1. Evidence suggests that developing and maintaining an effective immune system may be costly and that an organism has to make a trade-off between immune function and other fitness-enhancing traits. To test for a trade-off between growth and immune function we carried out a meta-analysis of data from lines of poultry that had been divergently selected for either growth (body mass) or an aspect of immune function. This is relevant to our understanding of the evolution of immune function, but also because the increased prevalence of antibiotic-resistant bacteria and calls to restrict the use of antibiotics in the agricultural industry has made immune function of livestock an important theme. Has the selection of animals for rapid growth unintentionally resulted in reduced immune function?
2. The lines selected for increased growth all showed a strong and significant decrease in immune function (standard difference in means = 0·8; P < 0·001). No difference was found between the effects on cellular or humoral immunity, although there were few data on cellular immunity, and hence this deserves more study. However, in the lines selected for immune function the effect on growth was heterogeneous and overall it was close to zero.
3. Testing for publication bias revealed that the effect of selection for body mass on immune function was robust. However, there was considerable heterogeneity in both body mass and immune function data. The heterogeneity in the growth-selected lines cannot be accounted for by gender or species: the only turkey line had an effect size between that of the two chicken lines.
4. In conclusion, we found that selection for growth does indeed compromise immune function, but selection for immune function did not consistently affect growth. This is in agreement with the supposition that the costs of growth are large relative to the costs of immune function, and on a practical level this suggests that it may be possible to breed animals for increased growth without loss of immune function.
To test whether there is a trade-off between growth and immune function, we studied experiments with commercial poultry. Fowl diseases, particularly those caused by Escherichia coli infection, present a significant problem to the poultry industry (e.g. Dziva & Stevens 2008). Could it be that by selectively breeding for animals that invest more resources into rapid growth, the industry also selected animals that allocated fewer resources towards developing immune function or with a changed immune function? There is some evidence in free-living birds suggesting a trade-off between immune development and growth rate (Nilsson 2003; Mauck et al. 2005). In experimental conditions, challenging the immune system also reduces growth (Klasing et al. 1987; Lochmiller & Deerenberg 2000), but these findings only take into account the deployment costs of the immune response. There are a considerable number of correlation studies on the relation between immune function and body mass, but unless one of the traits is manipulated these studies cannot prove a causal link. For example, if resources are scarce one expects both growth and immune function to be low, so a positive correlation between the two does not disprove the existence of a trade-off (Lochmiller, Vestey & Boren 1993; Birkhead, Fletcher & Pellatt 1999; Hoi-Leitner et al. 2001). To investigate the existence of a trade-off between growth and immune function, we review studies of animal lines that have been selected for either growth or immune function. By restricting the analysis to selection lines we can be reasonably certain that the differences in the non-manipulated trait are solely the result of selection for the manipulated trait. However, the results of one selection line can become skewed by accidentally co-selected traits. Furthermore, most experiments were limited to measuring one or two parameters (e.g. antibody-titre), and it would be naïve to assume that this would be sufficient to assess immune function. Poor performance after one immune-challenge does not prove the selection line is equally vulnerable to all pathogens, so the analysis must include multiple experiments of the same line if these investigate different aspects of the immune response. To integrate the results in one statistical test, we conducted a meta-analysis on the results of several selection lines.
Meta-analysis is a statistical technique that allows for quantifying, combining and comparing the outcome of multiple research articles, even when the original datasets are not available. The advantage is that while in a normal review the author has to compare different articles and give a subjective judgment on their weight, meta-analysis gives an objective and repeatable quantification of the outcome (Nakagawa & Cuthill 2007; Borenstein et al. 2009). We used meta-analysis to test for a trade-off between growth (measured as body mass) and immune function.
