Specialization directs habitat selection responses to a top predator in semiaquatic but not aquatic taxa

Habitat selectivity has become an increasingly acknowledged mechanism shaping the structure of freshwater communities; however, most studies have focused on the effect of predators and competitors, neglecting habitat complexity and specialization. In this study, we examined the habitat selection of semiaquatic (amphibians: Bufonidae; odonates: Libellulidae) and aquatic organisms (true bugs: Notonectidae; diving beetles: Dytiscidae). From each family, we selected one habitat generalist species able to coexist with fish (Bufo bufo, Sympetrum sanguineum, Notonecta glauca, Dytiscus marginalis) and one species specialized in fishless habitats (Bufotes viridis, Sympetrum danae, Notonecta obliqua, Acilius sulcatus). In a mesocosm experiment, we quantified habitat selection decisions in response to the non-consumptive presence of fish (Carassius auratus) and vegetation structure mimicking different successional stages of aquatic habitats (no macrophytes; submerged and floating macrophytes; submerged, floating, and littoral-emergent macrophytes). No congruence between habitat specialists and generalists was observed, but a similar response to fish and vegetation structure defined both semiaquatic and aquatic organisms. While semiaquatic generalists did not distinguish between fish and fishless pools, specialists avoided fish-occupied pools and had a preferred vegetation structure. In aquatic taxa, predator presence affected habitat selection only in combination with vegetation structure, and all species preferred fishless pools with floating and submerged macrophytes. Fish presence triggered avoidance only in the generalist bug N. glauca. Our results highlight the significance of habitat selectivity for structuring freshwater ecosystems and illustrate how habitat selection responses to a top predator are dictated by specialization and life history.


Methods
Study species. The common toad Bufo bufo (Linnaeus, 1758) (Anura: Bufonidae) is a widespread generalist. Highly poisonous bufotoxins in the skin of adults and tadpoles are an effective defense against vertebrate predators 36 , allowing spawning in large permanent water bodies with fish 26 . The green toad Bufotes viridis (Laurenti, 1768) inhabits warm and arid lowland regions, such as steppes, riverbanks or man-made structures (e.g., quarries, agricultural land, and urban areas). The spawning sites of this specialist include shallow, often ephemeral and warm water bodies, such as flooded fields, pools, or ditches, typically without fish and vegetation [42][43][44] .
The dragonfly Sympetrum sanguineum (Müller, 1764) (Odonata: Libellulidae) is a broadly distributed habitat generalist that inhabits all types of stagnant water bodies 45 . Its larvae develop dorsal and lateral spines on the abdomen that provide protection against fish 38 . The specialist Sympetrum danae (Sulzer, 1776) prefers acidic waters, such as peat bogs, fens, and moors; however, it may also inhabit shallow, densely overgrown ponds and ditches, particularly if associated with sedge, rush, and sphagnum moss 45 . Similar to other odonate species specialized in fishless habitats, the abdominal spines of its larvae are reduced, making them vulnerable to fish 46 .
The generalist aquatic true bug Notonecta glauca Linnaeus, 1758 (Hemiptera: Notonectidae) inhabits a wide range of habitats 47,48 , preferably with vegetation and including fish 48 . The specialist Notonecta obliqua Thunberg, 1787 is considerably less common and prefers bogs and fens with acidic water 49,50 . Its preference for fishless habitats, as well as dark body coloration, suggests high vulnerability to fish predation 51 .
