Vertical asymmetries in orb webs

Transcription

1 bs_bs_banner Biological Journal of the Linnean Society, 2015, 114, With 7 figures Vertical asymmetries in orb webs SAMUEL ZSCHOKKE 1 * and KENSUKE NAKATA 2,3,4 1 Department of Environmental Sciences, Section of Conservation Biology, University of Basel, St. Johanns-Vorstadt 10, CH-4056 Basel, Switzerland 2 Kyoto Women s University, Kitahiyoshi-cho 35, Higashiyama-ku, Kyoto, , Japan 3 Faculty of Human Environment, Nagasaki Institute of Applied Science, 536 Aba-machi, Nagasaki , Japan 4 Tokyo Keizai University, Minami-machi , Kokubunji, Tokyo, , Japan Received 21 August 2014; revised 22 October 2014; accepted for publication 22 October 2014 In almost all vertical orb webs the hub is above the geometric centre and consequently, the extent of the capture area is larger below the hub than above. In addition to this vertical web-extent asymmetry, orb webs show vertical asymmetries in number of spiral loops, mesh widths, and angles between radii. However, it was unknown whether these asymmetries are adaptations to the web-extent asymmetry or whether they are linked to gravity in a different way than through web-extent asymmetry. We reviewed known vertical asymmetries of orb webs, and we analysed the asymmetries of webs built by four different Cyclosa species, which show large intra- and inter-specific variation in web-extent asymmetry. We found all analysed structural asymmetries to be linked both to web-extent asymmetry and to gravity: Larger web extents below the hub and gravity both led to more sticky-spiral loops and to smaller angles between radii below the hub, whereas web-extent asymmetry and gravity had opposing effects on mesh width (mean and peripheral). Independent of web-extent asymmetry, almost all analysed webs had narrower peripheral meshes and smaller angles between radii below the hub than above. We interpret the narrow peripheral meshes along the web s lower edge as an adaptation to prevent tumbling prey from escaping, and the small angles between radii as an adaptation to prevent the sticky-spiral lines in these narrow meshes to come into contact with each other The Linnean Society of London, Biological Journal of the Linnean Society, 2015, 114, ADDITIONAL KEYWORDS: Araneidae Cyclosa gravity prey tumbling hypothesis spider web sticky-spiral spacing trap web architecture web geometry. INTRODUCTION The architecture of biological structures has been selected to increase the fitness of their bearer, or in the case of structures built by animals to increase the fitness of their builder. Animal-built structures can influence the fitness of their builder in several ways; among other things they can increase the builder s survival and food intake, or they can decrease the expenditures of their builder (Hansell, 2005). Orb webs built by araneoid spiders are a well known example of animal-built structures and they show, like other biological structures, a large intra- and *Corresponding author. samuel.zschokke@unibas.ch inter-specific variability (Witt, Reed & Peakall, 1968; Eberhard, 1986; Sandoval, 1994; Liao, Chi & Tso, 2009). Despite their structural variability, almost all vertical orb webs share the trait that they are vertically asymmetric with a larger extent of the capture area below the hub than above (e.g. Mayer, 1952; Witt & Reed, 1965; Risch, 1977; Endo, 1988; ap Rhisiart & Vollrath, 1994; Heiling & Herberstein, 1998). Empirical and theoretical studies suggest that this asymmetry is primarily an adaptation to the spider s prey-capture behaviour, reflecting the spider s ability to run downwards faster than upwards (Masters & Moffat, 1983; ap Rhisiart & Vollrath, 1994; Coslovsky & Zschokke, 2009; Maciejewski, 2010; Nakata & Zschokke, 2010; Zschokke & Nakata, 2010). 659

2 660 S. ZSCHOKKE and K. NAKATA In addition to this web-extent asymmetry, various structural traits of vertical orb webs have been found to show a vertical asymmetry: there are more sticky-spiral loops below the hub, the mesh widths (spacing between adjacent sticky-spiral loops) tend to be narrower below the hub, and the angles between adjacent radii are smaller below the hub. However, because only webs with larger lower extents were analysed until now, it was not possible to determine whether these vertical asymmetries of the structural traits are adaptations to the web-extent asymmetry, or whether they are linked to gravity in a different way than through web-extent asymmetry. Recently, Nakata & Zschokke (2010) described the behaviour and web asymmetry of several Cyclosa spiders from Japan, some of which face upwards in their webs and therefore build webs with larger upper extents (Fig. 1A). As these webs consequently show a large intra- and interspecific variability in web-extent asymmetry, they provide the unique possibility to separately examine the effects of web-extent asymmetry and of gravity on the structural asymmetries of vertical orb webs. The present study aimed: (a) to review the known vertical structural asymmetries of orb webs, (b) to formulate predictions on these structural asymmetries in orb webs with larger upper extents, and (c) to make a detailed analysis of these structural asymmetries in orb webs built by Cyclosa spp. spiders, which show a large intra- and inter-specific variability in web-extent asymmetry. When analysing these webs, we asked in particular whether the structural asymmetries are linked to web-extent asymmetry (by determining how the structural asymmetries change with the vertical and horizontal web-extent asymmetry) and whether the structural asymmetries are Figure 1. A, Web of Cyclosa argenteoalba, a species where the spider always faces upwards when waiting for prey at the hub of the web (unlike most other orb-web spiders). These webs are characterized by a positive vertical web-extent asymmetry (i.e. the web extent above the hub is larger than below the hub). Note also the area along the web s lower periphery with a very fine mesh. This particular web had a web-extent asymmetry of 0.082, a sticky-spiral asymmetry of 0.010, a mesh-width asymmetry of 0.112, a peripheral mesh-width asymmetry of 0.327, and a radii-angle asymmetry of Inset: a C. argenteoalba spider waiting for prey on the hub of its web. B, Definitions of measures used to assess the vertical asymmetries. The web-extent asymmetry is based on the vertical distances between the hub and the outermost sticky-spiral loops (blue). The sticky-spiral asymmetry is based on the number of sticky-spiral loops above and below the hub (green). The mesh-width asymmetry is based on the average sticky-spiral mesh widths above and below the hub (green). The peripheral mesh-width asymmetry is based on the average mesh width of the sticky spiral in the outermost quarter of the capture area. The radii-angle asymmetry is based on the average angle between radii in the upper quadrant and the lower quadrant (red). For comparison, the equivalent measures were also assessed horizontally, but they are omitted from this figure for reasons of clarity.

