Bottom‐up effects of plant quantity and quality on arthropod diversity across multiple trophic levels in a semi‐arid grassland

Plant quantity and quality can independently affect the diversity of the entire arthropod communities and multiple arthropod taxa in grassland ecosystems. However, it remains unclear how these effects on arthropod taxa at one trophic level propagate through food web to influence the diversity of higher trophic levels. We performed a monoculture experiment with 15 herbaceous species in the Inner Mongolian grassland to investigate how natural variations in plant productivity and host leaf traits affect herbivore taxon richness, which, in turn, affects predator taxon richness. For herbivores, plant productivity indirectly promoted herbivore taxon richness by increasing herbivore biomass, which was attributed to the increases in the richness of dominant sucking herbivores and endophytes with high food requirements. However, the high plant quality indicator (e.g. high leaf protein, phosphorus and water contents, and high leaf protein to carbohydrate ratio) directly increased, whereas the low plant quality indicator (e.g. high leaf lignin content) directly decreased herbivore taxon richness. Taxon richness of chewing and sucking herbivores with specific feeding modes (tearing or sucking mouthparts) showed strong positive responses to increasing plant quality. For predators, herbivore taxon richness, rather than herbivore biomass, mainly mediated the positive effects of plant productivity and the high plant quality indicator, but the negative effect of the low plant quality indicator, on predator taxon richness. At the feeding guild level, the taxon richness of parasitoids, other predators and spiders exhibited positive responses to different herbivores, which was attributed to their different diet preferences. Predator diversity could be promoted by prey partitioning among predator guilds facilitating species coexistence. At the family level, the taxon richness of most predator families was positively correlated with that of more than one herbivore family, suggesting that high predator diversity may be caused by balanced diets owing to high prey diversity. Synthesis. Natural variations in plant quantity and quality can substantially affect the diversity of herbivores and cascade up the food web to affect predators. Specificity and mechanisms of feeding have a large impact on the responses of arthropod guilds at each trophic level.

However, it remains unclear how these effects on arthropod taxa at one trophic level propagate through food web to influence the diversity of higher trophic levels.
2. We performed a monoculture experiment with 15 herbaceous species in the Inner Mongolian grassland to investigate how natural variations in plant productivity and host leaf traits affect herbivore taxon richness, which, in turn, affects predator taxon richness.
3. For herbivores, plant productivity indirectly promoted herbivore taxon richness by increasing herbivore biomass, which was attributed to the increases in the richness of dominant sucking herbivores and endophytes with high food requirements. However, the high plant quality indicator (e.g. high leaf protein, phosphorus and water contents, and high leaf protein to carbohydrate ratio) directly increased, whereas the low plant quality indicator (e.g. high leaf lignin content) directly decreased herbivore taxon richness. Taxon richness of chewing and sucking herbivores with specific feeding modes (tearing or sucking mouthparts) showed strong positive responses to increasing plant quality.
4. For predators, herbivore taxon richness, rather than herbivore biomass, mainly mediated the positive effects of plant productivity and the high plant quality indicator, but the negative effect of the low plant quality indicator, on predator taxon richness. At the feeding guild level, the taxon richness of parasitoids, other predators and spiders exhibited positive responses to different herbivores, which was attributed to their different diet preferences. Predator diversity could be promoted by prey partitioning among predator guilds facilitating species coexistence. At the family level, the taxon richness of most predator families was positively correlated with that of more than one herbivore family,

