Host ecology shapes geographical variation for resistance to bacterial infection in Drosophila melanogaster


*Present address and correspondence : Department of Ecology and Evolutionary Biology, University of Arizona, PO Box 210088, Tucson, AZ 85721, USA. E-mail:


  • 1Geographically distinct host populations often experience very different ecological conditions. These variable ecological conditions impact the strength of selection that these hosts experience from their parasites.
  • 2Numerous studies have characterized geographical patterns of resistance to infection among natural populations in the context of host–parasite local adaptation, but what other factors might contribute to these differences?
  • 3Here, we determined whether 20 naturally isolated populations of Drosophila melanogaster collected along the East Coast of the United States varied for survival after being inoculated with one of two species of bacteria – Lactococcus lactis and Pseudomonas aeruginosa. We then asked whether host environment accounted for the observed patterns of resistance.
  • 4Resistance to both types of infection varied spatially. The hosts’ natural environment was predictive of the observed spatial variation in resistance to L. lactis, but not P. aeruginosa, infection. Specifically, hosts exposed to species-rich bacterial communities were more likely to survive the infection.
  • 5We conclude that biotic characteristics of the host environment, specifically the number of species of bacteria hosts encounter, shape host resistance to bacterial infection in nature. We discuss our results in the context of what is known about the evolutionary ecology of resistance in invertebrate systems.


Ecological variation can affect the strength of natural selection that individuals experience. Geographically distinct populations often experience very different ecological conditions and selection intensities, leading to spatial variation in fitness-related traits (Thompson 1999; Brodie, Ridenhour & Brodie 2002; Baucom & Mauricio 2007). Work in the field of ecological immunology demonstrates that invertebrate resistance to infection can be highly variable in response to both biotic (Schmid-Hempel 2003, 2005) and abiotic forces (Ferguson & Read 2002; Blanford et al. 2003; Mitchell et al. 2005). Some of these forces cause distinct geographical patterns of resistance, so that some host populations are more robust than others in fighting infection. Consider the numerous examples of parasites locally adapted to their hosts (Kaltz & Shykoff 1998; Dybdahl & Storfer 2003; Kawecki & Ebert 2004). A classic example comes from Lively's work in the snail (Potamopyrgus antipodarum)–trematode (Microphallus sp.) system, where reciprocal transplants demonstrated that parasites were more infective on sympatric hosts relative to allopatric hosts (Lively 1989). Apart from the specific one-host, one-parasite interactions characteristic of local adaptation, however, invertebrates deal with a multitude of other environmental stressors. Many of these stressors are spatially variable and might produce distinct patterns of resistance among host populations.

Drosophila melanogaster is a cosmopolitan generalist that is distributed widely throughout the globe (Powell 1997; Markow & O’Grady 2006). Owing to their wide distribution, they encounter a diverse set of environments. Across spatial scales, such environmental variation could cause spatial patterns in the ability of hosts to fight infection. For example, Kraaijeveld & van Alphen (1995) found that D. melanogaster populations collected throughout Europe varied for resistance to attack by the parasitoid Asobara tabida. These patterns were not explained by local adaptation (Kraaijeveld & Godfray 1999), and the authors suggested that differences in either the frequency of alternative hosts or pressure from other parasites might play a role. In contrast to what is known about parasitoid infection, there is no evidence of variable resistance to bacterial infection among D. melanogaster populations. However, because these hosts are distributed widely (Powell 1997; Markow & O’Grady 2006) and model systems for studying immunity (Vodovar et al. 2004), we can easily ask key questions about whether and why resistance to bacterial infection is variable across invertebrate populations. Motivated by the empirical examples below, we discuss three factors that could cause spatial patterns of resistance: temperature, precipitation and variation in the structure of bacterial communities interacting with D. melanogaster.