The large diving beetle Dytiscus marginalis Linnaeus, 1758 (Coleoptera: Dytiscidae) is found in numerous aquatic habitats but prefers relatively deep, open waters. Because of its large body (27-35 mm), short lifespan as larvae, a hard cuticle, and defensive secretions, this generalist species is resistant to fish predation 37 . The specialist Acilius sulcatus (Linnaeus, 1758) is a smaller dytiscid species (15-18 mm). In spite of secreting defensive vertebrate-type steroids 52 , adults are highly susceptible to fish predation 53 and readily respond to fish chemical stimuli 54 . This species prefers larger water bodies with rich submerged vegetation 55 but may also colonize temporary habitats or those in early successional stages to escape predation by fish 54,56 . Design of the mesocosm experiment. The  www.nature.com/scientificreports/ sions of 12 × 6 × 3 m; steel construction covered with polyamide netting with mesh size 2 × 2 mm) located in the botanical garden of the University of Ostrava, Czech Republic (49.8274 N, 18.3259 E). Tanks within a block were arranged in two rows (three pools per row), spaced approximately 2 m apart (see Supplementary Fig. S2 online), and filled with well water. Tanks were surrounded by grass, upright branches that served as perches for dragonflies, and evenly distributed toad shelters composed of old wood and stone. Colonists/ovipositors were sampled according to a fully randomized 2 (fish or fishless pools) × 3 (no macrophytes; only submersed and floating macrophytes; submerged, floating, and littoral macrophytes) factorial design. The six treatments (presence or absence of fish × one of three vegetation types) were randomly assigned to tanks within each block. All tanks contained plastic predator cages (40 cm diameter × 40 cm height) covered with a polyethylene screen (mesh size of 5 × 5 mm), allowing larger prey to pass through while providing visual and chemical cues indicating the presence of fish to experimental organisms, but preventing fish from consuming them. Fish were represented by three 15-20-cm-long individuals of the crucian carp Carassius auratus (Cyprinidae). This invasive, omnivorous predator of nymphs and adults of aquatic insects, as well as eggs and early amphibian larval stages, is typically found in stagnant water bodies in Europe 57 . Submerged and floating macrophytes were represented by Nymphaea alba, Nuphar lutea, Elodea canadensis, Trapa natans, and Potamogeton natans, which were distributed evenly throughout the particular pools. Littoral (emergent) macrophytes were distributed along the pool edges, and consisted of Iris pseudacorus, Eleocharis palustris, Juncus spp., and Carex spp. The macrophytes were collected in the field, thoroughly washed, and carefully examined to prevent uncontrolled colonization. Vegetation levels, composition, and arrangement remained constant in fish and fishless pools.
Prior to starting the experiment (March, 2019), each pool was inoculated with detritus and organisms collected from aquatic habitats near the experimental site to provide prey for diving beetles and bugs, according to Briers and Warren 58 . A second inoculation was performed in July 2019, prior to starting the experiment with true bugs. Prey for adult dragonflies (flying insects, mainly Diptera and Lepidoptera) was captured in the adjacent meadows using a sweep net, and released evenly into each block approximately twice a week throughout the experimental period. Toads were fed by releasing laboratory-reared crickets (2 L per block) at the beginning of the experiment. Fish were fed common pelleted fish food.
Animal experimental setup and data sampling. Habitat selection of diving beetles was monitored from May 27 to June 30, 2019, which included the period of dispersal colonization flights of the study species 59 . Prior to sampling, 52 D. marginalis individuals were released into two blocks (26 per block), and 106 A. sulcatus in the other two blocks (53 per block) to avoid D. marginalis preying on smaller A. sulcatus 60 . The beetles were randomly divided into three equally populous groups, each of which was released onto one of three shallow trays (approximately 20 × 20 cm) placed between each pair of pools within a block. The trays held only a small amount of water to promote the dispersal of beetles. Habitat selection was examined approximately every three days, for a total of 11 sampling events, by removing all macrophytes and carefully checking for beetles using hand nets (0.5 cm and 1 mm mesh). Beetles were counted, transferred to a single container, and after examining all pools within a block, they were released following the same procedure as during initial stocking to allow for de novo selection. After the fifth sampling, D. marginalis individuals were relocated to the blocks originally inhabited by A. sulcatus and vice versa, to ensure a balanced experimental design. All blocks were examined on the same day.
Habitat selection of true bugs was monitored from August 8 to September 9, 2019 during the period of epigamic activity and colonization flights of study species 61 (nine sampling events). The blocks were stocked with 84 N. glauca individuals (21 per block) and 84 N. obliqua individuals (21 per block). As intrageneric predation is unlikely among similar-sized true bugs 62 , both species were kept in all four blocks simultaneously. Release and sampling were as in the case of beetles, except for the unnecessary species switch.
