Phenotypic plasticity vs. local adaptation in quantitative traits differences of Stipa grandis in semi-arid steppe, China

Whether plants are able to adapt to environmental changes depends on their genetic characteristics and phenotypic plastic responses. We investigated the phenotypic responses of 7 populations of an important dominant species in semi-arid steppe of China - Stipa grandis, and then distinguished which adaptive mechanism(s), phenotypic plasticity or local adaptation, was/were involved in this species to adapt to environmental changes. (1) All traits were significantly influenced by the interaction of population and growth condition and by population in each condition, and inter-population variability (CVinter) was larger in the field than in the common garden for 8/9 traits, indicating that both phenotypic plasticity and genetic differentiation controlled the phenotypic differences of S. grandis. (2) From a functional standpoint, the significant relationships between the values of traits in the common garden and the environmental variables in their original habitats couldn’t support local habitat adaptation of these traits. (3) Low CVintra, low quantitative differentiation among populations (QST), and low plasticity shown in the western populations indicated the very low adaptive potential of S. grandis to environmental changes. (4) From the original habitats to the common garden which is far away from S. grandis distribution region, positive phenotypic responses were found in several populations, indicating that some original habitats have become unfavorable for S. grandis.

selection of environmental changes [11][12][13] . What's more, how a trait varies within and among populations is critical to determine the potential of a species/population to perform along environmental gradients 8 .
Due to climate changes and anthropogenic activities, steppes are becoming fragmented and degraded, especially in arid and semi-arid areas 14 . Comparing with tree species 2,[15][16][17][18] , only a few studies paid attentions to phenotypic plastic responses of steppe species and local adaptation is less common 10,19,20 . Given steppe species could not adapt to the rapid environmental changes, their distributions will be greatly influenced. Therefore, more studies are needed to know about the phenotypic plastic responses of steppes species to environmental changes in order to protect the structures and functions of grassland communities.
Stipa grandis steppe is the most common, representative and stable community of typical steppe in Euro-Asian Steppe 21 . However, the distribution region of S. grandis has rapidly changed due to fragmentation and degradation by climate changes and anthropogenic activities in the past decades, showing a pattern of eastward migration 22 . Mode of reproduction can influence distribution shifts by affecting evolutionary potential and dispersal capacity 3 . S. grandis is self-compatible 23 , therefore, the fragmentation and degradation of habitats would enhance its inbreeding and enlarge population genetic drift, decrease population genetic diversity, then affect its evolutionary potential to environmental changes 24 . In our pervious study, amplified fragment length polymorphism (AFLP) markers were used to analyze its population genetic characteristics based on 7 populations across its distribution region in semi-arid steppe of China 25 . In this study, exactly the same 7 populations (Table 1) were chosen to analyze its phenotypic plastic responses because the combination of genetic and phenotypic analysis could help us to forecast a species' full potential to adapt to rapid environmental changes 26 . Nine quantitative traits of individual plants in these populations were measured in a field (in situ) and in a common garden, and the environmental variables including geographic and bioclimatic variables in their original habitats were collected to use for traitenvironment correlation analysis. In order to test the evolutionary potential and distinguish phenotypic adaptive mechanism(s) of S. grandis to the environmental changes across its main distribution region, we tested reaction norms of these quantitative traits from their original habitats to the common garden, estimated traits differences among populations in each condition, calculated intra (inter) -population variability (CV intra and CV inter ) for every trait, related the values of quantitative traits with the environmental variables in their original habitats in each condition, and calculated quantitative differentiation among populations (Q ST ) for every trait examined in the common garden.