The presence and prevalence of host-associated bacteria associating with D. melanogaster vary across this host species’ range (Corby-Harris et al. 2007). Where infection risk is high, there should be positive selection for resistance to a novel infection. Conversely, if resistance is costly (Kraaijeveld, Ferrari & Godfray 2002), resistance to a novel infection should be detrimental to host fitness in areas where the risk of infection is low. It is currently unclear whether the uneven distribution of D. melanogaster-associated bacteria contributes to geographical variation for host resistance. However, a recent study by Scharsack et al. (2007) in vertebrates may provide some clues. Scharsack et al. (2007) exposed river-adapted sticklebacks, which normally live with fewer parasites than lake-adapted individuals, to parasite-dense lake conditions. They found that river-adapted fish had higher parasite loads and reduced immune response, both innate and adaptive, relative to lake-adapted fish (Scharsack et al. 2007). Such patterns suggest that the risk of encountering parasites shapes host immunocompetence (Scharsack et al. 2007). For some D. melanogaster populations, we know how many bacterial species associate with these hosts in nature (Corby-Harris et al. 2007). We can therefore use these data to generate a testable prediction motivated by Scharsack et al.'s (2007) work: hosts that associate with a species-rich bacterial community are more likely to resist a novel infection.

Temperature varies across a host species’ range and is an important regulator of invertebrate host immunocompetence over both ecological and evolutionary time-scales. Over shorter ecological time-scales, fluctuations in ambient temperature can greatly affect host resistance to fungal, bacterial and viral infections (Thomas & Blanford 2003). Temperature also affects long-term host–pathogen dynamics by impacting pathogen virulence (Mitchell et al. 2005), the ability of host populations to adaptively respond to infection (Ferguson & Read 2002), or the presence of symbionts that afford protection from pathogens (Bensadia et al. 2006). Mitchell et al. (2005) studied whether temperature mediated interactions between Daphnia magna and its sterilizing pathogen, Pasteuria ramosa. At high temperatures, the pathogen was much more virulent and sterilized more hosts relative to low-temperature conditions, suggesting that the strength of parasite-mediated selection was temperature-dependent (Mitchell et al. 2005). D. melanogaster inhabit a wide geographical range and are probably influenced by temperature shifts. These temperature fluctuations may regulate host–bacteria interactions and contribute to spatial variation in D. melanogaster immunocompetence.

Levels of precipitation often vary across a host's range. Although the research on the relationship between humidity and host–parasite interactions is scant, it suggests that precipitation may cause among-population resistance variation. In birds, for example, low humidity causes a reduction in the number of ectoparasites that hosts encounter (Moyer, Drown & Clayton 2002). In the aphid Acyrthosiphon pisum, there is strong evidence that the endosymbiont Regiella insecticola confers resistance to fungal infection under laboratory conditions (Scarborough, Ferrari & Godfray 2005). Researchers in Japan find that R. insecticola tend to be more frequent in drier environments (Tsuchida et al. 2002), suggesting a potential link between precipitation and resistance. Although this relationship has not been tested explicitly, such data suggest that precipitation could be important in regulating the evolution of host resistance through changes in parasite-mediated selection pressure.

In this study, we investigated whether D. melanogaster populations collected along the East Coast of the United States varied for resistance to bacterial infection. We then sought to explain this spatial variation in resistance by asking whether differences in temperature, precipitation or bacterial species richness explained variation in resistance among host locations. Because resistance to parasitoid attack shows evidence for geographical variation (Kraaijeveld & van Alphen 1995; Kraaijeveld & Godfray 1999), we expected resistance to bacterial infection to vary among populations. Further, because differences in pathogen species richness (Scharsack et al. 2007), temperature (Ferguson et al. 2002; Mitchell et al. 2005; Bensadia et al. 2006) and precipitation (Moyer et al. 2002; Tsuchida et al. 2002; Scarborough et al. 2005) are related to differences in parasite-mediated selection pressure in a variety of host systems, each of these factors could correlate with among-population differences in resistance. Interestingly, our data suggest that one factor in particular, bacterial species richness, shapes geographical variation for resistance in naturally isolated D. melanogaster populations.


fly collections

Flies were collected from 20 sites along a latitudinal transect on the East Coast of the United States between June and September 2005 (Fig. 1, Table 1) by sweep-netting behind fruit stands and over fruit bucket traps. Following collection, flies were anaesthetized over ice. Male D. melanogaster were separated from other Drosophilid species on the basis of morphological characteristics (Ashburner 1989; Markow et al. 2006). These males were set aside and preserved in groups of five in 70% ethanol for subsequent analysis of their bacterial communities. After returning to the laboratory from the collection sites, the ethanol-preserved samples were kept at –80 °C. The D. melanogaster-like females were separated from the group of field-collected flies and were maintained in individual vials containing standard cornmeal, molasses and yeast fly medium at 24 °C on a 12 : 12 h light : dark cycle. Females were cleared from their vial after laying eggs. Approximately 12 days later, their male progeny were examined further to determine which lines were D. melanogaster. D. melanogaster isofemale lines were then established by first mating F1 full- or half-sibs, followed by subsequent generations of full-sib matings.

