The hump-shaped effect of plant functional diversity on the biological control of a multi-species pest community

Plant taxonomic and functional diversity promotes interactions at higher trophic levels, but the contribution of functional diversity effects to multitrophic interactions and ecosystem functioning remains unclear. We investigated this relationship in a factorial field experiment comparing the effect of contrasting plant communities on parasitism rates in five herbivore species. We used a mechanistic trait-matching approach between plant and parasitoids to determine the amount of nectar available and accessible to parasitoids. This trait-matching approach best explained the rates of parasitism of each herbivorous species, confirming the predominant role of mass-ratio effects. We found evidence for an effect of functional diversity only in analyses considering the ability of plant communities to support the parasitism of all herbivores simultaneously. Multi-species parasitism was maximal at intermediate levels of functional diversity. Plant specific richness had a negligible influence relative to functional metrics. Plant communities providing large amounts of accessible nectar and with intermediate levels of functional diversity were found to be the most likely to enhance the conservation biological control of diverse crop herbivores.


Supplementary methods: Design of species assemblages with contrasting species and functional diversities
We constructed the high functional diversity -medium species richness diversity (HFMS) assemblages by choosing species from each of the 12 functional groups identified in Table S1. The low functional diversity -medium species richness (LFMS) assemblages were obtained by reducing the number of functional groups to seven and increasing the number of species per group, so as to keep species richness constant. In the HFLS assemblages, we retained only the species of the HFMS assemblages belonging the most highly contrasting functional groups, to obtain the highest diversity possible. In the HFHS assemblages, we increased species richness and kept functional diversity as constant as possible by adding extra species with trait combinations closely resembling those already present in the HFMS assemblages.
We constructed the medium species richness -high functional diversity assemblages (HFMS) by choosing one species from each of the nine functional groups identified in Table S1. The low functional diversity assemblages (LFMS) were then obtained by reducing the number of functional groups to seven and by increasing the number of species per group, so as to keep species richness constant. In the HFLS assemblages, we retained only the species of the HFMS assemblages from the most strongly contrasting functional groups, so as to obtain the highest diversity possible. In the HFHS assemblages, we increased species richness and kept functional diversity as constant as possible by adding supplementary species, with trait combinations as similar as possible to those already present in the HFMS assemblages. Thus, at high functional diversity, the increase in species richness was associated with an increase in functional redundancy.
This process was performed for two different lists of dicotyledonous species, resulting in eight different assemblages (Table S2). We added the same three tussock grass species, in similar amounts, to each of the assemblages (Table S2). We selected common and native species and preferred perennial species over annuals, to ensure the durability of the sown communities. Within each assemblage, the species were present in similar proportions, on the basis of seed numbers and the thousand-seed weight of each batch of seed. . x x Hesperis matronalis L.
x x Leucanthemum vulgare Lam.
x x Medicago sativa L.
x x Securigera varia (L.) Lassen x x Trifolium pratense L.
x x Achillea millefolium L. x x x x x Cyanus segetum L.
x x x Trifolium repens L.
x x x Veronica hederifolia L.
x x x Centaurea scabiosa L.
x x Euphorbia cyparissias L.
x x Hypericum perforatum L.
x x Tanacetum vulgare L.
x x Ajuga reptans L.
x Bellis perennis L.
x Echium vulgare L.
x Malva sylvestris L.
x Potentilla reptans L.
x x Cynoglossum officinale L.
