Litter conditioning phase
Six grass species were selected to represent fast- (Arrhenatherum elatius L. (P. Beav.), Holcus lanatus L., Lolium perenne L.) and slow- (Agrostis capillaris L., Deschampsia flexuosa L. (Trin.), Festuca ovina L.) growing species (Elberse and Berendse 1993; Heinen et al. 2020; Scheurwater et al. 2002). Despite grasses having inherently lower concentrations of secondary metabolites (Geisen et al. 2022) and generally lower molecular richness (Defossez et al. 2021) as compared to forbs and the fact that these compounds are often affected by herbivores (Gatehouse 2002; Kaplan et al. 2008; Karban 2011; van der Putten 2003), we opted to investigate only grass species because grasses are the dominant functional group in grassland systems. Seeds were obtained from Cruydt-Hoeck (Nijeberkoop, the Netherlands) or Pratensis AB (Lönashult, Sweden). The grasses were sown on 15 November 2017 directly into 5-litre plastic pots in soil that consisted of 90% gamma irradiated soil (Synergy Health, Ede, The Netherlands) characterised as holtpodzol sandy loam (84% sand, 11% silt, 2% clay, ~ 3% organic matter, pH 5.9, 1150 mg kg− 1 N, 61 mg P2O5 100 g− 1, 2.4 mmol K kg− 1) collected from a grassland near Lange Dreef, Driebergen, The Netherlands (52° 02’ N, 5° 16’ E) and 10% live field soil that was collected from a restored grassland site abandoned from agricultural use in 1996 (“De Mossel”, Ede, The Netherlands, 52° 04´ N, 5° 45´ E) characterised as holtpodzol sandy loam (94% sand, 4% silt, 2% clay, ~ 5% organic matter, 5.2 pH, 1060 mg kg− 1 N, 75 mg P2O5 100 g− 1 P, 1.9 mmol K kg− 1). After the seeds sprouted, each pot of grass seedlings was thinned to obtain a similar visual density. During the entire growing period, the grasses were grown in a glasshouse with climate control (light regime 16:8 h day:night, day temperature 21 °C, night temperature 16 °C) and watered and weeded as needed. D. flexuosa was re-sown due to poor germination rates. Treatments consisted of aboveground herbivory, belowground herbivory, both above- and belowground herbivory and a no herbivory control (n = 5 for each treatment per plant species, yielding a total of 120 pots). All plants were grown in hanging plastic mesh sleeves (BugDorm, Taiwan) for the duration of the herbivory treatment.
Belowground herbivory: The highly polyphagous root-feeding larvae of click beetles (Traugott et al. 2008) (c. 75% Agriotes lineatus and c. 25% A. obscurus), were chosen as the belowground herbivore (hereafter: wireworms). The wireworms were collected near Lelystad (52° 54' 50.35" N, 5° 53' 68.28" E) in marine sandy loam (c. 7% clay) a few weeks before the start of the experiment and stored at 4 °C until they were used in the experiment. On 20 December 2017, four holes approximately 2–3 cm deep were made in each corner of each pot receiving the belowground herbivory treatment and one wireworm was placed into each hole and covered with soil. Holes were also made in the same manner as described above in the remaining pots (i.e., those that did not receive the belowground herbivory treatment) in order to control for artefact effects. Addition of wireworms to D. flexuosa was delayed by 2 weeks due to re-seeding.
Aboveground herbivory: Caterpillars of the highly polyphagous cabbage moth (Mamestra brassicae), which is ubiquitous within the grasslands from which the plants chosen here originate (Wu et al. 2015), were placed on the plants receiving the aboveground herbivory treatments. The eggs from M. brassicae were obtained from the Department of Entomology at Wageningen University, The Netherlands. The colony has been maintained for years on Brassica oleracea var. gemmifera cv. cyrus and the larvae were originally collected from a cabbage field near Wageningen. Previous work has shown that M. brassicae performs well and sometimes even prefers grasses over forbs (Heinen et al. 2020). Upon hatching, the larvae were reared in separate groups of 200–300 larvae and provided with artificial diet (140 g agar dissolved in 5 L of boiling water with addition of 800 g maize flour, 250 g beer yeast, 250 g wheat germs, 10 g sorbic acid, 8 g nipagin (methyl-4-hydroxybenzoate), 40 g ascorbic acid and 0.5 g streptomycin), which was regularly refreshed. Caterpillars were placed on the plants in three successive rounds to ensure the plants were sufficiently damaged and thereby the quality of the litter they produced was affected. In the first round of herbivory (20th December 2017), two early L3 larvae were selected and placed on the grass monocultures using a fine-hair brush. In the second round of herbivory (26 December 2017), five late L1 larvae were added to the monocultures, followed by a third round of herbivory (3rd January 2018) in which an additional five L2 larvae were added to the monocultures. Larvae at different stages were added to increase the chances of successful establishment. Addition of caterpillars to D. flexuosa was delayed by 2 weeks due to re-seeding.
