Skip to main content

Data for Rapid phenotypic differentiation and local adaptation in Japanese knotweed s.l. (Reynoutria japonica and R. × bohemica, Polygonaceae) invading novel habitats

Cite this dataset

Richards, Christina (2022). Data for Rapid phenotypic differentiation and local adaptation in Japanese knotweed s.l. (Reynoutria japonica and R. × bohemica, Polygonaceae) invading novel habitats [Dataset]. Dryad.


PREMISE: Many plant invaders like the Japanese knotweeds are thought to colonize new habitats with low genetic diversity. Such species provide an opportunity to study rapid adaptation to complex environmental conditions.

METHODS: Using replicate reciprocal transplants of clones across three habitats, we described patterns of phenotypic response and assessed degree of local adaptation.

KEY RESULTS: We found plants from beach habitats had decreased height, number of leaves, leaf area, and biomass allocation to roots and shoots compared to plants from marsh and roadside habitats when grown in their home habitat. In the marsh habitat, marsh plants were generally larger than beach plants, but not different from roadside plants. There were no differences among plants from different habitats grown in the roadside habitat. Despite this evidence of differentiation in beach and marsh habitats, we found mixed evidence for local adaptation. In their “home site” plants from the marsh habitat had greater biomass than plants from the beaches but not compared to plants from roadsides. Biomass comparisons in other habitats were either maladaptive or not significant. However, plants from the roadside had greater survival in their “home site” compared to foreign plants. There were no differences in survival in the other habitats.

CONCLUSIONS: We found phenotypic differentiation associated with habitats despite the low reported genetic diversity for these populations. Our results partially support the hypothesis of local adaptation in marsh and roadside habitats. Identifying whether these patterns of differentiation result from genetic or heritable non-genetic mechanisms will require further work.


Collection sites and experimental gardens ­–

In mid-May 2005, we collected Japanese knotweed s.l. rhizomes for reciprocal transplant studies between beach, marsh and roadside sites across Suffolk County, Long Island, New York (Table 1; four sites of each habitat type for 12 total sites were used as sources for plant material and as locations of transplants). We created four groups of plants for reciprocal transplant in order to maximize the ability to replicate each genet in one site of each of the habitat types while still testing for superior performance in the original “local” site.

Due to the natural topography of Long Island, the beach sites are all located on the northern shore of Long Island, while the salt marsh sites and roadside sites are more evenly distributed around Suffolk County. The beach sites are separated by 1-65 km, the marsh sites are separated by 14–40 km, and the roadside sites are separated by 20–32 km. At each site, we collected approximately one meter of rhizome from seven (Beach 1) or eight genets (all of the other 11 sites) that were approximately 10 m apart, to maximize the chances of sampling different genotypes and to represent plant distribution at each site. We refer to each of these rhizomes as a separate “genet” because replicates cut from the same rhizome should have the same genotype. However, our previous studies show that most “genets” within a site also have the same AFLP haplotype and most likely belong to the same individual. Rhizomes were brought to the Stony Brook University greenhouse and cut into pieces of 4-8 grams fresh-weight (12 sites x 7 or 8 rhizomes per site for a total of 95 genets x 18-25 replicates = 2225 rhizome pieces). We planted the rhizome pieces in individual wells in 24-well flats with Pro-mix potting medium (Pro-mix Bx, Quakertown, Pennsylvania, USA), and approximately one teaspoon of slow release fertilizer (15-9-12 Osmocote Plus 8-9 month, Marysville, Ohio). The flats were placed in a temperature-controlled greenhouse under conditions approximating mid-summer in Suffolk County, Long Island and watered as needed to keep the soil moist. Day temperature was maintained at 30˚C and night temperature at 25˚C. We grew the plants in the greenhouse for approximately six weeks to allow for shoots to emerge from the rhizomes and grow to a height of approximately 10-15 cm.

The 12 sites were organized into four reciprocal transplant groups, each with one beach, one marsh and one roadside location. Based on the number of plants that emerged, we assigned an equal number of replicate pieces of each rhizome collected at each source site to be transplanted into its home site and into one site of each of the other two habitat types. Therefore, three to eight replicates of each rhizome were assigned randomly across five blocks for each of the three transplant habitats (4 replicate studies x 3 transplant habitats x 3 source habitats x 5-8 genets x 2-8 replicates = 1287 plants; Table 1). We prepared the transplant gardens by removing only above ground vegetation (typically other knotweed plants) to a height of less than 2 cm with a machete. The cleared area included a border of 15-30 cm outside of the transplant blocks.