x x Daucus carota L.
x x Jacobaea vulgaris L.
x x Lotus corniculatus L.
x x Barbarea vulgaris R. Br.
x x x x Cota tinctoria L.
x x x x Pastinaca sativa L.
x x x x Medicago lupulina L.
x x x Stellaria media (L.) Vill. x x Glechoma hederacea L.
x Lamium album L.
x Ranunculus repens L.
x Poaceae Arrhenatherum elatius (L.) P. Beauv. the calendar week number at which flowering began and the duration of flowering (number of weeks), averaged over the three years of field observations. We also checked the entire plant for the presence of extrafloral nectar (0 = none, 1 = present).
Flower traits were measured at peak flowering, on a minimum of five or ten flowers collected from different plants, for spontaneous and sown species, respectively. Flowers were sampled early in the morning and placed in water for at least one hour before observations. Small flowers were examined under a binocular microscope (Leica M80, 60×) linked to a video camera (Moticam 10, Motic).
As a proxy for flower attractiveness (Fiedler and Landis, 2007), flower diameter (maximum flower size at the tip of the petals) or inflorescence diameter (for grouped flowers) was measured under a binocular microscope or with a ruler (for flowers > 20 mm in diameter).
The amount of floral nectar was assessed in a qualitative manner on a three-point scale (0 = none, 1 = a little, 2 = plenty). We assessed the accessibility of the nectar to parasitoids (as described in Appendix 4), by measuring four specific flower traits: -Flower opening diameter or the width of the narrowest constraint limiting the access to the flower. Opening diameter was generally measured at the extremity of unfused petals or at the point of fusion of the petals (or tepals or sepals if no petals were present), as appropriate.
Alternatively, we measured the narrowest constraint limiting access to the flower. For example, the structures measured included fringes of dense hairs (Lamium album, Malva sylvestris) or the extremity of the petals of closed flowers (many members of the Fabaceae, Fumaria officinalis). In each case, we measured the narrowest zone, in which the insect head might become blocked. FGor petals with inward corolla protuberances restricting corolla width (e.g. Myosotis arvensis), we measured the diameter of the opening at this location.
-Corolla height, measured from the corolla opening to the base of the perianth -Nectar depth is the distance between the corolla opening (or the narrowest constraint) and the top of the zone in which nectar is found -Nectar tube diameter at the top of the zone in which nectar is present.
We also extracted measurements for other traits from previous studies. We assessed the attractiveness of flowers, using data for basic flower colour and the presence of a UV reflection pattern extracted from the BiolFlor database (Kühn et al. 2004). Missing data were replaced with our own observations (for colour) or with data from the Floral Reflectance Database (Arnold et al. 2010). As a proxy for the provision of habitats to insects, we also collected trait data for leaf distribution (rosette, semi-rosette or leaves distributed along the stem), vegetative and flowering heights from the LEDA database (Kleyer et al., 2008).
We then calculated the functional diversity of the plant assemblages on the basis of the We calculated (1) functional dispersion, defined as the abundance-weighted mean distance of individual species from the centroid of all species in the trait space (Laliberté and Legendre, 2010) and (2) Rao quadratic entropy, i.e. the abundance-weighted sum of pairwise functional distances between species (Rao 1982). Functional dispersion and Rao quadratic entropy were strongly correlated (marginal R 2 = 0.95, p < 10 -4 , n=168, result obtained with a mixed model with treatment number as a random effect). We therefore considered only functional dispersion, which best reflects niche diversity, as a measurement of functional diversity. These metrics were calculated with the FD package (Laliberté et al., 2014). Supplementary Table S3. List of the traits involved in plant-parasitoid interactions taken into account in this study and associated hypotheses based on Gardarin et al. (2018).