Results
Principal component analysis (PCA) for bioclimatic variables. The first 2 principal components summarized 89.86% of the overall variation among the 19 layers. PC-1 and PC-2 explained 53.56% and 36.12% variance, respectively (Table 1). PC-1 could be thought of as precipitation component because the variables with high loadings (> 0.9) on PC-1 were annual precipitation, precipitation of wettest and driest month, precipitation of wettest and warmest quarter, and PC-2 could be thought of temperature component because the variables with high loadings (> 0.9) on PC-2 were isothermality, temperature seasonality, and mean temperature of coldest quarter (Table 1).
Phenotypic plasticity and reaction norms. The phenotypic plastic responses of different S. grandis populations were expressed by their slopes from original habitat to the common garden and their plasticity was shown by the absolute values of the slopes. The interaction of population (P) and growth condition (C) showed significant effects (P < 0.05) on all 9 traits, that is to say, there were significant different reaction norms among populations (Figs. 1A-I), indicating that different S. grandis populations showed different phenotypic plasticity to adapt to the changing conditions and that there was a genetic basis for their phenotypic plasticity. Both negative and positive responses were found for 6 traits ( Three growth related traits showed significantly (P < 0.01) positive relationships with longitude, height of reproductive shoot showed a significantly negative relationship with PC-1 score (precipitation component) (P < 0.05), and height of vegetative shoot and length of the maximum leaf showed significantly negative relationships with altitude (P < 0.05). No significant relationships were found between seed related traits and any environmental variable (P > 0.05) ( Table 2). In addition, non-significant relationships were found between field-quantitative and geographic distances (R 2 = 0.022, P = 0.302) ( Fig. 2A), between field-quantitative and climatic distances (R 2 = 0.113, P = 0.090) by Mantel's tests (Fig. 2B).
CV intra ranged from 0.065 to 0.192 and CV inter ranged from 0.070 to 0.264. CV intra was a little lower than CV inter for all traits, with significant differences (P < 0.001) for 3 growth related traits and non-significant differences for 6 seed related traits (Table 3).
Phenotypic differences in the common garden. All quantitative traits examined in the common garden showed significant differences (P < 0.05) among populations (right in Figs. 1A-I). Q ST of these traits ranged from 0.033 to 0.274 (Table 3).
CV intra ranged from 0.053 to 0.166 and CV inter ranged from 0.063 to 0.227. CV intra was a little lower than CV inter for all traits, with significant differences (P < 0.05) for height of reproductive shoot, length of the maximum leaf, seed, and the first segment of awn (Table 3). CV inter in the common garden was lower than in the field for all traits except the height of reproductive shoot (Table 3).

Discussion
From original habitats to common garden, reaction norms of S. grandis were significantly (P < 0.05) different for all 9 traits as shown by the significance of the interaction of population and growth condition, traits differences among populations were significant (P < 0.05) in each condition (Fig. 1), and CV inter in the field was larger than in the common garden for 8/9 traits (Table 3). These results indicated that both phenotypic plasticity and genetic differentiation controlled the phenotypic differences of different S. grandis populations and suggested the genetic basis of phenotypic plasticity of S. grandis 11,27 . But, we did not provide determination proofs for local adaptation (adaptive genetic changes) of S. grandis populations although some significant trait -environment relationships were found. For example, regarding 9 quantitative traits measured in the common garden, five traits showed significantly negative relationships with latitude and 7 traits showed significantly positive relationships with PC-2 score (temperature component) by Spearman's correlation analysis (Table 2). Moreover, significant relationships were found between common garden-quantitative and geographic (climatic) distances by Mantel's tests (Fig. 2). From a functional standpoint, smaller sizes may be favored in drier habitats, as growth related traits tested in the field were shown ( Fig. 1; Table 2), because smaller leaves provide less surface area for transpiration water loss and smaller organ and plant size can reduce developmental time 26,28 . However, the significant relationships mentioned above suggested that the organs or plant sizes of S. grandis increased with the increase of the temperature. That is to say, S. grandis had larger organs or plant sizes in relatively drier habitats ( Fig. 1; Tables 1 and 2). Therefore, these significant trait-environment relationships did not show ecologically meaningful trends to support that local adaptation (adaptive genetic changes) helped S. grandis populations to adapt to their local conditions. Maladaptive or non-adaptive genetic changes could occur as a result of genetic drift or founder effect, or as a result of stress, nutrient limitation 7 . Both fragmental habitats and distribution shifts could contribute to non-adaptive genetic changes by increasing population genetic drift or environmental stress. In recent decades, because of less raining and intense human activities, S. grandis communities were fragmented and degraded, and as a result, they were replaced by other communities (e.g. S. krylovii community) and the distribution region of S. grandis has eastward shifted 22 . In the present study, the common garden site was chosen as an unfavorable or a hostile condition because it is beyond the distribution region of S. grandis 22 . From their original habitats to the common garden, the eastern populations, such as Bayantuhai, showed negative phenotypic plastic responses for most traits (negative slopes in Fig. 1), demonstrating that the common garden condition was not as favor as their original habitats; however, populations from the western region, such as Bieligutai and West-Xilinhot, showed positive phenotypic plastic responses for most traits (positive slopes in Fig. 1), indicating that their original habitats were more unfavorable (hostile) than the common garden condition. These results provided some proofs for the possibility of maladaptive or non-adaptive genetic changes affecting S. grandis' phenotypic difference among populations as well for the eastward shift of S. grandis distribution region 7,22 . In addition, according to the theory of Merilä and Crnokrak 29  Variability within and between populations could also help plant to track environmental changes. In the present study, several CV intra showed significantly lower than CV inter (Table 3), and both CV intra and CV inter were relatively lower than other species reported 8,10 . Furthermore, plasticity of the western populations of S. grandis was lower than that of the eastern populations. The most eastern population -Bayantuohai had the highest absolute values of slopes for all traits (Fig. 1). Besides, compared with other outcrossing or perennial grasses, S. grandis had a relatively low population genetic diversity 25 . Relatively low CV intra of quantitative traits, low plasticity of some populations and low population genetic diversity would seriously hamper the adaptive capacity of S. grandis to environmental changes, such as climate changes and intense anthropogenic activities.
Summarily, phenotypic plasticity rather than local adaptation (adaptive genetic changes) played an important role in helping S. grandis populations to adapt to environmental changes. Bearing in mind non-adaptive genetic changes and low adaptive capacity of S. grandis populations, some measures should be carried out to protect their habitats in order to decrease environmental stress or unfavorable/hostile environmental conditions, and then gradually decrease population genetic drift and enhance population genetic diversity, finally improve population evolutionary potential to environmental changes and maintain ecological functions of the communities.

