The return of the trophic chain: fundamental vs realized interactions in a simple arthropod food web

The mathematical theory describing small assemblages of interacting species (community modules or motifs) has proved to be essential in understanding the emergent properties of ecological communities. These models use differential equations to study pairwise interactions between species. However, as community modules become more complex, it is not certain that all potential interactions will be effectively realized. Here, we use community modules to experimentally explore whether the number of trophic links among species scales with community complexity (i.e., by adding species known to feed on each other from pair-wise trials). To investigate this question, we used a simple mite community present in avocado orchards (Persea americana), composed of two predators (Euseius stipulatus and Neoseiulus californicus), one herbivore as shared prey (Oligonychus perseae), and pollen of Carpobrotus edulis as an alternative food resource. These configurations enabled the potential for (intraguild) predation and (apparent) competition to be expressed. Using a series of controls, we assessed whether the presence of one species affected the survival of another, or its conversion of food into offspring). We found that increasing the number of potential interactions did not result in more complex realized community modules. Instead, all communities were reduced to one or two linear trophic chains. Our results show that trophic links assumed to occur when species are assembled in pairs do not necessarily occur when other components of the community are present. Consequently, food-web structure may be unrealistic in theoretical community modules that are parameterized based on pair-wise interactions observed when alternative prey is absent. This further suggests the need for empirical research to work in concert with theoretical approaches to develop more realistic and predictive food-web models.

, that is a small number of species (e.g. three to six) linked in a specified structure of interactions. Among those, intraguild predation (IGP), in which two consumers (the intraguild predator and the intraguild prey, hereafter IG-predator and IG-prey) not only compete for a shared resource but also engage in predator-prey interactions (Polis, Myers, & Holt, 1989), and apparent competition, in which two non-competing prey share a common predator (Holt, 1977(Holt, , 1997, are the most common (Bascompte & Melián, 2005). Whether and how often species engage in intraguild predation or apparent competition strongly affects the long-term persistence of communities (i.e. the "temporal stability in community composition", Pimm, 1984). Theory predicts that intraguild predation destabilizes communities because it reduces the parameter space where coexistence of the IG-predator, IG-prey and shared prey is possible (Holt & Polis, 1997), compared to that of a predator, a prey and a resource in trophic chain models (Oksanen et al., 1981).
Some theoretical studies predict that the inclusion of some factors may reduce this instability (reviewed in Novak, 2013, appendix S1).
Thus, a prevailing outcome of the ecological theory is that domains of persistence of communities with IGP increase as the strength of trophic interactions between predator species decreases. Indeed, weak interactions have long been recognized to stabilize ecosystems by dampening oscillations between consumers and resources, thus promoting community persistence (Gellner & McCann, 2016;McCann, Hastings, & Huxel, 1998).
As communities become more complex, predator species can interact trophically with a higher number of potential prey. Community network building is typically done using the species fundamental trophic niches, which includes all the pairwise trophic interactions that this focal species can establish with others. However, it is known that the number of potential interactions in food webs tends to become much higher than the number of realized interactions (Beckerman, Petchey, & Warren, 2006). The factors affecting the ratio between who can eat whom and who actually eats whom (i.e. connectance) are similar to those leading to reduced IGP. Indeed, a few studies indicate that connectance is linked to structure of the habitats in which communities occur (Beckerman et al., 2006;Tylianakis, Tscharntke, & Lewis, 