Behavioural Response in an Asilid Fly: Inuence of Ecological and Environmental Factors on Spatial Density Dependence

Behavioural response of a parasitoid shows the effect on host parasitism patterns at a given host distribution resulting in an increase or decrease of parasitism intensity according to local host densities. This relationship could be proportional, positive, or negative, as a consequence of foraging of parasitoids searching for hosts. Mallophora rucauda is a y parasitoid of Cyclocephala signaticollis scarab beetle larvae and a predator of honeybees. Females search and place egg-clusters overground in open grasslands near beehives. Larvae actively searching for host underground following chemical cues arising from the host itself. The parasitism pattern is a result of this complex host-searching strategy which is shared between both stages of the y. In this work we carried out a study at four spatial scales in apiaries located in the Pampas region of Argentina. We found that parasitism is inverse density-dependent at high female activity and direct density-dependent at low female activity at the larger spatial scale. We found a direct density dependent pattern associated to substrate height at intermediate spatial scale that is lost when the habitat has abundant oviposition substrates. Conversely, parasitism is inversely density-dependent at both smaller spatial scales, associated to oviposition substrate distance and saturation of healthy host by larvae attacking. Additionally, M. rucauda does not select the oviposition substrates according to the abundance of Cyclocephala signaticollis inhabiting underground. This work shows the importance of a proper scale for analysis of factors that inuence population dynamics and how environmental characteristics mould parasitism patterns in this dipteran parasitoid.


Introduction
The term behavioural response of a parasitoid has been de ned by Hassell (Hassell 1966) as a behaviour response to the distribution of hosts that, being distributed unevenly, results in an increase or decrease of parasitism intensity according to local host densities. Hence, studying different dynamics of hostparasitoid systems has been subject of great interest (Walde and Murdoch 1988, Gunton and Pöyry 2016). In particular, mechanisms leading to stability are useful not only from a theoretical view but also in applied science given the importance of parasitoids as biological control agents (Waage andHassell 1982, Bernstein 1987, Fernández-arhex and Corley 2003, Jervis 2005. Spatial density dependence has been rendered as an important factor leading to stability as well as host speci city through its in uence on the functional response of parasitoids and on density dependence (Bernstein 1987, Hassell 2000. Spatial density dependence in parasitism is the outcome of parasitoids responding to differences in host density among patches leading to changes in the intensity in parasitism (Walde and Murdoch 1988).
Past work has shown that spatial density-dependent parasitism plays a role on population persistence and stability of the host-parasitoid systems (May et al. 1981, Murdoch et al. 1984, 1985, 2005 Hassell and May 1988, Murdoch and Stewart-Oaten 1989, Godfray and Pacala 1992, Briggs 1996, Teder et al. 2000). When parasitoids aggregate as a response to high host densities, with a consequent increase in parasitism percentage then direct density dependence occurs. On the contrary, if parasitism percentage decreases with increasing host abundance, inverse density dependence occurs.
Many studies have dealt with establishing the relation between parasitism and host abundance and examples of direct, inverse or independent density dependence have been found (Stiling 1987, Walde and Murdoch 1988, Gunton and Pöyry 2016. Nonetheless an important remark should be made. In a leading paper, Heads and Lawton (1983) noted the importance of spatial scale in this process. They showed that expected patterns of prey mortality imposed by a population of natural enemies aggregating in response to victim densities can vary from exponential curves when the sample area is smaller than a "patch" through transient intermediate relationships to an increasing relationship when the "patch" size is recognised by the natural enemy (Heads and Lawton 1983).
In a more recent update, Gunton and Pöyry (2016) introduced the "scale-speci c foraging hypothesis" implying that the nature of observed correlations between local host densities and parasitism rates is the result of methodological artefacts of the observational scales used. They propose that parasitoids discriminate among host patches according to their density at a "foraging grain size" that normally creates a positive density-parasitism relationship. But, as they point out, this relationship will be detected as long as the size of the study units within which densities are calculated is comparable to the foraging grain size.