Figure 1.

D. melanogaster were collected at twenty different locations along the East Coast of the United States. Numbers on the right side of the figure correspond to the population numbers in Fig. 2 and in Table 1. The grey box (left) indicates the range encompassed by the enlarged map on the right. Single points with two corresponding numbers represent separate locations ≤ 20 km apart. Map courtesy of

Table 1.  Characteristics of the 10 host populations assayed for bacterial species richness
Population no.1Collection locationLatitudeDate collectedMean annual precipitation (cm)2Mean summer temperature (°C)3Mean January low temperature (°C)4Chao15
  • 1

    Population number as referred to in Fig. 1.

  • 2

    30-year average of mean annual precipitation (in centimeters) recorded at that location by the NOAA (NOAA 2002a,b,c; d, e).

  • 3

    30-year average of mean summer temperatures (in °Celsius) recorded at that location by the NOAA (NOAA 2002a,b,c; d, e).

  • 4

    30-year average of mean low temperature (in °Celsius) recorded in the month of January at that location by the NOAA (NOAA 2002a,b,c; d, e).

  • 5

    Chao1 estimate of bacterial species richness (Chao 1984; Chao, Ma, & Yang 1993), using 1% sequence divergence ± one standard error around the estimate. Calculations for standard error were calculated according to Chao (1984), Chao, Ma, & Yang 1993).

1Oakland, New Jersey41·0128 Aug 2005123·925·0–7·5521·8 ± 0·78
6Inwood, West Virginia39·2213 Jul 2005100·024·3–5·2252·9 ± 3·2
8Woodstock, Virginia38·872 Sep 200595·322·7–6·6138·9 ± 1·8
9Daleville, Virginia37·412 Sep 2005107·922·7–3·0034·0 ± 3·2
10Daleville, Virginia37·412 Sep 2005107·922·7–3·0018·3 ± 1·6
12Hillsborough, North Carolina36·0719 Jul 2005122·025·8–2·3343·5 ± 9·7
13Raleigh, North Carolina35·8220 Jul 2005118·121·1–1·0617·5 ± 5·8
14Watkinsville, Georgia33·8611 Aug 2005127·024·2 1·39  28 ± 6·3
16Bishop, Georgia33·8111 Aug 2005127·025·1 1·39 9·5 ± 0·81
17Macon, Georgia32·8310 Aug 2005114·321·8 1·3933·2 ± 3·0

After 10 generations of inbreeding, for each of the 20 populations, 10 isofemale lines were selected randomly and combined to form an outbred population. Because only 10 D. melanogaster isofemale lines were collected from two of the populations, this method allowed us to keep the genetic variance relatively constant among each outbred population while including the maximum number of populations in the study. Each of these randomly selected lines was represented equally in the new outbred population by placing two males and two females from each of these 10 isofemale lines in duplicate 17 mL culture bottles containing standard medium under the conditions described above.

bacterial stocks

We acquired strains of Pseudomonas aeruginosa and Lactococcus lactis from B. P. Lazzaro in January 2004. L. lactis was isolated from a wild population of D. melanogaster in Pennsylvania and has been described previously (Lazzaro, Sackton & Clark 2006). 16S rDNA gene sequence analysis of this strain of L. lactis (kindly provided by BP Lazzaro) shows that it is not related closely to 16S rDNA gene sequences isolated from bacterial communities that co-occurred naturally with the flies used in this study at the time they were collected (V. Corby-Harris, unpublished data). We therefore assume that this bacterium, although present in some natural populations of D. melanogaster, is a novel infection from the perspective of the hosts we assayed. P. aeruginosa is a laboratory stock (strain PA01) and was not derived from wild flies. However, it is a common insect pathogen (Lacey 1997) and Pseudomonas species have been identified in wild D. melanogaster (Corby-Harris et al. 2007). Working stocks of bacterial cultures were maintained in nutrient broth at 4 °C in a 10−2 dilution (of an unmeasured dense culture) until ready for use. The day before the infections, these working stocks were grown for approximately 18 h at 37 °C, which corresponds to the log phase of growth (V. Corby-Harris, unpublished data).