2007). Others show that connectance is best explained by the intrinsic value of food items. For example, wide differences in resource quality are predicted to decrease connectance (Beckerman et al., 2006). Similarly, flexible foraging behaviour may decrease connectance when food web complexity increases. This was elegantly shown in a study that compared several plant-pollinator interaction networks differing in size (Spiesman & Gratton, 2016). The authors found that niche partitioning was stronger (ergo connectance was weaker) in highly diverse networks (i.e. networks with more plants and, thus, with more interspecific competitors) likely because pollinators adjusted their foraging strategies to minimize interspecific competition. Therefore, it is becoming clear that the fundamental trophic niches of species (i.e. with all their potential interactions; Elton, 1927) may not always be realized (Hutchinson, 1957).
Here, we test whether fundamental and realized trophic niches of species concur, by exploring, in a simple community, how pairwise trophic interactions between species are modified by the inclusion of other species. We focus on predation rate (here, number of individuals consumed per day) as a proxy for trophic interaction strength. Specifically, we mimicked different community modules of increasing complexity using a community composed of two predatory mite species (Euseius stipulatus and Neoseiulus californicus, Acari: Phytoseiidae), one herbivore mite species as their prey (Oligonychus perseae, Acari: Tetranychidae) and pollen as alternative food (González-Fernández et al., 2009), all of which inhabit avocado plants (Persea americana) in south-eastern Spain ( Figure 1A). Previous pairwise experimental studies showed that the interaction between N. californicus and O. perseae is stronger (i.e. predation rates are higher) than that between E. stipulatus and this same prey (González-Fernández et al., 2009). Moreover, pollen is an optimal food source for E. stipulatus but not for N. californicus (González-Fernández et al., 2009). Finally, E. stipulatus and N. californicus engage in size-dependent predator-prey interactions (Abad-Moyano, Urbaneja, & Schausberger, 2010). This knowledge was used to generate predictions on realized trophic links occurring in this system across community modules of increasing complexity ( Figure 1B). These predictions were then tested through a series of experimental treatments to assess a) whether (IG-)predators feed on each prey type; b) whether predation of (IG-)predators on one prey type is affected by the presence of the other; c) whether predation of (IG-)predators on both prey, and of IG-prey on the herbivore, is affected by the presence of alternative food; and d) whether the presence of alternative food affects predation of (IG-)predators on the two types of prey when they are together. With this set of experimental treatments specific planned F I G U R E 1 A: Fundamental community modules included in this study. (a) trophic chain, (b) apparent competition, (c) intraguild predation and (d) intraguild predation and apparent competition. From (a) to (d) the complexity of the community is increased via increasing the number of species and the number of interactions among them. B: Predicted trophic links that have been observed using pairwise experimental settings. C: Realized trophic links occurring across community modules of increasing complexity, obtained from the experiments presented here, where interactions are measured in the presence of other components of the community. SC stands for secondary consumer, PC for primary consumer, PP for primary producer and AF for alternative food. SC1 and SC2 are phytoseiid predatory mites, that is Euseius stipulatus and Neoseiulus californicus, respectively, PC is the tetranychid herbivore mite Oligonychus perseae, AF is pollen of Carpobrotus edulis and PP is the avocado Persea americana. Solid arrows indicate negative direct interactions (who eats whom), whereas dotted and dashed arrows in Figure 1a indicate negative indirect interactions (apparent competition and competition) comparisons allowed unravelling which trophic interactions within each community module were realized, thus providing a relatively simple test of how realized trophic niches can be narrower than fundamental trophic niches when network complexity increases.