Between June 16-21, we transplanted the five blocks for each transplant garden into the field (between 80-125 ramets per transplant garden). Plants were left to grow in the field from the summer of 2005 through the fall of 2006. We measured final height and final number of leaves on all plants and harvested above ground and below ground for all plants between 11-18 September 2006. Roots were harvested by carefully digging to unearth the entire root system. Roots were shaken to remove loose dirt in the field and thoroughly washed.

Traits measured –

We measured traits related to salt tolerance and overall performance for each plant: height, total number of leaves, total leaf area (Li-Cor Model LI-3100 Leaf area meter: Li-Cor, Inc., Lincoln, Nebraska, USA), succulence (g water in all leaves/ cm2 total leaf area), shoot dry biomass, root dry biomass, root:shoot ratio based on dry biomass, and total biomass at final harvest. For each plant, all live leaf tissue at final harvest was used for calculating total leaf area and succulence. Plants were dried in a forced air oven at 60° C for at least 72 h to determine shoot, root and total dry biomass. We evaluated survival and biomass (as proxies for fitness) to assess the degree of adaptation. These taxa have extensive clonal growth and many individuals may not flower at all in the field, but persist and spread from year to year so biomass is an important indicator of fitness (de Kroon and Groenendael, 1997).

Data Analysis –

We performed all statistical analyses in the R statistical programming environment version 4.0.0 (R Core Team, 2020), using the Linear-Mixed-Model (LMM) or Generalized Linear-Mixed-Model (GLMM) framework (lme4 package, Bates et al., 2015) and the Bayesian simulation package arm (Gelman and Su, 2020). We checked the residuals to assess the appropriateness of the model and performed data transformations on traits as appropriate: we did not transform succulence and final height, but we performed log 10-transformation on leaf area, and log 2 transformation on shoot, root and total biomass. For these traits we fitted LMM with the model:


We used GLMM to model the number of leaves, with the negative binomial distribution with the model:

modlfnum<glmer.nb(lf.number~SOURCE.type+GARDEN.type+SOURCE.type:GARDEN.type+ (1|| + (1|genetfactor),data=data, REML=F).

We did not model root to shoot ratio directly, but instead we used the ratio of the estimates of mean and variance for root and shoot to assess significance within the Bayesian framework (Korner-Nievergelt et al., 2015).

In the LMM and GLMM models, “SOURCE.type” is the origin habitat type (beach, marsh , road), “GARDEN.type” is the transplant habitat type (beach, marsh, road). These effects as well as their interaction terms were modeled as fixed effects. The origin site, the transplant site, and the individual genets (“genetfactor”) were initially included as random terms. To avoid overfitting, we removed random effect terms that effectively explained no variance. This was true for the genet term for all traits and for the “” term, which was removed for number of leaves, succulence and shoot biomass (see Table S1 for final models).

We examined the correlation matrix for each model to evaluate auto correlation between terms. To test the significance of the fixed effects of “SOURCE.type” and “GARDEN.type”, we used 95% credible intervals (CrI), a Bayesian analogue of confidence intervals (Bolker et al., 2009). For each response variable, we obtained the model estimates from the back-transformed effect sizes. We calculated the associated 95% Credible intervals (CrI) for the modeled effects by performing 10,000 iterations of Bayesian simulation of the mean and variance of each estimate, using the sim function in the R package arm with non-informative priors (Korner-Nievergelt et al., 2015). For each model, we examined distributions of simulated fitted values predicted by the fixed effect terms. We used the values corresponding to the 2.5 and 97.5 quantiles of the distribution to designate the lower- and upper-boundary of the 95% CrI. If the CrI of a group did not overlap with the mean of the other group within transplant gardens, we considered the difference between these groups to be significant (sensu Bucharova et al., 2016, 2017).

In order to understand how much of the variance was explained by the random and fixed effects in our model, we used several approaches. First, we used the package r2_nakagawa: Nakagawa's R2 for mixed models (Nakagawa and Schielzeth, 2013; Nakagawa et al., 2017) to determine the conditional R2 (the variance explained by both the fixed and random effects) and marginal R2 (the variance explained by the combined fixed effects). The random effect variances calculated in this package are the mean random effect variances, and appropriate for mixed models with nested random effects (Johnson, 2014). We also used the package rptR (Stoffel et al., 2017) to further evaluate the components of variance for each of the random effects separately (i.e., “Origin site” and “Transplant site”. We used the package commonalityCoefficients (Nimon et al., 2008) to examine the amount of variance explained by the separate fixed effects of the habitats of the source and transplant gardens (i.e., “SOURCE.type” and “GARDEN.type”). This approach does not include information about the random effects of origin site and transplant site which are nested within the source type and garden type. However, this approach is valuable for evaluating the relative contribution of each separate fixed effect.

