Drought-exposure history increases complementarity between plant species in response to a subsequent drought

Experimental design

To test whether an 8-year treatment of recurrent summer droughts would change biodiversity effects and species interactions of grassland plants when facing a subsequent drought event, we grew ambient- vs. drought-selected plants of 12 species in a glasshouse. The plants were grown from seeds collected from 40 plots (Supplementary Data 2) under 8-year treatments of yearly summer droughts vs. ambient precipitation in a biodiversity field experiment in Jena, Germany^11,41.

The Jena Experiment was established in 2002 using a common seed pool of 60 grassland species, with 80 (20times 20,{{{{{rm{m}}}}}}) large plots of species richness levels of 1, 2, 4, 8, 16, and 60 species⁴⁰. Most of the species are perennial and capable of outcrossing (Supplementary Table 1). The Jena Drought Experiment^11,41 was initiated in 2008. Two (1times 1,{{{{{rm{m}}}}}}) subplots were set within each large plot, designated as either drought treatment or ambient control. For the drought treatment, rainout shelters were set up to exclude natural rainfall in mid-summer for 6 weeks. The ambient control treatment got the same shelter construction but rain water was reapplied to not confound the results with artifacts from the shelter⁶⁰. We repeatedly harvested the aboveground biomass per year, once before and once after the summer drought treatment^11,41. The design of the Jena (Drought) Experiment did not allow the exclusion of cross pollination or gene flow between subplots or large plots in the field. Such gene flow may have reduced the possibility for genetic differentiation and for the observed effect sizes of the selection treatment²³. We collected seeds from drought and control subplots throughout the 2016 growing season (Fig. 1). We obtained seeds of 17 species, but only used 12 of them, because the other five species had either few seeds or low germination rates. Seeds per species per selection treatment were collected from 4 to 23 (interquartile range: [8.50, 17.00]) maternal plants distributed across 2–10 (interquartile range: [4.75, 9.00]) large plots in Jena Experiment, in which the functional group richness ranged from 1 to 4 (Supplementary Data 2). The 12 plant species represented four functional groups (grass, small herb, tall herb, and legume) (Supplementary Table 1). The detailed classifications of the functional grouping can be found in the design of the Jena Experiment⁴⁰. Eleven of the 12 species were perennial, and one was annual (Trifolium dubium). The average longevity of the perennial species in the Jena Experiment has been estimated at 3–4 years⁶¹, so that multiple generations and sexual reproduction cycles could occur during the 8-year drought treatment. Although each subplot was small, population sizes of each species were estimated to range from 100 to 1000 individuals m⁻² in ambient and drought subplots at the beginning of the drought treatment in the field⁶².

We germinated the seeds in Petri dishes and transplanted the seedlings into pots in February 2017 in a glasshouse (day temperature range 20–25 °C, night temperature range 15–21 °C, and humidity range 60–80%) at the University of Zurich, Switzerland. Seedlings were planted individually, in monocultures, or in 2-species mixtures in the glasshouse (Fig. 1). In the glasshouse experiment, both monocultures and mixtures contained four plants within a pot. The pots were (11times 11times 11.5) cm in size and filled with soil composed of 50% collected from a sugar-beet field, 25% sand and 25% perlite. We randomly assigned the pots into four blocks in the glasshouse. To test the effects of drought-induced selection on plant traits, we planted individual seedlings of the 12 species in a fifth block. Within the first 2 weeks, dead individuals were replaced, thereafter dead individuals were not replaced anymore. In total, we established 958 pots: 257 pots of mixtures, 217 pots of monocultures, and 484 pots of individual plants (244 pots of individuals in blocks 1–4, and 240 pots of individuals in block 5; Supplementary Methods). For mixtures, there were 21 species pairs (Supplementary Table 1). Species pairs composed of Crepis biennis or Lotus corniculatus had low numbers of replicates (Supplementary Table 1). However, including or excluding these communities produced qualitatively similar results. Thus, we present the results including these two species in this paper. We provide detailed explanations on the choices of species pairs and regarding the biodiversity treatment history in the Jena Experiment in Supplementary Methods.