| MATERIAL S AND ME THODS
All cultures and experiments were done in a climate chamber at 25 ± 1ºC, 65 ± 5% RH and 16:8h L:D (Light:Dark).

| Community modules
Experimental arenas to test the outcome of community modules have been described in detail in Guzmán et al. (2016). Briefly, a hole (6.5 cm Ø) was cut in a Petri dish (9 cm Ø), turned upside down and then filled with an avocado leaf disc (7.5 cm Ø). The borders were glued to a clay ring. Inside the Petri dish, wet cotton wool ensured enough humidity to keep leaves turgid. Petri dishes were then sealed with parafilm ® . To prevent individuals from escaping, a ring of Tanglefoot ® was applied along the outer margin of the leaf disc.
We performed experiments using two community blocks, according to the identity of the top predator or IG-predator ( Figure 1).
Because IGP interactions are size-dependent, IG-predators and IGprey consisted of adult gravid females (10-14 days old after egg hatching) and heterospecific juveniles (2-3 days old since hatching), respectively. Individuals of known age were obtained from cohorts prepared prior to the start of the experiments. Throughout the text, the identity of (IG)-predator and (IG)-prey will be indicated using the subscripts "ES" for E. stipulatus and "NC" for N. californicus. Predator females were randomly taken from these cohorts, and starved for 16 hr, to standardize hunger levels among individuals, and to ensure that egg production in tested females was not obtained from food ingested prior to the experiment. Arenas containing the herbivore were established as follows: ten females of O. perseae were let to build nests and lay eggs on experimental arenas during 4 days. The number of nests and eggs per nest on each arena was counted at the onset of the experiment. Pollen in arenas assigned to treatments with alternative food was supplied ad libitum, using a fine brush.
Increased complexity was mimicked through the combination of the presence/absence of 4 trophic positions: (IG-)predator, IGprey, herbivore and alternative food. This resulted in the community modules (Sensu Holt, 1997)  Arenas containing either one E. stipulatus or one N. californicus female without herbivores (treatment # 2), and others containing 10 O. perseae females without predators (treatment # 3) were done as controls for predator oviposition rate and prey natural mortality, respectively. 2. Apparent competition: arenas consisted of one female of either E. stipulatus or N. californicus, 10 O. perseae females, and pollen of C. edulis supplied ad libitum (treatment # 4). Similar arenas but without herbivores (treatment # 5) were made as controls for oviposition rates of predators on pollen only, and without the IGpredator (treatment # 6) to assess potential effects of pollen on the survival of the herbivore. 3. Intraguild predation: arenas consisted of 10 O. perseae females, either one E. stipulatus or N. californicus female, as IG-predators, and 10 heterospecific juveniles, as IG-prey (treatment # 7). Control treatments were done to evaluate: the predation/mortality rate of O. perseae in the presence of IG-prey but not of IG-predator (treatment # 8); the mortality rate of IG-prey in the absence of both IG-predator and prey (treatment # 9), and in the presence of IG-predator but not of herbivores (treatment # 10). 4.
Intraguild predation: Apparent competition: arenas consisted of 10 O. perseae females, either one E. stipulatus or N. californicus female, 10 heterospecific juveniles and pollen of C. edulis as alternative food, supplied ad libitum (treatment # 11). Similar arenas to those above but (a) without IG-predators (treatment # 12), (b) without herbivores (treatment # 13), and (c) without IG-predators and herbivores (treatment # 14) were done to evaluate predation of IG-prey on the herbivore in the presence of pollen, predation of IG-predators on IG-prey in the presence of pollen, and mortality of IG-prey in the presence of pollen, respectively.
Twenty-four hours later, the number of dead herbivores/IG-prey (predation/mortality rate) and the number of eggs laid by predators/ IG-predators (oviposition rate) were recorded. Each treatment was replicated 10 to 18 times.

| Data analyses
Statistical analyses were performed using the computer environment R (R Core Team, 2017). Analyses were done separately for each community block. The effects of the presence/absence of each of the trophic groups in the community module, as well as the presence of alternative food on predation rates on herbivores and on IG-prey, and rates of oviposition of IG-predators, were analysed using generalized linear models (GLM) assuming a Poisson distribution and a Log-link function, as no overdispersion of the data was detected. All the analyses assumed 3 full-factorial designs and followed a backward elimination procedure: when the interaction among the three explanatory variables was not significant and the model had higher or not a substantially smaller AIC (i.e. by at least two units) than that without the interaction, the latter was removed from the model. Subsequently, the same procedure was followed for second-order interactions, keeping as final model that with only significant interactions or no interactions at all (additive model). Using the significant terms of the above general models, we performed a series of planned comparisons using the "contrast" R package, to detect the presence or absence of specific trophic links based on the patterns of mortality in the herbivore and the IGP-prey and on the oviposition rates of the IG-predator. When specific sets of data were used in multiple comparisons, their significance was corrected using the sequential Bonferroni method correction (Holm, 1979;Rice, 1989). Alpha levels after Bonferroni correction are indicated in the text as α Bonf .
Mortality of O. perseae females was analysed using data from treatments containing this species. The 3 main factors in the model were presence/absence of IG-predators, IG-prey and alternative food.
IG-prey mortality was analysed using data from treatments containing IG-prey (i.e. predator juveniles). The 3 main factors in the model were presence/absence of IG-predators, herbivores and alternative food.
Oviposition rates were analysed using data from treatments containing IG-predators (i.e. adult predators). The 3 main factors in the model were presence/absence of herbivores, IG prey and alternative food.
F I G U R E 2 Mortality rates (average ± SE) of (a) herbivore prey (Oligonychus perseae females) and (b) IG-prey (Neoseiulus californicus juveniles), and (c) oviposition rates (average ± SE) of IG-predators (Euseius stipulatus females), in 14 different treatments defined by presence or absence of either IG-predators, IG-prey, herbivores or alternative food (pollen), depicted in the lower part of the figure, that mimicked four different community configurations and their respective controls