Materials and Methods
Species and sampling sites. S. grandis is the most important dominant and constructive species of the climax community in semi-arid steppe of China, therefore, its distribution shifts and population changes have great effects on community structure and function. We have studied its genetic characteristics based on 7 S. grandis populations which covers its main distribution region (115-120°E, 43-50°N), and in this study, we selected exactly the same 7 populations to analyze their phenotypic plastic responses from their original habitats to a common garden. A detailed description of sampling sites could be found in Wu et al. 's literature 25 .   (Table 1) were analyzed for this study. These bioclimatic variables could be obtained from the WorldClim database freely by geographical coordinate 30  Statistical analysis. Quantitative data meet assumption of normality and homogeneity of variance, therefore, they do not have to be transformed before data analysis. First, two-way analysis of variance (IBM, Armonk, NY) was conducted to investigate the effect of block on values of quantitative traits examined in the common garden, with block and population as fixed factors. Results showed that values were not influenced by block and population × block interaction. Therefore, we did not have to think about the block factor when we analyzed data examined in the common garden. Second, in order to examine the differences of phenotypic plasticity and reaction norms among populations, two-way analysis of variance (IBM, Armonk, NY) was conducted to investigate effects of population, growth condition and their interaction on quantitative data, with population and growth condition as fixed factors. Third, based on significant interactions of population and growth condition, we further analyzed trait differences among populations in each condition (field or common garden) by one-way analysis of variance (IBM, Armonk, NY), and got within-population variance (σ w ) and between-population variance (σ P ), then calculated quantitative differentiation among populations (Q ST ) by formula Q ST = σ P 2 /(σ P 2 + 2σ w 2 ) 29,32 . Fourth, intra-population variability (CV intra ) and inter-populations variability (CV inter ) were calculated as the ratio of SD within population to population mean and the ratio of SD between population to overall mean, respectively. "SD" is the abbreviation of "standard deviation". Significant difference between CV intra and CV inter was tested by one sample t-test, with CV inter as test value (IBM, Armonk, NY).
In order to reduce dimensionality from initial 19 bioclimatic variables by geographical coordinate, principal component analysis (PCA) was used, and variable factor loadings, cumulative proportions of the total variance and scores of the first 2 principal components for each population were calculated (IBM, Armonk, NY). Furthermore, relationships between values of quantitative traits and the environmental variables in their original habitats were analysed by Spearman's correlation analyses (IBM, Armonk, NY). It should be noted that environmental variables included geographic data (longitude, latitude, altitude) and climatic data (the first two principal components scores for 19 bioclimatic variables) in this study.
Population pair-wise distance matrix based on 19 bioclimatic variables or quantitative data collected in each condition were calculated by Euclidean's distance coefficient after standardization of data, respectively (IBM, Armonk, NY). Population pair-wise geographic distance matric were estimated in Google Earth. Relationships between quantitative and geographic distances, and between quantitative and climatic distances were examined by Mantel's tests (3000 permutations) in NTSYS-pc software 33 .  Table 3. Intra -population variability (CV intra ) and inter-population (CV inter ) of 9 quantitative traits of different S. grandis populations measured both in the field and in the common garden and quantitative differentiation among populations (Q ST ) of these 9 traits measured in the common garden.