details of infection

Each of the 20 outbred D. melanogaster populations was maintained for approximately six to 10 non-overlapping generations before being experimentally infected over a period of 7 days between January and February 2006. Adult male flies were infected manually in the lateral thorax using a fine stainless steel needle (Fine Science Tools, Foster City, CA, USA) dipped in a standardized bacterial culture (Tzou, Meister & Lemaitre 2002). Inocula were standardized by diluting an overnight culture with sterile nutrient broth until they measured 0·200 ± 0·050 A at 600 Λ using a UV spectrophotometer (Thermo Electron, Rochester, NY, USA). This concentration delivers approximately 60 bacterial cells into the individual (V. Corby-Harris, unpublished data).

On each of the 7 infection days, for each of the 20 host populations, we inoculated a random mixed sample of approximately 13 mated and virgin male flies that were between 24 and 48 h old. Of the 13 individuals infected per population per day, five were inoculated with a standardized P. aeruginosa culture, five with a standardized L. lactis culture and three with a negative control (sterile nutrient broth) to account for the effects of mechanical trauma. Across the seven dates when infections were performed, therefore, a total of 35 flies per population were infected with each of the two bacterial treatments and 21 were infected with the control. Immediately following infection, individuals were placed into clean, unyeasted vials with 5 mL of standard cornmeal, molasses and yeast fly medium in groups of five individuals, and monitored for 48 h at 24 °C with a 12 : 12 h light : dark cycle. The number of dead flies in each vial was recorded at 3 h post-infection, hourly from 16 to 31 h post-infection, and then again at 48 h post-infection.

On each of the seven dates of infection, summed over the 20 host populations, a total of 260 individuals (divided into 60 vials) were infected with one of the three treatments within one afternoon. Clearly, all 260 individuals could not be infected at the same time and we could not follow each of the 60 vials from the exact time of infection to exactly n hours post-infection. Previous work, however, suggests that there is an effect of time of day on resistance (Lazzaro, Sceurman & Clark 2004). We therefore took the following precautions in order to manageably account for the time of infection. First, on each date, the sequence of infections was randomized with respect to treatment and population. Secondly, the exact time of infection was recorded and all vials infected within a 1h time period were binned together into within-date time blocks. Measurements of mortality were staggered with respect to this within-date time block. Therefore, a death recorded at a particular time during the observation period would mean, in actuality, death at a later time-point for individuals infected earlier and death at an earlier time-point for those infected later. This data collection scheme was a compromise between accuracy of measurement and the reality of the workload.

Mortality from the control inoculation was negligible (see Results). Additionally, previous observations suggest that flies dying from the experimentally applied bacterial infection do so between 16 and 48 h post-inoculation (V. Corby-Harris, unpublished data). Flies dying within 16 h post-inoculation presumably died from the mechanical trauma of the infection, and those surviving past 48 h post-inoculation either cleared the bacteria or were tolerant to the negative effects of the infection. We therefore limited analyses of geographical variation for resistance to flies infected with either of the two species of bacteria that survived past 16 h post-inoculation. Infected individuals surviving past 48 h post-inoculation were censored in our analyses.

statistical analysis of resistance data

We determined an individual's likelihood of dying from infection given a set of predictor variables by analysing the Cox proportional hazards (CPH) model (Cox 1972; Allison 1995; Therneau & Grambsch 2000). This model is particularly useful in analysing resistance data because it allows for a time-dependent dichotomous response variable and for censored observations (Allison 1995). To facilitate the interpretation of our data, we include here a short review of the CPH model as it applies to the analysis of infection resistance data. More detailed discussion of these models can be found in Cox (1972), Allison (1995) or Therneau & Grambsch (2000).

