Unique behavioural modifications in the web structure of the cave orb spider Meta menardi (Araneae, Tetragnathidae)

In the last decade there has been a renewed interest in the study of behavioural adaptations to environmental constraints with a focus on adaptations to challenging habitats due to their reduced ecological complexity. However, behavioural studies on organisms adapted to nutrient poor subterranean habitats are few and far between. Here, we compared both morphological traits, in terms of relative leg lengths, and behavioural traits, captured in the geometry of the spider web, between the cave-dwelling spider, Meta menardi, and two aboveground species from the same family (Tetragnathidae); Metellina mengei and Tetragnatha montana. We found that the webs of the cave spider differed significantly from the two surface-dwelling species. The most dramatic difference was the lack of frame threads with the radii in the webs instead attaching directly to the surrounding rock, but other differences in relative web size, web asymmetry and number of capture spiral threads were also found. We argue that these modifications are likely to be adaptations to allow for a novel foraging behaviour to additionally capture walking prey within the vicinity of the web. We found only limited evidence for morphological adaptations and suggest that the cave orb spider could act as a model organism for studies of behaviour in energy-poor environments.


Results
We encountered a total of 19 spiders and webs of Meta menardi from caves near Nottingham, U.K, a total of 29 Metellina mengei from woods near Oxford, U.K. and a total of 25 Tetragnatha montana near rivers in Oxford, U. K.
Size and leg length. In terms of overall size, all three species differed from each other as is evident from the principal component analysis (PCA) ordination plot (Fig. 1A). The cave orb spider, even though the spiders measured in this study were late instar juveniles, was larger than the adults of two non-cave tetragnathid species in terms of cephalothorax width (mean ± SEM, M. menardi: 2.57 mm ± 0.10 vs. M. mengei: 1.56 mm ± 0.03 and T. montana: 1.72 mm ± 0.04), but was similar in total length to T. montana (8.28 mm ± 0.32 vs. 8.81 mm ± 0. 16) with both being longer than M. mengei (4.47 mm ± 0.10). However, when we control for size by looking at relative leg lengths (total leg length divided by cephalothorax width), the cave orb spider does not stand out compared to the aboveground species (Fig. 1). While there was a significant difference in the relative lengths of both leg I (Linear mixed model (LMM): F = 310.3, df = 2, P < 0.0001) and leg III (LMM: F = 162.1, df = 2, P < 0.0001), the outstanding species for either measure was a non-cave one (T. montana for leg I and M. mengei for leg III). A posthoc test showed that the cave spider had longer relative leg lengths than M. mengei for both leg I (t = 6.16, df = 10.25, P = 0.0003) and leg III (t = 14.73, df = 6.42, P < 0.0001). Significant differences were found also between M. menardi and T. montana for leg I (t = − 13.51, df = 9.46, P < 0.0001), but not leg III (t = 0.91, df = 5.18, P = 0.659), and between M. mengei and T. montana for leg I (t = − 24.84, df = 12.64, P < 0.0001) and leg III (t = − 15.293, df = 10.93, P < 0.001) (Fig. 1).