Materials and methods
Search methods and criteria of selection
Articles were collected by searching ISI Web of Science with the queries ‘selected for immune function’, ‘selected for growth’ and ‘selected for body mass’, and doing cited-reference searches on selected articles. Articles were included if they used a line that had been selected for at least three generations on the focal trait. Commercial lines were excluded because the exact selection criteria are not known and appropriate control lines are lacking. The search was not limited to birds, but the few articles that dealt with non-poultry species did not meet our selection criteria. Articles presenting least-square means only and no raw means were also excluded. Finally, in vitro immune function measures were excluded.
The statistics used in the meta-analysis were group size, group mean, standard deviation and P-value. If the test statistic was not reported and the P-value in the source article was indicated as <0·05, <0·01 or <0·001, we used 0·049, 0·009 and 0·0009 respectively. The meta-analysis was carried out with the program Comprehensive Meta Analysis, version 2 (CMA, v2·2·048; Biostat, Inc. Englewood, Cliffs, NJ, USA). We used selection lines rather than articles as the unit of analysis and treated articles reporting on the same line as different outcomes of the same experiment. Note however that using studies as units of analysis in no case changed the results. Since most of the selected articles were repeat experiments, using selection line prevented the lines on which multiple experiments had been carried out from having undue weight. Further analyses were carried out with gender, type of challenge used or immune parameter measured as moderator. Moderators are variables used to group the results, in order to compare the outcome of different types of experiments (e.g. experiments on males and females), and calculate an overall effect size based on this comparison. Using selection lines rather than articles as the unit of analysis had the added benefit of allowing us to bypass CMA’s limitation to calculations with one moderator. We could now set a different moderator (e.g. gender) while still taking into account the various lines. For calculating the combined effect size we used a random-effects model (fully random-effects model when moderators were used). As effect size measure we used the standard difference in means (Borenstein et al. 2009).
To evaluate possible publication bias we used three tests: Rosenthal’s failsafe N, Orwin’s failsafe N and Duval & Tweedie’s trim and fill. Publication bias is the phenomenon that experiments with negative or inconclusive outcomes are less likely to be published, and less likely to be cited, making them harder to find. Publication bias can significantly skew the results of a meta-analysis. Rosenthal’s failsafe N tests the robustness of the data by calculating how many additional studies with a insignificant outcome are required to reduce the overall effect to zero. Orwin’s failsafe N is similar, except that it calculates the number of studies necessary to reduce the overall effect to non-significance (as defined by the user). A high failsafe N (relative to the number of studies included in the analysis) indicates that the results are robust, because an exceptional number of experiments must have been left unpublished in order to invalidate the results.
Duval & Tweedie’s trim-and-fill test tests for bias by assuming a symmetrical distribution of experimental outcomes around the true effect size. If the distribution of data points is asymmetrical, the algorithm will remove extreme outcomes on one side and place them on the other to normalize the symmetry. The higher the asymmetry, the higher the trim-and-fill number is (note that this does not necessarily mean the corrected overall effect size is insignificant) (Borenstein et al. 2009).
Heterogeneity is the dispersal of effect sizes due to true variation rather than random error. Heterogeneity is expressed with three parameters: Q, τ and I2. Q is the weighed sum of squares (WSS). Testing it against the expected WSS (based on the assumption that all studies share a common effect) yields the chance that dispersal is real rather than random error, and an estimate of true variation. The two other parameters give more information on the true variation. τ is an estimate of the standard deviation of true (as opposed to observed) effects. I2 is the percentage of observed dispersion that is real (Borenstein et al. 2009).
Selection for body mass and immune function
Data from 14 studies on three different poultry lines were available for the meta-analysis of the effect of selection for body mass (Table S1, Supporting information). All studies except one reported that selection for growth decreased resistance/immune function (Fig. 1). The average effect size was −0·80. Although confidence intervals are wide, the summarized effect of each line is highly significant (P < 0·001). The only study that yielded a positive effect was Sacco et al. (1994) and only for the antibody titre against the Newcastle-disease virus (NDV). However, another article reported that NDV mortality was higher in the same selection line despite higher antibody levels (Tsai et al. 1992). Gender had no influence on the effect size: if male and female results were separated the graph looked almost identical, except for the greater confidence intervals for males (Fig. 2).