The CPH model is semiparametric; it depends on a non-parametric baseline hazard function (h0) that remains unspecified and a linear combination of k predictor variables that is exponentiated. For each replicate, i, the ‘hazard’ of dying from infection at time t is:


Strata variables can also be included in the model to account for parameters whose effects one is not interested specifically in estimating, but none the less must control for (Therneau & Grambsch 2000).

Two types of CPH models were analysed. Vial was the unit of replication in each case. We were first interested in the effects of host population, date of infection and within-date time block. As such, for each of the two bacterial treatments, we determined whether host population, date of infection, within-date time block, the interaction between date of infection and host population and the interaction between within-date time block and host population affected significantly hours to death post-infection. In subsequent analyses of the relationship between host resistance and host environment, we were interested primarily in estimating the susceptibility of each population to infection while controlling for the effects of date of infection and within-date time block. Here, for each of the two bacterial treatments, a model was analysed where host population was the only fixed independent variable and hours to death was the dependent variable. We controlled for the effect of both within-date time block and date of infection by including them as strata variables. All survival analyses were performed using proc tphreg in sas version 9·1 (SAS Institute 2005).

bacterial richness assays

The species richness of the bacterial communities associating with D. melanogaster hosts was assayed at 10 of the 20 sites where D. melanogaster were collected from June to September 2005. A full description of the methods used to isolate, amplify and sequence the bacterial DNA inside the Drosophila hosts is provided in Corby-Harris et al. (2007). Operational taxonomic unit (OTU) groupings were determined using 99% sequence identity and the Chao1 non-parametric species richness estimator (Chao 1984; Chao, Ma & Yang 1993) was obtained for each of the 10 populations.

statistical tests to determine environmental factors contributing to resistance variation

We investigated whether aspects of the hosts’ environment, such as temperature, precipitation and the number of host-associated bacterial species, accounted for the observed spatial variation in resistance to L. lactis and P. aeruginosa. Data sets for each type of bacterial treatment were analysed separately. For each species of bacteria, we first analysed a full multiple regression model that asked whether the independent variables, including mean annual temperature, mean summer temperature (defined as the average temperatures between the months of July, August and September), mean January low temperature, mean annual precipitation and bacterial species richness at each of the 10 collection sites, contributed to variation in host mortality post-infection. Temperature and precipitation data represented 30-year averages recorded by the National Oceanic and Atmospheric Administration (NOAA) at each collection site (NOAA 2002a,b,c,d,e). For each population, we estimated the mean resistance to either species of bacteria as the hazard ratio from the proportional hazards model for each treatment. This ratio measures the likelihood of a population dying from the inoculation relative to all other populations infected with that bacterium. Large hazard ratios indicate susceptibility to infection. For both bacterial treatments, this estimate was natural log-transformed to improve normality, which was evaluated using plots of residuals vs. the predicted values of the response (Quinn & Keough 2002). The Chao1 non-parametric estimates of species richness generated from the 16S rRNA gene sequence data were used to estimate the number of bacterial OTUs co-occurring with each of the 10 D. melanogaster populations. We next used collinearity diagnostics to ensure that the predictor variables were not confounded with one another (Quinn & Keough 2002; Graham 2003). Under the full model, variance inflation factors (VIFs) were calculated from the regression of each predictor variable against the remaining predictor variables (Quinn et al. 2002). Variables with a VIF greater than 10 were examined more closely to determine whether they could be collapsed into one predictor variable. In such cases, the predictor variable with the strongest correlation with the response was retained in the reduced regression model. A reduced model was then generated, tested using multiple regression and again checked for collinearity. All multiple regression analyses were performed using proc reg in sas version 9·1 (SAS Institute 2005).


Among the 371 negative control flies infected with sterile broth, eight died within 48 h (0·02%). Host population (logistic regression Wald inline image = 0·0015, P = 1·0), date of infection (Wald inline image = 4·6 × 10−6, P = 1·0) and the interaction between date of infection and population (Wald inline image = 4·7 × 10−4, P = 1·0) did not significantly predict these eight individuals’ mortality.