| Community block with E. stipulatus as the (IG-) predator
Mortality rates of the herbivore were significantly affected by the interaction between the presence of IG-predator ES and IG-prey NC and between the presence of IG-prey NC and pollen (Table 1a).
Indeed, more prey died in arenas with both the IG-prey NC and the IG-predator ES than with the IG-predator ES alone (Figure 2a, compare bar 1 to 7), but not than with the IG-prey NC alone (Figure 2a, compare bar 8 to bar 7). Also, the presence of pollen led to reduced herbivore mortality rates, but only in the absence of IG-prey NC (Figure 2a, compare bars 4 and 6 to bars 11 and 12), suggesting that IG-prey NC were not feeding on pollen in the presence of herbivores and that the IG-predator ES stopped feeding on herbivores when pollen was present.
Mortality rates of the IG-prey NC were affected by all the double interactions except that between the herbivore and pollen (Table 1b) F I G U R E 3 Mortality rates (average ± SE) of (a) herbivore prey (Oligonychus perseae females) and (b) IG-prey (Euseius stipulatus juveniles), and (c) oviposition rates (average ± SE) of IG-predators (Neoseiulus californicus females), in 14 different treatments defined by presence or absence of either IG-predators, IG-prey, herbivores or alternative food (pollen), depicted in the lower part of the figure, that mimicked four different community configurations and their respective controls  Figure 2b, compare bar 7 to 11), suggesting that IG-prey NC were feeding on herbivores while the IG-predator ES fed mostly on pollen.
Oviposition rates of IG-predators ES were only affected by the presence of pollen (main factor pollen, Table 1c) and indeed treatments with pollen resulted in much higher oviposition than those without pollen (compare bars 4, 5, 11 and 13 to bars 1, 2, 7 and 10).

| Community block with N. californicus as the (IG-)predator
Herbivore mortality was affected only by the interaction between IGpredator NC and IG-prey ES (Table 2a). Indeed, mortality of herbivores was drastically affected by the presence of IG-predators NC (Figure 3a, compare bar 1 to 3), but this effect was lower in the additional presence of IG-prey ES (Figure 3a, compare bar 1 to 7). Mortality of IGprey ES was only affected by the presence of pollen (Table 2b). Oviposition rates of IG-predators NC were affected by the main factor herbivore and the interaction between the IG-prey ES and pollen (Table 2c) Figures   2a and 3a confirmed that trophic links between both species TA B L E 1 Results of generalized linear models applied to (a) herbivore mortality rates, (b) IG-prey (juveniles of Neoseiulus californicus) mortality rates and (c)  (1)*(2)*(3) NS* Notes: All the analyses were 3 full-factorial designs. When interactions among the three explanatory variables were not significant, and if the new model yielded a lower AIC, they were removed from the model. Subsequently, the same procedure was followed for double interactions. These cases are shown in the table as NS*.

Trophic chain: comparisons between bars 1 and 3 in
of predator mite and the herbivore were realized (Figure 1 3. Intraguild predation: when intraguild prey was added to the trophic chain community, E. stipulatus preyed on the IG-prey (compare bars 7 and 8, Figure 2b), but not on the herbivore: bars 7 and 8 in Figure 2a indicate that mortality of the herbivore was inflicted by the IG-prey NC , which is supported by comparing oviposition rates of the IG-predator ES with and without IG-prey (bars 1 and 7, Figure 2c). This resulted in the realized food web configuration depicted in Figure 1, c.2.1. In the presence of the IG-prey ES , N. californicus ceased foraging on the herbivore (compare bars 1 and 7, Figure 3a), which translated into no predator fecundity (compare bars 1 and 7, Figure 3c). Instead, herbivore mortality was inflicted by the IG-prey ES (compare bars 7 and 8, Figure 3a).
This resulted in the realized food web configuration depicted in 4. Intraguild predation and apparent competition: when pollen was added to the IGP community module with E. stipulatus as the IG-predator, herbivore mortality was mainly inflicted by the IG-prey NC (compare bars 11 and 12, Figure 2a). IG-predators ES ceased attacking the IG-prey NC (compare bars 7 and 11, Figure   2b) and foraged exclusively on pollen, its optimal food (compare oviposition rates, bars 7 and 11, and bars 5 and 11, Figure 2c). This resulted in the realized food web configuration depicted in Figure Figure 3a (predation rates) and in Figure 3c (oviposition rates)], and on the survival of IG-prey ES increasing in the presence of its optimal food (compare bars 7 and 11 in Figure 3b). This resulted in the realized food web configuration depicted in Figure 1, d.2.2.