Orb web geometry. A visual comparison of the overall structure of webs of the three species, strongly indicates a difference between them. The webs of the cave spider M. menardi were smaller with fewer spirals and most noticeably had few or no frame threads (Fig. 2). When analysing the web characteristics individually, significant differences were found for relative web area (total web area divided by cephalothorax width squared), where M. menardi webs were smaller than the webs of the two aboveground species ( However, only non-significant differences were found between the species for the number of radii (Fig. 3B, LMM: F = 3.26, df = 2, P = 0.076) and relative mesh height (total mesh hight divided by cephalothorax width) (Fig. 3C, LMM: F = 4.37, df = 2, P = 0.068). As stated, the most obvious visual difference between the webs is the lack of frame threads with the radii attaching directly to the substrate resulting in the radii, in effect, acting as mooring threads in the cave spider M. menardi. This dramatic difference is further highlighted when visualizing the linear relationship between the number of radii and mooring threads (Fig. 4A) and number of frame threads (Fig. 4B). Unsurprisingly, the statistical analysis supports this as M. menardi had significantly more mooring threads (LMM: F = 23.5, df = 2, P < 0.0001; M. menardi-M. mengei posthoc: t = 6.73, df = 13.0, P < 0.0001; M. menardi-T. montana posthoc: t = 5.64, df = 12.3, P = 0.0003) and significantly   (Fig. 4). Significant differences between the species were also found in vertical web asymmetry, absolute web area and number of capture spiral turns, but not in web shape ( Table 1)

Discussion
We observed, in accordance with our hypothesis, that webs of the cave-dwelling orb spider Meta menardi showed statistically discernible and visually apparent differences to the webs of related aboveground species. Most notable was the lack of the frame that usually enclose the capture spiral and radii of orb webs. This resulted in the radii acting as mooring threads by attaching directly to the surrounding rock substrates. Thus, these cave webs had a much larger number of mooring threads than standard orb webs, which could provide information to the spider about prey walking on the rock surface in vicinity of the web. To our knowledge, this study, utilizing the unique opportunities for quantifying foraging behaviour offered by spider webs, is one of very few studies on cave-dwelling animals that demonstrate a significant difference in behaviour compared to closely related surface-dwelling counterparts 6,38 .
Morphological adaptations to the subterranean environment have been studied in much greater details than behavioural adaptations with examples of convergent evolution in a number of traits including loss of eyes, loss of pigments and limb elongation 3,5,39 . To complement our behavioural studies, we therefore also compared limb elongation in the cave orb spiders given that they show no evidence of eye or pigment loss 15,40 . However, in our comparative study, M. menardi, contrary to our hypothesis, did not have the longest relative leg length for either leg I or III, although it did have statistically significantly longer relative leg lengths than its closest surface relative, Metellina mengei. The facultative cave spider Metellina merianae has relative leg lengths in between these two species 41 , so more detailed studies are needed to determine if some degree of limb elongation might be present  www.nature.com/scientificreports/ in cave orb spiders. The large overall size of M. menardi could potentially also be viewed as an adaptation to the subterranean environment as an increased body size is a frequent morphological adaptation seen in cave arachnids 3 . Meta spiders are generally among the largest members of the family Tetragnathidae 15 , so their size could indicate some degree of morphological adaptation, although it is worth noting that large surface-dwelling araneid spiders are relatively common in the tropics.
In the Japanese cave orb spider, Meta japonica, a very similar web structure to the one we found for M. menardi was observed, showing an omission of a complete frame (number of frame sections, mean ± SD: 1.1 ± 0.6) and a similar number of radii which attached directly to the wall (14.6 ± 5.3) 31,42 . However, the number of capture spiral turns (17.0 ± 4.3) was considerably greater than recorded for M. menardi in this study (6.0 ± 1.7), suggesting that the mesh height would be lower. A similar number of radii (16.9 ± 3.0) and spiral turns (13.0 ± 3.0) were observed in the facultative cave spider M. merianae, but this species does not show any reduction of frame threads and only a few mooring threads connect the web to the cave wall 41 . Interestingly, the webs of M. menardi were more symmetrical than the webs of the two surface-dwelling species. However, as web asymmetry in tetragnathid spiders is closely linked to inclination with more vertically inclined webs being more asymmetric 43,44 , the symmetry of the cave webs is possibly due to these webs being more horizontally inclined (D. Simonsen, Pers. Obs.) than an adaptation to subterranean conditions per se. The webs of M. menardi were smaller than those of its surface relatives, although the webs of M. mengei and M. menardi were more similar to each other than either one was to Tetragnatha montana. Given that web area tends to increase with spider size 45,46 , our results are diametrically opposed to what was expected with the relative web area of M. menardi being significantly smaller than found in the other two species.
A relatively smaller web combined with the lower number of capture spirals and the slightly increased relative mesh height, suggests that the webs of M. menardi are not as important for prey capture as in other species. However, there was no difference in the number of radii, resulting in M. menardi having a higher density of radii and potentially more stopping pontential, which, combined with their relatively smaller capture area, may suggest that they rely on capturing fewer, but larger prey 47,48 . This is though not supported by the prey captured in their web, which seems to consist mainly of small gnats, mosquitoes and caddisflies 15,32 , whereas woodland Tetragnatha and Metellina are known to feed on relatively larger prey including large mosquitoes and tipulids 49,50 , although Metellina mengei from Wytham Woods seems to predominantly feed on small aphids 51 . Finally, aerial prey (mainly mosquitoes) has been observed to escape quickly from M. menardi webs 52 , suggesting they might not successfully retain larger and stronger insects should such prey be intercepted.