Other important factors that have been studied that in uence spatial density dependence can be summarized in host characteristics. For instance, host distribution can in uence profoundly parasitoid's response size where optimal strategies can vary for highly aggregated or randomly distributed hosts (Walde and Murdoch 1988). Within hosts characteristics, exotic hosts and large bodied parasitoids seem to be associated to negative parasitism responses (Gunton and Pöyry 2016). Although there are many theoretical approaches and experimental studies trying to shed light on the effects and mechanisms of spatial density dependence on population dynamics, there are still needs for multi-scale studies. Even more if we consider that most of the studies (~ 85%) refer to hymenopteran parasitoids that differ greatly in the searching and oviposition strategies with other parasitoids (e.g. dipterans) (Godfray 1994, Feener Jr and Brown 1997, Gunton and Pöyry 2016.
In this work we analyse spatial density dependence parasitism in a system of a dipteran parasitoid and its coleopteran host. Mallophora ru cauda Wiedemann (Diptera: Asilidae) is a pestiferous robber y common in the open grasslands of the Pampas region of Argentina. Adults are predators of insects and larvae are solitary koinobiont ectoparasitoids of the second and third instar of scarab beetle larvae Cyclocephala signaticollis Burmeister (Coleoptera, Scarabaeidae) (Castelo and Capurro 2000, Castelo and Corley 2010, Crespo and Castelo 2010. Mated M. ru cauda females deposit eggs in clusters that are placed away from the host on elevated places, typically tall grasses, or arti cial supports higher than 1.25 m, in areas close to bee hives Corley 2004, 2010). Emerging robber y larvae are wind dispersed, drop to the soil from the oviposition site, and bury themselves searching for hosts (Castelo and Lazzari 2004, Castelo et al. 2006, Crespo and Castelo 2008. It has been established that the selection of oviposition height by the M. ru cauda female contributes to larval dispersal and, as a result, the parasitism success is maximal when egg-clusters are placed on substrates between 1.25 and 1.50 m in height (Castelo et al. 2006 Regarding host location, this parasitoid has a split strategy in which both female and larvae are involved. Firstly, female places its eggs on tall substrate as mentioned before. After being wind dispersed, rst larval instar moults to the next instar and then active host searching occurs. In order to nd its host, M. ru cauda larva orientates to its host through detection of cues produced in the host's posterior intestine (Castelo and Lazzari 2004, Crespo and Castelo 2008, Groba and Castelo 2012 Martínez et al. 2017). Given the split strategy in this system involved in host location, second instar larva can be claimed to be the ecological equivalents of female in hymenopteran parasitoids. Hence, it is justi ed to ask if spatial density dependence of parasitism exists in this system.
Previous studies on this host-parasitoid system have shown contradictory results when analysing spatial density dependence at a large spatial scale, but both were consistent nding an inverse density dependence pattern at the smaller spatial scale (Castelo and Capurro 2000, Castelo and Corley 2010). One of them found evidence of direct density dependence at a spatial scale compatible to adult patches (Castelo and Capurro 2000) while the other study, which included increased sample sites, found no relation at the larger spatial scale (Castelo and Corley 2010). Both studies were performed including the potential host species that make up the community of rizophagous Scarabaeidae larvae in the study area, introducing a potential confounding factor because M. ru cauda has a marked preference for C. signaticollis larvae (Castelo and Crespo 2012). This work is motivated in the clari cation of the results found in previous studies with the addition rstly of more sample sites and secondly much more information on host speci city in this species and other ecological features regarding oviposition height and distance to the hosts, substrate availability, and activity of females. So, the goal of this work is to show the results of our studies on the spatial density dependence of parasitism by M. ru cauda on its preferred host, C. signaticollis at different spatial scales. We also show how the inclusion of information from habitat use by females and host use by larvae combine to properly determine the spatial density patterns at different spatial scales.

Field sampling methods
Field studies of host abundance and parasitism were carried out in six geographical localities of the Pampas region of Argentina: Luján, Mercedes, Moreno, Pigüé, Pilar and Victoria. Sampling was done during June to August between 1997 and 2006. These localities are within the major beekeeping region of Argentina, where adult robber ies feed mainly on honeybees ( Figure 1).
