Host genetic architecture in single and multiple infections

Authors


K. Mathias Wegner, Experimental Ecology, Institute of Integrative Biology, ETH Zürich Universitätstrasse 16, 8092 Zürich, Switzerland.
Tel.: +41 44 633 6036; fax: +41 44 632 1271; e-mail: mathias.wegner@env.ethz.ch

Abstract

Hosts are often target to multiple simultaneous infections by genetically diverse parasite strains. The interaction among these strains and the interaction of each strain with the host was shown to have profound effects on the evolution of parasite traits. Host factors like genetic architecture of resistance have so far been largely neglected. To see whether genetic architecture differs between different kinds of infections we used joint scaling analysis to compare the genetic components of resistance in the red flour beetle Tribolium castaneum exposed to single and multiple strains of the microsporidian Nosema whitei. Our results indicate that additive, dominance and epistatic components were more important in single infections whereas maternal components play a decisive role in multiple infections. In detail, parameter estimates of additive, dominance and epistatic components correlated positively between single and multiple infections, whereas maternal components correlated negatively. These findings may suggest that specificity of host–parasite interactions are mediated by genetic and especially epistatic components whereas maternal effects constitute a more general form of resistance.

Introduction

In nature, hosts are rarely infected by a single parasite alone and multiple infections seem to be the rule rather than the exception (Read & Taylor, 2001). Diverse infections can be caused from infections by different pathogen species or from infections by genetically different strains of the same pathogen species (Frank, 1996; Graham et al., 2005; de Roode et al., 2005). If common, genetically diverse infections of the same parasite species can have a profound effect on the evolution of parasite traits like virulence and infectivity (van Baalen & Sabelis, 1995; Frank, 1996), because such infections usually affect the same tissues, leading to conflict and competition between co-infecting strains. The outcome of such conflicts depends on the relatedness among parasite strains (Frank, 1992; Chao et al., 2000) or the magnitude of host inflicted damage (Schjorring & Koella, 2003). Commonly, virulence is predicted to increase with frequency of multiple infections (May & Nowak, 1995; Frank, 1996) but this might depend on how resource extraction from the host, i.e. damage caused to the host, is determined (Brown et al., 2002). Empirical research indeed reached conclusions about parasite virulence in multiple infections that are not always in agreement with earlier standard theory. Some studies demonstrated an increased virulence (Davies et al., 2002), whereas others found no significant difference (Imhoof & Schmid-Hempel, 1998; Wedekind & Ruetschi, 2000) or even reduced virulence (Ben-Ami et al., 2008; Rauch et al., 2008).

Here we take another route by taking into account the obvious fact that the outcome of an infection also depends on host resistance (Gandon et al., 2002). In particular, host resistance can be based on specific interactions between host and parasite genotypes (G × G interactions). Such G × G interactions are prevalent in most cases studied to date ranging from invertebrate (Carius et al., 2001; Schmid-Hempel, 2005) to vertebrate hosts (Grech et al., 2006; Rauch et al., 2006; for review see Lambrechts et al., 2006) and generally involve traits with complex underlying genetic architecture on the host side (Kover & Caicedo, 2001; Wilfert & Schmid-Hempel, 2008). Importantly, genotypic interactions can also alter the type of competition between co-infecting parasite genotypes. In rodent malaria for example, competitive exclusion between parasite strains when infecting resistant host genotypes changed to strain co-existence in infections of susceptible host genotypes (de Roode et al., 2004). Despite the importance of a trait’s genetic architecture for the response to selection in general (Fisher, 1930; Wright, 1931; Hansen, 2006), the genetic architecture of host resistance has mainly been characterized for single infections representing a simple case of a single G × G interaction. Potential differences in what kind of genetic architecture underlies multiple infections with multiple G × G interactions are so far largely unknown.