3 VERTICAL ASYMMETRIES IN ORB WEBS 661 linked to gravity in another way (by determining whether the vertical structural asymmetries persist in vertically extent-symmetric webs). REVIEW OF VERTICAL STRUCTURAL ASYMMETRIES In the following we summarize the current knowledge, which is based on orb webs with a larger lower extent of the capture area. NUMBER OF STICKY-SPIRAL LOOPS There are more sticky-spiral loops below the hub than above in almost all vertical orb webs (Reed, Witt & Jones, 1965; Witt & Reed, 1965; ap Rhisiart & Vollrath, 1994; Vollrath, Downes & Krackow, 1997; Heiling & Herberstein, 1998). Spiders achieve this asymmetry by inserting partial sticky-spiral loops below the hub (ap Rhisiart & Vollrath, 1994; Zschokke, 2011). Exceptions to this almost universal trait are very rare. Possibly the most striking exceptions are the ladder webs built by spiders of the genus Scoloderus (Eberhard, 1975; Stowe, 1978). These webs are much elongated above the hub, most likely as an adaptation to capture moths: moths avoid adhering to orb webs by shedding the scales of their wings. Thus they tumble down along the face of orb webs, leaving a telltale trace of scales, until they can escape at the bottom of the web. However, if the web is long enough for the moths to loose enough scales so that their thus de-scaled wings adhere to the sticky silk, they can be caught by the spider (Stowe, 1978). In Scoloderus webs, the necessary length of the web to capture moths is achieved by adding hundreds of partial sticky-spiral loops above the hub. Consequently, Scoloderus webs always have many more sticky-spiral loops above the hub than below. More sticky-spiral loops above the hub have also been described for Zygiella x-notata webs, but only for the rare cases where the retreat is below the hub, in which case the spider shifts the hub downwards, to be nearer to the retreat (Le Guelte, 1967), leading to a larger number of sticky-spiral loops above the hub. Finally, the non-vertical orb webs of Gasteracantha spp. and Micrathena spp. always have roughly the same number of sticky-spiral loops above and below the hub, because they have no partial sticky-spiral loops (Zschokke & Vollrath, 1995). MESH WIDTH One of the most relevant structural traits of orb webs is their mesh width (i.e. the spacing between subsequent loops, Figure 1B) of the sticky spiral. Narrow meshes increase the web s prey interception and prey retention capabilities (Rypstra, 1982; de Crespigny, Herberstein & Elgar, 2001; Blackledge & Zevenbergen, 2006) but require higher building costs per web area (Eberhard, 1986), increase the drag of the web (Lin, Edmonds & Vollrath, 1995) and make the web more susceptible to damage in windy conditions (Zaera, Soler & Teus, 2014). Consequently spiders build orb webs with wide meshes when they are exposed to wind (Vollrath et al., 1997; Liao et al., 2009; Wu et al., 2013) but see (Hieber, 1984). A closer look at orb webs reveals that mesh widths are not the same throughout the entire web (cf. Figure 1A), but there seems to be no general rule, whether the average mesh width is narrower above or below the hub. Tilquin (1942), mainly working with Z. x-notata, found the sticky-spiral meshes below the hub to be narrower than those above. Peters (1939) described a web of Araneus diadematus with narrower meshes below the hub, while Zschokke (2011) found no difference in mesh width between the lower and the upper parts in A. diadematus webs. Finally, Eberhard (2014) found the meshes in non-vertical orb webs to be narrower below the hub in Micrathena gracilis and narrower above the hub in Leucauge mariana. MESH-WIDTH VARIABILITY While it is not entirely clear, whether there is a vertical asymmetry in average mesh width, it is undisputed that the mesh width tends to increase from near the centre of the web to the periphery for most orb-web spiders (Zschokke, 2002; Eberhard, 2014). Peters even suggested a constant mesh shape throughout the web, i.e. that the mesh width increases linearly with mesh length (= distance between neighbouring radii) (Peters segment rule; Peters, 1947, 1951, 1954). Later studies found the mesh width to be more uniform below the hub than above (Le Guelte, 1967; Witt, Rawlings & Reed, 1972), or, in other words, that Peters segment rule applies, at least to some extent, to the upper part and to both sides (left and right), but not to the lower part of the web (ap Rhisiart & Vollrath, 1994). The reason for the increasing mesh width towards the periphery could be that the two functions of stopping prey (by the radii) and retaining prey (by the sticky spiral) should be matched and that consequently, since mesh lengths increase towards the periphery in most orb webs, mesh widths should also increase (radius density hypothesis, Zschokke, 2002). This could explain why in orb webs with parallel radii (webs of Nephila spp. and some ladder webs) the mesh width does not increase towards the periphery (Eberhard, 1975; Eberhard, 2014; Kuntner, Coddington & Hormiga, 2008; Robinson & Robinson, 1972; Wiehle, 1931), but it cannot explain why the

4 662 S. ZSCHOKKE and K. NAKATA mesh width does not increase towards the bottom in other orb webs. PERIPHERAL MESH WIDTH The reason why the mesh width does not increase towards the bottom of orb webs is probably related to the observation that some webs have an area along the web s lower edge with particularly narrow meshes (de Crespigny et al., 2001; Eberhard, 1975; own observations: Fig. 1A). After impact, prey sometimes tumbles down the web under gravity (Eisner, Alsop & Ettershank, 1964; Stowe, 1978; Forster & Forster, 1985; Eberhard, 1989; Zschokke et al., 2006). Consequently, the narrow-meshed area along the lower edge of vertical orb webs can provide the last opportunity to ensnare prey that would otherwise tumble off the web (prey tumbling hypothesis, de Crespigny et al., 2001; Eberhard, 2014), since prey retention is best in areas with narrow meshes (Rypstra, 1982; Blackledge & Zevenbergen, 2006). ANGLE BETWEEN RADII The average angle between radii is smaller below the hub than above (Peters, 1937; Tilquin, 1942; Mayer, 1952; Le Guelte, 1967; ap Rhisiart & Vollrath, 1994; Heiling & Herberstein, 1998), probably to achieve ideal lengths of sticky silk between two radii, which can be achieved by smaller angles between radii where the web extent