The study was carried out in 17 elds with apiaries where robber y activity was registered in the previous summer.
Some elds were sampled repeatedly in different years, so the combination apiary/year was de ned as the scale "site" (see Table 1). In each site, three plots with different agricultural or cattle breeding management practices and vegetation (from now on "sub-site" scale) were sampled. On each sub-site a grid was placed next to the wire fence. Grids were made up of 10 lines of 5 samples parallel to the wire fence ("Line level"). Samples were taken every 2.5 m within each line. Lines were placed every 5 m covering a total of 50 m. Each sample (small scale) consisted of the extraction of a soil block of 0.35 m side by 0.30 m depth obtained with a shovel (volume: 36.8 L; surface area: 0.12 m 2 ). In sum, samples were grouped in lines (5 samples per line) with 10 lines per sub-site obtaining 50 samples in total. The largest scale, "site", consisted of 3 sub-sites, hence 150 samples ( Figure 2).
From each block of soil, all scarab beetle larvae were collected by manually breaking the soil and identi ed to the species level in the laboratory with a dichotomic key (Alvarado 1980)</C>. A stereomicroscope was used to register the number of larvae of M. ru cauda attached to the host cuticle. Only C. signaticollis larvae were counted as host abundance since it is the preferred host for M. ru cauda (Crespo and Castelo 2010, Barrantes and Castelo 2014).
Scales were chosen because we believe they represent the true complexity in this system. As mentioned before, M. ru cauda adults belong to a genus of robust dipterans that feed on other ying insects like honey bees (Bromley 1930, 1946, Cole and Pritcharkd 1964, Corley and Rabinovich 1997. Asilids have an important ying capacity, near 1-2km, easily covering a site area (Kanmiya 2002, Londt 2020. Once the female places its eggs on the substrate, larvae will be wind dispersed, so the sub-site together with the line levels capture mainly the in uence of wind on larval dispersal and a possible effect of distance to the oviposition site but no further in uence of females. Finally, the sample scale captures host-searching performed by the larva itself after dropping and burying into the soil.

Density dependence analysis
We considered in the analysis only sites where parasitized scarab beetle larvae were found (n = 25). We carried out sampling in different years because parasitoids move freely and frequently among localities as a consequence of the host population dynamics and food availability. Cyclocephala signaticollis larvae abundance might be very variable among years due to different causes (crop management, eld conditions, parasitism outcome itself) introducing variability in the presence of M. ru cauda and parasitism levels at a given site. Due to this scenario, it was necessary to rede ne sampling places every year.
For each of the scales analysed, (i) site (apiaries); (ii) sub-sites ( eld lots); (iii) lines (transects within eld lots) and (iv) samples (unitary block of soil), proportion of parasitized hosts was calculated as the ratio between the number of parasitized C. signaticollis and the total number of C. signaticollis found. To avoid overestimating the proportion of hosts bearing no parasitoids, those sites with 0% parasitism were excluded from analysis, assuming that parasitoid larvae may have not arrived to the soil in these places or adult parasitoids did not oviposit in those speci c places the previous summer.
We analysed the proportion of parasitized C. signaticollis through generalized linear models. For each scale we generated a model that included different predictors informative of that scale (site, sub-site, line and sample models). All models were generated with a Binomial distribution and a logit link function.
After modelling a full model, model selection was performed. For every model, the effect of dropping an interaction or a predictor variable (with drop1 function) was evaluated through the Likelihood ratio test and the AIC value. After obtaining the minimal model, signi cant terms were evaluated with the anova function. Finally, interaction plots of the estimated marginal means were done to explore the relation between the predictors and the response variable. For the site model host abundance and amount of egg-clusters in place were used as predictor variables.
Host abundance was included as a discrete predictor variable while amount of egg-clusters was included as a categorical predictor variable with two levels (low and high). Egg-clusters was included as a predictor variable since the abundance of M. ru cauda cannot be directly calculated.