Here, we wanted to quantify potential differences in the genetic architecture underlying resistance either in single or multiple infections. In particular, we studied the genetic architecture of resistance of the red flour beetle Tribolium castaneum against multiple isolates of its specific, natural microsporidian parasite Nosema whitei by means of a ‘Joint scaling’ analysis (Hayman, 1958; Mather & Jinks, 1982; Demuth & Wade, 2006). Joint scaling has recently been proven successful to demonstrate nonadditive genetic interactions in population crosses (Lair et al., 1997; Kelly, 2005; Demuth & Wade, 2007a) and for the Tribolium–Nosema system and single infections we have already shown that a major proportion of the genetic architecture of two correlated resistance traits, i.e. host survival and parasite infection intensity, is due to epistatic and maternal effects (Wegner et al., 2008). Here we present the results from the same six line crosses used in this earlier study but with multiple infections to describe the respective fitness landscape relative to that of single infections. Based on the large contribution of epistatic effects in single infections (Wegner et al., 2008) we hypothesized that the extension of the single G × G interaction to more complex, multiple G × G interactions will result in a quantitative reduction of additive and especially epistatic effects. This is to be expected as each specific G × G interaction will rest on its own genetic architecture and, thus, multiple infection should reveal the net overlay of all different genetic effects amplifying the effect of shared genetic components (to which epistasis is expected to add little owing to the very specific effects associated with each single infection). These genetic components are likely to show the strongest response to selection making them the decisive factors determining resistance in multiple infections of a genotypically diverse host–parasite system.

Materials and methods

Beetle crosses

We used the same beetle populations as described in (Wegner et al., 2008) as the basic experimental material. The four basic beetle populations were derived from four natural populations of a worldwide collection (line nr. 32: San Bernadino, USA., nr. 40: Kampala, Uganda, nr. 43: Kyushu Island, Japan, nr. 44: Chiang Mei, Thailand). Prior to our experiments, these had been kept at large population sizes (> 200 adult unsexed beetles) on standard medium (type 550 ‘Knospe’ organic flour containing 5% dried yeast) under standard environmental conditions (24 h dark, 32 °C, 70% humidity) for more than 50 generations.

We set up joint scaling crosses for all possible combinations between these four lines (i.e. all possible pairings of four populations yielding a total of six line crosses, Fig. 1a). Within each of the six so defined line crosses we produced nine different types of joint scaling crosses (Fig. 1b). Crosses were set up as mass matings between six virgin males and six virgin females, and each cross was replicated three times, except for crosses including the reciprocals (F1, BC1, BC2) where only two replicate crosses were set up. Outbreeding was maintained throughout all crosses by choosing the appropriate parents from different crosses or unrelated individuals from the stock lines. In total, we generated (six line combinations) × (nine JS-crosses) × (two or three replicates) = 126 groups of beetles.

Figure 1.

 Crossing scheme used for joint scaling crosses. (a) Crosses between the four populations used and all six possible combinations. (b) Principle of a joint scaling cross using a cross between lines nrs. 32 and 40 as an example. Each joint scaling cross consisted of nine cross types: crosses between the pure parental lines P1 and P2 – (i) paternal P1 × maternal P1 = P1; (ii) maternal P2 × paternal P2 = P2; two F1 hybrids – (iii) paternal P1 × maternal P2 = F1, and its (iv) reciprocal: paternal P2 × maternal P1 = rF1; (v) a hybrid intercross F1 × F1 = F2; and four types of F1 backcrosses – (vi) paternal P1 × maternal F1 = BC1; (vii) paternal F1 × maternal P1 = rBC1; (viii) paternal P2 × maternal F1 = BC2; (ix) paternal F1 × maternal P2 = rBC2.