| D ISCUSS I ON
In this study, we tested the effect of community structure on the realized interactions within a community of predatory and herbivorous mites. We show that adding species to a community increases the number of potential trophic interactions, but not necessarily their occurrence. Indeed, despite the potential for module configurations of communities with apparent competition and intraguild predation, all modules could be described by linear food chains in our system ( Figure 1c).

| Basic properties of the experimental system and implications for population dynamics
All the community modules considered in this study naturally occur in the avocado orchards of south-eastern Spain. Field samplings done on avocado trees during four consecutive years revealed that the population dynamics of phytoseiids typically has two maxima, one in spring and the other in summer. In spring, the phytoseiid population growth is strongly linked to the dynamics of pollen con- This suggests that E. stipulatus is the most efficient predator converting food into eggs, but that N. californicus is more efficient at reducing herbivore populations. Moreover, unlike N. californicus, E. stipulatus fed and oviposited on pollen. This allows the latter to remain in the field when animal prey is scarce, as observed in field surveys in springtime (González-Fernández et al., 2009).
Our results also revealed asymmetry in intraguild predation between E. stipulatus and N. californicus, with adults of the former preying upon juveniles of the latter, but not the reverse. Because N. californicus is likely the best competitor for the shared prey, coexistence between predators is thus possible in this system (Holt & Polis, 1997). Yet, the simultaneous presence of the two predators is likely to have little effect upon the densities of the shared prey.
Indeed, whereas adding N. californicus adults to an arena with E. stipulatus juveniles results in higher shared prey densities as compared to the presence of N. californicus adults alone with the shared prey, the reverse is not true when adding adult E. stipulatus to an arena with juveniles N. californicus. Thus, the net effect of these interactions upon prey density is probably negligible. This is corroborated by field studies showing that natural population control of the persea mite when the two species of predators are present is not successful (Montserrat et al., 2013). However, the presence of alternative food (i.e. pollen) contributed to reduce trophic interactions between predator species resulting in community configurations that could enhance pest control. Thus, supplying alternative and preferred food to the IG-predator is probably detrimental to populations of O. perseae. Again, this finding is in line with field observations (Montserrat et al., 2013). In this work, the authors spread commercial bee pollen dissolved on water onto the avocado trees, resulting on a better control of O. perseae populations.
Optimal foraging theory predicts that species engage in trophic interactions on more than one food source when these are available (Pulliam, 1974). Here, we show that E. stipulatus acting as intraguild predators feeds on the herbivore, O. perseae, on the intraguild prey, N. californicus, and on the alternative food, pollen, when each of these are presented alone. However, in the presence of pollen, E. stipulatus stops feeding on both prey species. This may be explained by the fact that pollen is the most profitable food for this species (Ferragut, Garcia-Mari, Costa-Comelles, & Laborda, 1987).
Similarly, N. californicus adults and juveniles ceased foraging on other food sources in presence of the herbivore. These results suggest that realized interactions hinge on the presence of the most profitable food source. Indeed, in the most complex community studied here, with all 5 species present, the presence of the optimal food source for each predator species originated the split of the community into two trophic chains, one with E. stipulatus feeding on pollen and the other with N. californicus feeding on the herbivore (Figure 1 d).
Another factor that contributed to the linearization of the food web was that, when both the IG-prey and the shared prey were together,  (González-Fernández et al., 2009). Therefore, the realized community was that of a 4-level trophic chain (Figure 1

