Data from: Are all forms of defense lost on islands? Persistence of an indirect defensive trait in six island colonists from New Zealand

(a) Data collection Data collection was conducted between June 2018 and April 2019. The six study taxa (Coprosma repens [Rubiaceae], C. rhamnoides, C. robusta, C. lucida, Elaeocarpus dentatus [Elaeocapraceae] and Vitex lucens [Lamiaceae]) were chosen as they are the most widely distributed across New Zealand’s north-eastern islands. All taxa produce pit domatia, except E. dentatus which produces tent domatia. Mainland field sampling was conducted in the Kaimai-Mamaku Forest Park in Tauranga (37°41ʹS, 175°45ʹE). This site was chosen because the Kaimai Ranges span a large latitudinal extent of the north-eastern corner of New Zealand, and therefore represent the probable source pool for many island populations. Island field sampling was conducted on seven islands: Cuvier, Great Mercury, Red Mercury, Otata, Ruamahuaiti, Ruamahuanui, and Waiheke. Individuals were chosen haphazardly while walking through easily accessible forest sections. Only fully expanded, mature leaves were measured (1-3 leaves per individual from at least 30 individuals per site per species). Leaf length was measured as the longest linear distance from the most proximal to the most distal point of the leaf lamina using a digital calliper. Leaf width was measured as the widest distance across the leaf lamina perpendicular to the leaf length measurement. Leaf area was calculated as the product of leaf length and leaf width. Although more accurate methods of estimating leaf area exist [e.g., leaf scanners, image recognition software, or using ad hoc leaf shape correction factors 43], leaf x width calculations sufficed for the purpose of this analysis as it is not concerned with among species differences in leaf size per se. Couched in other terms, any errors associated with leaf area estimates are consistent between island-mainland comparisons. In a prior study incorporating the same species. I trialled ad hoc correction factors and found they provided little utility. Further, a leaf scanner could not be used as most of the islands included in this study are protected by the Department of Conservation, which does not allow destructive sampling protocols. Domatia were counted systematically in a basipetal direction with the aid of a USB microscope (Toolcraft DigiMicro 2.0 Scale). It should be noted that while this study quantifies defense investment by domatia count, experimental work with Grape cultivars has demonstrated that mite abundance on the phylloplane can scale with domatia size. To expand upon data gathered in the field, pressed herbarium specimens from a further 53 islands and 58 mainland sites from the Auckland War Memorial Museum Herbarium (AK) were measured. The Herbarium houses an extensive collection of high-quality, preserved specimens of both indigenous and exotic plants spanning the full geographic extent of the New Zealand landmass – including its offshore islands. To standardize sampling across specimens, 3 leaves per specimen were measured using the same methodology outlined above. Leaves were chosen haphazardly, and care was taken to not damage specimens (i.e., gloves and minimal calliper–specimen contact). 1,129 observations from 60 islands and 59 mainland sites comprised the final dataset. (b) Data analysis To test for mainland-island differences in domatia production, while simultaneously accounting for leaf size and exploring differences in domatia–leaf size scaling, I performed a linear mixed effects models of number of domatia production (total count per leaf) against leaf area (cm2), insularity (island, mainland), and their interaction. Both domatia count and leaf area were logarithm transformed to conform to linearity. One was added to counts of domatia before transformation to avoid biologically nonsensical values (i.e., negative infinity). Species was included as a random effect permitting both intercept and slope to vary among species. Analysis and data visualization were conducted in R version 4.2.2 (R Core Team 2022) using the ‘lme4’ and ‘tidyverse’ packages.