Six of the P. aeurginosa-infected and 15 of the L. lactis-infected individuals died within 16 h post-inoculation. Mortality within 16 h was unaffected by infection date (Wald inline image = 0·0025, P = 1·0), host population (Wald inline image = 0·0031, P = 1·0), bacterial treatment (Wald inline image = 1·04 × 10−7, P = 0·99), population × treatment (Wald inline image= 0·0012, P= 1·0), population × date of infection (Wald inline image = 0·17, P = 1·0) and treatment × date of infection (Wald inline image = 0·012, P = 1·0).

Forty (5·9%) of the P. aeruginosa-infected and 22 (3·2%) of the L. lactis-infected flies survived past the 48 h observation period. These censored individuals were distributed randomly with respect to infection date (Wald inline image = 0·0035, P = 1·0), host population (Wald inline image = 0·0028, P = 1·0), bacterial treatment (Wald inline image = 9·5 × 10−5, P = 0·99), population × treatment (Wald inline image = 0·0042, P = 1·0), population × date of infection (Wald inline image = 5·84, P = 1·0) and treatment × date of infection (Wald inline image = 0·17, P = 0·99).

Of the 683 D. melanogaster infected with P. aeruginosa, 643 (94·1%) died within 48 h post-inoculation. Overall, P. aeruginosa-induced mortality increased between 15 and 24 h post-infection before levelling off. The median time to death of these individuals was 24 h, and the mean survival time was 23·94 h ± 0·14 standard error (SE). For the L. lactis treatment, 661 (96·8%) of the 684 died within 48 h. Mortality rates for the L. lactis treatment increased between 15 and 21 h post-infection before plateauing. The median time to death for these individuals was 22 h, and the mean survival time was 23·00 h ± 0·14 SE. Pooled across host populations, mortality was affected by within-date time block; individuals infected later each afternoon generally had higher mortality post-inoculation for both bacterial treatments.

There was significant variation for resistance among outbred D. melanogaster host populations infected with P. aeruginosa and L. lactis (Fig. 2). Host population (Wald inline image = 206·49, P < 0·0001, Fig. 2), date of infection (Wald inline image = 93·23, P < 0·0001), within-date time block (Wald inline image = 57·00, P < 0·0001), date of infection × host population (Wald inline image = 264·48, P < 0·0001) and within-date time block × host population (Wald inline image = 266·47, P < 0·0001) were significant predictors of the observed variation in time to death following inoculation with P. aeruginosa. Host population (Wald inline image = 60·50, P < 0·0001) remained significant for the reduced data set (n = 10) subsequently used for testing the relationship between host environment and population mean resistance. Host population (Wald inline image = 42·67, P = 0·0014, Fig. 2), date of infection (Wald inline image = 18·99, P = 0·0042), within-date time block (Wald inline image = 28·52, P < 0·0001), date of infection × host population (Wald inline image = 189·58, P < 0·0001) and within-date time block × host population (Wald inline image = 115·37, P < 0·0001) were significant predictors of the observed variation in mortality post-inoculation for the L. lactis treatment across the 20 populations assayed for geographical variation for resistance. The effect of host population (Wald inline image = 17·22, P = 0·0454) remained significant across the 10 populations subsequently tested for the effects of host environment on resistance.

Figure 2.

Populations of D. melanogaster varied for how well they resisted both P. aeruginosa (black) and L. lactis (grey) infection. Outbred host populations (x-axis) are arranged from left to right in order of decreasing latitude. Mortality post-inoculation (y-axis) represents each population's natural log transformed hazard ratio post-inoculation under a proportional hazards model. Values below the x-axis indicate resistance, while values above indicate susceptibility. Error bars represent the standard error around the transformed hazard ratio estimate. Host population was a significant predictor of time to death following inoculation for both the P. aeruginosa (Wald inline image = 206·49, P < 0·0001) and L. lactis (Wald inline image = 42·67, P = 0·0014) treatments.

Bacterial richness varied across host populations as measured by the Chao1 estimator (Table 1). The mean number of bacterial OTUs associating with D. melanogaster hosts was 29·8 ± 13·3 standard deviation (SD).

