Reduced defense against large herbivores has been suggested to be part of the “island syndrome” in plants. However, empirical evidence for this pattern is mixed. In this paper, we present two studies that compare putative physical and chemical defense traits from plants on the California Channel Islands and nearby mainland based on sampling of both field and common garden plants. In the first study, we focus on five pairs of woody shrubs from three island and three mainland locations and find evidence for increased leaf area, decreased marginal leaf spines, and decreased concentrations of cyanogenic glycosides in island plants. We observed similar increases in leaf area and decreases in defense traits when comparing island and mainland genotypes grown together in botanic gardens, suggesting that trait differences are not solely driven by abiotic differences between island and mainland sites. In the second study, we conducted a common garden experiment with a perennial herb—Stachys bullata (Lamiaceae)—collected from two island and four mainland locations. Compared to their mainland relatives, island genotypes show highly reduced glandular trichomes and a nearly 100-fold reduction in mono- and sesquiterpene compounds from leaf surfaces. Island genotypes also had significantly higher specific leaf area, somewhat lower rates of gas exchange, and greater aboveground biomass than mainland genotypes across two years of study, potentially reflecting a broader shift in growth habit. Together, our results provide evidence for reduced expression of putative defense traits in island plants, though these results may reflect adaptation to both biotic (i.e., the historical absence of large herbivores) and climatic conditions on islands.

Study 1: Chaparral Shrub Sampling

We selected five pairs of taxa characteristic of the chaparral plant community that occur on both the California Channel Islands and the nearby southern California mainland. Pairs were chosen because they are common representatives of the chaparral flora and also to match the taxa sampled in Bowen and Van Vuren (1997). Sampling consisted of either congeners or conspecifics from three plant families: Rosaceae (Cercocarpus, Prunus, Heteromeles), Papaveraceae (Dendromecon), and Rhamnaceae (Ceanothus). We collected leaf tissue in February and March of 2016 for use in morphological and chemical analysis. In total, we sampled 291 individual plants from five taxonomic pairs across six sites (three island, three mainland), for an average of approximately 10 plants per site.

We collected leaf tissue for morphological analysis from focal plants by clipping branches containing variable numbers of leaves. When possible, we collected a branch from both the lower (< 1 m in height) and the upper (> 2 m in height) portion of the plant canopy to capture morphological differences associated with accessibility to mammalian herbivores. For analysis of cyanogenic glycosides, we collected individual leaves from the lower portion of the plant canopy for three species (Heteromeles, Prunus, Cercocarpus) and, when possible, included both fully mature/expanded leaf tissue as well as young/actively expanding leaf tissue. Leaf chemistry samples were immediately frozen on dry ice and were later transferred to a −80 °C freezer until processing. For each sampled plant, we recorded its GPS coordinates, elevation, and slope aspect (when relevant) using a handheld Garmin GPS device, and we also recorded the approximate stem diameter at 0.25 m above the ground using a digital caliper.

For each sampled branch, leaves were removed and imaged using a flatbed scanner (CanoScan LiDE 120, 2400 × 4800 dpi²) with a scalebar. We recorded the following measurements from each leaf: leaf area (without petiole), marginal leaf spinescence, and percent of leaf tissue missing due to herbivory. All measurements were taken using ImageJ v. 1.51. For a visual depiction of our measurement protocol, see Figure S2. Non-fully expanded leaves (n = 809) were measured but were excluded from subsequent analyses. We also measured specific leaf area (SLA) at the level of branches by taking the cumulative area of all fully expanded leaves (in cm²) and dividing this by their cumulative mass (in g).

To measure cyanogenic glycoside (CNglc) content, we followed a modified version of the evolved hydrogen cyanide (HCN) protocol described in Experiment 2 of Gleadow et al. (2011). We only collected tissue for species in the Rosaceae (Cercocarpus, Heteromeles, Prunus), which are known to produce CNglcs, and included paired samples of mature (“old”) and expanding (“young”) leaf tissue from each plant, where possible. For a full description of methods used to quantify CNglc content, see Supplemental Materials. In total, we generated 194 measurements of CNglc content from 108 individual plants.

We also sampled leaf tissue from two botanical gardens (Santa Barbara Botanic Garden and Rancho Santa Ana Botanic Garden) on the mainland that featured island and mainland genotypes of the species of interest, grown from either seed or cuttings. All leaf tissue collection, morphological analysis, and chemical analysis were conducted in the same way as described above, although SLA was not measured for botanical garden plants. In total, we sampled an additional 40 plants (18 island and 22 mainland genotypes) from these common environments.