is larger, i.e. towards the bottom of the web (ideal segment length hypothesis, Zschokke, 2002). Recently, Eberhard (2014) formulated the sticky-spiral entanglement hypothesis to explain the relationship between mesh width and mesh length. When two adjacent sticky-spiral loops are close together, there is the danger that they will come in contact with each other when they are displaced (e.g. by wind), causing them first to adhere to each other and then even to merge with each other (Zschokke, 2004). The longer the segment of sticky silk hanging free, and the smaller the distance between the sticky silk lines, the larger the danger for the sticky silk to come into contact with each other and thus causing this damage (Eberhard, 2014). Since the meshes along the bottom of the web are narrower than along the top of the web (see above), this hypothesis gives a further reason why the angles between radii should be smaller below the hub than above. DOUBLED RADII To compensate the larger forces on the radii above the hub (Wirth & Barth, 1992), a few spider species are known to double the radii above the hub, either along their entire length, like Caerostris darwini (Gregorič et al., 2011), or partially, but to a greater extent than below the hub, like Zilla diodia (Zschokke, 2000). However, as we had no indication that the Cyclosa species used in our study double the radii, we did not further evaluate this trait. Among the Cyclosa species, only C. caroli has been described to show a behaviour indicating partially doubled radii (Eberhard, 1981). PREDICTIONS Based on the structural asymmetries described above, we formulated the following predictions for upsidedown orb webs, i.e., for webs with larger upper extents (cf. also Fig. 7). NUMBER OF STICKY-SPIRAL LOOPS As we expected no or only small changes in mesh width with different web-extent asymmetries (see first hypothesis below), we expected the number of sticky-spiral loops to change proportionally with web extent above and below the hub, and therefore the sticky-spiral asymmetry to increase proportionally (i.e. slope = 1.0) with web-extent asymmetry. MESH WIDTHS The variability of mesh widths within vertical orb webs is probably largely an adaptation to prey capture (prey tumbling hypothesis, see above). Based on this postulate, we formulated two hypotheses, which both predict the peripheral meshes below the hub to be always smaller than above. The first hypothesis assumes no influence of spider orientation on the distribution of mesh widths, and consequently predicts the distribution of mesh widths to be independent of web-extent asymmetry and spider orientation. The second, alternative hypothesis takes into account the conclusion by Zschokke & Nakata (2010: 51): Waiting head up can be advantageous for spiders whose downwards upwards running speed asymmetry is small and who experience a high prey tumbling rate. This implies that spiders facing upwards are those experiencing a higher prey tumbling rate, and consequently, assuming that the area along the web s lower edge with narrow meshes is an adaptation to prevent tumbling prey falling out of the web (prey tumbling hypothesis), we expected that spiders facing upwards should build webs with even narrower meshes along the web s lower edge compared with webs built by downwards facing spiders. In addition, based on geometrical relationships, we expected the average mesh widths to be positively correlated with peripheral mesh widths (since peripheral meshes are included in the average mesh size) and negatively correlated to the number of spiral

5 VERTICAL ASYMMETRIES IN ORB WEBS 663 loops (the more spiral loops there are, the shorter must be the distance between them). ANGLES BETWEEN RADII For the angles between the radii we again formulated two predictions. Firstly, based on the ideal segment length hypothesis, we expected the angles between the radii to be smaller wherever the web extent is larger, as it has been described for Zygiella x-notata (Le Guelte, 1967) and for Eustala illicita (Hesselberg, 2013), and furthermore, we expected the average mesh lengths (= distance between two neighbouring radii) along the web periphery to be similar in all web quadrants and consequently that symmetric webs have similar angles between radii above and below the hub. Alternatively, based on the sticky-spiral entanglement hypothesis, we expected because of the narrower meshes along the web s lower edge that in symmetric webs the angles between radii are smaller below the hub, and furthermore, we expected the average mesh shapes along the web periphery to be similar in all parts of the web, as described for Z. x- notata (Peters, 1951). MATERIAL AND METHODS SPIDERS We analysed the webs of four species of the genus Cyclosa: Cyclosa ginnaga Yaginuma 1959, Cyclosa argenteoalba Bösenberg & Strand 1906, Cyclosa confusa Bösenberg & Strand 1906 and Cyclosa octotuberculata Karsch 1879, all of which are diurnal orb-web spiders in the Western part of Japan, building vertical or near vertical webs. These species were selected for our study because they are common and because they differ in their orientation when waiting for prey: C. ginnaga and C. argenteoalba always face upwards on their webs, whereas C. octotuberculata always face downwards and C. confusa individuals face in various directions: some face upwards and others face downwards or sideways (Nakata & Zschokke, 2010). In C. argenteoalba and C. confusa, which are both multivoltine, adult spiders can be found from May to November. C. ginnaga adults mainly occur in July and August, and C. octotuberculata adults occur from late May to early August. We collected C. ginnaga (adult females) and C. octotuberculata (adult females) in 2002, and C. confusa (adult and subadult females) in 2001 and 2002, during their respective adult season, at Mt. Inasa Park, Nagaski, Japan, and we collected C. argenteoalba (sub-adult males, and adult and sub-adult females) in 2001 and 2002 at the Botanical Garden of Kyoto University, Kyoto, Japan. DATA COLLECTION Spiders were kept individually in Perspex frames ( cm, which is large enough for these spiders to expand their webs, Nakata, 2012) Each frame was put between two transparent cover sheets, thus ensuring that the spiders were not exposed to any wind, which is known to influence mesh widths (Vollrath et al., 1997; Liao