| The return of the trophic chain: fundamental versus realized trophic interactions
By combining data of mortality and oviposition at different community structures, we could determine who eats whom in a simple food web. Although this approach is powerful, it does have its limitations.
Indeed, it assumes additive effects of conversion efficiencies of pairwise interactions. For example, if feeding on a prey item allows predators to better convert the food provided by another prey, this cannot be detected in our approach. Furthermore, it may be largely unfeasible to extend this approach to more complex food webs.
Indeed, these full-factorial studies are extremely rare in the literature (but see Otto, Berlow, Rank, Smiley, & Brose, 2008;Schmitz & Sokol-Hessner, 2002). Still, it is becoming clear that we need to know how food is transformed into predator offspring in order to fully understand food webs in nature (Neutel & Thorne, 2014).
Connectance is a fundamental measure of food web complexity that describes the proportion of realized interactions amongst all possible ones (May, 1972). Connectance is generally much lower than the number of potential interactions (Beckerman et al., 2006).
Identifying trophic links in food webs, however, is not a simple task.
Molecular methods are useful to process field data and they deliver reliable information on who eats whom, but such tools currently only provide semi-quantitative estimates of predation, and they are expensive (Birkhofer et al., 2017 cies. Yet, one-on-one approaches may ignore emergent indirect effects of having several species together (Wootton, 1994). For instance, Cancer productus, a crab native to the Northwest Pacific, consumes equal amounts of native oysters and of invasive drill oysters when each type of prey is offered alone, but when these prey are offered together, crabs interact with the native oyster species only (Grason & Miner, 2012). Therefore, if trophic links are not evaluated in presence of all species in the community, one may overestimate connectance in food webs. Here, we show that all communities ended up becoming a sum of one or more trophic chains ( Figure 1c). Thus, the fundamental trophic niche of species in this system (i.e. the food items that species are potentially able to feed on) is larger than the realized trophic niche [i.e. the food items that species actually feed on when present in combinations exceeding the individual pairwise interactions (Hutchinson, 1961)].
This indicates that indirect interactions, such as IGP and apparent competition, may be weak or absent. Therefore, our results suggest that some food webs may be less complex than previously thought.
Theoretical models exploring persistence in communities with IGP find a limited parameter space for three-species coexistence (e.g. Mylius et al., 2001), but field observations show that IGP is actually widespread (Polis, 1991). Our results suggest that IGP in some systems might actually be occasional, as predators will tend to forage on the most profitable food, which is generally not the IG prey (Polis et al., 1989). In line with this, some natural systems have shown that communities with IGP show dynamics that are compatible with linear food chains, rather than with IGP (Borer, Briggs, Murdoch, & Swarbrick, 2003). Therefore, predators may coexist because they rarely engage in IGP, and complexity may be over-estimated (Magalhães et al., 2005).
Alternatively, species persistence may be achieved because predators consume their preferred food when the latter is available, but may switch to less-preferred items when their preferred food is depleted (Wei, 2019). This could well be the case in our system. Both these alternatives are compatible with food web theory stating that weak trophic interactions promote the persistence of communities (Gellner & McCann, 2016;McCann et al., 1998, among others). Our results suggest that increasing the number of potentially interacting species results in most species interactions becoming weaker. Indeed, the structure of interactions among species in natural communities is characterized by many weak and few strong interactions (McCann et al., 1998;Paine, 1992), and such skewedness towards weak interactions is crucial to food web persistence (Neutel, Heesterbeek, & Ruiter, 2002;Neutel et al., 2007).
Furthermore, trophic interaction strengths are unlikely to be constant over time. For example, seasonal changes in species composition cause temporal variation in the strength of interactions (Carnicer et al., 2009;Gabaldón et al., 2019;Wang et al., 2019), as well as changes in food web topology and structure (McLaughlin, Jonsson, & Emmerson, 2010). In agroecosystems, temporal variability in species interaction strength is detectable even within crop seasons, as recently shown by Roubinet et al. (2018) in a barley field. Therefore, because a species' fundamental trophic niche (all of its potential interactions) is unlikely to be realized at a particular place or time, it is crucial to determine the resources which species in a community actually feed upon, and under what circumstances. Thus, unravelling realized food webs (i.e. interaction strengths across different nodes and trophic levels, including indirect effects) may be key to understanding these ecological networks and their persistence.

ACK N OWLED G EM ENTS
This preprint has been reviewed and recommended by Peer Community Ecology In (https ://dx.doi.org/10.24072/ pci.ecolo gy.100008). The authors deeply thank Francis J. Burdon and two anonymous reviewers of PCI Ecology for improving the manuscript with their contributions. We are also indebted to two anonymous reviewers and the editors of Functional Ecology for their crucial contributions, which significantly improved the quality and clarity of this manuscript. We deeply thank Rosa María Sahún Logroño for her valuable help in maintaining the experimental popula-