Invasion away from roadsides was not driven by adaptation to grassland habitats in Dittrichia graveolens (stinkwort)

Invasive plants along transportation corridors can significantly threaten ecosystems and biodiversity if they spread beyond anthropogenic environments. Rapid evolution may increase the ability of invading plant populations to establish in resident plant communities over time, posing a challenge to invasion risk assessment. We tested for adaptive differentiation in Dittrichia graveolens (stinkwort), an invasive species of ruderal habitat in California that is increasingly spreading away from roadsides into more established vegetation. We collected seeds from eight pairs of vegetated sites and their nearest (presumed progenitor) roadside population. We assessed differentiation between populations in roadside and vegetated habitat for germination behavior and for response to competition in a greenhouse experiment. We also tested for increased performance in vegetated habitat with a grassland field experiment including a neighbor removal treatment. Germination rates were slightly reduced in seeds from vegetated sites, which may indicate lower seed viability. Otherwise, plants did not show consistent differences between the two habitat types. Competition strongly reduced performance of D. graveolens in both the greenhouse and in the field, but plants originating from vegetated sites did not show enhanced competitive ability. Our findings show no evidence of adaptive differentiation between D. graveolens populations from roadside and vegetated habitats to date, suggesting that invasiveness in grasslands has not been enhanced by rapid evolution in the 40+ years since this species was introduced to California. Evolutionary constraints or potentially high levels of gene flow at this small scale may limit adaptation to novel habitats along roadsides.

Methods
Plant community survey: Data was collected at 8 sites of Dittrichia graveolens in Santa Clara County, CA, USA, from July 1st through August 14th, 2020. Each site had a roadside and vegetated population pair. At each population, we laid a 50 m transect tape along the longest axis of the population (for roadsides, transects were always parallel to the road) and placed a 0.5 x 0.5 m quadrat at three equidistant points along the axis. We visually estimated percent cover within each quadrat for D. graveolens, other vegetation, and bare ground (sum equaling 100%). We identified species within the three quadrats for each population and then walked the area to search for additional rare species. Taxa were identified to species when possible using The Jepson Manual: Vascular Plants of California (Second Edition). We calculated the average percent cover of bare ground, D. graveolens, and other vegetation per population by taking the mean of the three quadrats along each transect. Species richness was the total number of species found at a population (the three quadrats + surrounding rare species survey). We evaluated differences in percent cover and species richness between source habitats (roadside and vegetated) using paired t-tests (N = 8 sites with pairs of roadside and vegetated populations at each site).

Dittrichia graveolens seed collection for lab, greenhouse, and field experiments: In September and October of 2020, we sampled seeds from each of the 16 populations. We collected from at least 10 individuals, 3 m apart, for each population, along a randomly-placed transect. We combined seeds from all individuals in a population. 

Lab - seed behavior: In the summer of 2021, we compared germination behavior of seeds from roadside and vegetated habitat types. We did three studies on different substrates: moist filter paper, engineered fill, and field topsoil (collected from a site on the UC Santa Cruz campus). We germinated 50 seeds from each population in Petri dishes (80 Petri dishes; 5 replicates with 10 seeds each) for each substrate (filter paper, engineered fill, and field topsoil). Seeds were visually inspected beforehand to ensure that only fully developed seeds were used for all experiments. Petri dishes were sealed with Parafilm M ™ and placed in a randomized block design in an incubation chamber with a daytime temperature of 23 ºC from 0900 - 0100 h and a nighttime temperature of 19 ºC from 0100 - 0900 h. We scored germination daily until no further germination was observed, then 7 more days (a total of 23 d on filter paper, 12 d on engineered fill, and 11 d on field topsoil). Signs of germination included the first emergence of the root radical or the cotyledon. Petri dishes were misted with DI water, and germinated seeds were removed once scored. We also took one homogenized sample of 30 seeds from each of the 16 populations and weighed them to the closest 0.001 g. We analyzed the germination rate on each of the three substrates (filter paper, engineered fill, and field topsoil) using a mixed-effects Cox proportional hazards model (coxme and survival packages), with source habitat as a fixed effect and site, population, and dish number as nested random effects. We evaluated the main effect of source habitat using a Type II partial-likelihood-ratio test (car package). We calculated average seed mass for each source habitat using a Welch Two Sample t-test.