The diet of M. menardi has been confirmed, from several field studies, to consist of between 36 and 69% non-flying prey including slugs, spiders and millipedes 15,32,40,53 compared to a diet predominantly consisting of flying prey observed in other orb spiders 18 . It is generally accepted that Meta spiders combine both traditional on-web prey capture with novel off web hunting to achieve their broadened diet 15,30,31,40 . However, a foraging behaviour involving hunting independent of the web seems unlikely as this would require evolving a completely new additional prey capture behaviour 15 . Instead, with the radii attaching to the wall and M. menardi residing in the hub of its web, as seems to be its usual position (D. Simonsen. Pers Obs) 52 , it would be able to detect the presence and direction of both prey flying into the web and walking prey colliding with the radii through the same sensory mechanism of vibrations transmitted through the radii 54,55 . This hypothesis is supported by the observation from Novak et al. 40 that M. menardi "swiftly traced prey from its web onto the wall" after the prey had collided with the radii and by an unpublished pilot study by Peter Smithers (Pers. Comm. 2019) that indicated that M. menardi in the laboratory catch woodlice by running down and attaching silk threads to them after they walk into a radial thread. Thus, the observed lack of frame threads in the webs of cave orbs spiders from this study is likely to be a behavioural adaptation to broaden the diet in response to the rarity of flying prey in subterranean habitats. To confirm this hypothesis, either careful field work or laboratory experiments, which result in the direct observation of prey capture from the wall would be required combined with a comparative analysis of web structure and prey capture in one of the confirmed surface-dwelling Meta species; for example the endemic M. stridulans from the laurel forests of Madeira 56 .
The comparative approach adopted in this study was somewhat skewed as all cave spiders measured were late instar juveniles, while the aboveground spiders measured were all adults. However, while this may have impacted some of our quantitative results, it is unlikely to have had any influence on the qualitative differences and conclusions presented in this paper as orb webs generally do not undergo strong ontogenetic changes 57 . The changes they do show either relate to web asymmetry, as older and larger spiders build webs with larger lower parts due to faster gravity-assisted downwards running or to derivative web features such as free sectors not present in juvenile spiders [58][59][60] . The tetragnathids studied here, all built standard orb webs with the exception of the lack of frame threads in subadult cave spiders, and as they built inclined webs, the link between size and asymmetry is expected to be weak 43,44 . Thus, the major differences such as number of frames, number of radii and relative web size between the cave spider and the two aboveground spiders found in this study are therefore likely to be due to habitat rather than ontogeny, especially as observations on adult Meta menardi suggest they construct similar webs to those in our study 30,61 . However, we cannot rule out that the differences found in asymmetry are caused by ontogeny.