Infection and mortality measurements

Nosema whitei is a highly virulent parasite infecting early larval stages of the beetle (Blaser & Schmid-Hempel, 2005), where it induces gigantism and usually kills the host before pupation (Bass & Armstrong, 1992; Blaser & Schmid-Hempel, 2005). Infection in adults has severe fitness effects on longevity, body condition and fecundity (Armstrong & Newton, 1985; Armstrong & Bass, 1986). Parasite inoculated medium was generated following standard procedures (Blaser & Schmid-Hempel, 2005; Fischer & Schmid-Hempel, 2005). As a source of N. whitei spores we used isolates originating from another experiment where a mix of different parasite isolates were co-evolved for two generations with a variety of beetle lines (Greef, 2007). The results of that and previous experimental evolution studies (Fischer & Schmid-Hempel, 2005) suggest that in these experiments substantial genetic variation in the parasite population exists and that parasite populations diverge with time as they adapt to their own host lines. To carry out the single infections, out of these diverged parasite isolates we chose one (nr. NL5) that showed strong differences in virulence (i.e. infection-induced host mortality) when infected in several beetle populations in a fully crossed infection experiment (Otti, 2007). We mixed this isolate with two others from the same experiment (nrs. NL7 and NS3) in equal proportions for the mixed infections (i.e. one third per isolate). NL7 and NS3 showed similarly variable infection profiles in single infections (Otti, 2007). The virulence of these isolates (i.e. parasite- induced mortality rates) did, on average, not differ between the original beetle isolates used here to found the joint scaling crosses (average difference between survival infected with NL5 and expected survival in co-infections: 4%, paired t-test, t = 1.179, d.f. = 3, P = 0.323); thus, average virulence in the genetic backgrounds used here was comparable. As mentioned above, the chosen isolates did however show significant G × G interactions when used to infect different beetle populations (Otti, 2007) indicating that previous co-evolution selected for different parasite lineages specific to the coevolving host line. Nosema spores can be stored at 4 °C for extended periods of time without losing infectivity (Milner, 1972), which enabled us to reduce environmental noise by infecting all crosses with the same sources of spores. As infective dose we used a final concentration of 5 × 103 spores g−1 medium (Greef, 2007; Wegner et al., 2008) for both single and multiple infections. Approximately 0.125 g of inoculated medium was then filled into each well of a 96-well cell culture plate (Sarstedt, Sevelen, Switzerland) resulting in a final average infective dose of ∼625 spores per individual host. We used the same infection dose for single and multiple infections because previous experiments did not find significant differences in hatching rate or spore load between multiple infections with single or triple infection doses (Otti, 2007). For the infections, we used 24 freshly hatched larvae (sampled after 1 week of egg laying by the parent beetles), from each replicate of the nine joint scaling crosses per line combination and split them equally but randomly between single and multiple infection treatment. The resulting 12 larvae (per treatment) were further split into two groups of six individuals. These two groups were then randomly distributed over single and multiple infection plates taking care that replicates of one cross ended up in two different plates per treatment to control for plate effects. After 7 weeks, host survival was assayed in each group; a total of 2440 beetles could be assayed in this way.

Analyses

With the crossing scheme used here (consisting of nine types of crosses nested within each of the six line combinations) we can maximally fit an eight-parameter model using a weighted least square regression (Mather & Jinks, 1982). As we were mainly interested in comparing the results of multiple infections with those of single infections we applied the same strategy used before (Wegner et al., 2008). In short, we first fitted a model for additive and dominance components, including the population mean (m) plus additive (d) and dominance (h) coefficients (‘Model AD’).

The value of the simple AD model was calculated with the goodness of fit against a chi-squared distribution with six degrees of freedom (d.f. = 9 cross means – three estimated parameters; distributed as chi-square). If a test for goodness of fit rejected this simple model after sequential Bonferroni correction for multiple testing within each cross, the full eight-parameter model (‘Model ADME’) was fitted that included two additional maternal effect coefficients, dm and hm, plus three epistatic components labelled by their coefficients, i, j, and l. Expected contributions of all parameters in this model can be found in Table 1. All genetic components detected this way have to be regarded as conservative estimates because multiple differences may cancel each other out, and any significant genetic component out of such cross will represent a net effect summarizing all genetic components simultaneously. Size and direction of significant parameter estimates were analysed with standard least square regressions and least square means from whole models are given where not indicated differently. Absolute contribution of significant parameters was assessed as the absolute value, inline image, of each parameter irrespective of its sign. All tests and analysis scripts of the joint scaling analysis were implemented in r statistical software package (R Development Core Team, 2007).

