Intraspeci c Trait Variability Determines Understorey Plant Community Assembly

Piotr Olszewski (  thecla@wp.pl ) Nicolaus Copernicus University in Torun: Uniwersytet Mikolaja Kopernika w Toruniu https://orcid.org/0000-0002-8538-4183 Radosław Puchałka Nicolaus Copernicus University in Torun: Uniwersytet Mikolaja Kopernika w Toruniu Piotr Sewerniak Nicolaus Copernicus University in Torun: Uniwersytet Mikolaja Kopernika w Toruniu Marcin Koprowski Nicolaus Copernicus University in Torun: Uniwersytet Mikolaja Kopernika w Toruniu Werner Ulrich Nicolaus Copernicus University in Torun: Uniwersytet Mikolaja Kopernika w Toruniu https://orcid.org/0000-0002-8715-6619


Introduction
Traditionally, studies on plant community assembly have been species centred (Götzenberger et al. 2012) focusing on three important processes, dispersal, competition, and habitat ltering (Callawy and Walker 1997; Maire et al. 2012;Ulrich et al. 2016). This work revealed major habitat dependent trade-offs between these drivers (D'Amen 2018). Additionally, the spatial distribution of individual plants was found to affect competitive interactions and to increase community stability (Ives 1991; Stoll and Prati 2001).
However, species based approaches have well-known shortcomings as they do not cover intraspeci c trait and niche differences (Jung et al. 2010), and might lack generality beyond the focal community (Zakharova et  Increasing evidence points to the major impact of intraspeci c variability in species characteristics on community assembly (e.g. Jung et al. 2010;Bolnick et al. 2011;Siefert et al. 2015;Hausch et al. 2018).
Theoretical studies predict major effects of intraspeci c variability on various ecological processes (cf. Bolnick et al. 2011;Westerland et al. 2021). The trade-off of intra-and interspeci c variability in key traits have been identi ed to determine the probability of local plant competitive success and persistence . Intraspeci c trait variation has also been shown to in uence the spatial distribution In general, we might analyse sources of intraspeci c trait variability from the endpoints of a stochasticdeterministic gradient. The stochastic endpoint includes, for instance, the genetic diversity or phenotypic plasticity, partly re ecting random environmental variation (Forsmann 2015). On the other side, average trait expression might change deterministically along ecological gradients, particularly environmental (Violle et al. 2007). In the latter case the distribution of traits with respect to ecological drivers should be not random. Allometry might have consequences for community assembly. For example, competition theory predicts a negative correlation between resource use in important niche dimensions and community richness (Chesson 2000). However, if high trait variability promotes diversity (Proß et al. 2021;Dallas et al. 2021) these contrasting effects might affect community assembly, possibly masking single effects. Such contrasting impacts of mean trait expression and the respective variance have so far mainly been studied with respect to spatial aggregation of individuals (Yan et al. 2018) and resources (Fajardo and Siefert 2018). Aggregation, equivalent to a high variance in resource distribution, tends to increase local diversity (Hartley and Shorrocks 2002; but see Veech et al. 2003). To our knowledge, the impact of intraspeci c trait variability on community assembly has not been studied so far.
Here, we try to ll these knowledge gaps and link the intraspeci c variability of important functional traits in Polish mixed temperate forest understorey plants to small scale community assembly and functional diversity. We focus on the small balsam Impatiens parvi ora, a dominant species in these forests (Ulrich et al. 2021). As I. parvifora is not only dominant but also one of the largest herbaceous species it locally accounts for more than half of the total understorey plant biomass (Ulrich et al. 2021). Therefore, we expected to see detectable effects of trait expression and trait variability of this species on the abundances, community composition, and species richness of other plants. Particularly, we hypothesise that 1) the variability in important functional traits of I. parvi ora scales according to a power function, 2) this scaling affects community assembly, 3) higher trait variability decreases community diversity, and 4) trait variability is mainly determined by environmental variability. Importantly, we look at small scale community assembly within the interaction horizons of single I. parvi ora rametes. At such spatial scales the impact of interspeci c competition should be most detectable, while environmental lter effects should be comparatively small.

Study sites and sampling
In June and July 2020 understorey plant samples were undertaken in 25 4 × 4 m2 plots each at two seminatural mixed temporal forest sites near Toruń (53.2°N, 18.5°E) in Northern Poland (cf. Ulrich et al. 2021 for detailed descriptions of the study sites). Plots within each site were selected for homogeneous soil and microclimate conditions. For the present study we randomly selected 41 subplots (1 m2 each) from these plots and identi ed all understorey herbaceous species and woody plant saplings. For the study of intraspeci c variability, we chose the dominant small balsam (Impatiens parvi ora), which accounted for 1398 of all 1930 individual plant rametes (from 37 species) in these subplots. Other species were generally too rare within single subplots for a su ciently precise analysis of individual trait variability.
The complete set of raw data is contained in Appendix A and has been uploaded to gshare (to be uploaded after acceptance).

