing of host breadth in local populations. Here we outline four and argue that two constitute unlikely explanations. One evolutionary process that could reduce host breadth is selection on the pathogen strain for avirulence on host species it initially infected. Such selection could be driven by costs of virulence. In particular, consider a new pathogen strain that migrates into a community. Our results suggest that, initially, that strain would be virulent on two or more local hosts. A mutation that caused the pathogen to become avirulent on one of the hosts would presumably incur a strong fitness penalty because mutant spores landing on that host would be essentially dead. As described above, colonization of morning glories in the spring and early summer occurs by a “rain” of aeciospores from pine trees, which means that the probability of any particular mutant spore landing on a particular host species is approximately proportional to the abundance of that host species in the community. If the three host species are approximately equally abundant, then the mutant would incur a fitness decrement of 1/3–1/2, depending on whether the pathogen initially infected three or two hosts, respectively. It is presumably this type of selection that has led to the persistence of pathogen genotypes that attack multiple resistance genotypes in systems with only one host species (7–10).

In order for the mutant to increase in frequency, costs of virulence would have to offset this increased mortality. Such costs would have to be very large. Although costs of virulence appear to be frequent (25), the few studies that have quantified costs in fitness units have found costs ranging up to ∼0.25 (26–28), which would not be sufficient to offset the mortality costs described above. Moreover, our observation that most inocula exhibit virulence to nonnative hosts suggests that costs of virulence are typically small in this system. Nevertheless, without actually measuring these costs, we cannot rule out the possibility that some of them are substantial enough to compensate for the increased mortality of a mutant that reduces host breadth.

Under this scenario, selection would favor the mutant because of higher reproductive success on the host(s) it can infect. Passage through these hosts would increase the frequency of the mutant, which more than compensates for the reproductive success of the nonmutant on the host species the mutant does not infect. It is also possible that another mutation could arise that caused avirulence on a different host species. If the cost of virulence on that host were sufficiently high, this mutation would also increase in frequency. To the extent that pathogen populations are regulated independently on different hosts, two specialist mutant strains could persist, each with a narrower host range than the original strain, which would be eliminated (29). An alternative possibility is that selection against unnecessary virulence occurs on the primary host during the winter and spring.

A second, related explanation for our results is that they are an artifact of the time during the season when pathogens were sampled: Mating and recombination on the primary host generates genotypes that can infect several host species, but asexual passage through individual secondary hosts selects for specialists by the end of the summer. In other systems, passage experiments of pathogens on particular hosts frequently results in loss of virulence on other hosts, presumably because of costs of virulence (30). There is certainly the potential for this type of selection for specialization because C. ipomoeae can cycle through as many as 10–15 asexual generations on secondary hosts. This explanation seems unlikely, however, because we collected inocula at the very beginning of the season before asexual passage. Host specificity is thus present immediately after sexual reproduction and is not an artifact of measuring host breadth at the end of the summer season.

A third explanation for the existence of host–specialist pathogen genotypes within local populations is that resistance and virulence are determined not by a GFG system but by some sort of matching allele, or matching genotype, system. There are two types of such systems (31). In type matching-allele 1 (MA1), infection would be successful unless there is an exact genotypic match between pathogen and host genotypes at a set of specificity loci (32). This type of system produces generalist pathogens. In type MA2, infection does not occur unless there is an exact match (33). MA2 systems produce highly host–genotype-specific pathogen strains, and evolutionary models of this type of system predict that multiple genotypes can coexist stably within both the pathogen and the host within a population (34–37), yielding different pathogen genotypes infecting different host genotypes—a clear parallel to our results.

We believe that this explanation is unlikely for two reasons. First, although GFG systems are common in plants, demonstrable MA2 systems are rare (31). Second, it is difficult to reconcile this mechanism with the observation that most pathogen inocula tested can infect multiple host species from populations outside their local community. If an MA2 system alone causes high local host specificity, we would expect “matching” genotypes to occur as frequently locally as nonlocally, and thus the ability to infect two or more hosts locally to be common.