The sub-site full model was constructed with host abundance as discrete predictor variable and vegetation height as a categorical predictor variable (low or high). Vegetation height is an indicator of oviposition substrate availability for M. ru cauda (Castelo and Corley 2004). In elds with low vegetation height, only wire fences are available for oviposition while many other oviposition substrates (e.g. tall grasses, stems, sticks) are also available in elds with high vegetation. If only wire fences are available for oviposition, egg-cluster aggregation could occur as a result of availability of oviposition substrates. However, egg-cluster aggregation could still occur in vegetation if females are attracted to odours from damaged plants, hosts or other egg-clusters favouring oviposition in speci c substrates. Attraction to damaged plants has already been discarded since oviposition on dry plants and wire fences are frequent (Castelo et al. 2006). In order to discard female attraction to host odours, we studied if female M. ru cauda places more egg-clusters on plants and wire fences associated with hosts at a small scale. For this, we registered the position of between 28 and 35 plants or wire fences with egg-clusters, in six apiaries with proven presence of M. ru cauda in the previous summer. From each plant and wire fence, the total number of egg-clusters placed by M. ru cauda females was registered. Wire fences were 2m portions of longitudinal wire randomly chosen. Plants geographic positions were registered because the following step of the study was performed during autumn and many plants were gone by then. Hence, in autumn, soil samples were obtained using the same technique as previously described. Soil samples with previous plant positions were taken using the geographic position as the centre. For samples from wire fences, soil samples were obtained from the midpoint of the wire longitude. Soil samples were analysed to quantify the number of larvae of C. signaticollis present. With data from number of egg-clusters and number of hosts, a model was constructed with the former as the response variable and the latter as a discrete numerical explanatory variable. The in uence of the number of C. signaticollis hosts on the number of egg-clusters was evaluated with a glm with a Gamma distribution and log link function with the ID of each site as a random factor. Next, line model included the same predictor variables as sub-site model (host abundance and vegetation height) and distance to the wire fence as another discrete predictor variable. This variable accounted for any distance effect that could be introduced in elds with low vegetation height. Finally, sample model included the same predictor variables as the line model (host abundance, vegetation height and distance to the wire fence). We assumed that dependency on sites and sub-sites would not introduce a notorious effect on the results since females are not able to place two egg-bouts on a single day (M. Castelo, personal observation). Given the fact that at maximum only 3 sub-sites per site could be included as random effects, they were not included as random variables because at least ve replicates is suggested to estimate variance.
All the statistical analyses were performed using R version 3.6.3 "Holding the Windsock" (R Core Team 2020). Models were performed with the function glmmTMB from the glmmTMB package (

Density dependence analysis
The four models showed information that can be separated in three (site model, subsite-model and linesample models together). For every model we found support for either direct or inverse density dependence.
For the site model we found that the relationship between the proportion of parasitism and the number of hosts depended on the estimated abundance of M. ru cauda through the abundance of egg-clusters found. In sites where estimated abundance of M. ru cauda was high, we found inverse density dependence. On the contrary, in sites where the estimated abundance of M. ru cauda was low, a direct density dependence relation was found (Fig. 3A, Table 2).
For the sub-site model, we found that the relationship between the proportion of parasitism and the number of C. signaticollis hosts depended on the vegetation height. In sub-sites where vegetation height was high, we found density independence. On the contrary, in sites where the vegetation height was low (only wire fences as oviposition substrates), a direct density dependence relation was found (Fig. 3B, Table 3). We also found no relation between the abundance of C. signaticollis hosts and the number of egg-clusters in substrates indicating that oviposition is not related to plants and wire fences associated with hosts at a small scale (Chisq 3 = 1.936, P = 0.586).
For the line and sample models, we found an inverse density dependence between the proportion of parasitism and the number of hosts (Figs. 3C and 3D, Tables 4 and 5). In both models no in uence of distance to the wire fence or vegetation height was found.