Plant traits and soil characteristics
For each I. pari ora ramet we measured three plant traits (speci c leaf area SLA, leaf dry weight LDW, and stem height SH). Additionally, we counted the total number of ramets NI and the number of owers NF. For all other species we counted total abundances Nother and species richness Sother in each subplot.
We determined ve basic soil parameters using standard methods described in Ulrich et al. (2021): soil water (SWC), total organic carbon (TOC), nitrogen (N) content, the respective C / N ratio, and exchangeable basic cation (Ca2+, Mg 2+ , K + , Na + ) content from topsoil samples taken in the center of each plot. All trait and soil data are contained in Appendix A.

Data analysis
For each subplot we calculated the arithmetic mean µ and total trait expression Σ of the four I. parvi ora traits measured, as well as the respective variances σ 2 , and skewness γ. The skewness is often positively correlated with the variance and quanti es the importance of outliers, a neglected aspect in trait analyses. Trait variability ΔV within each subplot was assessed from eq. 1 and the residuals of observed variance σ o 2 and the variance σ p 2 predicted by eq. 1 .
We compared ΔV and γ for plots having <3 or ≥3 non-I. parvi ora species and <10 or ≥10 non-I. parvi ora rametes We also used general linear modelling to link trait expression, ΔV and γ to soil and community characteristics. Variables were Z-transformed prior to analysis. Errors always refer to standard errors.
We used principal components analysis (PCA) to reduce the dimensions of soil variables. The rst two components (PC1, PC2) accounted for 53.2% and 44.0% of variance and loaded highest with SWC (r = -0.72) and Ca (r = 0.70), respectively. The correlation matrix for all variables used here is contained in Table S1. Bray-Curtis dissimilarity based PERMANOVA is a well-introduced method to assess changes in community composition across one or two categorical variables. Here, we used two way PERMANOVA to assess whether high or low variability in trait expression affects subplot species composition. For this task we used Z-transformed Σ, σ 2 , ΔV, and γ-values with codes 1 for positive and 2 for negative Z-values.

Results
Average values of LDW, SH, and SLA were only marginally correlated with soil characteristics and community size (Table S2). SWC was positively correlated with N I , N other and S other and negatively with N F (Table S2). N I was signi cantly positively correlated with N other explaining 26% of variance in N other (Fig. 2a). In turn, N I and S other were not signi cantly related (Fig. 2b). Numbers N other (Table 1) and also S other (Table S3) were negatively linked to PC1, interpreted as soil water axis, and positively linked to PC2, the soil mineral axis (Table 1). Interestingly, N other was signi cantly negatively correlated with N F (Table  1). Variability in trait values as quanti ed by ΔV and γ did not signi cantly depend on soil characteristics (Tables S4, S5). Soil variables explained less than 10% of variance in ΔV and γ (Tables S4, S5).  Except for SLA, abundances of non-I. parvi ora species signi cantly decreased within increasing variance in I. parvi ora trait expression (Fig. 4). This was not the case with respect to non-I. parvi ora species richness (Fig. S1), where richness and trait variance were not signi cantly correlated. However, ΔV and γ differed between subplots of lower and higher species richness or abundance of non-I. parvi ora species (Fig. 5). Higher ΔV in LDW and N F was associated with increased subplot richness and abundance ( Fig. 5) although these patterns explained less than 10% of variance when accounting for soil covariates (Table 1). Two-way PERMANOVA detected at most marginal effects of I. parvi ora trait variability on community composition ( Table 2). The highest impact had the variance σ 2 of SLA explaining approximately 7% of variance in community composition (Table 3).