A final possible evolutionary mechanism that could lead to host-specific pathogen genotypes is evolution of resistance by the host species. A possible scenario is depicted in Fig. 4, which portrays a system in which the pathogen produces two potential elicitors (α and β), and each of three host species, each of which has two R genes (A and B; these are not necessarily homologous genes in the different host species) that potentially “recognizes” one of the elicitors. Subscripts designate alleles, and virulence occurs unless one of the R genes matches (has the same subscript as) the corresponding elicitor. Initially (Fig. 4A), a pathogen genotype α₁ β₁ immigrates into the community. Because there are no matching specificities, it is virulent on all three hosts. Subsequently (Fig. 4B), in hosts 1 and 3, one receptor evolves to match one of the elicitors, which results in the pathogen being restricted to one host species. A new pathogen strain, α₂ β₁, then immigrates or arises by mutation (Fig. 4C). Initially this strain is virulent on hosts 1 and 2, but receptor A in host 2 evolves to recognize α₂, yielding two host-specific pathogen strains (Fig. 4D). Although recombination on the primary host may produce a diploid that is α₁ α₂ β₁ β₁, this genotype will not infect any of the host species. Finally, a third pathogen strain α₃ β₃ immigrates into the community (Fig. 4E). Initially, this strain can infect all three host species. Subsequently, hosts 1 and 2 evolve B₃ alleles, which results in three host-specific pathogen strains (Fig. 4F). Moreover, any recombinant genotypes can infect at most one host species (Table S4).

Fig. 4.

Model of evolutionary changes in pathogen and host specificities leading to high host specificity in the pathogen. α_i and β_j represent genotypes of two pathogen elicitors. A_i and B_j represent genotypes of R genes in a given host. The genes in the three hosts are not necessarily orthologous. Subscripts represent specificities. Pathogen genotypes are diploid, but are shown as haploid for convenience and should be interpreted as homozygous at each locus. If the subscript of α matches the subscript of A, or if the subscript of β matches the subscript of B, then the pathogen is virulent (plant is susceptible). Otherwise the pathogen is avirulent (plant is resistant). A–F represent successive evolutionary changes. The specific changes are indicated by red subscripts. Pathogen genotypes in red indicate genotypes introduced to the community either by immigration or mutation. Pathogen genotypes directly above host genotype are virulent on that host. (A) No pathogens are originally present, and a new pathogen genotype is introduced by immigration. This genotype is virulent on all host species. (B) Host species 1 and 3 evolve resistant genotypes, leading to avirulence of the pathogen on these hosts. (C) A new pathogen genotype immigrates or is produced by mutation. (D) Evolution of a novel resistance allele in host 2 makes host 2 resistant to the new pathogen genotype. (E) A new pathogen genotype immigrates, and is initially virulent on all three host species. (F) Host species 1 and 2 evolve new resistance genotypes, making the immigrant pathogen genotype avirulent. At this stage, the three pathogen genotypes are each virulent on only one host species. Table S2 shows that pathogen genotypes derived from these by recombination are also all virulent on at most one host species.

Table S4.

Virulence of pathogen genotypes on the three host species in Fig. 4

Pathogen genotype	Host species
	1	2	3
α1 α1 β1 β1	0	1	0
α1 α1 β1 β3	0	0	0
α1 α1 β3 β3	0	0	1
α1 α2 β1 β1	0	1	0
α1 α2 β1 β3	0	0	0
α1 α2 β3 β3	0	0	1
α1 α3 β1 β1	0	1	0
α1 α3β1 β3	0	0	0
α1 α3 β3 β3	0	0	1
α2 α2 β1 β1	1	0	0
α2 α2 β1 β3	0	0	0
α2 α2 β3 β3	0	0	1
α2 α3 β1 β1	1	0	0
α2 α3 β1 β3	0	0	0
α2 α3 β3 β3	0	0	1
α3 α3 β1 β1	1	0	0
α3 α3 β1 β3	0	0	0
α3 α3 β3 β3	0	0	1

Genotypes of host species are: host 1 A₁B₃, host 2 A₂B₃, and host 3 A₀B₁ (as depicted in Fig. 4F). 0, avirulent; 1, virulent.

Although this model indicates how the evolution of host-specific pathogen strains could occur via the evolution of host resistances, there are potential difficulties with this mechanism as well. For example, for this type of scenario to consistently produce pathogen genotypes that are virulent on just one host, the rate of host resistance likely must be substantially faster than the immigration rate, and yet not too fast so as to cause local extinction of the pathogen. It is unclear whether this is true in this system. It is also not clear how restrictive conditions are for the operation of this type of process, because explicit models of it are lacking. Moreover, because genes conferring virulence and resistance have not been identified in this system, it is unclear whether they conform to the assumptions of this model. For example, this model assumes that a given elicitor can exist in any of several states, and that host R genes can evolve to recognize (match) a particular state. This variation in elicitor state might occur, for example, if the different elicitor states represent different protein sequences (e.g., refs. 1 and 38). However, in some plant–pathogen systems, virulence is achieved by eliminating the expression of an elicitor (39, 40). In this case, there would likely be no way for an R gene to recognize a down-regulated elicitor, violating the model’s assumptions.