Table 1.   Expected generation means of the full additive-dominance, digenic epistasis model with maternal effects (ADME).
GenerationContribution to generation mean*
  1. m, cross mean (intercept); d, additive components; h, dominance; dm, dh, maternal additive and dominance components; i, additive × additive epistasis; j, additive × dominance epistasis; l, dominance × dominance epistasis.

  2. *Adapted from Mather & Jinks (1982), Lair et al. (1997).

P1m + d + dm + i
P2m − d − dm + i
F1m + h + dm + l
rF1m + h − dm + l
F2m + ½ h + hm + ¼ l
B1m + ½ d + ½ h + dm + ¼ i + ¼ j + ¼ l
rB1m + ½ d + ½ h + hm + ¼ i + ¼ j + ¼ l
B2m − ½ d + ½ h − dm + ¼ i − ¼ j + ¼ l
rB2m − ½ d + ½ h + hm + ¼ i − ¼ j + ¼ l

Results

Phenotypic variation in population crosses

By using offspring originating from the same parents our tests were essentially paired between treatments. We therefore analysed our data as repeated measure anova with infection type (single/multiple) as repeated measure and fitted the model shown in Table 2. Figure 2 shows the mean survival rates observed in multiple and single infections. On average, mean survival per cross was significantly higher in the multiple infection treatment than in the single infection treatment (Fig. 2, also see ‘Infection type (I)’ term in Table 2), indicating that virulence was lower in multiple infections. Despite lower survival in the single infection, survival rates in multiple infections also correlated strongly with survival in single infections on the level of the within line joint scaling crosses (R2 = 0.55, F1,44 = 54.91, P < 0.001). This might signify that host genetic factors did not vary systematically between infection treatments and that the observed reduction in virulence was rather caused by parasite effects than by host effects. We found no significant differences between different cross types in general (that is, P, F1, F2 and backcrosses), nor a significant interaction between cross type and type of infection (single vs. multiple) (Table 2). We did however find a significant difference between single and multiple infections when considering the reciprocal crosses set up for backcrosses and F1 crosses, which only differed by contributions from mothers of the original parental lines. This indicates that maternal effects might play a different role in single as compared to multiple infections (see ‘× R[CT]’ interaction in Table 2).

Table 2.   Repeated measures anova table using infection type (I, i.e. single and multiple infection) as repeated measure demonstrating the effect of infection type and reciprocal pairings on phenotypic differences between single and multiple infections.
SurvivalFactord.f.FP
  1. Statistically significant effects are printed in bold.

  2. *Cross types include parental (P), F1, F2 and backcrosses (BC).

  3. †Survival in single infections was 0.365 on average whereas it was 0.455 in multiple infections.

Between subjectsAll between5, 380.3720.865
Cross type (CT)*3, 380.4260.735
Reciprocal [CT] (R[CT])2, 380.3010.742
Within subjectsAll within5, 381.6180.179
Infection type (I)†1, 389.9150.003
I × CT3, 380.3450.793
I × R[CT]2, 383.6720.035
Figure 2.

 Joint scaling plots for survival in all experimental crosses (mean ± SE). The left column shows survival in the multiple infection treatment and the right column survival in the single infection treatment [data correspond to (Wegner et al., 2008)]. Parental lines are shown by circles, pooled back-crosses by diamonds, pooled F1 crosses by upward triangles and F2 crosses by downward triangles. Labels indicate line combinations. Above each treatment-cross panel the goodness of fit for the additive-dominance (AD) model, and where applicable, also the full model including additive, dominance, maternal and epistatic components (ADME) is indicated. Statistical inferences is indicated as follows: model rejected with: ***: P < 0.001, **: 0.001 < P < 0.01, *: 0.01 < P < 0.05, ns: P > 0.05. Regression lines show the expected values from the AD model. Dotted lines show the overall average survival for each line combination in both treatments indicating higher survival in all multiple infection treatments.