| INTRODUC TI ON
Arthropods comprise the majority of animal biodiversity in terrestrial ecosystems, and their growth rate and reproductive success depend largely on primary producers (Siemann, 1998;Uchida & Ushimaru, 2014;Wilson, 1987). For example, a previous study showed that increased plant productivity elevated herbivore and predator diversity simultaneously, with predator diversity showing the strongest response (Wimp et al., 2010). Lu et al. (2021) recently reported that changes in plant quantity (e.g. biomass production) and quality (e.g. leaf nutrient content) across different plant species can independently affect the diversity of the entire arthropod communities and different taxonomic groups in grasslands. Both herbivores and predators can be assigned to different guilds based on their feeding habits (Carmona et al., 2011;Pratt et al., 2017). Changes in plant biomass and host plant traits may influence herbivore diversity, which, in turn, affects predator diversity through trophic interactions between feeding guilds (Christenson, 1984;Jacquot et al., 2019;Scherber et al., 2010).
However, it remains unclear how effects of plant quantity and quality on arthropod taxa at one trophic level propagate through the food web to influence the diversity of higher trophic levels (Awmack & Leather, 2002;Forbes et al., 2017;Perkins et al., 2004).
Plant quantity and quality can strongly regulate herbivore diversity, but their effects are feeding-guild specific (Behmer, 2009;Wimp et al., 2010). As explained by the plant quantity hypothesis, plant species with high biomass production could directly expand the diversity and abundance of the feeding niche and allow both common and rare herbivorous guilds and species to persist, thereby increasing arthropod richness (H1; Table 1; Figure 1; Siemann, 1998). Plant production can also indirectly affect herbivore richness by affecting herbivore biomass (Simons et al., 2014).
In grassland ecosystems, high plant production can increase the growth rate of dominant sucking herbivores (Family: Aphididae) by meeting their heavy food demands, and herbivore populations with high biomass are less prone to extinction due to environmental disturbance (Lu et al., 2021). According to the plant quality hypothesis, plant nutrient concentration could exert bottom-up control on herbivore richness (H2; Table 1; Figure 1; Awmack & Leather, 2002;Loranger et al., 2012), because different herbivorous guilds may use specific feeding modes to access plant nutrients (Novotny et al., 2010). For instance, an increase in host foliar nitrogen and phosphorus contents could increase herbivore diversity by promoting taxon richness of chewing and sucking herbivores, because such guilds often use tearing or sucking mouthparts to suggesting that high predator diversity may be caused by balanced diets owing to high prey diversity. 5. Synthesis. Natural variations in plant quantity and quality can substantially affect the diversity of herbivores and cascade up the food web to affect predators.
Specificity and mechanisms of feeding have a large impact on the responses of arthropod guilds at each trophic level.

K E Y W O R D S
arthropod biomass, arthropod taxon richness, food web, herbivore, multi-trophic interactions, plant leaf traits, plant productivity, plant-herbivore interactions, predator TA B L E 1 Proposed hypotheses regulating the taxon richness of herbivores and predators in grasslands