et al., 2009; Wu et al., 2013). The frames were placed under a natural D: L cycle in the laboratory, where the spiders started building webs within several days after collection. The webs were then photographed against a black background (Zschokke & Herberstein, 2005), and the photographs were printed out on A4 paper. From each web, the positions of selected points were digitized, i.e., their x, y coordinates were entered into the computer with the aid of a digitizing tablet. To assess the mesh widths in the upper, lower, left and right quadrants of the web, a line was drawn from the hub in the middle between two radii as close as possible to vertically upwards and downwards (light green lines in Figure 1B), as well as horizontally left and right. We then digitized the points, where these lines intersected the sticky-spiral loops. To assess the angles between the radii and the web-extent asymmetry (of the capture area), the attachment points of the outermost sticky spiral on every radius were also digitized. Finally, the centre of the hub (the place where the radii would meet) was visually determined and also digitized. For the calculations, this point was taken as centre of the coordinate system; angles of the radii were measured clockwise, with straight up being 0. Nomenclature of orb-web elements follows Zschokke (1999). MEASUREMENTS For each quadrant (upper, lower, left and right), we calculated based on the digitized x, y coordinates the web extent (i.e. the vertical or horizontal resp. distance from the hub to the outermost sticky-spiral loop), the number of sticky-spiral loops, all mesh widths (i.e., the spacing between two subsequent sticky-spiral loops), and all angles between neighbouring radii (Fig. 1B). The angles were grouped into four quadrants, delimited by those radii whose angles were closest to 45, 135, 225 and 315 and the average angle between radii (including subsidiary radii) was then calculated for each quadrant. For the mesh widths, we calculated for each quadrant the average mesh width, the peripheral mesh width (i.e. the average mesh width of the outermost quarter of the sticky-spiral loops), as well as the mesh-width increase (i.e. the slope of the linear regression of mesh widths against the distance from the hub), and the mesh-width variability (measured as coefficient of

6 664 S. ZSCHOKKE and K. NAKATA variation). The average peripheral mesh lengths were calculated trigonometrically for each quadrant based on the average angle between the radii and the distance between the hub and the middle of the outermost quarter of the capture area (Fig. 1B). ANALYSES As measure for the vertical asymmetries, we used the ratio (upper lower)/(upper + lower), where upper was the value above the hub (e.g., extent of web above hub, number of sticky-spiral loops above hub or average mesh width above hub) and lower was the equivalent value below the hub. These ratios yielded the asymmetries for web-extent, sticky spiral, mesh width, peripheral mesh width, and radii angle. Webs with equal values above and below the hub had an asymmetry of 0.0 for this trait, and webs with smaller values above the hub thus had a negative asymmetry. Our calculation of web-extent asymmetry is identical to that used by Coslovsky & Zschokke (2009), Hesselberg (2010), Nakata & Zschokke (2010) and Zschokke (1993; 2011). In addition, we calculated the horizontal asymmetries using the ratio (right left)/(right + left) to assess how the different structural traits are affected by web-extent asymmetry in horizontal direction, where gravity plays no role. For the analysis, webs with fewer than 18 stickyspiral loops were omitted to ensure enough data points when analysing mesh-width variability within the web. In total, we analysed the webs built by 93 different spiders of four different species: C. ginnaga (12 spiders with 24 webs, all facing upwards), C. argenteoalba (23 spiders with 32 webs, all facing upwards), C. confusa (47 spiders with 113 webs, 73 facing downwards, 17 facing sideways, 23 facing upwards), and C. octotuberculata (11 spiders with 22 webs by, all facing downwards). To ensure independence between samples, we used individual spiders as experimental units. For spiders, which had built several webs, we calculated the average for all analysed measures. STATISTICAL ANALYSES For all structural asymmetries, we calculated a linear regression with web-extent asymmetry as the independent variable. In addition, we performed a generalized linear model (glm, type I) for each structural asymmetry with web-extent asymmetry and species as independent variables. The interaction was omitted since it was not significant in preliminary tests (P > 0.35). To evaluate, whether the orientation of the spider (facing downwards vs. facing upwards) affected the peripheral mesh-width asymmetry, we compared the residuals of the linear regression with an unpaired t-test between upwards facing and downwards facing spiders (spiders facing sideways were omitted from this comparison). Comparisons between the quadrants were made with a mixed model with species as factor and quadrant as repeated measure, with interaction. To determine which quadrants differed from each other, Tukey-Kramer post-hoc tests were applied. To compare the two hypotheses, that radii angles were either optimized for a constant mesh length or for a constant mesh shape along the web s periphery, we first calculated the average mesh lengths as well as the average mesh shape (length/width) for the outermost quarter of each quadrant. To test which of the two was more uniform, we calculated the coefficient of variation (CV) for both measures among the quadrants of each web, and compared the two coefficients of variation with a mixed model with species as factor and CVs as repeated measure (without interaction, since it was not significant in a preliminary test). All statistical analyses were performed with SAS University Edition 9.4 (SAS Institute, 2014). Unless indicated otherwise, the sample size was N = 93 (i.e. all spiders) for all analyses. RESULTS NUMBER OF STICKY-SPIRAL LOOPS The asymmetry of the number of sticky-spiral loops increased with web-extent asymmetry (sticky-spiral asymmetry = * web-extent asymmetry; r 2 = 0.66, P < ; intercept: P < ; continuous black line in Fig. 2), implying that the number of sticky-spiral loops increased with web extent and, as indicated by the negative intercept, that symmetric webs had more sticky-spiral loops below the hub than above. The glm revealed no significant difference between species (F 3,88 = 2.36, P = 0.077; coloured lines in Fig. 2). Horizontally, the asymmetry of the number of sticky-spiral loops increased in a similar way with web-extent asymmetry (horizontal sticky-spiral asymmetry = * horizontal web-extent asymmetry; r 2 = 0.59, P < ; intercept: P = 0.89; dashed black line in Fig. 2), but with no intercept, implying that horizontally symmetric webs had similar numbers of sticky-spiral loops on both sides. AVERAGE STICKY-SPIRAL MESH WIDTH The asymmetry of average sticky-spiral mesh widths increased with web-extent asymmetry (mesh-width asymmetry = * web-extent asymmetry; r 2 = 0.52, P < ; intercept: P < ; Fig. 3),