Discussion
Our results con rm the general tendency for power function VMRs of ecological variables (Fig. 2). In this respect, Taylor dσ 2 σ 2 = z dμ μ → σ 2 = cμ z . Given that traits values can vary equiprobably between xed upper and lower limits, any increase in average trait will automatically result in a proportional increase in variance. Consequently, we argue that the appropriate null hypotheses of species trait variability is a power function VMR. In the case of linearly equiprobable probability of trait expression between the upper and lower boundaries, the VMR exponent will have a value of z = 2 according to the equiprobable resampling model. This is exactly the pattern reported here with respect to LDW, SH, and N F (Fig. 3), by this answering positively to our rst starting question.
Lack of and also deviations from an allometric VMR, as calculated here by ΔV, point to ecological processes causing different patterns of trait variability. Similar to Ulrich et al. (2021) we found SLA not to obey the allometric VMR (Fig. 3). SLA does not stem from a single measurement but is the ratio of leaf area and dry mass. If area (A) and mass (M) variances scale allometrically to their respective means with similar exponents z, the quotient ( ) in accordance with our nding. Compound traits are obviously not suited for the study of allometric variability and community assembly.
Negative local species co-occurrences have traditionally been interpreted as evidence for interspeci c competition for limiting resources where a mass effect of one species (numbers of individuals) naturally reduces the numbers of other species (Chesson 2000). However, the ubiquity of allometric VMR offers the interpretation that it is not the mean number that is decisive but the variance in the occurrence of these numbers or, as in the present case, the variability in the expression of important function al traits that de ne niche dimensions. As this variability was directly negatively linked to the occurrences in the individuals (Fig. 4), the variability in trait expression among individuals translates into a spatial variability in trait expression. Trait variability as an agent in community assembly has only received minor interest Our third starting question asked whether higher trait variability decreases community diversity. In this respect, Hart et al. (2016) argued that higher trait variability should increase abundances of the dominant species. These authors analysed classical two species competitive models including intraspeci c trait variation and found trait variation to decrease diversity by widening the niche space of the stronger competitor. However, whether such simple models also refer to the more complicated assembly dynamics of multi-species communities remains unclear. Indeed, neutral models (Hubbell 2005;Fridley et al. 2007) predict trait variation to increase local diversity. Moreover, models and empirical studies that include intransitive species competition also predict co-existence (Soliveres et al. 2015, Ulrich et al. 2017).
One mechanism generating competitive intransitivity might be intraspeci c niche variability. Our results (Fig. 4) (Fig. 4) but these trends were not accompanied by a respective decrease in species richness as predicted by theory (Fig. S1). Trait variability was also not signi cantly related to community composition ( Table 2). As both of our study forests are old-grown with comparatively low anthropogenic in uence successional dynamics should not in uence our results. Further, abundances of I. parvi ora and the abundances and species richness other all other species were positively correlated (Fig. 2). Therefore, we conclude that the potential negative impact of lower abundances due to a high trait variability of the dominant species does not signi cantly impact local richness. Apparently, counterbalancing mechanisms exist that stabilize diversity. One such mechanism might be the constant colonization input from the surrounding forest plants making local communities dynamically stable despite of marked dominance orders. We have to remember that our plots were arti cial units in a continuous matrix of plants.
Interestingly, deviations in abundance and species richness from the null expectation were in contrast to the trait variance -abundance pattern showing that deviations from a statistical standard (Fig. 4) might contain additional information not obvious from the patterns in raw data (Fig. 3). This fact is well known in community assembly theory (Ulrich and Gotelli 2013). Higher I. parvi ora trait divergence ΔV was associated with higher local abundance and species richness (Fig. 4). Again, this nding contradicts the models of Hart et al. (2016) and Dallas et al. (2021). Importantly, higher diversity and abundances were linked to higher skewness of the trait distributions of LDW and N F . To our knowledge the in uence of trait skewness has so far not been studied with respect to community assembly. Our results call for new theoretical and empirical approaches to the dependence of local community pattern on trait variability. Such approaches should include the study of traits of all species and not only of dominant once.
With regard to our fourth starting question we did not observe that trait expression and the respective variances were signi cantly determined by averaged values of soil factors (Figs. S2, S4). In our study sites higher water content correlated negatively and higher mineral content positively with abundance and species richness (Tables 1 and S3). Therefore, we expected to see respective in uences of soil characteristics on trait expression as soil nutrient availability is known to in uence trait expression and spatial variability either by plant plasticity (e.g. Borowy and Swan 2020) or genetic differentiation (e.g. Born and Michalski 2019). Apparently, our study plots located in two forests were environmentally too homogeneous to reveal marked in uences of soil factors. Consequently, we argue that habitat lter effects should not signi cantly in uence the above results on trait variability.
Our study has two major shortcomings that do not allow for a more functional interpretation of our results. First, due to the fact that we studied only one vegetation season, temporal species turnover of the subdominant species was not assessed. Even in seemingly stable semi-natural old-grown forest temporal turnover might be an important factor in small scale community assembly, thus masking the effects of competitive superiority and spatial distributions (Ulanova 2000). Second, due to limitations in manpower, our studies was done at a one m 2 resolution (still including 1398 single rametes to be analysed morphometrically). A higher resolution and a correspondingly increased sample size might have revealed species-and ramete-pair trait relationships. Such an approach should directly reveal competitive interactions and, applied to the whole community, should allow for the construction of species ×species competitive interaction matrices, necessary to reveal multispecies competitive hierarchies and loops (Soliveres et al. 2015). We call for respective individual based studies of local plant trait and species cooccurrences (cf. Ulrich et al. 2021).

Declarations
Author contribution WU devised the study, analysed the data and wrote the draft text. PO did the eld and laboratory work, the image processing, and nalised text. PS analysed the soil data. RP and MK provided plant trait data and conceptual input. All authors contributed to the nal text.
Con ict of interest The authors declare that they have no con ict of interest.
Ethics approval Not applicable.
Consent to participate Not applicable.
Consent for publication All authors read the nal manuscript version and agreed to publish the data contained in this study.
Funding This work was supported by a grant from the Polish National Science Center (2017/27/B/NZ8/00316).
Availability of data All raw data contained in this study are contained in the supplementary le Appendix A and B.