Hypothesis name Prediction
Plants to Herbivores H1 Plant quantity hypothesis Plant biomass could directly and indirectly promote herbivore richness by increasing herbivore biomass 1,2 H2 Plant quality hypothesis Plant nutrient content could directly and indirectly promote herbivore richness by increasing herbivore biomass 3,4 Herbivores to Predators H3 More individuals hypothesis Herbivore biomass could promote predator biomass, leading to an abundance-driven accumulation of predator species 5,6 quickly access nutrients from leaf or phloem sap for their growth and reproduction (Novotny et al., 2010;Welti et al., 2020;Wilson et al., 2018). However, the diversity of endophytes (e.g. fruit flies and leaf miners) may be less likely to reflect plant leaf quality since they often feed within flowers and fruits and are more sensitive to carbohydrate than protein (Peguero et al., 2017). Alternatively, an increase in plant protein content could increase the fecundity and biomass of arthropods such as flies (Family: Tephritidae) and aphids (Family: Acrididae), which would indirectly increase herbivore richness because of the reduction in resource competition among herbivores (Awmack & Leather, 2002).
Herbivores could mediate the effects of plant quantity and quality on the diversity of higher trophic levels via prey and predator interactions (Jacquot et al., 2019;Siemann, 1998;Welti et al., 2020).
Two main hypotheses have been proposed to describe the effects of herbivores on predator diversity. First, the more individuals hypothesis predicts that increases in the biomass of herbivores promote predator biomass and thus increase predator richness (H3; Table 1; Figure 1; Simons et al., 2014;Srivastava & Lawton, 1998).
If the increase in plant biomass and host leaf nutrient content promotes biomass of sucking and chewing herbivores, it may increase the number and range of prey available to parasitoids and predators and thus elevate total predator diversity (Hawkins et al., 1997;Petermann et al., 2010;Welti et al., 2020). Second, the resource heterogeneity hypothesis predicts that an increase in herbivore richness promotes predator richness through increased prey niches and diet variety (H4; Table 1; Figure 1; Hutchinson, 1959;Lewinsohn & Roslin, 2008). On one hand, increases in herbivore diversity under high plant resource conditions could increase opportunities of niche specialization for specialist predators, thereby promoting predator richness via the niche partitioning effect (Gamfeldt et al., 2005;Jacquot et al., 2019). For example, increased leaf nutrient content is expected to promote the diversity of sucking herbivores, which may, in turn, promote the diversity of the parasitoids reliant on this specific prey item (Hawkins et al., 1997;Petermann et al., 2010). On the other hand, increases in herbivore richness could enhance predator richness by providing more diverse and balanced prey resources for generalist predators (Dassou & Tixier, 2016;DeMott, 1998). In grasslands, spiders prey not only on sucking herbivores but also on chewing grasshoppers (Uetz, 1992). Hence, arthropod communities with diverse herbivorous feeding guilds may contain a greater variety of nutrient biomolecules essential for spiders, possibly leading to high spider richness (Yip et al., 2008). In fact, positive relationships are commonly observed between herbivore and predator diversity in terrestrial ecosystems (Dassou & Tixier, 2016;Jacquot et al., 2019;Siemann, 1998 Although both plant quantity and quality could affect multitrophic arthropod diversity, little is known about their independent roles in grassland ecosystems (Lu et al., 2021;Wimp & Murphy, 2021).
First, previous studies manipulated plant communities by watering, fertilization and grazing, largely ignoring the effects of natural variations in plant productivity and host nutrient content on arthropod diversity (Simons et al., 2014;Zhu et al., 2017). Therefore, changes in plant biomass were often tightly linked with concomitant shifts in host leaf traits, confounding investigators' ability to isolate the effects of plant quantity and quality on the richness of arthropod communities (Wimp & Murphy, 2021). Second, many previous studies examined the effects of plant production and/or host plant traits only on a subset of taxonomic or functional groups (Peeters et al., 2007;Sagers, 1992). Nonetheless, multiple trophic groups of arthropods such as herbivores and predators are mainly regulated by vegetation and are critical for ecosystem function as pollinators and pest control (Joern & Laws, 2013). Third, although a few studies have attempted to elucidate the effects of plant biomass and host leaf traits on the diversity of multiple trophic groups (e.g. herbivores and predators), they did not break each trophic group into different feeding guilds according to their feeding modes and diet preferences (Simons et al., 2014). The specific interactions between plants and herbivorous feeding guilds (Welti et al., 2020) and between herbivorous and predatory guilds could be an important mechanism underlying the effects of plant quality and quantity on the diversity of higher trophic levels (Haddad et al., 2009;Jacquot et al., 2019). Thus, experimental designs that explore the effects of plant quantity and quality on multi-trophic arthropod diversity in realistic communities are needed.
To address these uncertainties, we used replicated monocultures  (Table S1). All necessary permits were obtained from IMGERS prior to establishing of the monoculture experiment.

| Plant productivity and leaf trait measurements
For each monoculture plot, the standing biomass of plants was sampled from a 50 cm × 50 cm area in mid-August 2018. Plant material was sampled by cutting all plants in each quadrant at the soil surface and then was oven-dried at 65°C for 48 h and weighed. Plant biomass was calculated from dry biomass measurements (g m −2 ) of each plot, as the standing above-ground biomass of these steppe communities reached the annual peak at mid-August (Bai et al., 2004).
In August 2018, five undamaged, fully expanded leaves were collected from each of the 15 plant species from each monoculture plot. Plant leaf traits, including leaf non-structural carbohydrate, protein, phosphorus, water and lignin contents were measured.
These leaf traits are important determinants of plant nutrient quality for arthropods (Awmack & Leather, 2002;Forbes et al., 2017;Joern et al., 2012). Fresh leaves were weighed and dried for 24 h at 65°C, and leaf water content was expressed as the difference between fresh and dry weight, divided by dry weight. A ball mill was used to grind dry leaf material to a fine powder, and samples of 10 mg were analysed for leaf phosphorus content using an elemental analyzer (VarioEL Element Analyzer; Hanau, Germany). We measured plant protein content with using the Bradford assay and analysed non-structural carbohydrate content using the phenol-sulphuric acid method, following the protocol of Clissold et al. (2006). The leaf protein:carbohydrate ratio was defined as the ratio of leaf protein to leaf non-structural carbohydrate content. Leaf lignin content was measured following sequential extraction analysis of acid detergent lignin (Forbes et al., 2017). We averaged the data of each leaf trait for each plant species and used the mean leaf traits values for data analysis. These 15 plant species encompassed a broad range of leaf trait variation ( Figure S1).