During a first phase of 3 months in the glasshouse (Fig. 1), pots were watered regularly (“before drought”). After 14–16 weeks, when most of the species had reached peak aboveground biomass, we harvested all individuals in each pot by cutting them 3 cm above the ground, allowing regrowth from the left plant bases (first harvest, “before drought”). The time span for the first harvest included both the time for trait measurements (section “Plant traits” below) and for the immediately following biomass harvest. We completed the biomass harvest of each block within 1–2 days. This allowed us to account for the larger time differences between blocks by fitting block effects in the statistical analyses. After the first harvest of each block, plants were watered regularly and allowed to regrow until the 18th week from planting. This was followed by a second phase of 2 weeks without watering. Soil moisture decreased from more than 40% to less than 10% after 10 days since drought initiation. At the end of the second phase, that is after 20 weeks from planting, we made a second aboveground harvest at 3 cm above the ground (second harvest, “during drought”). During a third phase of 7 weeks, pots were watered regularly again for recovery until most plants reached a new aboveground biomass peak again. At the end of the third phase, that is after 27 weeks from planting, we harvested both above- and belowground plant biomass (third harvest, “after drought”). We checked and confirmed that most plants had reached the full-grown state and peak biomass before each harvest by monitoring their flowering. After each harvest, we cleaned and dried the harvested plant material at 70 °C for 48 h to obtain the dry biomass. We used the aboveground biomass as a proxy for productivity. Although clipping may affect plant responses to the experimental drought in the glasshouse, clipping had the advantage that all plants were “standardized” in height before the experimental drought, thus reducing carry-over effects of differential growth before the experimental drought.

Additive partitioning

We used the additive partitioning approach (Eq. 1)¹⁷ to decompose the net biodiversity effect (NE) on aboveground biomass into the complementarity effect (CE) and the sampling effect (SE):

where (triangle Y) is the NE; ({Y}_{O}) is the observed yield (productivity) in a mixture; ({Y}_{E}) is the expected yield in the mixture, calculated from the observed yield in monocultures and their corresponding species proportions planted in the mixture, here 0.5; the two additive terms at the right side of the equation represent CE and SE, respectively; N is the number of species in the mixture, here 2. The partitioning is based on the observed and expected relative yield (RY) of species in the mixture. The expected RY of species in the mixture is the proportion planted. ∆({{{{{bf{RY}}}}}}) is the difference between observed and expected RY of species in the mixture; (overline{triangle {RY}}) is the average of ∆({{{{{bf{RY}}}}}}). A positive (overline{triangle {RY}}) indicates a positive CE; a positive covariation between monoculture yield (M), and ∆({{{{{bf{RY}}}}}}) indicates a positive SE. More details about the calculation can be found in Loreau and Hector¹⁷. We conducted the partitioning separately for each harvest, selection treatment, and block. We did not perform the partitioning for mixtures with zero biomass⁶³. For monocultures with zero biomass in the second or third harvest, we kept the ones which had positive biomass in the previous harvest but excluded the ones which had zero biomass in the previous harvest. For example, when performing the partitioning for the second harvest, we kept the monocultures that had zero biomass in the second harvest but non-zero biomass in the first harvest; we excluded the monocultures that had zero biomass already in the first harvest. This was to assure that communities that died before the drought could not reappear during or after the drought, and communities that had died during the drought could not reappear after the drought.

We used mixed-effects models to assess the influences of drought vs. ambient-selection treatments on biodiversity effects (NEs, CEs, and SEs) separately for each harvest (Fig. 2; Table 1). Block and selection treatment were set as fixed-effects terms, while species composition (identity of species pair) and its interaction with selection treatment were set as random-effects terms. This conservative approach was used to allow for generalizations across all possible species compositions, although a more liberal approach with species composition and its interactions as fixed-effects terms could also have been applied (see Schmid et al.⁶⁴ for a discussion of defining terms as fixed- vs. random-effects terms, including a justification of preference for treating block as a fixed-effects term). We square-root transformed the CEs and SEs with sign reconstruction (({{{{{{rm{sign}}}}}}}(y)sqrt{y})) prior to analysis to improve the normality of residuals¹⁷. The mixed-effects model did not converge in the analysis with CE after the drought event. In this case, we used a general linear model, in which we fitted block, species composition, selection treatment, and species composition by selection treatment interaction in this order. Then we tested the significance of selection treatment using its interaction with species composition as an error term. This procedure is an alternative to mixed-effects models that estimate variance components for random-effects terms with maximum likelihood⁶⁴.

To test whether biodiversity effects on productivity differed from zero, we additionally tested the significance of NEs, CEs, and SEs separately for each selection treatment and harvest (Supplementary Table 3). We set block and species composition as fixed- and random-effects terms, respectively. The model corresponding to CE for ambient-selected plants during the drought event did not converge so that we fitted it with a general linear model, in which we tested the significance of the overall mean (intercept) using species composition as an error term. All statistical analyses were conducted in R 3.6.3⁶⁵. The mixed-effects models were conducted with asreml-R package 4.1.0.110⁶⁶.