Mean annual temperature, mean summer temperature and mean January low were significantly correlated positively (data not shown) and the VIF of all three temperature variables was significantly greater than 10. This suggested strong collinearity among temperature variables (Quinn et al. 2002). To resolve this issue, we constructed reduced models containing the independent variables bacterial species richness, mean annual precipitation and either mean summer temperature or mean January low temperature for the L. lactis and P. aeruginosa data sets, respectively (see Methods for variable selection criteria). In contrast to the full models, these reduced models were no longer affected by collinearity among the predictor variables, as all variables had VIFs  2·10. Under the reduced model, among-population differences in L. lactis-induced mortality were negatively correlated with differences in bacterial species richness (β = –0·022, t1 = –2·57, P = 0·042; Table 2), while differences in mean summer temperature (β = –0·035, t1 = –0·94, P = 0·38; Table 2) and mean annual precipitation (β = –0·04, t1 = –1·33, P = 0·23; Table 2) were not. P. aeruginosa-induced mortality was not associated with differences in mean January low temperature (β = –0·18, t1 = –1·76, P = 0·1297, Table 2), mean annual precipitation (β = 0·31, t1 = 1·90, P = 0·1054, Table 2) or bacterial species richness (β = 0·0023, t1 = –1·25, P = 0·26).

Table 2.  Multiple regression analysis of the effect of host environment on mortality post-inoculation. For each species of bacteria with which hosts were infected, the results of the reduced model analysis are presented (see text); * P < 0·05
Bacterial treatmentFactorParameter estimatet1P
Pseudomonas aeruginosaBacterial species richness 0·002 0·05 
Mean January low temperature–0·180–1·76 
Mean annual precipitation 0·308 1·90 
Lactococcus lactisBacterial species richness–0·022–2·57*
Mean summer temperature–0·035–0·94 
Mean annual precipitation–0·040–1·33 


In summary, our data show that resistance to bacterial infection, like resistance to parasitoid infection (Kraaijeveld et al. 1995; Kraaijeveld et al. 1999), varies among D. melanogaster populations. Among population differences in L. lactis resistance were associated significantly with differences in bacterial species richness. In contrast, the among-population differences in host resistance to P. aeruginosa were not correlated with any of the parameters included in our models.

The data set we have analysed in this study has both strengths and limitations. A notable strength of this work is that it provides an increased understanding of the ecology of D. melanogaster, a model organism whose natural history is not well understood. In addition, as D. melanogaster (Tzou et al. 2002; Hultmark 2003; Vodovar et al. 2004) and now other species in the genus (Sackton et al. 2007) are used increasingly as models for studying the molecular and genetic basis of the immune response, an understanding of the natural forces shaping Drosophila resistance variation is warranted. There are four significant limitations of this study. First, in our bacterial species richness assay we sampled only five flies per location (pooled into one sample), and our estimates of species richness could change with increased sampling effort. Secondly, our ability to identify trends in bacterial species richness is potentially weakened by the fact that there are not multiple samples within each collection site. Thirdly, we did not explicitly test whether bacterial richness results in resistance variation through experimental manipulation of this parameter. Lastly, the hosts assayed for resistance in this study were assayed using septic injury, a method of infection that D. melanogaster hosts would probably not encounter in the wild. None the less, our results confirm that D. melanogaster is a powerful model organism not only for laboratory-based studies of immune function, but also to help us predict and understand the ecological forces that shape immunity.

Populations exposed to a species-rich bacterial community in nature are more resistant to a novel L. lactis infection under laboratory conditions. That hosts may respond adaptively to the population-level characterisitics of microbial communities has been acknowledged previously in reference to both invertebrate (Kraaijeveld et al. 1999; Kurtz & Hammerschmidt 2006) and vertebrate (Scharsack et al. 2007) host systems. Several processes could be operating to produce such a pattern. Below, we discuss each of these processes in further detail.