Flower or inflorescence attractiveness
Olfactory, gustatory and visual signals facilitate the recognition and detection of resources, but may also be repellent, depending on the preferences of arthropod groups. Visual signals, such as plant height, flower height, inflorescence size and colour, are involved in resource detection, and a high degree of visual attractiveness increases the abundance of natural enemies (Fiedler and Landis 2007).

Diameter
Color UV reflectance pattern

Temporal availability of nectar
The synchrony between the plant and arthropod cycles determines the likelihood of interaction (Welch and Harwood 2014). The seasonal availability of resources depends on plant phenology. The phenological match between flowering period and arthropod floral resource requirements is crucial for completion of the life cycles of both herbivores and their natural enemies.
Date of flowering onset Duration of flowering

Floral nectar accessibility
In flower-visiting arthropods, head or body size may physically restrict access to floral resources in flowers with a small corolla diameter. In such cases, the plant-arthropod interaction depends on the length of the mouthparts. A correlation between nectar holder depth and the proboscis length of the flower visitor has been observed in several insect groups, especially in pollinators. Short corolla flowers favour hoverflies, whereas bumblebees prefer long corollas (Campbell et al. 2012). As a result, flower size is one of the most important variables determining the abundance and diversity of flower visitors and their size (Ibanez 2012;Stang et al. 2006;van Rijn and Wackers 2016). By contrast, extrafloral nectar is generally produced on exposed nectaries, with no size constraints on its accessibility. In April 2017, we collected 100 (when possible) last instar larvae of each species from the petioles of basal leaves of oilseed rape. The collection of larvae was very time-consuming, and we were, therefore unable to sample these larvae at a distance of 20 m for all assemblages (only the LFMS2, HFMS2, HFHS1, HFHS2 and control treatments were sampled at this distance).
The larvae were placed in hermetic plastic boxes (9.5 cm in diameter, 7 cm high) two-thirds filled with sieved dry soil. The C. pallidactylus and for P. chrysocephala larvae were placed in separate boxes. The rearing boxes were placed at room temperature (about 20°C) for six months, then at 4°C for five months to simulate winter temperatures, before being returned to room temperature. We recorded the emerging adults of P. chrysocephala and C. pallidactylus (which developed from non-parasitized larvae) during the first two months after collection in the field. Parasitoids emerged after the release from low temperatures and were recorded over a period of two months, until no further emergences occurred. For both hosts, the parasitoids were morphologically similar to Tersilochus microgaster (Szépligeti, 1899) (Hymenoptera, Ichneumonidae) and T. obscurator (Aubert, 1959) (Hymenoptera, Ichneumonidae), according to published identification tools (Vidal, 2007;Barrari et al., 2005;Robert et al., 2019), and could not be assigned to either species with certainty. They are described as being specific of theirs hosts among oilseed rape pests (Ulber et al., 2010).
Finally, we searched the soil in the rearing boxes for any remaining cocoons of parasitoids or non-parasitized adults of the host species that did not emerge and died after metamorphosis because of non-optimal rearing conditions. Parasitism rates for C. pallidactylus and P.
chrysocephala were then calculated as the number of adult parasitoids that had emerged or were recovered from the soil divided by the sum of parasitoids and adult hosts that had emerged or were recovered from the soil. Parasitism rates were calculated as the number of larvae containing at least one parasitoid egg divided by the total number of hosts.
• Parasitism of larvae of Dasineura brassicae (Winnertz, 1853) (Diptera: Cecidomyiidae) Immediately after the oilseed rape harvest in July 2017, we collected soil samples with a spade, from an area of 20 cm × 20 cm and a depth of 5 cm. The soil samples were stored for two months at 5°C. They were then gently washed in a sieve with a 1-mm mesh, to remove all the fine soil particles. For each sample, a minimum of 150 white cocoons of Dasineura brassicae were recovered from the particles remaining in the sieve. We dissected the cocoons under a binocular microscope, and recorded the numbers of cocoons containing non-parasitized host nymphs or adult micro-hymenoptera (from one to three individuals).
We assumed that all were parasitoids, although we cannot exclude that morpho-species were hyper-parasitoids. Empty cocoons were not taken into account. The parasitism rate was calculated as the number of cocoons containing at least one micro-hymenoptera divided by the total number of non-empty cocoons.

Supplementary methods: Estimation of nectar accessibility to parasitoids
The accessibility of floral nectar depends on the morphological match between insects and flowers. We developed a geometric model, adapted from that described by van Rijn and Wäckers (2016), and parameterized with four flower traits and three insect traits ( Figure S1).
The flower traits included are generally measured on the corolla, but sometimes on the perianth, in situations in which the sepals and petals have similar functions: w is generally the width of the flower opening. It was generally measured at the extremity of unfused petals or at the point of petal (or tepal or sepal if no petals) fusion, as appropriate.
However, w could also be the width of the narrowest constraint limiting access to the flower.
For example, it could be fringes of dense hairs (Lamium album, Malva sylvestris) or the extremity of the petals of closed flowers (many members of the Fabaceae, Fumaria officinalis). We systematically recorded the width of the narrowest zone, in which the head of the insect might be blocked. For flowers with petals displaying inward corolla protuberances restricting corolla width (e.g. Myosotis arvensis), we measured the diameter of the opening at this location.
h is corolla height, measured from the opening of the corolla to the base of the perianth p is nectar depth, measured as the distance between the opening of the corolla (or the narrowest constraint) and the upper part of the zone containing nectar d is the nectar tube diameter in the upper part of the zone containing nectar.
Three insect traits (Table S3) were measured for each parasitoid species or morphospecies: r, the radius of the insect head (half the diameter, assuming a spherical head) z, proboscis length x, proboscis width.  The constraints limiting nectar accessibility are summarized on a decision tree (Figure 4). In particular, we needed to calculate the distance between the parasitoid head and the site at which the nectar was found. We distinguished two cases, according to the ability of the insect to penetrate the flower.

