Data collection
We followed PRISMA protocol in defining our research questions, conducting data collection, and also in reporting our results using a standardised approach (O’Dea et al., 2021). This was done to minimise potential biased associated with meta-analyses. We include a check list in which our studies followed the PRISMA protocol in our supplementary data (appendix 1) Our meta-analytical approach involved quantification of network patterns through collation of original data, and calculation of network metric using those data, rather than through collation of effect sizes from previous studies (Xing & Fayle, 2021). This manuscript is part of the LifeWebs project, a collaborative effort that aims to use a meta-analytic approach to investigate how interaction networks respond to global-scale environmental gradients (www.lifewebs.net). We collated existing ant-plant interaction networks by searching for publications on the Web of Science (WoS) online database. These published datasets were supplemented by direct requests to authors of papers in which the data had not been published, and furthermore through requests to data authors identified through snowballing (see below). We excluded datasets in which the links between ant and plant species were not available (i.e. where we were unable to build a bipartite network). The use of a single search engine may have led to some studies being missed but allowed us to focus on high quality network in a systematic manner. Data collection activities were carried out throughout 2021, with no date limits imposed. We searched the Web of Science database using the keywords ”ant-plant interaction” and then refine this search using the additional keywords, ”extrafloral nectaries” (EFN) OR ”food bodies” (FB), OR ”myrmecophily”, OR “Co-occurrence”, OR ”ant-plant foraging,” OR ”myrmecochory”. We only selected networks that consisted of at least three plant species and at least three ant species. We excluded networks that consisted of only presence/absence data, as analyses on binary data are more sensitive to sampling effort (Miranda et al., 2019), which varied between studies in our collated data. We also excluded networks collated from museum collections because these were not quantitative and often lacked geographical data. Lastly, we excluded studies that pooled ant-plant data across a whole region or country, for example,Macaranga spp. and Crematogaster spp. in Southeast Asia (Fiala et al., 1999). Where necessary, we contacted authors to provide metadata. In some cases, additional more recent data (post-2021) were contributed to the network as a result. In addition, we conducted a snowball search by identifying relevant references in all collated data papers. Our decision tree for including/excluding network is presented in Figure 1. It has to be noted that our studies has some geographical limitation as there are not many studies in Africa fulfil our criteria.
Our protocol resulted in a total of 74 ant-plant interaction networks. These included 18 myrmecophytic, 17 myrmecophilic, 14 myrmecochorous, and 25 foraging networks. The networks spanned an absolute latitudinal range of 1.8° to 49° and an elevational range from 4 m to 2800 m above sea level (asl) with 41 networks in undisturbed areas, and 33 networks in disturbed areas (Figure 2 & 3). In some cases, the network data were not published with the corresponding paper and the author did not respond to our request. Hence these networks were excluded. However, since these occurrences were a minority (10 / 74) we believe our collated networks represent a substantial proportion of those available.
We classified the networks into four types: (i) myrmecophytic : networks where plants provide nesting space (domatia) for ants; (ii)myrmecophilic : networks where plants offer food to ants (extrafloral nectaries (EFN) and food bodies (FB) but without domatia); (iii) myrmecochorous : networks involving ant dispersal of plant seeds, or ant consumption of fruits; (iv) foraging : networks in which interactions between ants and plants were recorded, with ants being found foraging on plants, but without utilising any plant-provided resources. Latitude, elevation, and presence/absence of anthropogenic disturbance were recorded as metadata for each network. If elevation was not included in the article or provided by the author, we determined this from the geographic coordinates in the study site description using google earth (elevation determined in this way for 9 of 74 networks). If the elevational range across multiple sites within a study was less than 300 meters, we combined these sites into a single network and used the mean elevation value (14 of 74 networks resulted from such merging). Where sites within a study were > 300 m apart in terms of elevation we retained the data as multiple separate networks (10 of 74 networks). In these cases, study identity was retained as a random factor in all models, to account for greater similarity in sampling effort and methodology within studies than between studies. Each network was classified as being in either anthropogenically disturbed (gardens, recently cut secondary forests or production forests) or undisturbed (primary forests or nature reserves), based on the original study description. We combined different kinds of disturbed habitats into one category because sample sizes for finer grained categories were not large enough for statistical analyses. Finally, we updated the species names of plants using taxize in R with the gbif database (https://www.gbif.org/), accessed January 2022. We manually fixed typos when names were classified as “fuzzy” according to taxsize and rechecked the updated plant names on the WFO database (http://www.worldfloraonline.org/) when names were classified as “doubtful “ Meanwhile, ant taxonomic names were checked manually using the AntWeb (https://www.antweb.org/) and AntWiki databases (https://www.antwiki.org/), accessed July 2022. We mapped the network sampling locations using QGIS 3.16.15 to visualize the distribution of our sites. We also chose the network with the median species richness from each interaction type to visualize the differences in structure across each network type. These networks were plotted using the plotweb function in thebipartite package.