Discussion
In this work we show the results of our studies on spatial density dependent parasitism by the robber y M. ru cauda on C. signaticollis scarab beetle larvae at different spatial scales. We found two different scenarios, i.e. at large spatial scales either direct density dependence or inverse density dependence was found. At smaller spatial scales only inverse density dependence was found. The patterns we found in this system might be related to the two-step stages involved in host searching. In this kind of parasitoids (dipteran, coleopteran, lepidopteran, neuropteran, strepsipteran and trichopteran; Mills 2009) where the female cannot access directly to the concealed host, females place their eggs close to hosts and it is the larva that must locate the host (Godfray 1994, Feener Jr andBrown 1997). At large spatial scales, generalist adult parasitoids can produce a density independent pattern of parasitism when parasitism rates vary among sites as a consequence of variable local abiotic factors. As host density can be variable, foraging parasitoids may not be able to distinguish areas with high host density. However, when the sample area is bigger than the effective patch, then density dependence patterns might be lost (Heads and Lawton 1983). In M. ru cauda we observed inverse density dependent parasitism when the abundance of adults was high and direct density dependence pattern was observed when the adult abundance was low. This result lends support to both previous studies that found contradictory results Capurro 2000, Castelo andCorley 2010). Although previous studies found support for opposed density dependent patterns, both studies included as potential hosts, larvae of other species than C. signaticollis because it was thought that M. ru cauda could develop on several Scarabaeidae host species. However, it has been already shown that M. ru cauda can only develop to imago on C. signaticollis as its host albeit positive orientation toward other Cyclocephala species occur (Barrantes and Castelo 2014). Also, previous studies did not include information regarding vegetation or wire fences height, distance to the wire fences or adult activity. This information, as shown through our analysis, proved relevant for M. ru cauda given its split host location strategy.
Another important difference that might in uence the results found at the site scale could be related to the size of the site area that included elds with many different characteristics. The Pampas region where M. ru cauda is typically found has been highly modi ed for livestock and agricultural purposes. Site scale therefore includes elds with crops and livestock culture that introduces high variability. The abundance of hosts in elds with livestock culture or soybean crops is much lower than abundance in cereal or potato crops and soy or sun ower cultivars. Another possible explanation for the difference found with previous studies at site scale might be that Asilidae adults are territorial (Onsager and Rees A particular type of competition occurs in this species since oviposition often occurs on substrates where a previous eggcluster had been placed by another female (M. Castelo, personal observation). This particular behaviour could introduce competition where high amount of egg-clusters produce an inverse density pattern since not all larvae from egg-clusters will be able to reach a host or several larvae could reach the same host incurring in ine cient superparasitism.
The results so far could indicate that female M. ru cauda could lay eggs in environments with high host density and may have some skill to qualify environments according to host abundance supporting conclusions drawn in previous studies (Castelo and Capurro 2000). However, Castelo and Corley's (2010) study included a larger dataset and found a density-independent parasitism at the site scale. This highlights on the importance of adequate sampling (replication and spatial scale) in detecting true density-dependence.
We then analysed the sub-site scale that matches the surface of elds with an homogenous type of land use. At this spatial scale, we found both direct and independent density patterns depending on the vegetation height of the eld was low and high respectively. This apparent difference in the results could be explained by the oviposition behaviour that M. ru cauda has Corley 2004, Castelo et al. 2006). Females select the higher available substrates maximizing larval dispersal and minimizing superparasitism (Castelo et al. 2006). In elds with low vegetation height the only suitable oviposition substrates were the wire fences (~ 1 m), while on elds with high vegetation heights many other oviposition substrates were available. Given the sampling grids that were settled starting at the wire fence, elds with lower vegetation might be revealing the true process in this system. Since larva are dispersed by the wind, the distance travelled by the larva will depend on environmental conditions like wind velocity and height of the oviposition site. The parasitism pattern found at this scale will have contributions of the females that decided where to place the egg-clusters and will be the result of larval dispersion. Hence, this spatial scale could be truly representing the patch size recognized by M. ru cauda showing a direct density dependence. On the contrary, in elds with high vegetation, density independence was found. This result should not be surprising since the availability of many oviposition sites could generate a confounding effect tarnishing the pattern.