Supplementary
• Case 1: the parasitoid has a head too large to enter the corolla (2r ≥ w). In this case, we calculated the distance l between the extremity of the head and the base of the corolla: l = JC = h -OJ.
h, the total corolla height, is a measured trait. We needed to calculate the distance OJ.
Given that RJ = r = OR + OJ, we obtain: The nectar is accessible when z, the length of the proboscis, is greater than or equal to the distance JK between the parasitoid head and the upper part of the nectar-containing zone.
We have: l = JK + (hp), where p is nectar depth, measured from the top of the corolla (or constraint).
And finally: z ≥ JK therefore implies that z ≥ lh + p.
• Case 2: the head can penetrate the corolla (2r < w). We needed to calculate the distance l between the extremity of the head and the base of the corolla.
As in case 1, the nectar is accessible when z, the length of the proboscis, is greater than or equal to the distance between the parasitoid head and the upper limit of the nectar-containing zone, i.e. when z ≥ lh + p.

Assessment of the populations of crop herbivores
In oilseed rape, we characterized the pressure of the adult herbivore populations colonizing the field and at the origin of the new generation, in which we measured the larval parasitism. The damage is caused by the adults of this generation (for B. aeneus) or by the larvae of the following generation (all other studied herbivores).
Five yellow pan traps (28 cm diameter, filled with water and a drop a scentless detergent) were placed in the 6.5 ha-experimental field, with one trap at 20 m from each field corner and one trap at the centre of the field. The pan traps were monitored weekly during the whole oilseed rape cycle in 2016-17.
In P. chrysocephala, there were 265 ± 70 adults per trap captured from September to November on the experimental field, which is substantially higher than in conventional fields in the region (45 adults over a network of 41 fields monitored by agricultural extension services, Concerning B. aeneus, we captured 1072 ± 408 adults on average per trap in March and in April, which is clearly higher than what can be observed in the literature. In Hiiesaar et al. (2003), there were about 180 adults on average per trap during the equivalent period, and in Hatt et al. (2015), there were about 60 adults on average per trap during the equivalent period).  Supplementary Table S8. Effects of the composition and structure (percentage plant cover providing accessible nectar, species richness and functional dispersion) of the flower strip plant communities on rates of parasitism in five herbivorous insect pests and on global multi-species parasitism (quantified by a multi-threshold approach). We show here the best models (ΔAIC from the best model < 2) obtained with the multi-model inference procedure comparing all possible combinations of the fixed effect variables (linear and quadratic terms) and their interactions. Generalized linear mixed effect models were used, assuming a binomial (parasitism rates) or Poisson (multi-species parasitism) error distribution, and including the strip as a random effect. All explanatory variables were scaled.   197.5 197.6 197.7 198.0 198.2 198.2 198.6 198 Table 1) Effects of the composition and structure (proportion of plant cover providing accessible nectar, species richness and functional dispersion) of the flower strip plant communities on the rates of parasitism in five herbivorous crop pests and on global multi-species parasitism (quantified via a multi-threshold approach). All possible combinations of the plant community variables (nectar resources, species richness and functional diversity, with both linear and quadratic effects) and their interactions were compared. The best models were ranked according to their AIC (Table S8) and we present the results for the conditional average of best models. Generalized linear mixed effect models were used, assuming a binomial (parasitism rates) or Poisson (multi-species parasitism) distribution, with strip as a random effect. All explanatory variables were scaled. We report the relative importance for each predictor (weight). "n" is the number of observations for each response variable.  Table 1 and Table S9) Figure S3. Effect of the plant assemblages making up the flower strips on parasitism rates in five herbivorous insect pests in the adjacent crop. The eight plant assemblages have a low or high functional diversity (LF or HF, respectively), a low, medium or high species richness (LS, MS or HS respectively) and are composed of species from two different lists (1 or 2), and are compared with a control plot (C), on which the strip was sown with the same crop species as the field and managed in a similar manner. The different letters indicate significant differences between groups (P<0.05) between plant assemblages, within each category of distance (5 and 20 m from the strip). Not all treatments were studied at the 20 m distance. This figure was made using R version 3.6.3 (https://www.R-project.org/). Figure S4. Dynamics of the functional diversity (a) and species richness (b) of the eight sown plant assemblages (means ± standard errors). The data for 2013 are based on the seed mixtures sown, and the data for the next four years were obtained from field observations. The eight plant assemblages have a low or high functional diversity (LF or HF), a low, medium or high species richness (LS, MS or HS) and are composed of species from two different lists (1 or 2). This figure was made using R version 3.6.3 (https://www.R-project.org/).