We also took advantage of a series of herbivore exclosures on Santa Catalina Island—which still has introduced deer and bison present—to test for the potential effects of herbivore-mediated plasticity in plant traits. Because of the relatively small number of intact exclosures available, our sampling across species was somewhat uneven, though we were able to sample a total of 24 plants inside of exclosures and 35 plants outside of exclosures (Table S1). Note that plants in the herbivore exclosures experience gene flow from genotypes outside the exclosure through pollen and seed.

Study 1: Abiotic Variation between Sites

Island and mainland sites have generally similar climates, although island locations may have more frequent nocturnal fog that reduces summertime evaporative water loss. To formally measure climatic differences between island and mainland locations, we used recorded coordinates from each plant to extract bioclimatic variables from the WorldClim2 database at 1 km resolution. We used principal component analysis (PCA) to explore variation in climate data and found that the first PC axis explained more than 83% of overall variation and was dominated by a single bioclimatic variable, temperature seasonality (BIO4). This axis separated island sites from the two more inland mainland sites (Stunt Ranch, Santa Monica Mtns.), which have higher temperature seasonality, while the third mainland site (Gaviota) had lower temperature seasonality and was more akin to island sites. The second PC axis explained 13% of the overall variation and included loadings for precipitation-related variables; this axis separated the drier Santa Catalina Island from all remaining sites. Island and mainland sites may also differ in soil properties; however, we did not attempt to quantify this potential variation.

Study 1: Statistical Analyses

We analyzed our data using multilevel linear mixed models implemented in the lme4 package in R version 4.1.3 to account for the hierarchical nature of our data. Response variables of interest were leaf area, SLA, marginal leaf spinescence, and leaf CNglc content. Leaf area and CNglc content were log-transformed to ensure that model-estimated confidence intervals were above 0; SLA and marginal leaf spinescence were untransformed. For marginal leaf spinescence, we only included Heteromeles and Prunus since these were the only species with stiff, rigid spines. Likewise, because CNglc levels in Cercocarpus were ~100x lower than in Prunus and Heteromeles (and often below our detection limit), CNglc analysis was restricted to the latter two species. Covariates that were included in each model included the site of collection, canopy position (upper versus lower), north/south slope aspect, and east/west slope aspect. We considered including elevation and stem diameter (as a proxy for plant age) as covariates, though because of limited within-site and within-species variation in these measures, we ultimately omitted them from analyses. Furthermore, we attempted to include bioclimatic variables as covariates in these models, but because of the relatively limited spatial scale of sampling across sites and within sites and the 1 km² resolution of the bioclim dataset, we captured relatively little overall climatic variability for most bioclimatic variables.

For each response variable, we fit an overall model that included all samples collected in situ across all species (n = 4,096 leaves from 291 plants). These models were of the form (in lme4 syntax):

where IM corresponds to whether samples came from an island or mainland site. Plant species interacts with island vs. mainland status to allow for variation in the magnitude of island vs. mainland contrasts across species. The collection site was included as a random intercept, with plant ID nested within the site and branch ID nested within the plant ID. Since specific leaf area was calculated by pooling leaves from within branches, the SLA model does not include a branch ID term. To assess within-species differences in trait expression between islands and mainland locations, we used the emmeans package.

For two of the response variables (marginal spinescence, CNglc content), we included additional parameters based on a priori hypotheses. In the model considering marginal spinescence, we included an interaction between island/mainland status and canopy position to allow for the degree of spinescence heteroblasty to vary across locations. In the model considering CNglc content, we included a term for leaf age (old vs. young) based on our sampling scheme and predictions from optimal plant defense theory that younger leaf tissue should be more heavily defended against herbivores .

To test for genetically based differences in trait values, we analyzed samples collected from botanic garden plants in a separate set of linear mixed models. These models were similar to those described above and were of the form:

where Source_IM refers to whether each plants’ original provenance was an island or mainland location. As above, we also estimated within-species differences between island and mainland locations using the emmeans package.

Because we sampled the same species as Bowen and Van Vuren (1997), we can compare insularity effects across studies (our field sampling, our botanic garden sampling, and the field sampling from Bowen and Van Vuren). To do so, we calculated standardized effect sizes (Cohen’s d_s) for each measured trait. Since Bowen and Van Vuren only report t statistics and sample sizes, we used the following formula for Cohen’s d_s: where t corresponds to the mean of their reported t statistics, and n₁ and n₂ correspond to sample sizes from island and mainland locations. To generate effect size estimates and corresponding confidence intervals from our in situ and botanic garden sampling, we used the effect size package.