7 wider meshes below hub Mesh-width asymmetry wider meshes above hub more sticky-spiral Sticky-spiral asymmetry more sticky-spiral loops above hub VERTICAL ASYMMETRIES IN ORB WEBS C. ginnaga C. argenteoalba C. confusa C. octotuberculata 0 loops below hub Web-extent asymmetry larger extent below larger extent above Figure 2. Relationship between vertical web-extent asymmetry and sticky-spiral asymmetry, showing that the number of sticky-spiral loops increased with web extent, and that symmetric webs had more sticky-spiral loops below the hub. Each point represents the web(s) built by one spider. The continuous black line is the linear regression based on all points, the dashed black line shows the regression line of the same relationship in horizontal direction (points not shown). The coloured lines represent the species-wise regression lines as modelled by the glm C. ginnaga C. argenteoalba C. confusa C. octotuberculata Web-extent asymmetry larger extent below larger extent above Figure 3. Relationship between vertical web-extent asymmetry and mesh-width asymmetry, showing that the average mesh width of the sticky spiral increased with web extent, and that symmetric webs had narrower meshes below the hub. Each point represents the web(s) built by one spider. The continuous black line is the linear regression based on all points, the dashed black line shows the regression line of the same relationship in horizontal direction (points not shown). The coloured lines represent the species-wise regression lines as modelled by the glm. implying that the mesh width of the sticky spiral increased with web extent and, as indicated by the positive intercept, that symmetric webs had wider meshes above the hub than below. The glm revealed no differences between species (F 3,88 = 2.07, P = 0.11). Horizontally, the asymmetry of average stickyspiral mesh widths also increased with web-extent

8 wider peripheral meshes above hub Peripheral mesh-width asymmetry 666 S. ZSCHOKKE and K. NAKATA C. ginnaga C. argenteoalba C. confusa C. octotuberculata 0.3 asymmetry (horizontal mesh-width asymmetry = * horizontal web-extent asymmetry; r 2 = 0.59, P < ; intercept: P = 0.009; dashed line in Fig. 3), but with only a very small intercept, implying that horizontally symmetric webs had meshes of similar average widths on both sides. The analysis of the sticky-spiral mesh widths for the non-peripheral meshes gave similar results to those of the average sticky-spiral mesh widths (nonperipheral mesh width asymmetry = * web-extent asymmetry; r 2 = 0.48, P < ; intercept: P < ). PERIPHERAL MESH WIDTH The peripheral mesh widths were smaller below the hub than above in all spiders (and in all but 2 of the 191 analysed webs). The asymmetry of the peripheral sticky-spiral mesh widths increased with web-extent asymmetry (peripheral mesh-width asymmetry = * web-extent asymmetry; r 2 = 0.40, P < ; intercept: P < ; Fig. 4), implying that the mesh width of the peripheral sticky spiral increased with web extent and, as indicated by the positive intercept, that symmetric webs had narrower peripheral meshes below the hub than above. The glm revealed differences between species (F 3,88 = 3.32, P = 0.024). The residuals of the regression did not differ between webs with downwards Web-extent asymmetry larger extent below larger extent above Figure 4. Relationship between vertical web-extent asymmetry and peripheral mesh-width asymmetry, showing that the relative width of the peripheral sticky-spiral meshes increased with web extent and that all spiders had built webs with larger peripheral mesh widths above the hub than below. Each point represents the web(s) built by one spider. The continuous black line is the linear regression based on all points, the dashed black line shows the regression line of the same relationship in horizontal direction (points not shown). The coloured lines represent the species-wise regression lines as modelled by the glm. facing spiders and webs with upwards facing spiders (unpaired t-test, d.f. = 84, t = 0.08, P = 0.94). Horizontally, the asymmetry of peripheral stickyspiral mesh width also increased with web-extent asymmetry (horizontal peripheral mesh-width asymmetry = * horizontal web-extent asymmetry; r 2 = 0.20, P < ; intercept: P = 0.57; dashed line in Fig. 4), but with no intercept, implying that horizontally symmetric webs had peripheral meshes of similar widths on both sides. The average mesh width correlated positively with the peripheral mesh width (Pearson correlation; above the hub: r = 0.83, P < ; below the hub: r = 0.91, P < ) and correlated negatively with the number of spiral loops (above the hub: r = 0.90, P < ; below the hub: r = 0.89, P < ). MESH-WIDTH VARIABILITY The mesh-width increase towards the periphery differed significantly between the web quadrants (F 3,356 = 123.6, P < ), and between species (F 3,356 = 6.67, P = ) with a significant interaction (F 9,356 = 6.14, P < ). It was strongest above the hub (average 0.022), intermediate towards either side (0.009 and resp.), and smallest below the hub ( 0.010; Tukey Kramer post hoc tests: P < for all comparisons, except between the web s left and right side: P = 0.94). Below the hub, the mesh width