About one month after the herbivory treatments were initiated (19th January 2018), aboveground damage to the plants was assessed visually as an estimate of total surface area consumed by the caterpillars, and expressed as a percentage of the total surface area. (Note: Both M. brassicae and Agriotes spp. larvae were still present on the plants and under the ground, respectively.) Assessment of D. flexuosa was delayed by 2 weeks. Regrowth of biomass after herbivory was substantial in A. capillaris and F. ovina, whereas in H. lanatus, A. elatius and L. perenne, regrowth was comparatively less. As a result, in A. capillaris and F. ovina, the estimates of herbivory were comparatively low. Aboveground herbivore visual estimation of damage on the plants exposed to aboveground and above-belowground herbivory ranged between 5–25%. Respectively, aboveground and above-belowground herbivory damage values were as follows: A. capillaris: 5 ± 0% and 5 ± 0%; A. elatius: 16 ± 2% and 15 ± 3%; D. flexuosa: 9 ± 2% and 13 ± 3%; F. ovina: 9 ± 2% and 9% ± 2%; H. lanatus: 21 ± 2.2% and 17 ± 7%; and for L. perenne 8 ± 1% and 10 ± 2%. No aboveground herbivore damage was observed on plants from the control and belowground herbivory treatments. It was not possible to assess belowground herbivore damage, but given that Agriotes spp. are polyphagous and voracious feeders (Hermeziu 2021; Traugott et al. 2008), it is very likely they caused severe damage to the roots of the plants in the belowground and above-belowground herbivory treatments.
After damage was assessed, we stopped watering the plants so that they senesced and their litter could be collected. Again, for D. flexuosa this was delayed by 2 weeks. It is well known that drought causes changes to the chemical composition of plant root and shoot litter (He and Dijkstra 2014; Varela et al. 2016). However, obtaining the litter for this experiment via drought was the only feasible way to ensure the production of enough dead litter in a timely manner. After c. 3 weeks (5 February 2018) the plants were fully senesced and root and shoot litter were harvested. A subsample of litter from each plant was taken, freeze dried and set aside for elemental analyses (see below). Shoots were carefully detached from roots and placed in a paper bag. Roots were then washed clean and left to air-dry overnight, then placed in a paper bag. Both roots and shoots were oven-dried at 40 °C for a minimum of 72 h, and weighed to determine total biomass. Six randomly selected subsamples of 0.5 g of both shoot and root litter were collected from each of the 120 pots. These samples served as litter sources in the decomposition phase of the experiment.
Litter chemistry analyses
In order to make mechanistic links between litter properties and changes to plant growth during the litter feedback phase (see below), analyses on litter chemistry were performed. After harvest, a subsample of the shoot and root tissue from each pot was also analysed for total carbon (C) and nitrogen (N) content. Each subsample was ground with a ball mill (Schwingmühle Qiagen Tissue Lyser II, Hilden, Germany), placed into a tin capsule and then analysed using a Flash EA1112 elemental analyser (Thermo Fisher Scientific, Inc., Waltham, MA, USA). An additional subsample of ground shoot and root tissue was analysed for total polyphenolic concentrations using the Folin-Denis method (Folin and Denis 1915; Hagerman and Butler 1989). Briefly, 25 mg of freeze-dried root and shoot litter was extracted in 5 ml of a 50:50 2.4 M HCl:MeOH solution heated to 90 °C for 2 h. Extracts were then centrifuged for 10 m at 5000 rpm and the top 2 ml pipetted into an Eppendorf tube and stored at 20 °C until analysis. Upon analysis, extracts were warmed to room temperature. Then, 200 µL was placed into a 2 mL Eppendorf tube along with 200 µL of Folin-Denis reagent and 1 mL of 1.6 M sodium carbonate. Tubes were vortexed for 10 s and then allowed to incubate on the lab bench for 30 m before being centrifuged for 5 min at 14000 rpm. The top layer of liquid was pipetted into a 96-well plate and absorption was read at 750 nm using on plate reader with Gen5 software (version 1.11.5, BioTek Instruments, Inc., Winooski, Vermont, USA).