Our design is constrained by the fact that origin site and transplant sites are nested within levels of “Transplant group” (see discussion here Long, 2021). To examine the importance of this design constraint, we also reran the LMER and GLMER models for each trait with the fixed term “Transplant group”. To properly test for the effects in this nesting design, we should ideally fit random intercepts for the sites nested within groups, but we did not have enough replication within groups to do so. We assume that fitting the fixed effect of the “Transplant group” also controls for the non-independence of the origin site and transplant site within groups (Long, 2021). By comparing the modeling with and without the fixed term of Transplant group, we evaluated how these random terms impact the main effects of interest which are the fixed effects of the habitats of the source and transplant gardens (i.e., “SOURCE.type” and “GARDEN.type”).

We tested for local adaptation with two fitness proxies: total biomass and survival. We ran a “local vs. foreign” test using the Bayesian fitted values for biomass obtained from the same LMM model:


For each garden type we performed random pairwise contrasts as the differences between Bayesian fitted values of the local plants with those from foreign habitats. For survival, we performed random pairwise contrast by calculating the log of odds ratio between local and foreign plants.  We reported the mean, 95% CrI, and percentage of contrasts showing superior performance of plants grown in their home site compared to plants from each of the other habitats as magnitude of local adaptation.

Literature Cited

Bates, Douglas, Martin Mächler, Ben Bolker, and Steve Walker. 2015. “Fitting Linear Mixed-Effects Models Usinglme4.” Journal of Statistical Software 67 (1).

Bolker, Benjamin M., Mollie E. Brooks, Connie J. Clark, Shane W. Geange, John R. Poulsen, M. Henry H. Stevens, and Jada-Simone S. White. 2009. “Generalized Linear Mixed Models: A Practical Guide for Ecology and Evolution.” Trends in Ecology & Evolution 24 (3): 127–35.

Bucharova, Anna, Walter Durka, Norbert Hölzel, Johannes Kollmann, Stefan Michalski, and Oliver Bossdorf. 2017. “Are Local Plants the Best for Ecosystem Restoration? It Depends on How You Analyze the Data.” Ecology and Evolution 7 (24): 10683–89.

Bucharova, Anna, Mark Frenzel, Karsten Mody, Madalin Parepa, Walter Durka, and Oliver Bossdorf. 2016. “Plant Ecotype Affects Interacting Organisms across Multiple Trophic Levels.” Basic and Applied Ecology 17 (8): 688–95.

Gelman, A., and Y-S Su. 2020. “Arm: Data Analysis Using Regression and Multilevel/Hierarchical Models. R Package Version 1.11-2.”

Johnson, Paul Cd. 2014. “Extension of Nakagawa & Schielzeth’s R2GLMM to Random Slopes Models.” Methods in Ecology and Evolution / British Ecological Society 5 (9): 944–46.

Korner-Nievergelt, Franzi, Tobias Roth, Stefanie von Felten, Jérôme Guélat, Bettina Almasi, and Pius Korner-Nievergelt. 2015. Bayesian Data Analysis in Ecology Using Linear Models with R, BUGS, and Stan. Academic Press.

Kroon, Hans de, and Jan-Marie Groenendael. 1997. The Ecology and Evolution of Clonal Plants. Backhuys Publishers.

Long, Robert. 2021. “Crossed Random Effects: How Do We Model Multiple Reciprocal Transplants in lme4?” 2021., Crossed random effects: how do we model multiple reciprocal transplants in lme4?, URL (version: 2021-06-16):

Nakagawa, Shinichi, Paul C. D. Johnson, and Holger Schielzeth. 2017. “The Coefficient of Determination R2 and Intra-Class Correlation Coefficient from Generalized Linear Mixed-Effects Models Revisited and Expanded.” Journal of the Royal Society, Interface / the Royal Society 14 (134).

Nakagawa, Shinichi, and Holger Schielzeth. 2013. “A General and Simple Method for obtainingR2from Generalized Linear Mixed-Effects Models.” Methods in Ecology and Evolution / British Ecological Society 4 (2): 133–42.

Nimon, Kim, Mitzi Lewis, Richard Kane, and R. Michael Haynes. 2008. “An R Package to Compute Commonality Coefficients in the Multiple Regression Case: An Introduction to the Package and a Practical Example.” Behavior Research Methods 40 (2): 457–66.

R Core Team. 2020. “R: A Language and Environment for Statistical Computing.” Vienna, Austria. : R Foundation for Statistical Computing.

Stoffel, Martin A., Shinichi Nakagawa, and Holger Schielzeth. 2017. “rptR: Repeatability Estimation and Variance Decomposition by Generalized Linear Mixed-effects Models.” Methods in Ecology and Evolution / British Ecological Society 8 (11): 1639–44.

Usage notes

R scripts are anotated with results that we obtained.


New York Sea Grant

Federal Ministry of Education and Research, Award: 306055

Research Foundation for the State University of New York