Urban environments represent a theatre for life history evolution. Species able to survive in cities can adapt to the local and often divergent environmental conditions compared to rural or (semi-)natural environments. Dispersal determines establishment, gene flow, and thus the potential for local adaptation. Since habitats in urban environments are highly fragmented, and show substantial turnover, contrasting adaptive effects on dispersal are expected. Fragmentation selects against dispersal while patch turn-over is expected to promote the evolution of dispersal. We here show both processes to act in concert when different scales are considered. Dispersal behavior of juvenile, lab-reared garden spiders from three mid-sized European cities were tested under standardized conditions. While long-distance dispersal showed to be overall rare, short-distance dispersal strategies increased with urbanization at small scales, but declined when urbanization was quantified at large scales. We discuss the putative drivers behind these differences in natal dispersal and highlight its importance for urban evolution and ecology.

Spider collection and husbandry

Adult females A. diadematus were collected in the field in both Belgium and Denmark. In Belgium, 57 females were randomly sampled in the larger area of Ghent following a gradient of urbanization (see locations in Supplementary Figure 1) in October 2020 and 2021. At these dates, female A. diadematus have all mated but not yet laid eggs. In Denmark, 65 adult females were sampled at 4 locations following a paired sampling design: spiders were collected in the city (Aarhus and Aalborg, urban locations) and their close countryside (rural locations). Females were housed individually in the lab after which temperature was gradually reduced (to mimic natural conditions) to stimulate egg laying. Housing conditions were very similar in Belgium and Denmark (details can be found in supplementary information). To ensure successful hatching, A. diadematus eggs need a cold period of at least 4 weeks (Canard, 1984). After this cold period, the temperature was stepwise increased to 21°C (Supplementary Table 1 for exact rearing conditions). One egg sac can contain several hundred juveniles (i.e., spiderlings), and egg hatching typically starts within one week after reaching 21°C. After hatching, spiderlings stay together close to the egg sac forming aggregations or clutches. With time, spiderlings start to interact and diffuse, which increases the distance between spiderlings and thus the size of the aggregation. Aggregation size was measured for the Belgium spiders and used as a proxy for competition. 

Dispersal assays

We applied a modified experimental setup of Bonte et al. (2003). The setup consists of two pots with five wooden sticks fixed with a plaster layer that are placed on a platform which is subject to a chaotic airflow generated by a fan at one meter with the wind velocities ranging between 0.3 m/s and 2 m/s. The ambient temperature was 21°C, but the setup was additionally lighted by a 100W incandescent lamp which mimics the effect of the sun (Supplementary Figure 2). Five spiders are placed individually at the top of each stick with a soft brush. Individuals displaying either rappelling (i.e., spider climb on an anchored silk thread) or ballooning (spiders disperse aerially attached to a non-anchored silk thread) were directly removed from the setup. Individuals only showing pre-dispersal tiptoeing behavior or using silk from other spiders to move between sticks were left on the platform for ten minutes to display ballooning or rappelling. The observation was stopped after ten minutes. After each assay, all spider silk was removed to minimise the effects of the previous clutch on the frequency of dispersal in spiderlings in the new batch (see De Meester & Bonte 2010). After the assay, spiderlings were placed back in their clutch.

Dispersal experiments were executed approximately two to three weeks after hatching, an age at which spiderlings have usually all left the eggsac and formed characteristic aggregations. A total of 372 spiderlings from 65 clutches were tested once for Denmark. A total of 1682 spiderlings from 57 clutches were tested 2-3 times at intervals of 4–7 days for Belgium.

Quantification of urbanization

For the Ghent region (Belgium), we determined urbanization levels for the different sampling locations by calculating the percentage built-up areas (BUA). We used the object-oriented reference map of Flanders as a vectorial layer (Large-scale Reference Database 2018) that defines building contours. Roads, public squares and parking lots infrastructure are not included, therefore, an estimate of 15% built-up area can be considered highly urbanized. BUA was determined for multiple scales, using buffers of 50, 100, 250, 500, 1000, 2000 and 4000 m radius around the sampling location. BUA values for the sampling locations in the Ghent region are strongly correlated (Supplementary Figure 3). For Denmark, data on urbanization levels around the study sites was obtained in a similar way, this time using OpenStreetMap (OpenStreetMap contributors, 2022) as a data source. All polygons in the broader study areas with key="building" were extracted using the osmdata R package (Padgham et al. 2017) and used to estimate the percentage of built-up area around each site at the same spatial scales.

Statistical analyses

Dispersal at the individual level is a binary yes/no response and follows a binomial distribution at the clutch-level (proportion of offspring dispersing). Individuals were scored after displaying their first behavior (ballooning/rappelling), and responses were, in principle, not independent at the trial level. However, since individual spiderlings may also show no responses, the correlation between ballooning and rappelling frequencies at the clutch level is low (r = -0.37). Because spiderlings from the Danish locations did not show ballooning at all, we decided to analyze rappelling and ballooning frequencies separately. We used mixed logistic regressions with urbanization as a fixed effect and the clutch ID as random effects. Because of design differences between Ghent and Danish samples, we analyzed them in separate and slightly different models. First, because Ghent spiderlings were tested multiple times, aggregation size was added as a covariate and proxy for the current level of competition. Second, compared to the random sampling design in the Ghent data, the paired design from the Denmark data breaks down to four locations with different levels of urbanization. Therefore, for the Ghent data, we ran 7 models separately for the different spatial scales of urbanization and selected the best model according to AIC criteria, while for the Danish data we analyzed urbanization as a factor (2-levels) variable. We qualitatively contrasted this output with the Ghent results. Ripley-K statistics (Spatstat package, Baddeley et al. 2015) were used to quantify the level of spatial correlation of both the urbanization rates and dispersal responses. For dispersal traits showing spatial autocorrelation (ballooning) spatial regressions were done using the glmmTMB package (Brooks et al. 2017). These showed the absence of spatial correlation in urbanization rates at small spatial scales (Supplementary Figure 4), implying random sampling across the city of Ghent. Evidently, when scaling-up, correlations become prominent as we only have one city and samples show spatial correlation according to the maximal scale of urbanization. Raw data and urbanization estimates are available on Zenodo. Maps were constructed in QGIS (version 3.283., QGIS Geographic Information System. QGIS Association) using the OpenStreetMap plugin (OpenStreetMap contributors, 2022), and the contours of Belgium were outlined using geoBoundaries (Runfola et al. 2020).