Genetic architecture in single and multiple infections

Despite the good phenotypic correlation between survival rate in single and multiple infections, we found differences in the underlying genetic architecture in these crosses. For example, although we had to reject the ‘AD’ model (additive and dominance effects) in single infections of cross no. 40 × 43, the AD model was sufficient to explain variation in multiple infections for this same cross (Fig. 2). The opposite pattern was found for cross no. 32 × 43 where the ‘AD’ model sufficiently explained genetic architecture in single infections but failed to do so in multiple infections (Fig. 2).

Table 3 gives an overview of all parameter estimates for the respective best-fitting models (AD or ADME). The magnitude of the absolute parameter estimates was the highest for epistatic components in both infection treatments (Fig. 3). There was however a significant interaction term between genetic component and infection type. The interaction was mainly caused by a stronger contribution of maternal effects along with a lowered contribution of epistatic components in multiple infections (Fig. 3) supporting the phenotypic differences in survival observed in reciprocal crosses (Table 2).

Table 3.   Parameter estimates for survival rate and significance levels from joint scaling models.
CrossTreatmentmComponent†
Additive, dominance (AD)Maternal (M)Epistatic (E)
dhdmhmijl
  1. S, single infection; M, multiple infection.

  2. †For explanation of symbols for genetic contributions, see Table 1. Where no entry is shown for M and E components, the AD-model was sufficient to explain the observed variation.

  3. Numbers printed in bold indicate significant parameter estimates with ***P < 0.001, **0.001 < P < 0.01, *0.01 < P < 0.05, ns P > 0.05.

32 × 40S0.757***0.367***−0.613 ns−0.096***−0.028***−0.444***0.070*0.087 ns
M0.684**0.230***−0.522 ns0.125***0.125***−0.267**−0.126**0.074 ns
32 × 43S0.297***0.268***0.091***
M−0.114 ns0.283***−0.493 ns0.047***0.594***0.556 ns−0.886***1.176 ns
32 × 44S−0.267 ns−0.097***1.201 ns0.013***0.030**0.933**−0.107 ns−0.386 ns
M0.457***0.103***0.126***−0.134***−0.268***0.346***0.496*0.075 ns
40 × 43S0.332***0.016***−0.071 ns−0.009***−0.074***−0.298***−0.044 ns−0.222 ns
M0.109***−0.051***−0.009*
40 × 44S−0.660 ns−0.354***2.368 ns0.000 ns−0.083 ns1.056**−0.125 ns−1.319**
M−0.354 ns−0.049***0.868 ns0.000 ns0.25***0.444 ns0.097 ns−0.431 ns
43 × 44S−0.056 ns−0.418***0.685 ns0.057***0.083***0.444 ns−0.444***0.075 ns
M0.433***−0.355***0.319***
Figure 3.

 Absolute values of parameter estimates for significant genetic components (additive, dominance, epistatic, maternal) derived from joint scaling analysis after single and multiple infections. Open bars represent the single infection whereas hashed bars show the parameter estimates from multiple infections. Absolute parameter estimates for epistatic components were significantly higher than for other components in both infection types (component: F2,36 = 15.504, P < 0.001) but differed relative to maternal components between infection types (component × infection type: F2,36 = 4.189, P = 0.023).

When all parameter estimates were considered additive, dominance as well as epistatic components correlated positively between single and multiple infections (Fig. 4a,b). The slopes of these regression lines were virtually identical [AD, Fig. 4a: parameter(single) = 0.22 + 1.76 × parameter(multiple) compared to E, Fig. 4b: parameter(single) = −0.16 + 1.75 × parameter(multiple)] indicating that the contribution of these genetic components (additive, dominance, epistatic) was higher in single infections (Figs 3 and 4). Maternal components, on the other hand, correlated negatively indicating that the factors contributed by the mother were more important and of opposite effect in multiple infections than in single infections.