Regardless of whether genders were combined or not, there was no significant heterogeneity of effect sizes (combined Q = 2·30, P = 0·32; separated Q = 2·44, P = 0·88). Due to the limited number of data points it was not possible to carry out a trim-and-fill test for combined genders. However, Rosenthal’s failsafe N was 33, indicating that 33 selection lines with no effect are required to nullify the observed effect size. Orwin’s failsafe N set for a small effect (ES = 0·2, average ES of missing studies = 0·0) gave nine selection lines as result. It is unlikely that such a large number of unreported selection experiments exist, underlining the robustness of our result. No significant heterogeneity within sexes was found.
We tested whether the selection effect differed between different arms of the immune system (Fig. 3). Three studies investigated the humoral immune response (Martin, McNabb & Siegel 1988; Sacco et al. 1994; Li et al. 2000): they showed a clear and significant effect (ES = −0·65, P = 0·022). However, the majority of articles are concerned with mortality (ES = −0·76, P < 0·001). A single study checked the toe-web response (Bayyari et al. 1997b), a measure of cellular immunity, and reported a large effect (ES = −0·81, P = 0·09; note that this P-value is higher than reported by Bayyari et al. 1997b due to the random-effects model). The effect sizes of the different arms of the immune system did not differ significantly.
Another factor that may have influenced the effect is the antigen with which the animals have been presented. Of the six categories, the NDV (ES = −0·69, P = 0·108), Pasteurella multocida (ES = −0·54, P = 0·062) and toe-web (ES = −0·81, P = 0·082) challenges were not significant, while sheep red blood cells (ES = −1,2, P < 0·001), Erysipelas (ES = −1,2, P = 0·007) and Marek’s disease (ES = −0·80, P = 0·001) were. Again, no significant differences in effect sizes were found.
In conclusion, there was a large (standard difference in means ≈−0·8) and significant (P < 0·001) negative effect of selection for body mass on resistance/immune function.
Selection for immune function and body mass
For the second meta-analysis, we found five different selection lines (for the effect sizes of individual articles, see Table S2). Although the individual studies showed a clear and often significant effect on growth, the outcomes of different lines contradicted and the overall result was not significantly different from zero (ES = −0·23, P = 0·31; Fig. 4). The results from the Hiroshima, Virginia and Wageningen selection lines reported an effect in agreement with the results from the growth selection lines, i.e. reduced growth when increased immune function is selected for (ES = −0·63, −0·49 and −0·67 for all three P < 0·001), but the studies of the French and Israeli lines reported significant effects in the opposite direction (effect sizes: 0·27 and 0·34; P = 0·001 and 0·005 respectively). Separating studies based on gender did not affect these results (Fig. 5). It should be noted that the effect size of the French selection experiment was the combined effect of three selection lines compared with one control line: in the experiment only one of the lines (the line selected for PHA response, which indicates cellular immune responsiveness) differed significantly from the control line (Pinard-van der Laan 2002). Obviously, there was a high heterogeneity in the immune function results. Heterogeneity within lines was not significant for the Wageningen line, but it was significant for the Jerusalem line (Q = 20·64; P = 0·004) and the Virginia line (Q = 42·05, P < 0·001).
Experimental selection for accelerated growth had a large and significant negative effect on immune function. The results of selection for increased immune function are mixed however. No significant overall effect was found, but results within the selection lines are consistent, suggesting that it is possible to select birds for immune function while not negatively affecting growth. We found no consistent difference between the arms of the immune system, either when comparing the effects of selection for growth on different immune system components, or when comparing the effects of selection for growth on different components of the immune system. This is not unexpected because the effectiveness of different arms of the immune system need not be genetically correlated (Pinard-Van Der Laan 2002), so an increased antibody or toe-web response does not necessarily indicate increased overall immune function (it should further be noted that the value of the toe-web test as a measure of cellular immune function has been questioned). As such it is questionable to pool all immune function data into one effect size, as was done in our analysis, and indeed it is reason to question treating immune function as one trait. However, the effects of selection for growth were consistent with respect to the aspect of immune function that was measured, while the variation in growth response to immune function selection was independent of the type of immune response that was selected for.