| Arthropod sampling and identification
Using the sweep-net sampling, we collected arthropods from the monoculture plots between 10 AM and 4 PM on days with no rain-  (Guo et al., 2009), our sampling period typically coincided with the peak abundance of many arthropod taxa in the study area (Wang et al., 2020). We did not use pitfall trap sampling method to collect the ground-dwelling arthropods, which may have led to the omission of some predators (e.g. carabid beetles) playing key roles in grassland ecosystems (Andersen et al., 2019;Pringle & Fox-Dobbs, 2008). Therefore, arthropod diversity may be underestimated in this study. However, other studies have found that the number of arthropod species obtained from sweep-net sampling is highly obtained with that sampled using vacuum sampling for both vegetation-dwelling and ground-dwelling arthropods (Siemann, 1998). We conducted 50 sweeps by using a muslin net for each monoculture plot. We sampled the arthropods by sweeping at 180° arcs through the vegetation canopy, quickly turning, and reversing the direction at the end of each arc (Doxon et al., 2011). At the end of each arc, a quick but fluid upturn of the sweep net was used to prevent the escape of the captured arthropods. The contents of the sweep net were preserved in bottles containing ethyl acetate.
In the laboratory, all arthropod individuals were identified by optical microscopy at the genus and species levels as far as possible.
Some species were treated as reference specimens because they could not be identified to the genus or species level during the initial identification. These reference specimens were placed in vials containing 75% ethanol and sent to taxonomists for accurate identification to morphospecies. Each morphospecies was further placed into one of two trophic categories (Table S2): herbivores and predators (Perner et al., 2005), and then carefully assigned to one of the three feeding guilds based on published accounts for the taxonomic guilds (Carmona et al., 2011;Pratt et al., 2017). Herbivorous feeding guilds consisted of (1) sucking/piercing herbivore (species or genus from Pentatomidae, Rhyparochromidae, Coreidae, Tingidae, Piesmatidae, Mordellidae, Miridae, Curculionidae, Aphididae and Cicadellidae); (2) chewing herbivore (species or genus from Anthicidae, Acrididae and Chrysomelidae) and (3) (3) spiders (species or genus from Thomisidae; Pratt et al., 2017). We also measured the biomass of each arthropod family collected from the sampled monoculture plots. For herbivores, sucking and endophyte herbivores were the dominant feeding guilds (relative biomass [RB], >10% of total herbivore biomass), and chewing herbivores were the rare feeding guild (RB, <10% of total herbivore biomass; Figure S2a). For predators, parasitoids and other predators were the two dominant feeding guilds (RB, >10% of total predator biomass), and spiders were the rare predator feeding guild (RB, <10% of total predator biomass; Figure S2b). Consequently, we could establish gradients in taxon richness and biomass of the entire community and each herbivore and predator feeding guild across the monoculture plots of different plant species (Tables S3 and S4). richness. We considered all explanatory variables because we assumed no multicollinearity between the explanatory variables (all variance inflation factors were <10). The SEMs were implemented using the lavaan package. All variables were transformed to natural logarithms before SEM to mitigate departure from normality and linearity. Because of the potential nonlinear relationships between plants and arthropods, we used the nonparametric Bollen-Stine bootstrapping estimations to increase the robustness of our SEM (Bollen & Stine, 1992). A good model fit was indicated by a Bollen-Stein bootstrap p > 0.10.

| Data analysis
Finally, the relationships between herbivore richness and predator richness were further examined using linear regression at the feeding guild level and at the family level.