9 larger angles below hub Radii-angle asymmetry larger angles above hub VERTICAL ASYMMETRIES IN ORB WEBS C. ginnaga C. argenteoalba C. confusa C. octotuberculata Web-extent asymmetry larger extent below larger extent above decreased significantly towards the periphery (one sample t-test, t = 6.17, P < ), implying a mesh width decrease towards the web s lower edge. This decrease was also highly significant (P < ) for three of the four species when tested separately; in C. ginnaga, it was not significant with an average slope of 0.001, SD (d.f. = 11, t = 0.72, P = 0.49). Mesh-width variability (measured as coefficient of variation) was correlated with the absolute value of the slope of the mesh-width increase (Pearson correlation coefficients ranged from 0.31 to 0.56 in the quadrants). Consequently, mesh-width variabilities also differed between the web quadrants (above hub: 0.308; to the sides: and resp.; below hub: 0.267; F 3,368 = 10.94, P < ), but the Tukey Kramer post-hoc tests revealed differences only between the quadrant above the hub and the other quadrants (P < ), but not between the other quadrants (P > 0.95). ANGLE BETWEEN RADII The angles between the radii were larger above the hub than below in all but one of the 93 analysed spiders (and in all but 2 of the 191 analysed webs). The asymmetry of the angles between radii decreased with web-extent asymmetry (radii-angle asymmetry = * web-extent asymmetry; r 2 = 0.11, Figure 5. Relationship between vertical web-extent asymmetry and radii-angle asymmetry, showing that the angle between neighbouring radii decreased with web extent and that that all spiders except one had built webs in which the angles between the radii above the hub were larger than below the hub. Each point represents the web(s) built by one spider. The continuous black line is the linear regression based on all points, the dashed black line shows the regression line of the same relationship in horizontal direction (points not shown). The coloured lines represent the species-wise regression lines as modelled by the glm. P = ; intercept: P < ; Fig. 5), implying that the angle between neighbouring radii decreased with web extent and, as indicated by the positive intercept, that symmetric webs had smaller angles between radii below the hub. The glm revealed no difference between species (F 3,88 = 1.03, P = 0.38). Horizontally, the asymmetry of the angles between radii also decreased with web-extent asymmetry (horizontal radii-angle asymmetry = * horizontal web-extent asymmetry; r 2 = 0.20, P < ; intercept: P = 0.24; dashed line in Fig. 5), but with no significant intercept, implying that the angle between neighbouring radii decreased with web extent and that horizontally symmetric webs had similar angles between radii on both sides. To assess the possible reasons for the observed distribution of radial angles, we compared the CV of the peripheral mesh lengths with the CV of the peripheral mesh shapes. The mesh lengths were slightly more uniform (mean CV = 0.155) than the mesh shapes (mean CV = 0.179; F 1,181 = 8.23, P = ) with a significant difference between species (F 3,181 = 10.23, P < ). Within the web s peripheries, the length of the meshes differed significantly between the web quadrants (F 3,356 = 13.37, P < ), and between species (F 3,356 = 97.76, P < ) with no interaction (F 9,356 = 1.16, P = ). Meshes were longest at the

10 34.9 loops 38.3 loops 42.0 loops 28.3 loops 35.4 loops 42.7 loops 668 S. ZSCHOKKE and K. NAKATA w = 1.61 l = 7.78 s = 4.83 w = 1.62 l = 7.69 s = 4.73 w = 1.61 l = 7.56 s = w = 0.99 l = 1.80 s = 1.82 w = 1.39 l = 1.72 s = 1.24 top (7.58), intermediate on the sides (both 6.60), and shortest at the bottom (5.84; Tukey Kramer post-hoc tests: P = for comparisons between top and either side and P = 0.03 for comparisons between either side and the bottom; Fig. 6). The shape of the meshes also differed significantly between the web quadrants (F 3,356 = 35.73, P < ), and between species (F 3,356 = 68.41, P < ), but with a significant interaction (F 9,356 = 1.94, P = 0.046). However, in contrast to mesh length, mesh shapes were most elongated at the bottom (5.70) intermediate on the sides (5.24 and 5.03), and least elongated at the top (4.23; Tukey Kramer post-hoc tests: P < for comparisons between top and either side, and P < and P = resp. for comparisons between either side and the bottom; Fig. 6). DISCUSSION w = 1.07 l = 6.94 s = Our results showed that all analysed structural traits of vertical orb webs were linked both to w = 0.83 l = 1.50 s = 1.82 w = 1.09 l = 1.34 s = 1.24 w = 0.99 l = 5.76 s = w = 0.64 l = 1.45 s = 2.27 w = 0.89 l = 1.38 s = 1.55 w = 0.71 l = 5.03 s = 7.07 Figure 6. Typical structure of the central part of orb webs with larger lower extent (left), with similar sized upper and lower extents (centre) and with larger upper extent (right). These drawings are based on average values for number of sticky-spiral loops, mesh widths, mesh-width distribution and radii angles of webs normalized to the same overall vertical height (100 units) of the capture area and on the simplifying assumption of a linear change of mesh widths towards the periphery. The drawing on the left is based on 57 webs with web-extent asymmetries between 0.20 and 0.05, the one in the centre on 47 webs with asymmetries between and , and the one on the right on 43 webs with web-extent asymmetries between and The angles represent the average angles for the upper and the lower quadrant respectively, w denotes the mesh width, l the mesh length and s the mesh shape (length/width) for the outermost and innermost meshes in the upper and the lower quadrant. web-extent asymmetry and to gravity (Fig. 7). The number of sticky-spiral loops, average mesh-width and peripheral mesh-width increased with web extent, whereas the angle between radii decreased with web extent. In addition, vertically symmetric webs had more sticky-spiral loops, narrower meshes, narrower peripheral meshes and smaller angles between radii below the hub than above. As these structural asymmetries were observed in webs with the same extent above and below the web, we conclude that they are linked to gravity in some way; in particular, we suggest that they are direct or indirect adaptations to prey tumbling (see below). In horizontal direction, the effects of web-extent asymmetry on the structural traits were comparable to those in vertical direction for all analysed structural traits, and there were, as expected, no horizontal structural asymmetries in horizontally symmetric webs. No qualitative differences in these relationships were found between the four analysed Cyclosa species.