Finally, to test for the effects of introduced herbivores on plant traits on Santa Catalina Island, we separately analyzed all trait data from Santa Catalina and included a term to account for whether samples came from inside versus outside of an herbivore exclosure.

Study 2: Stachys bullata–Background

Stachys bullata (Lamiaceae) is a perennial herbaceous plant that occurs in coastal California from approximately Orange County to the San Francisco Bay Area, with populations present on Santa Cruz, Santa Rosa, and Anacapa Islands. It reproduces both clonally via rhizomes and sexually and is described as being glandular, with aromatic foliage that is characteristic of many plants in the Lamiaceae. However, island populations have been noted to have non-aromatic foliage as well as larger leaves and flowers than their mainland relatives [34], and densities of glandular trichomes appear to be much lower on island plants (Figure 2C).

Study 2: Stachys bullata Common Garden Experiment

To determine whether observed trait variation between island and mainland S. bullata populations is environmentally or genetically determined, we set up a multi-year common garden experiment in which we grew island and mainland S. bullata genotypes together at the Santa Barbara Botanic Garden (SBBG). Plants were collected in the field in late 2015 from two island (Santa Cruz, Santa Rosa) and four mainland locations as rhizomes, which were transported to UC Davis and shallowly planted in potting mix. Plants were grown in 1-gallon pots for approximately three months and were then split into clonal replicates that were grown in their own 1-gallon pots. In total, we collected 44 S. bullata genotypes that were separated into 112 individual plants.

In February 2016, we set up a common garden plot at the SBBG located on an east-facing slope that received partial or full sun throughout the year. Plants were spaced at a distance of 1 m apart from each other in a gridded pattern. The plot was surrounded by a 2 m fence to prevent browsing by deer, and each plant was enclosed in a cage made from hardware cloth to limit root herbivory by pocket gophers (Thomomys bottae), which were common at the site. We installed a drip irrigation system to assist with initial plant establishment. Plants were outplanted randomly with respect to island/mainland status in late February and early March of 2016 and received approximately 2L of water from a drip irrigation system at 1-week intervals between March and August 2016. In late August 2016, we ceased supplemental watering, and plants subsequently only received water from ambient precipitation. Plants became dormant in October 2016 and then subsequently began to regrow naturally in early February 2017. In addition, we set up a smaller common garden at the Santa Cruz Island Reserve, although due to concerns over the introduction of non-native genotypes, this common garden consisted of only genotypes from Santa Cruz Island.

We generated four categories of data from common garden S. bullata plants. Three measures (biomass, SLA, gas exchange) were related to plant growth, while one measure (leaf surface chemistry) was related to plant defense. For biomass measurements, we collected all above-ground biomass at the end of each growing season (in 2016 and 2017) and recorded its dry mass. For SLA, we collected leaf tissue from each plant in April 2017 and measured leaf area and dry mass from fully expanded S. bullata leaves. In April 2017, we used a Li-6800 portable photosynthesis system (LI-COR Biosciences Inc., Lincoln, NE) to measure gas exchange rates on the most recent mature leaves that were sun exposed. For details on gas exchange measurements, see Supplemental Materials.

For leaf chemistry, we focused on volatile organic compounds (VOCs) present on leaf surfaces and in glandular trichomes. Plants in the Lamiaceae are known for their exceptional diversity of terpenoid compounds, which are a diverse group of plant secondary metabolites thought to be involved in defense against herbivores and pathogens, plant communication, and modulating thermal and oxidative stress. We used a modified version of the protocol described in Pratt et al. (2014) for measuring terpenes in Artemisia californica (Asteraceae). Briefly, in April of 2017, after > 1 year of growth in the common garden, we used a hole punch to collect six leaf discs, each from a different leaf, from approximately 75 Stachys plants across all genotypes. Leaf surface chemistry was quantified using gas chromatography-mass spectrometry (GCMS) for a subset of 47 of these plants. For a full description of chemical methods, see Supplementary Materials.

Study 2: Stachys Data Analysis

We analyzed aboveground biomass using a linear mixed effects model of the form:

where IM refers to whether a given plant originated from an island or mainland site, and column and row refer to the location of plants within the common garden grid. To analyze gas exchange measurements, because of our smaller sample, we used a simple linear model with net carbon assimilation (A_net) as the response and provenance (island vs. mainland) as the predictor.

To analyze plant chemistry, we divided each integrated peak area by its corresponding internal standard peak area to standardize all values. We added all peaks together to achieve a cumulative compound abundance measure and also separated compounds based on their biochemical pathway (e.g., monoterpenes, sesquiterpenes, aromatic compounds). We used non-metric multidimensional scaling to visualize compositional differences between sites.