Habitat selection by the dragonfly S. danae was monitored from August 8 to 26, 2020 (nine sampling days; 32 tandem pairs, see below), and by S. sanguineum from August 28 to September 8 (eight sampling days; 31 tandem pairs). Two blocks were stocked with adult males (eight per block) and two blocks with adult females (eight per block) to avoid male sexual harassment impacting negatively on female fitness 30 . Habitat selection was assessed directly by observing ovipositing tandem pairs, whereby the female drops eggs directly into the water or sediment by performing abdominal dips in the air 45 . As the eggs within a clutch may be spread among several water bodies 30 , each such move is considered a habitat selection event. Observations were made around noon (between 10 and 14 h mean solar time), coinciding with the species peak epigamic activity 45 . The experiment was carried out one block at a time, with one tandem pair per observation. Each female from a "female" block was marked on the wings with a permanent marker and released into a "male" block. There, it was usually grasped almost immediately by one of the perching males, and copulation began, followed by oviposition into the pools. After oviposition, the female was returned to the female block. The mated male from the tandem pair was marked, released into the second "male" block, and replaced by an unmated male from that block to maintain constant numbers within a block and the same possible disturbance levels from other males. The same procedure was repeated until all females and males were mated.
Habitat selection by B. bufo was assessed from May 9 to 20, 2020 during the period of epigamic activity 42 and was preceded by the release of 12 individuals (six males and six females) into each block. The animals were evenly placed in their ground shelters. We were unable to obtain B. viridis females. However, males tend to select and occupy particular pools, and attract females through calling, usually leading to amplexus formation and oviposition 63 . Given the strong correlation between male calling and oviposition site 18 , male habitat selection was considered as determinative and was monitored from June 4 to July 5, 2020 (males called during the whole period) with 12 males per block. For each sampling event (six in B. bufo; 12 in B. viridis; sampling every 2-3 days), toads were caught by hand or by hand nets and placed in a single container. Habitat selection was www.nature.com/scientificreports/ considered to occur when an individual called in the immediate vicinity or from inside of the pool, or was present inside the pool without calling. After examining all pools within a block, individuals were released following the same procedure as during the initial stocking to enable de novo habitat selection. In B. bufo, oviposition events coincided with the habitat selection of males. In the subsequent analysis, only the habitat selection of males was considered (i.e., female choice and egg masses were not taken into account) so that the results were comparable with those obtained for B. viridis.

Statistical analysis.
Despite the large number of individuals used in the experiment, individual sampling events were not independent; each individual entered the experiment repeatedly and was only allowed to select among the six treatments within a block, without an opportunity to choose pools from other blocks (see Supplementary Fig. S1 online). Therefore, we used the generalized estimating equations (GEEs) for fitting marginal generalized linear models as they increase the model fit by accounting for correlations between variables 64 . The geeglm function, which has a syntax similar to glm but relies on a quasi-likelihood function instead of using full likelihood estimates, was applied to correlate datasets by fitting GEEs via the 'geese.fit' function of the 'geepack' package 65 . For all taxa, models with a Poisson distribution of errors (link = log) and exchangeable correlation structure were performed. In each model, predator presence (fish/no fish) and vegetation type (no macrophytes; submerged, floating; submerged, floating + littoral) and their interaction were always independent variables. The response variable was the number of individuals from each sampling event that selected specific pools, or the number of dips females performed during oviposition (in odonates). Identification of the sampling event (id) was used to specify individual clusters. An analysis of variance that compares models through Wald tests was used to get the most parsimonious model.

Discussion
The present study reports similar responses to fish and vegetation structure within semiaquatic and aquatic organisms, rather than the expected preferences sorted primarily by specialization. In semiaquatic taxa (toads and dragonflies), habitat generalists did not distinguish between fish and fishless pools, whereas species specialized in fishless habitats selected fishless pools with their preferred vegetation structure, in accordance with our first hypothesis. However, all aquatic taxa (true bugs and beetles) significantly preferred fishless pools with submerged and floating macrophytes, regardless of the level of specialization. Therefore, each group relies on a different mechanism of predator detection and a different strategy for habitat selection.
In toads, only the specialist B. viridis significantly avoided fish-occupied pools, which corroborates the findings of previous studies on anurans specialized in fishless habitats 14,17,18 . Although this species naturally uses pools without fish and vegetation, in our study, it preferred fishless pools with macrophytes. Therefore, instead of vegetation structure, it likely uses a different mechanism for predator avoidance, such as chemical detection 29 . In adults, this mechanism must be reliable and strongly selected during evolution as the larvae, which inhabit ephemeral pools and rarely encounter fish, are typically palatable and unable to detect fish cues and react adequately to the danger of being devoured 67 . The preference of the generalist B. bufo for pools with macrophytes was unsurprising, as this species attaches its eggs to vegetation to prevent them from being washed away 42 . Although the preference of B. viridis for this type of pool was unexpected, the presence of vegetation offers some additional benefits, such as promoting the survival of the offspring by enabling for food growth (e.g., periphyton) 68 , as well as offering shade and refuge 69,70 .
A different strategy may be applied by the generalist B. bufo, who did not distinguish between fish and fishless pools. Habitats with fish typically have more periphyton and phytoplankton due to lower levels of herbivorous zooplankton and aquatic insects 5 . Pools with fish may also entail fewer competitors; therefore, as B. bufo larvae are toxic and unpalatable to fish 36 , it may be desirable to oviposit in fish-occupied pools. However, an exclusive preference for fish habitats could lead to overcrowding and negative density-dependent effects on offspring fitness 6,71 . As certain amphibians tend to avoid conspecifics, especially those with cannibalistic larvae 9,17,67 , B. bufo may favor an ideal free distribution to avoid a competitive environment for its larvae 72 . Indeed, species that can detect predators and conspecific density might adopt a mixed oviposition strategy and, like B. bufo, lay eggs in both predator-free and predator-occupied patches 73 .
Fish avoidance by a specialist was also observed in dragonflies, complementing evidence from natural experiments with Libellulidae 21,22 . In our study, only ovipositing S. danae significantly avoided fish-occupied pools. As chemical detection of predator cues has not been documented in adult odonates 74 , polarotaxis has been suggested as the main mechanism for habitat selection 27,75 , even though no evidence suggests it could have a role in predator detection. Alternatively, the presence of fish may alter water surface polarization patterns 27 , as regular feeding of fish causes eutrophication 76 , in turn, affecting habitat selection 21 . Although turbidity did not differ visibly between predator treatments, it is possible that differences were detected by S. danae.
The specialist S. danae showed a considerable preference for pools with macrophytes, which aligns with its natural preferences. In odonates, the attraction to a particular vegetation structure has been proposed as another possible mechanism for habitat selection 25 , which may serve as an indicator of predator presence. However, considering the mismatch we created between the fish presence and vegetation structure, this mechanism seems irrelevant. Some taxa susceptible to fish but unable to detect them, such as Enallagma spp. damselflies 20 or the dragonfly specialist Sympetrum depressiusculum 77 , may rely on natal philopatry 78 . However, in our study, none of the animals emerged from the experimental pools. Fish detection and avoidance likely depend on more complex mechanisms, which will be determined by additional studies on other odonate species specialized in fishless habitats.
The generalist dragonfly S. sanguineum did not show predator avoidance or preference for a certain vegetation type. Based on evidence from the well-studied Leucorrhinia system, the larvae of dragonfly generalists may coexist with fish as they possess abdominal spines that provide defense against predation 38 and may further elongate during ontogeny when fish are actually present in the environment 41 . They may also use behavioral defenses, such as burst swimming or a reduction in activity 46 . Despite the lower abundance of prey for odonate larvae in habitats with fish 5 , and consequent impact on fitness 79 , generalist dragonflies may resemble B. bufo, and spread their reproductive effort among fish-free and fish-occupied patches to avoid negative density-dependent effects on offspring. Such behavior has been described in mosquitoes 19,80 in response to the actual presence of competitors, whereas in S. sanguineum the pools were completely free of competitors and there was only one ovipositing tandem pair at a time. Therefore, this could indicate risk-spreading 81 , whereby individuals unable to detect risk deposit their clutches among different habitat patches to increase the probability of offspring survival (i.e., bet-hedging). This was evidenced by tandem pairs of S. sanguineum ovipositing immediately after mating and spraying one clutch into several nearby pools. In contrast, oviposition of S. danae was preceded by flying around the net cage for a long time and careful selection of the suitable pool, in which the whole clutch was usually placed. Hence, spraying eggs among fish and fishless pools seems to be a general strategy to avoid predators and/or competitors only in species that can coexist with fish.
In both groups of aquatic insects, predator presence significantly affected habitat selection only in combination with vegetation structure: all species significantly preferred submerged and floating macrophytes in fishless pools but not in those with fish, regardless of their natural habitat preferences. In contrast, Binckley and Resetarits 1 found no interaction between habitat complexity and fish presence during habitat selection by aquatic beetles. Moreover, the same diving beetles significantly selected fishless pools, regardless of their complexity. Food availability and quality, as well as plant community type (i.e., complexity) largely define dytiscid habitats 82 . Habitat complexity and prey density play important roles also in coexistence among true bugs. Vegetation provides shelter 83  www.nature.com/scientificreports/ may hamper both backswimmers and beetles 5,13 . Surprisingly, only the generalist N. glauca was less frequent in pools containing fish. However, unlike the specialist N. obliqua, which can effectively exploit habitats with both low and high prey abundance, N. glauca needs high density of prey to achieve good feeding efficiency 84 . Similar principles may have driven the habitat selection of diving beetles. Both study species are fast swimmers and seem to prefer open waters 85 . Therefore, fishless pools with submerged and floating macrophytes, which they preferred, may offer plenty of food and shelter, plus more space for movement than pools containing also littoral macrophytes. As with semiaquatic taxa, negative density-dependent effects (see above) or predator dilution effect 86 may occur, whereby pools already containing other conspecifics may attract further colonists as adding prey reduces the overall predation risk. In contrast, large-bodied diving beetles such as D. marginalis employ secretions from their prothoracic and pygidial glands, which have narcotic and toxic effects on fish 37 , explaining the lack of selectivity for fish vs. fishless habitats.
In some taxa, both predator and dietary cues are needed to elicit full anti-predator responses 87,88 . In our study, the fish were caged; therefore, their chemical cues were present, but there was no signal of devoured conspecifics or heterospecifics. Thus, both backswimmer and beetle specialists would possibly avoid fish-occupied pools if fish posed a risk to them. This corroborates the finding that risk perception of certain aquatic taxa does not result from signals from predators alone, but from their consumption of prey 89 . As suggested by Åbjörnsson et al. 13 , this may not be true for the generalist N. glauca, which significantly avoided pools with fish cues.
In contrast, behavioral avoidance of some beetle taxa may be triggered by the mere presence of fish 1,5 (but see 6 ). Given that behavioral adjustments to the actual predator regime may be more important than complete avoidance of fish habitats 13 , it is possible that the consumptive effect of fish would elicit different behaviors in aquatic insects. Clear preference for pools with submerged and floating macrophytes without fish over those with fish points out to aquatic taxa relying mostly on chemical detection of the predator 13,19 . Habitat selection in aquatic taxa may be a complex process, as community assembly causes taxon-dependent feedback that alters fish avoidance behavior 6 . Tracking habitat selection behavior of marked individuals over time could help elucidate the underlying mechanism in these groups.
Our study expands current knowledge of habitat selection in response to a top predator by examining habitat selection behavior of taxa in relation to specialization and vegetation structure. Only specialists of semiaquatic taxa selected fishless habitats with vegetation structure matching their habitat preferences; whereas generalists relied on a bet-hedging strategy and/or responded to the actual presence of competitors. Therefore, oviposition habitat selectivity by semiaquatic specialists does not stem from their specialization to an inherent lack of fish, but from accurate predator recognition. In aquatic taxa, individuals probably respond to the actual risk of predation, regardless of specialization. Their preference for fishless pools with submerged and floating macrophytes probably stems from sufficient resources associated with this type of habitat. In specialists of semiaquatic groups, whose terrestrial adults use water only for breeding, the mere presence of a predator is sufficient to trigger avoidance; whereas in aquatic taxa, whose imagoes spend most of their lifetime in the water, the signal of consumed conspecifics/heterospecifics might be needed to elicit avoidance. Taxa with terrestrial adults and aquatic larvae obtain no feedback on the impact of adult decisions on the progeny; their mechanisms of habitat selectivity should therefore be faultless. The present results reinforce the importance of habitat selection for the colonization of aquatic ecosystems, and illustrate how taxa with different levels of specialization may respond differently to a top predator, depending on their life history. Future experiments using other generalists and a range of taxa from both sides of the specialist spectrum (e.g., species specialized in fishless as well as fish-heavy habitats) may help elucidate how widespread are the patterns found in this study, as well as which mechanisms animals use to avoid predators.

Data availability
The data that support the findings of this study are permanently archived in the figshare data repository under the link https:// doi. org/ 10. 6084/ m9. figsh are. 16627 561. v2.