In pathogen-rich environments, where hosts are likely to encounter novel infections, selection should favour an increased immune response that would decrease the likelihood of novel infection. This might lead, in turn, to among-population patterns of resistance similar to those we observed if hosts responded adaptively to bacterial species richness. Many Drosophila immune genes show signs of strong selection (Begun & Whitley 2000; Schlenke & Begun 2003; Lazzaro 2005; Jiggins & Kim 2006, 2007; Obbard et al. 2006; Sackton et al. 2007), suggesting that they coevolve with pathogens (Jiggins & Kim 2007). However, it is unclear whether Drosophila hosts respond to traits of individual bacteria (i.e. virulence) or the bacterial community overall (i.e. species richness). Interestingly, research in both D. melanogaster and Anopheles gambiae point towards a mechanism that may underlie adaptive changes in resistance in response to changes in bacterial species richness. An ‘adaptive’ immune gene, known as Dscam in D. melanogaster (Watson et al. 2005) and AgDscam in A. gambiae (Dong, Taylor & Dimopoulos 2006), responds specifically to infection via alternative splice variants (Dong et al. 2006). It is possible that diverse pathogenic environments could select for resistant invertebrate hosts if hosts that produce a more diverse set of these ‘adaptive’ gene variants have high relative fitness.

These hosts may be responding adaptively to the presence or absence of a particular species of bacteria, and not to bacterial species richness per se. For example, the presence of a particular Lactococcus species could select for increased resistance to subsequent L. lactis infection. This possibility of cross-resistance remains an open question. Our sequencing effort was not extensive enough to identify all the bacteria present in these populations, and some rare species probably remained undiscovered (Corby-Harris et al. 2007). Without a complete catalogue of all bacteria living with these flies in nature, it is difficult to investigate this relationship.

In cases where the host could be confronted with subsequent infections, mechanisms such as adaptive immunity that reduce the likelihood of secondary infection should be selected for (Kurtz 2004; Little & Kraaijeveld 2004). These adaptive immune responses would be activated in pathogen-rich environments, priming the host's immune response towards infection they could encounter subsequently. Kurtz & Hammerschmidt (2006) tested this expectation by subjecting copepod hosts to homogeneous or heterogeneous combinations of tapeworm parasites. They then tested their resistance to a novel subsequent tapeworm infection. Contrary to their expectations, the heterogeneous environment did not lead to increased resistance towards a subsequent challenge, suggesting that the effect of parasite heterogeneity on host immunity is not straightforward (Kurtz & Hammerschmidt 2006). Overall, we doubt that immune priming accounts for the relationship we observed between richness and resistance to L. lactis infection. Current examples demonstrate the positive effects of transgenerational priming over a maximum of two generations (Little & Kraaijeveld 2004; Sadd et al. 2005; Moret 2006). In contrast, the hosts studied here were removed from their natural environments for more than 15 generations before being assayed, suggesting that transgenerational effects are not operating.

Among-population variation for resistance was not influenced strongly by differences in either temperature or precipitation. This negative result might reflect our inability to capture fine-scale differences using 30-year temperature and precipitation averages. Alternatively, we may observe stronger environmental effects across a larger latitudinal gradient (Schmidt et al. 2005). Given the clear examples showing that climate (Mitchell et al. 2005; Bensadia et al. 2006) influences the strength of parasite pressure on hosts under laboratory conditions, we remain optimistic that these relationships exist in natural populations of D. melanogaster.

The work described here serves as an important first step towards understanding whether and why D. melanogaster populations differ in resistance. Overall, our data support the hypothesis that D. melanogaster host populations respond adaptively to characteristics of the bacterial communities with which they interact. We conclude with two important points. First, the experiments here serve as a way of identifying important broad-scale patterns. Only through experiments where bacterial species richness is manipulated under controlled conditions can the exact mechanisms underlying these patterns be deduced. Secondly, environmental factors other than those studied here are likely to influence variation in host resistance to bacterial infection. Those studied here, however, do provide a first clue as to which factors may be important to the evolution of resistance in nature. We hope this work identifies avenues of future research geared towards a better understanding of invertebrate immunity in a real-world context.


The authors wish to thank Jeff Bennetzen, Ana Clara Pontaroli and Larry Shimkets for assistance in obtaining the original bacterial species data, and the Promislow Laboratory and four anonymous reviewers for comments on a previous version of this manuscript. This work was supported by an NSF DDIG (DEB-0508785) and an Alton Fellowship in Genetics to VCH, and a Senior Scholar Award to DELP from the Ellison Medical Foundation.