Further, a subset of root litter samples (four herbivory treatments × six grass species × three replicates = 72) were analysed for (micro)nutrient concentrations. Root material was oven dried at 70°C for at least 48 h. Next, 20 mg of dried root material was transferred to glass digestion vials (MG5, Anton Paar GmbH). A mixture of 250 µL 69% HNO3 and 125 µL of 30% H2O2 was added to each vial. The vials were closed with special PEEK screw caps (MG5, Anton Paar GmbH) and disposable PFTE lip-type seals (Anton Paar GmbH) capable of tolerating high temperatures and pressures. Sample digestion was carried out in a microwave oven (Multiwave ECO, Anton Paar GmbH) mounted with a 64-position rotor (64MG5, Anton Paar GmbH). A 10 min ramping period was used to a maximum temperature of 140°C. The samples were kept at this temperature for 80 min after which the digested samples were left to cool for 10 min. The samples were then transferred to a − 20°C freezer for 30 min followed by the quick release of the pressure of all samples. This cooling step prevents the loss of volatile elements such as S. Finally, samples were diluted with Milli-Q water to a final concentration of 3.33% HNO3 and filtered using a Whatman Puradisc Aqua 30 filter with CA membrane. Samples were then analysed for Aluminum, (Al), Copper (Cu), Iron (Fe), Potassium (K), Manganese (Mn), Sodium (Na), Nickel (Ni), Phosphorus (P), Sulfur (S) and Zinc (Zn) by an inductively coupled plasma-optical emission spectrometer (ICP-OES, Thermo Scientific iCAP 6500 Duo Instrument with axial and radial view and CID detector microwave digestion system).
Litter feedback phase
On 5–8 March 2018, 1 L pots were filled with 1 kg of soil (a mixture of 90% gamma-irradiated soil and 10% live soil; live soil was collected from the field on 5 March 2018; see above). Using a randomized block design, pots were placed in the glasshouse under the same conditions as mentioned above and watered freely to allow the soil to settle and microbial activity to re-establish. On 12 March 2018, the collected six 0.5 g litter subsamples from all 120 pots from the herbivory treatment phase were placed into individual pots and allowed to decompose in preparation for the litter feedback phase. This resulted in a design as follows: four insect litter legacy treatments (aboveground herbivores, belowground herbivores, both above- and belowground herbivores, or no herbivores) ⋅ six ‘litter’ grass species (A. capillaris, A. elatius, D. flexuosa, F. ovina, H. lanatus, L. perenne) ⋅ two litter types (shoot, root) ⋅ six ‘response’ grass species (A. capillaris, A. elatius, D. flexuosa, F. ovina, H. lanatus, L. perenne) ⋅ five replicate blocks = 1,440 pots. In addition, each block included two control pots containing no litter ⋅ six response grass species (A. capillaris, A. elatius, D. flexuosa, F. ovina, H. lanatus, L. perenne) ⋅ five replicate blocks = 60 no litter control pots for a total of 1,500 pots (Fig. 1). Due to limited litter production of some species (i.e., D. flexuosa and F. ovina), some replicates were lost, leaving a total of 1,404 pots. Litter was placed onto the surface of each pot and gently pressed into the surface of the soil and then approximately 2 cm of fine quartz sand was placed on top of the litter in order to ensure that the litter was full covered and in contact with the soil substrate below. This helped to retain moisture to ensure decomposition took place and prohibited oviposition into the pots by fungus gnats (superfamily Sciaroidea). Pots were watered as needed over the next three weeks to ensure adequate moisture for decomposition.
On 15 March 2018, seeds from the six test species were sown onto sterilised glass beads, watered thoroughly, and placed into the glasshouse (same conditions as mentioned above) to allow for germination. Once species grew large enough for transplantation (approximately 2–3 cm in height), they were moved to the cool room and kept at 4 °C to arrest further growth. On 5 April 2018, a single seedling of the six grass species mentioned above was planted into each pot. All pots were checked daily and watered as necessary and dead seedlings were replaced up until ten days after the initial planting. After the last replacement of dead seedlings, a total of 1.1% of the plants died by the end of the experiment. Beginning on 16 April 2018, each pot was watered every other day with 50 mL of tap water until the harvest of the experiment.