Orb spiders are known to show an impressive flexibility in their orb web designs in response to a large number of biotic and abiotic factors 27,35,62 , including in response to spatial constraints 63,64 and to climatic variables 25 . Could the modification in web geometry in the current study therefore be explained by behavioural flexibility due to environmental differences rather than an evolved behavioural adaptation for foraging in caves? Would either of the terrestrial relatives produce similar webs if they found themselves in a subterranean habitat? The average temperature in the cave was 6 °C colder than in the wood and 10 °C colder than along the river. At colder temperatures, orb spiders build smaller webs with fewer capture spiral turns, but a similar number of radii 25 , which is similar to our findings. However, mesh height was also found to increase, contrary to our findings, and more Scientific Reports | (2021) 11:92 | https://doi.org/10.1038/s41598-020-79868-w www.nature.com/scientificreports/ importantly no changes in overall web structure are known to occur as a consequence of temperature. Similarly, wind, albeit not directly measured in this study, is likely to be lower in the caves than in the wood and along the river. Higher winds are known to lead to smaller webs with fewer capture spiral turns 25,65,66 , where we found the opposite with the smallest webs in the sheltered caves. Noteworthy is also that a field study on M. mengei did not find any differences in web design between webs built at the exposed woodland edge and webs built in the sheltered interior of the wood 67 . Again, wind is not known to affect the fundamental structure of the web. Yoshida and Shinkai 31 suggested that the reduction of frame threads in Meta japonica is a result of building webs in small cavities on the cave wall and ceiling, as M. menardi is also observed to do. Behavioural flexibility seems to play some role as when the frame threads were present, they were predominantly found at larger distances from the rock wall (D. Simonsen, Pers. Obs.), potentially suggesting that the presence of a frame is dependent on the topology of the environment. However, orb spiders under spatial constraints in the laboratory attempt to maintain overall web area by elongating their webs to match the available space and by reducing mesh height, contrary to our findings for M. menardi 25,63,68 . In addition, most species facing spatial constraints do not change the structure of their webs, although the araneid Eustala illicita and the tetragnathid Leucauge argyra have been observed to attach radii directly to the surface when building orb webs in very constrained artificial frames and tubes 63,64 . These webs, however, only had small sections of frame threads missing, contrary to our findings of an almost complete elimination of frame threads. The anapid Anapisona simoni shows more dramatic changes in frame, radial and capture threads that are somewhat similar to the responses observed in this study 69 , but anapid webs are very different from tetragnathid webs as they are more three-dimensional and lacking the non-sticky scaffolding spiral. Overall, comparisons between strongly constrained spiders in the laboratory and cave spiders in the wild have only limited validity as natural caves typically offers a range of potential web sites as demonstrated by the normal cave orb webs of the sympatric Metellina merianae 41 . Finally, it is also possible that the observed changes occur as a direct response to the low food availability in the subterranean habitats. However, while orb spiders are known to alter life history and silk properties due to starvation 70,71 , only minor quantitative changes in orb geometry, including increased area, but reduced number of capture spirals, were found 72,73 as opposed to the large-scale quantitative and qualitative differences observed in this study. In addition, Meta menardi does not show any specific physiological adaptations to starvation 74 . Overall, the changes we observed in the webs of M. menardi are unlikely to be caused by behavioural flexibility alone, although more detailed laboratory studies of web-building behaviour of the cave spider are needed to fully eliminate this possibility.
In conclusion, our findings that the cave orb spider showed significant departures in its web geometry compared to two related surface-dwelling orb spiders, especially in terms of a reduction of frame threads with radii attaching directly to the cave wall and ceiling, indicate that this behavioural modification could have evolved in response to the limited availability of flying prey and high relative abundance of walking prey in subterranean habitats. While more observations and experiments on the prey capture and web-building behaviour of cave orb spiders is needed, this is to our knowledge the first quantitative study to suggest novel behavioural adaptations in a terrestrial cave arthropod. This has broader implications since the cave spiders in our study only show limited evidence for morphological adaptations (no loss of eyes or pigment, but with some indication of minor limb elongation compared to their closest surface dwelling relative), which means these spiders could be in the initial stages of a full adaptation to the cave environment with behavioural adaptation preceding morphological adaptations (which might eventually evolve into a complete loss of the web and adaptations to off-web locomotion and prey capture). Thus, we suggest that Meta cave spiders could be a potential model organism for studying behavioural flexibility and the evolution of behaviour in energy-poor environments.