At smaller spatial scales we found the same result as in previous studies, i.e. inverse density dependence. These results may be explained by the fact that larvae of M. ru cauda are wind dispersed and cannot actively select landing place. Once larvae land to the soil they bury themselves and, after moulting, start searching for their hosts orienting to host chemical cues Lazzari 2004, Crespo andCastelo 2008). This type of anemophilic dispersal imposes dispersion limits into this system generating a top limit in the amount of parasitoid larvae that can land in any speci c spot. Given also that C. signaticollis hosts do not distribute randomly but tend to aggregate in space and are also attracted by the same chemical cues used by M. ru cauda, the resulting spatial pattern is inversely dependent. The spatial distribution of small particles wind dispersed in a heterogeneous habitat has been successfully used in modelling the spatial distribution of individuals from different species in forests (Shen et al. 2009). In this work, the authors showed that both dispersal limitation and habitat heterogeneity were important factors in determining spatial distribution of propagules. Also, soil factors could affect the risk of parasitism because of differences in host accessibility, as the depth to which hosts bury underground in a patch (Okuyama 2019). The idea that larval dispersion in M. ru cauda can be modelled as a seed dispersion phenomenon has already been explored with results that explained the oviposition strategy in this species (Castelo et al. 2006). Hence, an inverse density dependence at small spatial scales could be indicating that dispersal limitation in an heterogeneous habitat is the driving processes.
The dispersal of larvae is highly dependent on the female oviposition site selection. Hence, it is important to understand if females place more egg-clusters in substrates associated with more hosts underground that could be releasing attractive chemical cues. However, we found no relation between the amount of egg-clusters in a speci c plant and the abundance of C. signaticollis at the plant scale. This result reinforces the idea that M. ru cauda does not use host related cues associated to plants or wire fences. In fact, M. ru cauda females would be using substrates mainly based on its height which should minimize superparasitism and increase singly parasitism as it has been already proposed (Castelo et al. 2006). However, we have frequently registered egg-clusters clumped on the same plant in the eld (M. Castelo, personal observation). It is unknown if egg-cluster aggregation poses some adaptive advantage for females during oviposition site seeking or if female use some other characteristic of substrates indicative of good habitat quality. These characteristics could be other relevant factors related to their life cycle like host densities at different spatial scales, prey abundance or even human related activities like the agricultural management.
An important conclusion of our work could indicate that M. ru cauda actually could be considered of importance to control C. signaticollis population opposed to suggestions previously made considering several species of Scarabaeidae larvae (Castelo and Corley 2010). These results reinstates the discussion that actions against M. ru cauda to lessen their impact on beekeeping (e.g. egg cluster removal, adult mortality), may bring not negligible consequences on the population dynamics of C. signaticollis.
Our main conclusion is that in parasitoid-host systems with complex host searching strategies also show different parasitism patterns at different spatial scales. In fact, according to the 'scale-foraginghypothesis' of Gunton and Pöyry (2016), the nature of the observed correlations is strongly affected by the observational spatial scale and supports the idea that parasitoids discriminate among host patches according to their density at a 'foraging grain size'. However, when studying parasitism patterns and the in uence of density dependence it is important to analyse potential effects of ecological factors such as dispersion limitation or habitat heterogeneity that have important roles in shaping complex population dynamics.  Table 2. Estimated regression parameters, standard errors, z-values and P-values for the site model. abundance_CS is the abundance of C. signaticollis hosts. Low-egg-clusters are the sites where the estimated abundance of M. ru cauda was low. Table 3. Estimated regression parameters, standard errors, z-values and P-values for the sub-site model. abundance_CS is the abundance of C. signaticollis hosts. vegHeightLow are the sub-sites with low vegetation (only wire fences as oviposition substrates). Table 4. Estimated regression parameters, standard errors, z-values and P-values for the line model. abundance_CS is the abundance of C. signaticollis hosts. Table 5. Estimated regression parameters, standard errors, z-values and P-values for the sub-site model. abundance_CS is the abundance of C. signaticollis hosts. Figure 1 Map of the sampling sites.

Figure 2
Representation of how sampling of hosts and parasitoids was conducted.