Greenhouse - response to competition: We quantified response to competition in a greenhouse experiment with three treatments: D. graveolens grown alone, with Bromus hordeaceus, or with Festuca perennis. We collected B. hordeaceus seeds from Blue Oak Ranch Reserve and F. perennis seeds from the Terrace Lands of Younger Lagoon Reserve on the UC Santa Cruz Coastal Science Campus. We germinated D. graveolens seeds in the conditions described in Lab - seed behavior. We germinated grasses in trays with potting mix and placed them under fluorescent light banks for 16-hour length days and 8-hour length nights. Once radicles and cotyledons emerged, seedlings were transplanted in sets of three (one for each treatment). We grew plants in D16 Deepots (5 cm diameter, 18 cm height) in the greenhouse using field topsoil collected from a UC Santa Cruz campus site. Pots were then randomized into a blocked design with each block consisting of one D. graveolens seedling from each of the 16 populations for each of the three competition treatments, N = 48 per block x 8 blocks (384 total). After 4 months, we harvested D. graveolens aboveground biomass at the crown and dried it in a 60 ºC oven for 3 days before weighing it. We calculated response to competition as the log response ratio (LRR) of the aboveground biomass, LRR = ln(biomass with competitor / biomass alone), on a per-block basis (N = 8 blocks) for each of the 16 seed origins (vegetated or roadside habitat at each of the 8 sites). Therefore, each seed origin had 8 replicate LRR estimates for each competitor grass (Bromus hordeaceus and Festuca perennis). We fit a linear mixed effects model for each competitor with LRR as the response variable, source habitat as a fixed effect, and random effects for population nested in site, and block (lme4 package). Block was removed from the B. hordeaceus model because it did not explain sufficient variance, causing a singular fit. We tested for differences between source habitats using Type II Wald F-tests with Kenward-Rogers degrees of freedom (car package). To evaluate whether each competitor grass affected the biomass of D. graveolens, we tested whether the LRR intercept was significantly different from zero using t-tests with Kenward-Rogers degrees of freedom (pbkrtest and lmerTest packages).

Field - relative fitness: The field experiment was conducted at Blue Oak Ranch Reserve, San Jose, California, USA (37°22'54.89"N, 121°44'10.55"W). We tested whether rapid evolution during invasion into vegetated sites has enhanced fitness in the presence of grassland competitors. We established a 10 m x 26 m fenced field site and used a randomized block design with 10 blocks of 1.5 m2 plots. We had two treatments: grassland control (high competition) and complete competitor removal (no competition). We left the previous year’s thatch for the grassland control treatment and allowed vegetation to grow throughout the experiment. For the competitor removal treatment, we tilled the soil to completely remove below and aboveground biomass in December 2020 and then weeded to remove aboveground biomass throughout the growing season. In January 2021, we germinated seeds in Petri dishes in incubation chambers before transplanting them into soil collected in late December 2020 from Blue Oak Ranch Reserve. After about eight weeks, we planted seedlings into each plot from February 27 - March 24, 2021 (20 plots total). Each plot included one D. graveolens individual from each of the 16 populations, in a 4 x 4 grid centered on the plot. During the first month of growth, we replaced any D. graveolens that died. We surveyed plants weekly to assess D. graveolens survival and bud initiation until all plants had either produced buds or perished. We terminated plants at the first sign of budding to prevent reproduction of a noxious weed. As proxies for reproductive output, we measured height and biomass. We harvested aboveground biomass by cutting at the root crown and drying in a 60 ºC oven for three days before weighing. Height and biomass were strongly correlated (r = 0.74, N = 157), and results for the two response variables were similar. Therefore we present only the results for final biomass.

The field experiment had four response variables: survival (assessed both as total proportion surviving and time to death), final biomass at budding, and phenology (the survey date buds first appeared). We used a similar statistical approach for all response variables, fitting mixed effects models with source habitat, competition treatment, and their interaction as fixed effects; and initially including random effects for site, population nested in site, and block. Random effects that explained very low amounts of variance, causing singular fits, were removed. When interaction terms were not significant, they were removed and models were re-run with main effects only. Here we describe the structures of the final models. We compared total survival to budding with a generalized linear mixed model using a binomial family with a logit link function; fixed effects were source habitat and competition treatment, and random effects were population nested in site, and block (glmmTMB package). We evaluated the main effect of source habitat using a Type II Wald Chi-Square test (car package). Second, we analyzed survival using a mixed-effects Cox proportional hazards model (coxme and survival packages); fixed effects were source habitat and competition treatment, and random effects were population nested in site, and block. We evaluated the main effects of source habitat and competition treatment using likelihood ratio tests. We analyzed final biomass at the time of bud production using a linear mixed effects model (lme4 package); fixed effects were source habitat, competition treatment, and their interaction, and the only remaining random effect was site. We evaluated the main and interaction effects using Type II Wald F-tests with Kenward-Rogers degrees of freedom (car package). We used a log transformation of the biomass data to improve homoscedasticity. To assess changes in phenology, we compared the timing to bud for those plants that reached the reproductive state, using a mixed-effects Cox proportional hazards model (coxme and survival packages); fixed effects were source habitat and competition treatment, and random effects were population nested in site, and block. We evaluated the main effects using likelihood ratio tests.