It is unclear by what mechanism selection for growth compromises immune function. It may be down to simple energy allocation, in that resources spent on growth are no longer available for immune function and other energy-consuming processes (e.g. locomotion, thermoregulation), but possibly the limiting factor lies elsewhere. Selection for production traits in various domesticated animals has resulted in a number of metabolic changes (Rauw et al. 1998). It is also known that MHC haplotypes can affect growth (Warner, Meeker & Rothschild 1987; Lamont 1998), although this did not explain lowered immune function in the turkey line included in our analysis (Nestor et al. 1996). Furthermore, several pro-inflammatory cytokines reduce nutrient allocation to growth and induce anorexia (Klasing & Johnstone 1991; Lochmiller & Deerenberg 2000). Given the consistent effect of selection for growth on immune function these selection lines may be a promising setting to unravel the mechanism underlying the trade-off between growth and immune function.
The effect of selection for immune function on growth differed between selection lines, while there was no significant overall effect. This contrasts with the effects of selection for growth on immune function, which were uniform across selection lines and immune traits. There are several possible explanations for the difference in results between selection for growth or immune function, which are not mutually exclusive. First, the costs of growth are likely to be high in comparison to immune function (Klasing 1998). As a consequence, when growth is selected for, there is likely to be less leeway when compared with selection for a less resource-demanding process such as immune function, and this asymmetry may cause the different selection effect. For example: if we postulate that growth takes 20% of an organism’s energy intake and immune function takes 2%, then increasing the resource allocation to growth by 10% (to 22%) would require all resources allocated to immune function when that were the animal’s only option. Conversely, when increasing the energy allocated to immune function by 10% (to 2·2%), this would have a negligible effect (from 20% to 19·8%) on the resources available to growth. Thus, the potential effect of an increase in immune function on growth is substantially smaller than the potential effect of an increase in growth on immune function, which may explain why selection for growth affected immune function while selection for immune function did not affect growth. Secondly, because selection was for single immune traits rather than for a more comprehensive measure of immune function, it is possible that the selection success was achieved at the expense of other arms of the immune system, or through the selection of a more specific response. In this way it is possible that selecting for an increase in components of immune function did not result in an increase in the resource allocation to all the immune function, and hence no effect on growth. Thirdly, heritability of humoral immune parameters in chicken lines selected for humoral immunity were estimated at c. 0·2 (Wijga et al. 2009), implying that there may be a large environmental effect on immune function. Immune responsiveness is also subject to epigenetic effects, indicating that selection for either enhanced growth or enhanced immune competence does not result in consistent correlations between these traits due to non-inherited (epigenetic) maternal effects (Lemke, Hansen & Lange 2003; Lemke, Coutinho & Lange 2004).
In conclusion, selection for accelerated growth strongly and significantly reduced the response to a variety of immune challenges. This has implications for the agricultural industry, because our findings suggest that breeding animals for accelerated growth may unintentionally have resulted in poor immune function. Diseases present a major problem to the poultry industry (Dziva & Stevens 2008), more so due to the rising prevalence of antibiotic-resistant strains of bacteria. Public concern regarding the healthcare consequences of the large-scale application of antibiotics (Singer & Hofacre 2006; Larson 2007) and arsenic (Lasky et al. 2004) in livestock has prompted proposals to curtail their use. It may therefore be prudent to investigate breeding techniques that improve resistance against diseases. Selection for immune function can both inhibit and promote growth, but on average had no effect, which suggests it may be possible to select for enhanced immune function also in commercial settings without a reduction in the rate of growth.