Finally, we also tested whether the effects of drought selection on biodiversity effects (NEs, CEs, and SEs) in the glasshouse depended on the history of biodiversity treatment in the Jena Experiment. Most plants in the 2-species communities in the glasshouse originated from mixtures in the Jena Experiment (Supplementary Data 2; whether mixtures in the glasshouse composed of plants originating from monoculture field plots did not affect the effects of drought-selection on biodiversity effects on productivity (Supplementary Data 3)). To increase statistical power, we used functional group richness, ranging from 1 to 4, instead of species richness of the field plots as explanatory variable (Supplementary Methods). We fitted functional group richness either in linear (Supplementary Data 4) or log-linear (Supplementary Data 5) form. We did not detect significant effects of field treatment of functional group richness nor significant interactions between field treatment of functional group richness and the drought-selection history. Therefore, we excluded the history of biodiversity treatments in the field from further analyses.

Biomass stability to the drought event in the glasshouse

To assess the temporal responses of community aboveground biomass to the drought event, we calculated three indices representing different facets of stability: biomass resistance, recovery, and resilience (see van Moorsel et al.⁴³ for an example). We calculated resistance as the biomass ratio during vs. before the drought, recovery as the ratio after vs. during the drought and resilience as the ratio after vs. before the drought (see also Isbell et al.⁹). We log-transformed the indices (plus a half of the minimum positive value to allow taking logs of indices that were originally zero) prior to statistical analyses to improve the normality of residuals. Excluding index values that were originally zero produced qualitatively similar results.

To assess the effects of drought-selection on biomass stability, we fitted mixed-effects models with block and selection treatment as fixed-effects terms, and species composition and its interaction with selection treatment as random-effects terms (Supplementary Fig. 3; Supplementary Table 4). We fitted the models separately for mixtures and monocultures. We included the log-transformed biomass at the first harvest as a covariate because biomass stability in response to droughts often depends on plant performance under ambient conditions.

In the same way as net biodiversity effects on productivity were calculated for additive partitioning, we calculated biodiversity effects on biomass stability as the difference between each mixture and its corresponding monocultures. Then, we tested the influence of selection treatment on the biodiversity effects on biomass stability. Block and selection treatment were set as fixed-effects terms; species composition and its interaction with selection treatment were set as random-effects terms (Fig. 3; Supplementary Table 5). The log-transformed biomass at the first harvest was also included as a covariate⁴³. To assess the significance of biodiversity effects on biomass stability for each selection treatment, we fitted another set of simplified models, with block and log-transformed biomass as fixed-effects terms, and species composition as random-effects term (Fig. 3).

Neighbor interactions

We assessed interactions between neighboring plants within pots using the metrics of neighbor interaction intensity with multiplicative symmetry (NInt_M)⁴⁴:

where ({P}_{-N}) and ({P}_{+N}) are the productivities without (individual plant) and with neighbors (monocultures or mixtures), respectively; (triangle P={P}_{+N}-{P}_{-N}). Negative values of NInt_M indicate competition and positive values indicate facilitation. NInt_M is bounded between –1 (competitive exclusion) and 1 (“obligate” facilitation). For monocultures, we first calculated the per-plant biomass as the ratio between total biomass and planting density, and then used the per-plant value to compare with the corresponding individuals (without neighbor) of the same species with the same selection treatment in the same block. Note that under the reciprocal yield law⁴⁵, an individual grown alone in a pot should be four times larger than an individual grown with three others in a pot, resulting in a NInt_M of –0.75. For 2-species mixtures, we calculated the per-plant biomass separately for each species and took the average NInt_M of the two species to measure the interaction intensity of the mixture. We set zero biomass for dead plants in the calculation. Again, if mixtures would also follow the reciprocal yield law independent of species identity, then NInt_M = –0.75 would be expected. Values greater than –0.75 indicate some sort of overyielding due to higher density or higher density and higher diversity.

To assess how selection treatment modified interactions between plants, we tested the effects of selection treatment on neighbor interaction intensity separately for monocultures and mixtures. We included block and selection treatment as fixed-effects terms, species composition and its interaction with selection treatment as random-effects terms (Supplementary Fig. 4; Supplementary Table 6).

We calculated the difference between the heterospecific interaction in a mixture and the conspecific interactions in its two corresponding monocultures. A positive value of this difference indicates a weaker heterospecific than conspecific competition (i.e., niche differentiation) or stronger heterospecific than conspecific facilitation, which may lead to a positive complementarity effect. We tested the effects of selection treatment on interaction difference for each harvest by fitting block and selection treatment as fixed-effects terms, and species composition and its interaction with selection treatment as random-effects terms (Fig. 4; Supplementary Table 8). We also tested the significance of the interaction difference for each selection treatment by fitting block and species composition as fixed- and random-effects term, respectively (Fig. 4; Supplementary Table 7).