Methods
Study organisms. The European cave orb spider, Meta menardi (family Tetragnathidae), is a large spider with adult females measuring up to 15 mm in total length that inhabits the first 20-30 m of natural caves and artificial subterranean habitats including tunnels and mines, although it has also been found in cellars, tree crevices, under man hole covers and in cavities on scree slopes 15 . Since no species in the Meta genus, with the exception of the endemic Meta stridulans from Madeira, are known to occupy aboveground habitats, we compared the morphology and web geometry of M. menardi to two common sympatric aboveground tetragnathids; the closely related Metellina mengei (a fellow member of the Metainae subfamily) and Tetragnatha montana from the Tetragnathinae subfamily 15,33 . Metellina mengei is a small to medium-sized spider up to 6 mm in length that mainly inhabits humid woodland understory or shaded grassland, while T. montana is a medium to large-sized spider that can be up to 10 mm in total length and is found in humid woodland understory and in vegetation along streams and rivers. For all locations, the resident spiders of the webs were collected and taken back to the laboratory, where they were stored in 70% ethanol before their ID was confirmed, morphological data collected and their life stage determined based on the absence or presence of the epigyne. While all M. mengei and T. montana collected were adult females, the collected M. menardi turned out to be mid to late instar juvenile females. In general, adult Meta spiders are encountered much less frequently than juveniles 75 . Data collection: size and leg length. The collected spiders of all three species were brought to the laboratory where the following measurements were taken using a digital Vernier calliper under a stereo microscope to a precision of 0.1 mm: total length of the spider, cephalothorax width, as a measure of spider size and the combined patella-tibia length of leg I and III.
Data collection: web geometry. All webs located in the field were misted with water to make the silk threads more visible and hence easier to measure. Five different length characters (Fig. 5A) were measured using a digital Vernier calliper to the nearest mm: (1) The vertical ( d v ) and (2) horizontal diameters ( d h ) were measured from the outermost and opposite spiral threads (3) The upper ( r u ) and (4) horizontal radius ( r r ) were measured from the centre of the central hub to the outermost spiral thread. Finally, (5) the vertical diameter of the central hub and free zone ( H).
Count measurements (Fig. 5B) include the number of radii, the number of capture spiral turns along the mid radii in each of the 4 quadrants centres ( s Q1 , s Q2 , s Q3 , s Q4 ), the number of frame threads and the number of mooring points (i.e. the number of mooring threads connecting the web to the surrounding substrate).
The measured web parameters allowed us to derive these further web characteristics: Web area (the area of the capture spiral minus the free sector and hub) was estimated using the Ellipse-Hub formula76: The lower vertical radius ( r l ) and left horizontal radius ( r le ) were calculated as follows: Figure 5. www.nature.com/scientificreports/ Web asymmetry was characterized by a value between − 1 to 1, with 0 indicating a perfectly vertical symmetric web and a negative value indicating that the hub is located closer to the top than the bottom of the web and was found with the following formula 58 : Web shape was also characterized by a value between − 1 to 1, with 0 indicating a circular shape and a negative value indicating it is taller than it is wide and was found with this formula 77 : Mean mesh height, which is the distance between two adjacent capture spiral turns, was measured using this formula, which we adapted from the formula in Herberstein and Tso 76 in order to include the horizontal quadrants: Data analysis: morphology. To determine differences in morphology (total length, cephalothorax width, and patella-tibia length of leg I and leg III) between M. menardi and its surface-dwelling relatives, we performed a principal component analysis with the prcomp() function in R and used the ggplot2 package 78 to generate ordination plots. However, in order to investigate potential limb elongation in M. menardi, we followed Hesselberg and Simonsen 41 to control for spider size by using relative leg length (patella-tibia lengths of leg I and III divided by cephalothorax width) as the response variables in linear mixed models where species was the fixed factor with location (transect number and cave ID) as a random factor. To ensure normality, the relative leg length of leg I in the comparison of all species was transformed with the natural logarithm. In an attempt to separate phylogeny and habitat, we found pairwise contrasts with Kenward-Roger degrees of freedom and the Tukey Method for P value adjustment with emmeans() function from the emmans package 79 as a posthoc test, where the full model revealed significant differences between species. All models were built with the lmer() function from the lme4 package 80 in R 81 . P values were found with the Type II Wald F-test.
Data analysis: web geometry. Linear mixed models were developed following the approach described above for number of radii, absolute web area, number of mooring threads, web asymmetry, web shape, relative mean mesh height (mesh height divided by cephalothorax width) and relative web area (web area divided by cephalothorax width squared) with species as a fixed factor and location as a random factor. To ensure normality, the relative web area was logarithmic transformed. We used the same posthoc test described above for pairwise species comparisons, where the full model showed species was a significant variable. All models were built with the lmer() function from the lme4 package 80 in R 81 . P values were found with the Type II Wald F-test.
d v − r u = r l d h − r r = r le r u − r l r u + r l