Figure 4.

 Correlation of parameter estimates between single and multiple infections. (a) additive and dominance effects, (b) epistatic effects and (c) maternal effects. Plots include all parameter estimates and regression lines show significant correlations between single and multiple infections (a: R = 0.826, P < 0.001; b: R = 0.773, P = 0.015; c: R = −0.876, P = 0.022). Each dot represents the parameter estimates from a single line combination.

Discussion

The evolutionary consequences of genetically diverse infections have so far mainly been investigated with reference to the evolution of virulence (Frank, 1996; Read & Taylor, 2001; Graham et al., 2005; de Roode et al., 2005). How genetic components determining resistance traits behave under multiple infections, on the other hand, have so far received little attention. Consequently, most empirical studies that actually use different parasite species or genotypes apply separate single infections instead of multiple simultaneous infections and implicitly assume that the genetic components of host resistance traits behave in similar ways regardless of the particular infecting strain (e.g. Carius et al., 2001; Lazzaro et al., 2004; Grech et al., 2006; Rauch et al., 2006). This assumption is not justified even with single infections (Wilfert & Schmid-Hempel, 2008). Here we now show that the genetic architecture of a resistance trait of the red flour beetle T. castaneum, i.e. genetic components contributing to survival under infection with the microsporidian N. whitei, differs between infections with single isolates and simultaneous, multiple infections with the same isolates. The genetic architecture of a trait determines the response to selection and ultimately the evolution of the trait (Hansen, 2006). Hence, the observed differences in the relevant genetic architecture will, without doubt, change the evolutionary trajectory of resistance in single or multiple infections, respectively.

To begin with, variation in host resistance of some host line combinations could be sufficiently explained by additive and dominance components (model ‘AD’) only in one infection treatment but not the other (single or multiple). In these cases, the other treatment called for more complex models including maternal and epistatic components (model ‘ADME’; this was the case, for example, for line combinations nos. 32 × 43 and nos. 40 × 43, Fig. 1). Given the strong phenotypic correlation of survival rates between treatments (see Results, Fig. 1), such differences in model complexity demonstrate the power of the joint scaling principle to detect subtle differences in overall contributions of genetic architecture (Demuth & Wade, 2006). The rejection of the ‘AD’-model for the multiple infection in line no. 32 × 43 can for example be explained by large maternal effects as is visible in the large error bars in lines with reciprocal crosses (F1, BC1, BC2) and indicated by the high absolute values for the parameter estimates of maternal components (Table 3). Similarly, the higher variation in second generation crosses of line no. 40 × 43 in the multiple infection leads to lower weights for the associated means, and to the subsequent acceptance of the ‘AD’-model instead of the full ‘ADME’-model.

Overall, epistatic components had the strongest effect on phenotypic variation (Fig. 3), which supports previous results for this system (Wegner et al., 2008). This also fits the general pattern found with resistance quantitative trait loci (QTL) in plants (Kover & Caicedo, 2001) and animals (Wilfert & Schmid-Hempel, 2008) where epistatic interactions often explain a major proportion of the observed variance. Interestingly in this context, the relative contribution of additive and dominance components varied only little between single and multiple infections whereas the contribution of epistatic components were strongly reduced in multiple infections (Fig. 3).

Specific genotype × genotype interactions are widespread in nature (Lambrechts et al., 2006). A recent meta-analysis showed that the QTLs determining resistance against one parasite isolate were on average only recovered in 24% of the cases where infections with a different parasite isolate were examined (Wilfert & Schmid-Hempel, 2008). In other words, each particular host × parasite combination seems to be based on a different set of QTLs. In this light, it would not be surprising that genetic contributions from additive, dominance and epistatic effects are lower in multiple infections as joint scaling only detects the net contribution resulting from the overlay of specific genetic architectures in multiple infections (Demuth & Wade, 2006, 2007a). This does, however, not exclude that the contribution of these genetic effects could be substantial in the separate, single infections with Nosema isolates (NL7 and NS3).