Plant traits

To assess whether drought selection would change plant traits, we measured six traits (Supplementary Table 9) closely related to plant usages of water or carbon on plants in pots with one individual from blocks 1–5. We focused on the traits on individual plants without neighbor to evaluate the influence of selection treatment on traits without the impacts of plasticity induced by plant interactions. We measured leaf relative chlorophyll content, leaf area (LA), leaf mass per area (LMA) and leaf osmometric pressure before the drought; leaf stomatal conductance both before and during the drought; and dry biomass ratio between root and shoot after the drought (in the third harvest). Leaf relative chlorophyll content was measured for three mature, fully expanded leaves per plant by using a SPAD-502 Plus chlorophyll meter from Konica Minolta. LA was obtained by scanning 3–4 mature, fully expanded leaves per plant with a LI-3100C Area Meter from LI-COR. LMA was calculated as the ratio between leaf dry mass (oven-dried at 70 °C for 48 h, using the same leaves that for LA) and LA. Leaf osmotic potential at full hydration was considered as an important trait associated with plant tolerance to drought³⁰. We measured leaf osmotic potential with freeze-thaw leaf pieces cut from 1 to 2 mature, fully expanded leaves per plant by using a Wescor vapor pressure osmometer VAPRO (Model 5520) according to the method by Bartlett, et al.³⁰. Plants were fully hydrated 1 day before the leaf sampling for osmotic potential measurement. Leaf stomatal conductance is a measure of exchange rate of carbon dioxide and water vapor through the stomata⁶⁷. It was measured for 3–5 healthy mature leaves per plant by using a SC-1 Leaf Porometer from Decagon Devices. For grass species, 3 blades were placed adjacent to each other to have a large enough area for the measurement of stomatal conductance. For stomatal conductance during the drought event, we measured the individual plants from block 5 only due to limited time during the drought phase. We harvested aboveground and belowground plant biomass separately for alive individuals at the end of the experiment (after the complete recovery from the drought). The oven-dried (70 °C for 48 h) aboveground and belowground biomass were used to calculate the biomass ratio between root and shoot. We took the average value of each trait of each plant for statistical analyses. Each trait was measured for each block in turn.

We used linear mixed-effects models to assess the influence (generalized across species) of selection treatment on trait values (red lines in Supplementary Figs. 5–7). Block and selection treatment were set as fixed-effects terms; species and its interaction with selection treatment were set as random-effects terms. Alternatively, we set species, selection treatment and their interaction as fixed-effects terms to assess whether species responded differently to the selection treatment (Supplementary Table 9). To test whether effects of selection treatment on traits differed between the five trait groups (leaf relative chlorophyll content, leaf area, leaf mass per area, leaf osmometric pressure, and leaf stomatal conductance) measured before the drought event in the glasshouse, we conducted two alternative analyses. First, we performed a principal component analysis with all traits and retained the first two principal axes (PC1 and PC2), which accounted for 39.06% and 22.3% of the total variation, respectively. Then we used PC1 and PC2 as response variables in mixed-effect models, separately. We fitted the models with the same fixed- and random-effects terms as those using each separate trait as the response variable. Effects of selection treatment on PC1 or PC2 were not significant. Second, we pooled the five traits as a single response variable in a mixed-effect model (corresponding to multivariate analysis of variance). Block, trait group (a factor with five levels), selection treatment, and the interaction between trait group and selection treatment were set as fixed-effects terms; species and its interactions with trait group and selection treatment and their three-way interaction were set as random-effects terms. We did not detect significant effects of selection treatment nor its interaction with trait group. Therefore, we did not present the results associated with these multivariate analyses in this paper. LMA, LA, leaf osmotic potential, leaf stomatal conductance, and root-shoot biomass ratio were log-transformed to improve normality of residuals.

We also measured leaf relative chlorophyll content, LA and LMA in mixtures before the drought event (Supplementary Table 10) to evaluate the influence of selection treatment on trait dissimilarity between interacting species within communities. We calculated the absolute trait distance between two species in each mixture both separately for each trait and jointly with the three traits. For multi-trait-based dissimilarity, we standardized each trait to mean zero and unit standard deviation and calculated the Euclidean trait distance in standardized three-dimensional trait space.

We used linear mixed-effects models to assess the effects of selection treatment on trait dissimilarity in mixtures (Supplementary Table 10). Block and selection treatment were set as fixed-effects terms; species composition and its interaction with selection treatment were set as random-effects terms. The model for LA dissimilarity did not converge so that we fit it with a general linear model, in which we tested the significance of selection treatment using its interaction with species composition as an error term. For the models with LA, LMA, and the joint three traits as dependent variables, we removed one pot (B1P674) because the LA value of Alopecurus pratensis in this pot was extremely small (about 1/3 of the second minimum value of the same species in mixtures). However, including or excluding this pot produced qualitatively similar results.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Source: Ecology - nature.com