11 VERTICAL ASYMMETRIES IN ORB WEBS 669 Figure 7. Summary of the proposed adaptive relationships between different web traits. Arrows denote suggested relationships, with plus symbols indicating a positive influence on the trait shown on the following box, and minus symbols likewise indicating a negative influence. Red arrows show adaptations to web extent, whereas the adaptations to gravity or prey tumbling respectively are shown in blue. Numbers refer to the hypothesis and the empirical support shown below: 1 Gravity; empirical: (Masters & Moffat, 1983; ap Rhisiart & Vollrath, 1994) 2 Gravity; empirical: (Eberhard, 1989; Zschokke et al., 2006) 3 Classic web asymmetry hypothesis (Masters & Moffat, 1983); empirical: (ap Rhisiart & Vollrath, 1994) 4 Refined web asymmetry hypothesis; the dashed line indicates that this relationship is true only under special circumstances (Zschokke & Nakata, 2010) 5 Refined web asymmetry hypothesis (Zschokke & Nakata, 2010); empirical: (Nakata & Zschokke, 2010) 6 Prey tumbling hypothesis (de Crespigny et al., 2001; Eberhard, 2014) 7 Ideal segment length hypothesis (Zschokke, 2002); empirical: (Le Guelte, 1967; Hesselberg, 2013), this study 8 Geometry; empirical: this study 9 Spiral entanglement hypothesis (Eberhard, 2014); empirical: this study. NUMBER OF STICKY-SPIRAL LOOPS AND AVERAGE MESH WIDTH As predicted, the number of sticky-spiral loops increased with web extent in all quadrants. However, with a slope of 0.60 in the vertical and 0.66 in the horizontal direction, the number of sticky-spiral loops increased less than expected. It is therefore not surprising that the average mesh widths also increased with web extent. This implies that spiders achieved larger web extents by simultaneously increasing the number of sticky-spiral loops and increasing the mesh widths. PERIPHERAL MESH WIDTH AND MESH-WIDTH VARIABILITY One important reason why the meshes below the hub were on average narrower than above was that the meshes at the periphery, i.e. in the outermost quarter of the capture area, were narrower below the hub than above in almost all webs. We interpret this to be an adaptation to prevent tumbling prey from falling out of the web (prey tumbling hypothesis). In webs with larger upper parts, typically built by upwards facing spiders, the difference between the peripheral mesh widths was most pronounced. However, we found no effect of the spider s orientation per se on the peripheral mesh width, as could have been expected following the hypothesis that facing upwards was advantageous for spider experiencing a high tumbling rate (Zschokke & Nakata, 2010), combined with the prey tumbling hypothesis. We also found no support for Peter s segment rule sensu stricto (1947; 1951; 1954), which suggests that the mesh shape is constant throughout the web. Even towards the top, where the mesh width increased most towards the periphery, the average mesh shape (length/width) increased from 1.93 near the hub to 4.75 at the periphery (Fig. 6). For comparison, Peters (1951) reported average mesh shapes of 3.58 for Zygiella x-notata and 5.10 for Araneus diadematus. However, the mesh-width changes from centre to periphery we found were quite similar to those described for A. diadematus (ap Rhisiart & Vollrath, 1994): a strong increase towards the top and a weaker increase towards the sides. However, in contrast to the A. diadematus webs studied by ap Rhisiart & Vollrath, we even found a decrease of mesh widths towards the bottom of the web. The pattern of mesh-width variability we found in the quadrants was the same as found in earlier studies: largest above the hub, intermediate to the

12 670 S. ZSCHOKKE and K. NAKATA sides and smallest below the hub (Le Guelte, 1967; Witt et al., 1972). ANGLE BETWEEN RADII As predicted, the angles between radii changed with web-extent asymmetry. In addition, we found the angles below the hub to be smaller than above in almost all webs, even in webs with much larger upper extents. This indicates that there is an inherent tendency for angles between the radii to be smaller below the hub than above. We suggest that these smaller angles below the hub are adaptations to the narrow meshes along the lower edge of the web, and thus indirectly to prey tumbling. Our evaluation of the hypotheses that the spiders adjust the angles between radii with the aim to either obtain uniform mesh lengths along the web periphery (ideal segment length hypothesis) or, alternatively, to obtain uniform mesh shapes along the web periphery (sticky-spiral entanglement hypothesis) yielded no clear results. The variabilities of the peripheral mesh lengths and of the peripheral mesh shapes were fairly similar, but the two measures showed different asymmetries: whereas meshes were longer at the top of the web than at the bottom, mesh shapes were more elongated at the bottom of the web than above. These observations suggest that the spiders may have used a compromise between the two aims. SYNTHESIS To summarize, all traits of orb webs analysed in our study were affected by web-extent asymmetry and by gravity simultaneously. In normal webs, i.e., in webs with the hub above their geometric centre, web extent and gravity reinforced each other concerning the number of sticky-spiral loops (more towards the larger web extent and more towards the bottom) and concerning the angles between the radii (smaller towards the larger web extent and smaller towards the bottom). In contrast, web extent and gravity were found to be counteracting each other in normal webs concerning average mesh width and peripheral mesh width, i.e. larger meshes towards the larger web extent, but narrower meshes towards bottom of web (which is the larger web extent in normal webs). As there seem to be opposing factors on the optimal mesh width, and as we can expect these factors to be of different relevance in different species, it is not surprising that species differ in whether the average mesh width is larger above or below the hub (Tilquin, 1942; Zschokke, 2011; Eberhard, 2014). CONCLUSIONS Based on our results, we suggest that the distribution of the mesh widths within vertical orb webs is an adaptation to prey capture, i.e., an increase of mesh width towards the top and towards the sides so that the web s two functions of stopping prey (by the radii) and retaining prey (by the sticky spiral) are always matched (radius density hypothesis). In addition, there is a mesh width decrease towards the periphery below the hub to increase the stickiness of the web s lowest area in order to prevent tumbling prey from falling out of the web (prey tumbling hypothesis). Our study thus strongly supports the conclusion by Eberhard (2014), that different parts of orb webs differ in their characteristics and in their function. In addition, we suggest that the angles between the radii represent a compromise between homogeneous, ideal mesh lengths along the web s periphery (ideal mesh length hypothesis) and homogeneous mesh shapes along the web s periphery (sticky spiral entanglement hypothesis). Interestingly, the functional relationships we suggest, namely that the angles between the radii are an adaptation to the distributions of mesh lengths and shapes along the web periphery is opposite to the temporal relationship during web building: during web building, spiders first build the radii and later add the sticky spiral (Foelix, 2011). Clearly, more research is needed to determine the spider s behavioural rules, which define the mesh width during the building of the peripheral turns of the sticky spiral. The conclusions of our study require confirmation by further studies to examine to what extent our results apply to orb-web spiders of other genera. Prey tumbling, which is a central part of our reasoning, may not have the same importance in some other genera, because their webs differ in orientation and stickiness from those of Cyclosa spiders; the non-vertical web orientation of some genera reduces prey tumbling, because prey managing to free itself and falling down does not hit other sticky silk in non-vertical orb webs, and the sticky silk produced by spiders of some other genera is stickier than that of Cyclosa (Opell, 1997; Agnarsson & Blackledge, 2009), which may also lead to less prey tumbling. In addition, tumbling differs strongly between different kinds of prey (Zschokke et al., 2006), and consequently, the tumbling rate a spider experiences depends very much on the prey spectrum it encounters (which is unknown for most species). Nevertheless, we are convinced that not only the web extent but also structural traits of vertical orb webs are affected by gravity directly and indirectly in many ways, as much as gravity affects the behaviour of many other organisms whose life is not constrained to a horizontal plane (e.g. Moya-Laraño et al., 2009; Fraser, Coleman & Klein, 2010; Prenter, Fanson & Taylor, 2012).