| Responses of richness and biomass of herbivores to plant quantity and quality
Herbivore richness and biomass increased quadratically with in-   The high plant quality indicator The low plant quality indicator

| Responses of the richness and biomass of predators to plant quantity and quality
Predator richness and biomass showed linear relationships with plant biomass (Figure 3g,j; Table S5). For different feeding guilds, taxon richness of other predators and parasitoids ( Figure S3g; Table S6), and biomass of parasitoids increased with increasing plant biomass ( Figure S3j; Table S7). Predator richness increased quadratically and predator biomass increased linearly with increasing the high plant quality indicator (Figure 3h,k; Table S5). In particular, the taxon richness of other predators and the biomass of parasitoids increased linearly with increase in the high plant quality indicator ( Figure S3h,k).

| Pathways determining arthropod richness across trophic levels
Our SEM analysis showed that plant biomass increased herbivore biomass (standardized regression weight = 0.68), which, in turn, increased herbivore taxon richness (0.45; Figure 4). The altered herbivore richness, in turn, increased predator richness (0.28). Therefore, plant biomass had total positive effects (0.31) on herbivore richness and predator richness (0.09). The high plant quality indicator increased herbivore richness (0.23), which, in turn, increased predator richness, resulting in total positive effect on predator richness.

| Relationship between herbivore richness and predator richness
At the feeding guild level, sucking herbivore richness was positively related to the taxon richness of parasitoids and total predators ( Figure 5a); endophyte richness was positively related to the taxon richness of parasitoids, other predators and total predator (Figure 5b), and chewing herbivore richness was positively related to the taxon richness of spiders (Figure 5c). The biomass of sucking herbivores and endophytes were positively related to the taxon richness of both overall and each herbivore feeding guild ( Figure S4a,b).
The biomass of spiders, parasitoids and other predators were positively related to the taxon richness of overall and with each predator feeding guild ( Figure S5a-c).
At the family level, the taxon richness of Ichneumonidae, Crabronidae, Tiphiidae, Chrysopidae, Anthocoridae, Asilidae, Chironomia and Thomisidae (accounting for 53.3% of total predator families) were positively related to the taxon richness of more than one herbivore family (Figure 6a,b; Figure S6), indicating that these predators could be the generalist predators with more than one kind of edible prey. The taxon richness of Dryinidae and Meliinidae (accounting for 13.3% of total predator families) was positively related to the taxon richness of one herbivore family (Figure 6a,b; Figure S6), indicating that these predators could be the specialist predators with only one prey. However, the taxon richness of Scoliidae, Syrphidae, Dolichopodidae and Bombyliidae (accounting for 33.3% of total predator families) was not related to the taxon richness of any herbivore family, indicating the absence of edible prey for these predators (Figure 6a,b; Figure S6).