Between 22 and 28 May 2018, the experiment was destructively harvested. Shoots were clipped at the meristem and placed into a paper bag on the first day of the harvest. Pots were subsequently stored in the dark at room temperature until the roots could be washed. Roots were carefully washed clean of soil over a 4 mm sieve and placed on a paper towel overnight to air dry before they were placed into a paper bag. Remaining root litter from the litter treatments was separated from live roots (i.e., the colour and texture were different), while virtually all the shoot litter had decomposed during the course of the experiment and therefore posed no issue. All roots and shoots were placed into an oven and dried for a minimum of 72 h at 40 °C before dry weights were measured. (NB: for the species D. flexuosa when it was grown with root litter, there were numerous instances where the plants and their root systems were so small that it was not possible to disentangle the roots from the litter. In this case, root measurements could not be taken, but shoot measurements were still recorded. Further, due to contamination in the D. flexuosa seed batch, 12 individuals were actually a Poa spp. instead. These plants were dropped from subsequent analyses.)
Statistical analyses
All plant data from the glasshouse experiment were analysed using mixed effect models. Data collected after the litter conditioning phase on root and shoot litter characteristics carbon, nutrients and total polyphenols and root litter nutrients (i.e., Al, Cu, Fe, K, Mn, Na, Ni, P, S and Zn) were analysed in two models. To test for effects of herbivory on litter quality and plant growth speed on litter properties, the first model included herbivory (control (litter exposed to no insect herbivory), aboveground herbivory (Mamestra brassicae), belowground herbivory (Agriotes spp.), above- and belowground (Mamestra brassicae + Agriotes spp.)) and growth speed of the species from which the litter was obtained (fast versus slow) as fixed factors. The second model tested for the effects of herbivory and litter species identity on litter properties (i.e., the different grass species from which litter was derived), with both considered fixed factors. In both models, block (i.e., the randomized block design into which all the pots were placed in the glasshouse) was included as a random factor and all interactions were specified.
In order to standardize the response species data collected during the litter feedback phase, the root and shoot biomass of each response plant of each species was subtracted from the average respective root and shoot biomass of the no litter controls for the same species (i.e., two no litter control per species present in each of the five blocks for a total of ten no litter control units across all five blocks). For example, the root biomass of a L. perenne grown with litter exposed to belowground herbivory was subtracted from the average root biomass of the ten L. perenne grown without added litter designated for comparison to plants grown with root litter. To investigate our four hypotheses, we created three different models. All models allowed us to test our first hypothesis on the effects of above-belowground herbivory on plant growth response. In the first model, herbivory (as described above), litter type (root versus shoot litter) and response compartment (root or shoot of response the plants) were included as fixed factors, with all interactions specified. The response variable was the standardized value (see above) of the roots and shoots of the response plant; both root and shoot responses were included in the same analysis. Therefore, to account for autocorrelation, the sample identity (i.e., individual from which a particular pairing of root and shoot measurements originated) was included as a random factor. Block, litter species identity and response species identity (i.e., the different grass species that were used as response species) were also included as random factors. This model allowed us to investigate our second hypothesis regarding interactive effects between root versus shoot litter exposed to herbivory on the different compartments (i.e., roots versus shoots) of the response plant species. In the second model, herbivory, litter growth speed and response species growth speed (fast versus slow) were included as fixed factors, with all interactions specified. Random factors were as specified in the first model. The response variables were the standardized responses of the roots and shoots of the response plants. Plant roots and shoots that were grown with root versus shoot litter were analysed separately, resulting in four analyses: standardized responses of roots and shoots grown with root litter and standardized responses of roots and shoots grown with shoot litter. This model allowed us to interrogate our third hypothesis regarding the effect of plant growth speed on the PLSF pathway. In the third model, herbivory, litter species identity and response species identity were included as fixed factors and block was included as a random factor. Response variables were the same as in the second model. This allowed us to investigate our fourth hypothesis regarding species-specific effects of litter and herbivory on the response plants. In the three models described immediately above, herbivory was originally analysed as a binary response (i.e., aboveground herbivory yes/no, belowground herbivory yes/no), but these models generated results that were not drastically different from the models that considered herbivory as a single variable with four categories. Therefore, for the sake of simplicity and ease of interpretation, the above models were used.
All of the models described above included an a priori selection of random factors based on the experimental design. However, all possible combinations of random factors listed for each model were compared using the AICcmodavg package in R (Mazerolle and Mazerolle 2017) and the best selection of random factors for each response variable was selected. Please see footnotes in ANOVA Tables. All data were transformed as necessary to meet the model assumptions; see ANOVA tables for details. Restricted maximum likelihood (REML) estimation was used to produce an unbiased estimate of variation and covariation (Patterson and Thompson 1971) and Kenward-Roger degrees of freedom approximation was used to reduce bias introduced by a relatively small sample size (Kenward and Roger 1997). Analyses were performed using R software (R Core Team 2015) with the packages lme4/lmerTest (Bates et al. 2015; Kuznetsova et al. 2017).