What is surprising, on the other hand, is the large difference in maternal components between single and multiple infections, which becomes apparent not only by the magnitude of their absolute contribution (Fig. 3) but also by the opposite sign of effects (Fig. 4c). Maternal effects in immunity are well known for vertebrates (Grindstaff et al., 2003) but have recently also been described for invertebrates (Little et al., 2003; Sadd & Schmid-Hempel, 2007). In those studies, a pre-exposure of the mothers lead to a (facultatively) increased and also specific resistance against the parasite in offspring. As, the beetles used as breeders in this experiment were themselves not exposed to the parasite, different mechanisms must apply. One possibility is that specificity of host–parasite interactions in individuals, whose mothers have not been previously exposed to parasite, is mediated by genetic and especially epistatic components that are constitutive because of their presence in the genome. Maternal effects in unchallenged mothers may constitute a more general form of resistance probably resembling general vigour. Only upon exposure of the mother, maternal components conferring specificity might be up-regulated and transmitted to the offspring.

Another possibility is that the two parasite isolates (i.e. NL7 and NS3) that were added to the single infection isolate simply reveal higher contribution of maternal components and therefore reflect specific genetic architectures of each strain rather than a general difference between single and multiple infections. As we did not investigate the genetic architecture of survival with these two isolates in single infections we cannot rule out this possibility. Given the very small, albeit in the majority significant, values of the absolute parameter estimates for maternal components in the single infection treatment (Table 3) along with the reversal of the correlation from positive to negative (Fig. 4), we would argue that such a large shift seems unlikely for other single infections. Furthermore, we chose all isolates specifically so that their mean virulence did not differ in previous single infection experiments including all parental lines used to found the line crosses in this study (Otti, 2007), which rules out that differences in the observed architecture were based on different parasite virulence alone. In total, host mortality caused by multiple infections was significantly lower in the multiple infection treatment (Table 2). As the parasite isolates did not differ significantly in virulence in single infections (Otti, 2007), we would have expected a similar expression of virulence in the multiple infection treatment. Although we do not know if exposure with multiple parasite isolates really resulted in multiple infections, one explanation for the reduction in virulence therefore is competition within the host, which has previously been reported for a variety of experimental systems (Ben-Ami et al., 2008; Rauch et al., 2008). Yet, such competition might not necessarily result in the strains with lower virulence to prevail.

The presence of strong maternal, possibly X-linked, components is also in line with genetic effects governing other traits and male specific hybrid breakdown reported for T. castaneum (Demuth & Wade, 2007b), where interactions between X-linked and autosomal genes seem to be at least partly responsible. The T. castaneum X-chromosome is characterized by a high concentration of immune genes mainly involved in signalling pathways (Zou et al., 2007; Richards et al., 2008; KM Wegner, personal observation) and according to an earlier study in Drosophila melanogaster an interaction between signalling genes and genes located up- or downstream in the cascade is likely to bear the signature of epistasis (Lazzaro et al., 2004). Our results now suggest that resistance conferred by such epistatic interactions might represent the genetic architecture of a single infection. The combination of several genetic architectures in multiple infections lowers the overall effect of epistatic components (Fig. 3) suggesting that specific genetic architectures cancel each other out. Maternal components, on the other hand, gained more importance in multiple infections, which will result in a different evolutionary trajectory in multiple infections compared to single infections (Figs 3 and 4). Therefore maternal components and their potential feedback on host–parasite co-evolution in general should be considered further to identify the crucial genetic components that determine the topology of the fitness landscape in the more realistic case of multiple simultaneous infections.

Acknowledgments

We thank Lisa Shama, Oliver Otti and Christoph Hutter for assistance in the lab. This study was financially supported by an ETH-grant nr. TH-09 06-1 to PSH, and SNF grant 31-120451 to KMW; supported by the Genetic Diversity Centre of ETHZ (GDC) and CCES.

Ancillary