| Effects of plant quantity and quality on herbivore richness
Plant biomass production and leaf nutrient content each have been shown to independently influence the diversity of the entire arthropod communities and multiple arthropod orders (Lu et al., 2021).
Therefore, changes in plant quantity and quality may directly and/or indirectly affect the diversity of adjacent trophic levels, with effects cascading up to higher trophic levels (Barnes et al., 2020;Dyer & Letourneau, 2003;Scherber et al., 2010). We found that plant biomass was positively associated with herbivore taxon richness, which is consistent with our first hypothesis. However, previous studies, which focused on plant-herbivores interactions, have shown that the relationship between plant productivity and trophic diversity can be positive (Siemann, 1998), negative (Jepsen & Winemiller, 2002) or neutral (Post, 2002). These studies often manipulated plant community productivity with simultaneous changes in plant composition, confounding the effects of plant production on herbivore diversity.
After the effects of host plant traits were controlled, SEM showed that plant productivity could indirectly promote herbivore taxon richness by promoting herbivore biomass. Our finding is consistent with that of a previous study, which found that increases in plant production without changes in community composition increase herbivore richness (Wimp et al., 2010). With respect to different feeding guilds, we found that plant biomass promoted the biomass of dominant sucking herbivores and endophytes, which, in turn, increased their taxon richness and thus total herbivore richness.
Plant production, in contrast, has been shown to promote herbivore richness by allowing more rare species to persist (Siemann, 1998).
In our study, the strong positive responses of dominant herbivore feeding guilds to increase in plant biomass may be explained by their higher food requirements than those of the rare feeding guilds (Lu et al., 2021;Rich et al., 2013;Welti et al., 2020). For example, common sucking herbivores such as aphids often require larger amounts of food to maintain their growth rate (Lu et al., 2021;Perner et al., 2005). Consequently, the populations of dominant herbivores with high biomass and abundance would be less prone to extinction due to environmental disturbance, resulting in high arthropod richness. In addition, increased primary production could provide more habitat volume and energy for sucking herbivores such as leafhoppers (Prather & Kaspari, 2019), which may potentially contribute to their survival, growth and reproduction (Kansman et al., 2021).
For example, herbivores, such as Hemiptera, often choose plant vegetation with high plant biomass as oviposition sites due to the low temperature and high humidity environment (Clissold et al., 2013;Grevstad & Klepetka, 1992;Obermaier et al., 2008).
We found a direct positive effect of a high plant quality indicator (e.g. leaf protein: carbohydrate ratio, high leaf protein, phosphorus and water contents), but a direct negative effect of a low plant quality indicator (e.g. high leaf lignin content) on herbivore taxon richness. These results are consistent with our second hypothesis that high plant quality enhances herbivore richness (Awmack & Leather, 2002;Joern et al., 2012). Contrary to our findings, other studies have shown that plant food with a high phosphorus to carbon ratio can inhibit the growth of several herbivorous groups (e.g. chironomids) and thus decrease arthropod diversity (Boersma & Elser, 2006;Forbes et al., 2017). Possible mechanisms for these patterns include herbivores experiencing costs associated with the storage and excretion of excess macronutrients from their body, which could impair their development and growth (Boersma & Elser, 2006).
Additionally, other herbivorous groups (e.g. female locusts) have been shown to be sensitive to plant food with high protein:carbohydrate ratios because they require more soluble carbohydrate for reproduction and mobility (Cease et al., 2012;Le Gall et al., 2020).
However, in the present study, we found that the high plant quality indicator significantly increased, while the low plant quality indicator significantly decreased taxon richness of chewing and sucking herbivores, suggesting a bottom-up nutrient limitation for herbivores. In grassland ecosystems, chewing and sucking herbivores are two important functional groups with specific feeding modes for effectively accessing plant nutrients, and their diversity and abundance may depend largely on leaf quality (Welti et al., 2020;Wilson et al., 2018).
For example, chewing herbivores (Family: Acrididae) often use their tearing mouthparts to quickly consume leaf tissue, which provides them with essential proteins and thus their high growth and reproduction may be likely to reflect plant quality (Prather et al., 2021).
Unlike chewers, sucking insects such as aphids and whiteflies are The low plant quality indicator equipped with piercing and sucking mouthparts; this could facilitate their feeding on nutrient-rich leaf phloem (Petermann et al., 2010).
As a result, the density and diversity of sucking herbivores may be strongly influenced by the phloem sap quality of different plant species. Notably, we found that the taxon richness of endophytes did not change significantly with increased plant quality. This outcome is likely attributable to the fact that the host plant organs preferred by many endophytes (e.g. fruit flies) are flowers, fruits and seeds (Peguero et al., 2017), which would weaken the responses of endophytes to changes in plant leaf traits. Overall, our results suggest F I G U R E 5 Relationships between taxon richness (e.g. species or genera richness per m 2 ) of herbivores and predators at the arthropod feeding guild level. The average values with 95% confidence intervals for taxon richness of herbivores and predators are shown. Relationships between taxon richness of sucking (a), endophyte (b) and chewing herbivore (c) with parasitoid (blue line), other predator (green line), spider (orange line) and total predators (red line) are shown. Solid lines indicate fits of significant bivariate regression models (p < 0.05), and dashed lines were non-significant. R 2 values are shown for all regressions.

F I G U R E 6
Relationships between taxon richness (e.g. species or genera richness per m 2 ) of herbivores and predators at the arthropod family level. Solid red and blue lines indicate that bivariate relationships were positively and negatively significant, respectively (p < 0.05) (a). Based on the correlation analyses, the total number of herbivore families for which taxon richness is significantly correlated with the taxon richness of each predator family were calculated, and the potential proportions of generalists, specialist and unidentified predators are presented (b).

| Herbivore richness mediated the effects of plant quantity and quality on predator richness
In our study, herbivore biomass did not mediate the effects of plant nutrient quality indicators on predator taxon richness, which does not agree with the core predictions of our third hypothesis. This conclusion contradicts the findings of many previous experimental and observational studies in grasslands showing positive relationships between herbivore and predator biomass (Haddad et al., 2009;Welti et al., 2020). This unexpected result may be explained in several ways: First, previous studies that showed significant positive impacts of herbivore biomass on predator biomass were often conducted in fertilization and irrigation experiments, and the increased plant nutrient content could remarkably promote herbivore biomass and thus lead to biomass-driven accumulation of predator species (Simons et al., 2014;Zhu et al., 2019). However, under natural conditions, plants need to invest considerable carbon in physical and chemical defences against herbivores (Carmona et al., 2011); this could decrease plant nutrient content and thus nitrogen-rich herbivore tissues, which would weaken the positive correlation between herbivore and predator biomass (Awmack & Leather, 2002).
Second, the lack of significant positive relationships may be attributed to the fact that the biomass of predators is constrained by the presence of their own predators (Cronin et al., 2004). For example, many parasitoids and spiders are consumed by birds in grasslands (Tscharntke, 1992). Third, some spiders have relatively low metabolic rates, which allows them to burn less energy and thus to maintain relatively high biomass with limited available prey biomass (Welti et al., 2020). Finally, the increased plant biomass could provide additional habitat volume for spiders to hang their webs, and more flowers and pollen to feed parasitoid wasps, leading to high predator and parasitoid biomass and abundance (Prather et al., 2021). Therefore, herbivore biomass could be constrained by high predation rates, which would damp the relationship between herbivore and predator biomass (Barnes et al., 2020). Surprisingly, we found that taxon richness of predators increased significantly with predator biomass, both overall and within each feeding guild (parasitoids, other predators and spiders). This is easily explained by the increases in biomass of all feeding guilds, which can enhance the total number of predator individuals, and thus allow both common and rare species to persist locally (Siemann, 1998). However, we did not use pitfall trap method to sample the ground-dwelling arthropods, which may limit our ability to completely characterize the responses of predator richness to changes in plant resources (Hertzog et al., 2017). For instance, some ground-dwelling predators (e.g. carabid beetles) have been shown to prey at significant rates upon herbivores feeding on both C 3 and C 4 plants (Beckerman et al., 1997;Hawlena et al., 2012;Pringle & Fox-Dobbs, 2008). Future studies should determine the effects of plants on arthropod diversity using both pitfall trap and sweep-net sampling methods in grassland ecosystems.
Although many studies have shown positive effects of plant production and host plant traits on predator richness, they did not distinguish between direct and indirect effects (Dassou & Tixier, 2016;Haddad et al., 2009;Siemann, 1998). Based on the monoculture experiment, our SEM revealed that herbivore taxon richness mainly mediated the effects of plant productivity and host leaf traits on predator richness, supporting the fourth hypothesis. This is in contrast with the findings of previous studies showing that there was no effect of herbivore diversity on predator diversity (Fox, 2004;Jacquot et al., 2019). The absence of bottom-up diversity effects in those studies is partly explained by the lack of differentiation in predation for herbivores among predator feeding guilds (Gamfeldt et al., 2005). However, we documented a wide range of positive relationships between the taxon richness of herbivorous and predatory feeding guilds. The possible mechanism for these patterns lies in different feeding modes of predator guilds for consuming different herbivores (Hawkins et al., 1997).
Specifically, we found that plant biomass and host traits significantly promoted taxon richness of sucking herbivores and endophytes, which, in turn, were positively correlated with the taxon richness of parasitoids. Parasitoid wasps often attack sucking herbivores (e.g. aphids) and endophytes (e.g. leaf miners) by laying a single egg into the body of their host, and utilizing up most of the host's nutrients for development and growth of the parasitoid larva until it eventually pupates (Hawkins, 1992;Petermann et al., 2010). Therefore, it is possible that increasing plant quantity and quality could increase the number of specifically associated aphids and leaf miners, leading to an increase in the number of specifically associated parasitoid species. Additionally, we found that the taxon richness of endophytes was positively related to that of other predators. Although endophytes (e.g. leaf miners) feeding within plant tissues could be protected from general predation to a certain extent (Hawkins et al., 1997;Peguero et al., 2017), they can suffer high predation pressure from other types of enemies such as robber flies (Family: Asilidae), which have high mobility and actively search for food (Dennis & Lavigne, 2007). Moreover, we found that the taxon richness of chewing herbivores was positively related to that of spiders. Chewers such as grasshoppers usually have a larger body size and struggle longer than do other herbivores when they interact with spiders (Nentwig, 1985).
Communal or cooperative attacks on grasshoppers have been observed in many social spider species, which is a major driving factor maintaining sociality and diversity of spider communities (Yip et al., 2008). Our results suggest that herbivores governing predator richness differed among different predator guilds with feeding specificity, which could promote predator diversity through finely prey resource partitioning (Muller et al., 1999;Peguero et al., 2017). At the family level, we further found that taxon richness of most predators such as Ichneumonidae, Crabronidae, Tiphiidae, Chrysopidae, Anthocoridae, Asilidae, Chironomia and Thomisidae were positively associated with taxon richness of more than one herbivore family. We may speculate that the food web in our study is characterized by a high proportion of generalist predators (Jacquot et al., 2019). Hence, the positive effects of herbivore richness on predator richness may be attributed to a more balanced diet (Dassou & Tixier, 2016) since predator richness benefits from a mixed prey pool under high plant resource conditions (DeMott, 1998). We acknowledge that our conclusion here is based on correlation analyses; thus, more arthropod diversity experiments should be conducted in laboratory and/or in natural conditions to describe the trophic relationships between herbivores and predators.

| CON CLUS IONS
Our previous study from the same experiment demonstrates that plant quantity and quality can independently influence the diversity of arthropod communities, and with different patterns in each taxonomic group (Lu et al., 2021). Our findings in the present study provide new insights into the mechanisms underlying the bottom-up effects of plant quantity and quality on the taxon richness of multiple trophic levels of arthropods in grasslands. For herbivores, the increase in herbivore taxon richness was directly driven by the increase in plant nutrients and indirectly driven by higher herbivore biomass in plots where plant biomass was also greater. For predators, herbivore taxon richness, rather than herbivore biomass, mainly mediated the effects of plant quantity and quality on predator taxon richness. At the feeding guild level, taxon richness of parasitoids, other predators and spiders showed positive responses to different herbivores, which is probably attributed to their different diet preferences. At the family level, the taxon richness of most predator families was positively correlated with that of more than one herbivore family, suggesting that high predator diversity may be attributed to more balanced diets owing to the high prey diversity. Because the taxon richness of herbivores and predators depends largely on plants in natural communities, changes in plant quantity and quality may trigger cascading effects on the multiple services (e.g. pollination and pest control) that arthropods provide to humanity (Haddad et al., 2009;Lewinsohn & Roslin, 2008). Given the important role that plant diversity plays in sustaining arthropod diversity across trophic levels (Dyer & Letourneau, 2003;Haddad et al., 2009;Scherber et al., 2010), future experiments should aim to determine the effects of plant quantity, plant quality and plant diversity on the multi-trophic diversity of arthropods and their linkages to the functioning of grassland ecosystems.

ACK N OWLED G EM ENTS
We thank the staff at the Inner Mongolia Grassland Ecosystem Research Station (IMGERS), Chinese Academy of Sciences for their help in maintaining the field facilities and collecting the field data. This work was supported by the National Natural Science Foundation of China (32192461 and 32192464) and the Grant-in-Aid for Young Scientists A (no. 25712036) to T.S. from the Ministry of Education, Culture, Sports, Science and Technology of Japan, and the Asahi Glass Foundation to T.S.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are available from the Dryad Digital Repository https://doi. org/10.5061/dryad.jq2bv q8c7 (Lu et al., 2022).