Our field censuses demonstrated that, in local communities, one or more Ipomoea host species are not infected by C. ipomoeae. Both evolution of avirulence by the pathogen and evolution of resistance by the hosts could account for this pattern. For example, once local evolution has produced a set of host-specific genotypes such as those shown in Fig. 4F, evolutionary recruitment of a third R gene with a specificity matching β₁ would eliminate the pathogen from host 1. This host would remain pathogen-free until either a mutation or an immigrant pathogen strain arose that conferred virulence on host 1. By contrast, it seems less likely that selection for reduced virulence could cause elimination of highly host-specific pathogen genotypes from its host species. For example, a mutation during the asexual phase that prevented a host-specific genotype from infecting its current host would leave no offspring, and hence would be selected against. Conversely, if only one pathogen genotype is present, and that genotype is virulent on three host species, net selection favoring avirulence on one of the hosts would eventually result in that host being free of infection. As argued above, however, it seems unlikely that costs of virulence would be sufficiently high to provide a net benefit to avirulence.

These considerations suggest that the absence of infection on one or more host species locally points to the evolution of resistance in host species as the primary cause of the evolution of host-specific genotypes. However, it is also possible that absence of infection on some hosts may simply reflect local host species extinction, selection because of costs of virulence to eliminate genotypes virulent on the extinct host, and recolonization of the community by that host species. Although in some cases the remaining pathogen strains using other hosts in the community may be able to infect the immigrant host species (this case would be a category BD interaction), there is also a substantial probability that they would not be able to do so, leaving the host free of infection.

One potential limitation of our study is that we have inferred that if inocula consist of multiple genotypes, those genotypes have the same specificities. This inference is based on our failure, except in three cases, to find both successful infection and evidence of a hypersensitive response on the same plant. This inference may be suspect if successful infection by one genotype facilitates infection by a normally avirulent genotype or if a hypersensitive response results in a normally virulent genotype not infecting a plant. Studies on other pathogens have documented priority effects, in which, for example, inoculation by a virulent pathogen facilitates infection in a subsequent inoculation by a normally avirulent pathogen (41–43). However, when a virulent and an avirulent pathogen are simultaneously inoculated, such priority effects are generally absent (44–46). Because in our study different genotypes in an inoculum were introduced to a plant simultaneously, it seems unlikely that there would have been sufficient time for defenses to be activated to prevent at least some infection by the virulent genotypes. Similarly, it seems unlikely there would have been time for infections to produce biochemical changes that would completely prevent a hypersensitive response to the avirulent genotypes.

More important, however, is that even if inocula normally contained multiple genotypes with different specificities, our conclusions would not be substantially altered. Consider an inoculum collected from host A at site 1 that successfully infects hosts A and B at site 2, and suppose this pattern is because the inoculum contains two genotypes: one that infects host A at site 2 and one that infects host B at site 2 (they both infect host A at site 1). However, because the second genotype infects host B at site 2, the expected probability that it also affected host B at site 1 when it first arrived at site 1 was about 0.83 because it represents a category BS interaction. However, this pattern is never the case (i.e., a genotype from host A at site 1 never infects host B at site 1), implying that local evolutionary change in this genotype causes it not to infect host B; its host breadth is narrowed. The processes that could produce this narrowing are the same as those described above.

Theoretical investigations have suggested repeatedly that coevolutionary interactions between plants and pathogens will tend to produce pathogen genotypes that infect multiple host genotypes. In general, empirical investigations have supported these predictions. These results might be extrapolated to suggest that, in communities with multiple host species, selection should similarly lead to the persistence of pathogen genotypes with a broad host range. However, we have found just the opposite. Determining whether selection directly favors the evolution of host-specific genotypes in C. ipomoeae or, rather, favors genotypes with a broad host range but is counteracted by the evolution of resistance in the hosts will require both determining the magnitudes of costs associated with virulence as well as whether increased resistance to specific pathogen genotypes evolves in local communities.