13 VERTICAL ASYMMETRIES IN ORB WEBS 671 ACKNOWLEDGEMENTS We are grateful to Claudia List for digitizing the webs, to Atushi Ushimaru for his help to collect subject spiders, to Marie Herberstein and Thomas Hesselberg for fruitful discussions, to Peter Stoll for statistical advice, and to five anonymous reviewers for their helpful comments and suggestions. SZ thanks his family for bearing with him, even though he spent so much time working on this study. REFERENCES Agnarsson I, Blackledge TA Can a spider web be too sticky? Tensile mechanics constrains the evolution of capture spiral stickiness in orb-weaving spiders. Journal of Zoology 278: Blackledge TA, Zevenbergen JM Mesh width influences prey retention in spider orb webs. Ethology 112: Coslovsky M, Zschokke S Asymmetry in orb-webs: an adaptation to web building costs? Journal of Insect Behavior 22: de Crespigny FEC, Herberstein ME, Elgar MA The effect of predator-prey distance and prey profitability on the attack behaviour of the orb-web spider Argiope keyserlingi (Araneidae). Australian Journal of Zoology 49: Eberhard WG The inverted ladder orb web of Scoloderus sp. and the intermediate orb of Eustala (?) sp. Araneae: Araneidae. Journal of Natural History 9: Eberhard WG Construction behaviour and the distribution of tensions in orb webs. Bulletin of the British Arachnological Society 5: Eberhard WG Effects of orb-web geometry on prey interception and retention. In: Shear WA, ed. Spiders webs, behavior, and evolution. Stanford: Stanford University Press, Eberhard WG Effects of orb-web orientation and spider size on prey retention. Bulletin of the British Arachnological Society 8: Eberhard WG A new view of orb webs: multiple trap designs in a single structure. Biological Journal of the Linnean Society 111: Eisner T, Alsop R, Ettershank G Adhesiveness of spider silk. Science 146: Endo T Patterns of prey utilization in a web of orbweaving spider Araneus pinguis (Karsch). Researches on Population Ecology 30: Foelix RF Biology of spiders. New York: Oxford University Press. Forster LM, Forster RR A derivative of the orb web and its evolutionary significance. New Zealand Journal of Zoology 12: Fraser CML, Coleman RA, Klein JC Up or down? Limpet orientation on steeply sloped substrata. Aquatic Biology 11: Gregorič M, Agnarsson I, Blackledge TA, Kuntner M How did the spider cross the river? Behavioral adaptations for river-bridging webs in Caerostris darwini (Araneae: Araneidae). PLoS ONE 6: e Hansell MH Animal architecture. Oxford: Oxford University Press. Heiling AM, Herberstein ME The web of Nuctenea sclopetaria (Araneae, Araneidae): relationship between body size and web design. Journal of Arachnology 26: Hesselberg T Ontogenetic changes in web design in two orb-web spiders. Ethology 116: Hesselberg T Web-building flexibility differs in two spatially constrained orb spiders. Journal of Insect Behavior 26: Hieber CS Orb-web orientation and modification by the spiders Araneus diadematus and Araneus gemmoides (Araneae: Araneidae) in response to wind and light. Zeitschrift für Tierpsychologie 65: Kuntner M, Coddington JA, Hormiga G Phylogeny of extant nephilid orb-weaving spiders (Araneae, Nephilidae): testing morphological and ethological homologies. Cladistics 24: Le Guelte L La structure de la toile et les facteurs externes modifiant le comportement de Zygiella-x-notata Cl. (Araignées, Argiopidae). Revue du Comportement Animal 1: Liao C-P, Chi K-J, Tso I-M The effects of wind on trap structural and material properties of a sit-and-wait predator. Behavioral Ecology 20: Lin LH, Edmonds DT, Vollrath F Structural engineering of an orb-spider s web. Nature 373: Maciejewski W An analysis of the orientation of an orb-web spider. Journal of Theoretical Biology 265: Masters WM, Moffat AJM A functional explanation of top-bottom asymmetry in vertical orbwebs. Animal Behaviour 31: Mayer G Untersuchungen über Herstellung und Struktur des Radnetzes von Aranea diadema und Zilla x-notata mit besonderer Berücksichtigung des Unterschiedes von Jugend- und Altersnetzen. Zeitschrift für Tierpsychologie 9: Moya-Laraño J, Vinkovic D, Allard CM, Foellmer MW Optimal climbing speed explains the evolution of extreme sexual size dimorphism in spiders. Journal of Evolutionary Biology 22: Nakata K Plasticity in an extended phenotype and reversed up-down asymmetry of spider orb webs. Animal Behaviour 83: Nakata K, Zschokke S Upside-down spiders build upside-down orb webs: web asymmetry, spider orientation and running speed in Cyclosa. Proceedings of the Royal Society of London, Series B 277: Opell BD The material cost and stickinesss of capture threads and the evolution of orb-weaving spiders. Biological Journal of the Linnean Society 62: Peters HM Studien am Netz der Kreuzspinne (Aranea diadema). I. Die Grundstruktur des Netzes und Beziehungen zum Bauplan des Spinnenkörpers. Zeitschrift für Morphologie und Ökologie der Tiere 32: